├── LICENSE ├── README.md ├── REFIT_processed └── RAW_House1_1T_processed.zip ├── REFIT_source ├── RAW_READ_ME_081116.txt └── REFIT_RAW_081116 │ └── REFIT_RAW_081116.zip ├── TCN_w12.sh ├── TCN_w192.sh ├── TCN_w24.sh ├── TCN_w384.sh ├── TCN_w48.sh ├── TCN_w6.sh ├── TCN_w96.sh ├── data_process.ipynb ├── image ├── Cure.jpg ├── Dilated_Causal_Conv.png ├── ResultTable.jpg ├── SourceData.png ├── SourceDataTable.jpg ├── TCN.png ├── cure2.jpg └── line.jpg ├── model.py ├── requirements.txt ├── result.ipynb ├── tools.py ├── train_univariate.py └── vmd.py /LICENSE: -------------------------------------------------------------------------------- 1 | Apache License 2 | Version 2.0, January 2004 3 | http://www.apache.org/licenses/ 4 | 5 | TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION 6 | 7 | 1. Definitions. 8 | 9 | "License" shall mean the terms and conditions for use, reproduction, 10 | and distribution as defined by Sections 1 through 9 of this document. 11 | 12 | "Licensor" shall mean the copyright owner or entity authorized by 13 | the copyright owner that is granting the License. 14 | 15 | "Legal Entity" shall mean the union of the acting entity and all 16 | other entities that control, are controlled by, or are under common 17 | control with that entity. For the purposes of this definition, 18 | "control" means (i) the power, direct or indirect, to cause the 19 | direction or management of such entity, whether by contract or 20 | otherwise, or (ii) ownership of fifty percent (50%) or more of the 21 | outstanding shares, or (iii) beneficial ownership of such entity. 22 | 23 | "You" (or "Your") shall mean an individual or Legal Entity 24 | exercising permissions granted by this License. 25 | 26 | "Source" form shall mean the preferred form for making modifications, 27 | including but not limited to software source code, documentation 28 | source, and configuration files. 29 | 30 | "Object" form shall mean any form resulting from mechanical 31 | transformation or translation of a Source form, including but 32 | not limited to compiled object code, generated documentation, 33 | and conversions to other media types. 34 | 35 | "Work" shall mean the work of authorship, whether in Source or 36 | Object form, made available under the License, as indicated by a 37 | copyright notice that is included in or attached to the work 38 | (an example is provided in the Appendix below). 39 | 40 | "Derivative Works" shall mean any work, whether in Source or Object 41 | form, that is based on (or derived from) the Work and for which the 42 | editorial revisions, annotations, elaborations, or other modifications 43 | represent, as a whole, an original work of authorship. For the purposes 44 | of this License, Derivative Works shall not include works that remain 45 | separable from, or merely link (or bind by name) to the interfaces of, 46 | the Work and Derivative Works thereof. 47 | 48 | "Contribution" shall mean any work of authorship, including 49 | the original version of the Work and any modifications or additions 50 | to that Work or Derivative Works thereof, that is intentionally 51 | submitted to Licensor for inclusion in the Work by the copyright owner 52 | or by an individual or Legal Entity authorized to submit on behalf of 53 | the copyright owner. For the purposes of this definition, "submitted" 54 | means any form of electronic, verbal, or written communication sent 55 | to the Licensor or its representatives, including but not limited to 56 | communication on electronic mailing lists, source code control systems, 57 | and issue tracking systems that are managed by, or on behalf of, the 58 | Licensor for the purpose of discussing and improving the Work, but 59 | excluding communication that is conspicuously marked or otherwise 60 | designated in writing by the copyright owner as "Not a Contribution." 61 | 62 | "Contributor" shall mean Licensor and any individual or Legal Entity 63 | on behalf of whom a Contribution has been received by Licensor and 64 | subsequently incorporated within the Work. 65 | 66 | 2. Grant of Copyright License. Subject to the terms and conditions of 67 | this License, each Contributor hereby grants to You a perpetual, 68 | worldwide, non-exclusive, no-charge, royalty-free, irrevocable 69 | copyright license to reproduce, prepare Derivative Works of, 70 | publicly display, publicly perform, sublicense, and distribute the 71 | Work and such Derivative Works in Source or Object form. 72 | 73 | 3. Grant of Patent License. Subject to the terms and conditions of 74 | this License, each Contributor hereby grants to You a perpetual, 75 | worldwide, non-exclusive, no-charge, royalty-free, irrevocable 76 | (except as stated in this section) patent license to make, have made, 77 | use, offer to sell, sell, import, and otherwise transfer the Work, 78 | where such license applies only to those patent claims licensable 79 | by such Contributor that are necessarily infringed by their 80 | Contribution(s) alone or by combination of their Contribution(s) 81 | with the Work to which such Contribution(s) was submitted. If You 82 | institute patent litigation against any entity (including a 83 | cross-claim or counterclaim in a lawsuit) alleging that the Work 84 | or a Contribution incorporated within the Work constitutes direct 85 | or contributory patent infringement, then any patent licenses 86 | granted to You under this License for that Work shall terminate 87 | as of the date such litigation is filed. 88 | 89 | 4. Redistribution. You may reproduce and distribute copies of the 90 | Work or Derivative Works thereof in any medium, with or without 91 | modifications, and in Source or Object form, provided that You 92 | meet the following conditions: 93 | 94 | (a) You must give any other recipients of the Work or 95 | Derivative Works a copy of this License; and 96 | 97 | (b) You must cause any modified files to carry prominent notices 98 | stating that You changed the files; and 99 | 100 | (c) You must retain, in the Source form of any Derivative Works 101 | that You distribute, all copyright, patent, trademark, and 102 | attribution notices from the Source form of the Work, 103 | excluding those notices that do not pertain to any part of 104 | the Derivative Works; and 105 | 106 | (d) If the Work includes a "NOTICE" text file as part of its 107 | distribution, then any Derivative Works that You distribute must 108 | include a readable copy of the attribution notices contained 109 | within such NOTICE file, excluding those notices that do not 110 | pertain to any part of the Derivative Works, in at least one 111 | of the following places: within a NOTICE text file distributed 112 | as part of the Derivative Works; within the Source form or 113 | documentation, if provided along with the Derivative Works; or, 114 | within a display generated by the Derivative Works, if and 115 | wherever such third-party notices normally appear. The contents 116 | of the NOTICE file are for informational purposes only and 117 | do not modify the License. You may add Your own attribution 118 | notices within Derivative Works that You distribute, alongside 119 | or as an addendum to the NOTICE text from the Work, provided 120 | that such additional attribution notices cannot be construed 121 | as modifying the License. 122 | 123 | You may add Your own copyright statement to Your modifications and 124 | may provide additional or different license terms and conditions 125 | for use, reproduction, or distribution of Your modifications, or 126 | for any such Derivative Works as a whole, provided Your use, 127 | reproduction, and distribution of the Work otherwise complies with 128 | the conditions stated in this License. 129 | 130 | 5. Submission of Contributions. Unless You explicitly state otherwise, 131 | any Contribution intentionally submitted for inclusion in the Work 132 | by You to the Licensor shall be under the terms and conditions of 133 | this License, without any additional terms or conditions. 134 | Notwithstanding the above, nothing herein shall supersede or modify 135 | the terms of any separate license agreement you may have executed 136 | with Licensor regarding such Contributions. 137 | 138 | 6. Trademarks. This License does not grant permission to use the trade 139 | names, trademarks, service marks, or product names of the Licensor, 140 | except as required for reasonable and customary use in describing the 141 | origin of the Work and reproducing the content of the NOTICE file. 142 | 143 | 7. Disclaimer of Warranty. Unless required by applicable law or 144 | agreed to in writing, Licensor provides the Work (and each 145 | Contributor provides its Contributions) on an "AS IS" BASIS, 146 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 147 | implied, including, without limitation, any warranties or conditions 148 | of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A 149 | PARTICULAR PURPOSE. You are solely responsible for determining the 150 | appropriateness of using or redistributing the Work and assume any 151 | risks associated with Your exercise of permissions under this License. 152 | 153 | 8. Limitation of Liability. In no event and under no legal theory, 154 | whether in tort (including negligence), contract, or otherwise, 155 | unless required by applicable law (such as deliberate and grossly 156 | negligent acts) or agreed to in writing, shall any Contributor be 157 | liable to You for damages, including any direct, indirect, special, 158 | incidental, or consequential damages of any character arising as a 159 | result of this License or out of the use or inability to use the 160 | Work (including but not limited to damages for loss of goodwill, 161 | work stoppage, computer failure or malfunction, or any and all 162 | other commercial damages or losses), even if such Contributor 163 | has been advised of the possibility of such damages. 164 | 165 | 9. Accepting Warranty or Additional Liability. While redistributing 166 | the Work or Derivative Works thereof, You may choose to offer, 167 | and charge a fee for, acceptance of support, warranty, indemnity, 168 | or other liability obligations and/or rights consistent with this 169 | License. However, in accepting such obligations, You may act only 170 | on Your own behalf and on Your sole responsibility, not on behalf 171 | of any other Contributor, and only if You agree to indemnify, 172 | defend, and hold each Contributor harmless for any liability 173 | incurred by, or claims asserted against, such Contributor by reason 174 | of your accepting any such warranty or additional liability. 175 | 176 | END OF TERMS AND CONDITIONS 177 | 178 | APPENDIX: How to apply the Apache License to your work. 179 | 180 | To apply the Apache License to your work, attach the following 181 | boilerplate notice, with the fields enclosed by brackets "[]" 182 | replaced with your own identifying information. (Don't include 183 | the brackets!) The text should be enclosed in the appropriate 184 | comment syntax for the file format. We also recommend that a 185 | file or class name and description of purpose be included on the 186 | same "printed page" as the copyright notice for easier 187 | identification within third-party archives. 188 | 189 | Copyright 2023 梁显武 190 | 191 | Licensed under the Apache License, Version 2.0 (the "License"); 192 | you may not use this file except in compliance with the License. 193 | You may obtain a copy of the License at 194 | 195 | http://www.apache.org/licenses/LICENSE-2.0 196 | 197 | Unless required by applicable law or agreed to in writing, software 198 | distributed under the License is distributed on an "AS IS" BASIS, 199 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 200 | See the License for the specific language governing permissions and 201 | limitations under the License. 202 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # 基于深度学习的电力负荷预测算法研究 2 | > 作者：物联 192 梁显武 3 | > 指导老师: 曹忠 4 | 5 | 本代码为本人本科毕业设计的一部分。 6 | 7 | 8 | 9 | ## 基于TCN的电力负荷预测框架 10 | 11 | TCN 不像图像卷积那样通过池化层扩大感受野，而是通过增大扩张因子以及增加层数对感受野进行扩大，这使得它能够接受更长的历史时序信息，从而降低预测的误差、提高准确率。通过残差连接能够使得网络的层数加深而不丢失准确性,这种跨层连接的结构使得信息可以在神经网络的不同层之间直接传递，而不会受到层数的限制，从而提高了神经网络的训练效率和准确性。TCN 的结构主要包括扩张因果卷积以及残差连接组成。 12 | 13 | ![扩张因果卷积](./image/Dilated_Causal_Conv.png) 14 |
图1 扩张因果卷积
15 | 16 | ![基于TCN的电力负荷预测](./image/TCN.png) 17 |
图2 基于TCN的电力负荷预测
18 | 19 | 20 | 21 | ## 代码依赖 22 | 23 | * python3.8 24 | * keras==2.6.0 25 | * matplotlib==3.5.2 26 | * numpy==1.19.4 27 | * pandas==1.4.3 28 | * tensorflow==2.6.0 29 | 30 | 31 | 可以使用以下命令安装依赖项: 32 | `pip install -r requirements.txt` 33 | 34 | 35 | 36 | ## 实验平台 37 | 38 | CPU:Xeon(R) CPU E5-2620 v4 @ 2.10GHz 39 | GPU：NVIDIA TITAN V 40 | 41 | 42 | 43 | ## 数据 44 | 45 | 本文使用了一个名为 REFIT 的公开数据集。REFIT数据集是由斯特拉斯克莱德大学、拉夫堡大学和东安格利亚大学合作创建的，数据集包含了 2013 年至 2014 年期间在拉夫堡地区的 20 个家庭的负荷数据由于数据量过大，本文的工作仅仅基于House1的数据进行的。 46 | 47 | 本文所使用的原始数据可以在`REFIT_source`文件夹中找到，经过预处理（数据清洗）之后的数据可以在`REFIT_processed`文件夹中找到。然后在脚本`data_process.ipynb`中对数据进行了清洗工作。 48 | > 注意代码库里面的数据是用压缩形式保存的.zip，使用前需要将压缩包解压到当前文件夹中。 49 | 50 | 原始数据： 51 | ![原始数据](./image/SourceData.png) 52 | ![原始数据表](./image/SourceDataTable.jpg) 53 | 54 | ## 复现 55 | 56 | 所有的训练设置都已经在`.sh`脚本文件中配置好了，例如`TCN_w6.sh`表示使用处理好的数据对TCN、LSTM、DNN模型进行训练和预测，最终生成的结果将存放在`tuning_w6`文件夹中。如果要复现结果，直接在命令行中输入以下命令。 57 | 1. `TCN_w6.sh` 58 | 2. `TCN_w12.sh` 59 | 3. `TCN_w24.sh` 60 | 4. `TCN_w48.sh` 61 | 5. `TCN_w96.sh` 62 | 6. `TCN_w192.sh` 63 | 7. `TCN_w384.sh` 64 | 65 | 66 | 67 | ## 代码使用 68 | 69 | 代码文件的功能解释如下： 70 | 1. `data_process.ipynb` : 数据处理代码 71 | 2. `model.py` : 模型构建代码 72 | 3. `tools.py` : 模型评估方法和数据归一化方法 73 | 4. `train_univariate.py`: 训练代码 74 | 5. `result.ipynb`:查看结果代码 75 | 76 | 模型训练的代码使用如下： 77 | `python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Aggregate --window_size 6 --model TCN --layer 2 --patience 10 --save tuning_test/TCN_Aggregate --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 78 | ` 79 | 参数解释： 80 | 81 | 1. --data 训练数据的文件路径 82 | 2. --target 预测目标曲线 83 | 3. --window_size 数据窗口大小 84 | 4. --model 模型选择 85 | 5. --layer TCN模型的层数 86 | 6. --patience 早停止策略的容忍度 87 | 7. --save 训练结果保存的文件夹 88 | 8. --epoch 训练迭代轮数 89 | 9. --batchsize 批次大小 90 | 10. --lr 学习率大小 91 | 11. --gpu GPU编号 92 | 93 | 94 | 95 | ## 结果 96 | 97 | ### 训练结果表 98 | 每个训练结果都会保存到--save指定的文件夹中，例如`tuning_w6/`。使用`.sh`脚本所产生的结果如下： 99 | ![训练结果表](./image/ResultTable.jpg) 100 | 101 | 102 | ### 窗口的影响 103 | TCN 的对更长的历史窗口的信息捕获能力表现得更为明显，随着窗口增加，精度呈现出下降趋势，在窗口大小为 386 时预测误差达到了最小。 104 | ![窗口-精度](./image/line.jpg) 105 | 106 | 107 | 108 | 109 | ### 历史窗口大小为 48时，在曲线上模型预测的具体表现。 110 | 111 | 窗口为48时，Aggregate曲线 112 | ![Aggregate曲线](./image/Cure.jpg) 113 | 114 | 115 | 116 | 窗口为48时Appliance1曲线 117 | ![Appliance1曲线](./image/cure2.jpg) 118 | 119 | 120 | ## 参考论文 121 | [Bai S, Kolter J Z, Koltun V. An empirical evaluation of generic convolutional and recurrent networks for sequence modeling[J]. arXiv preprint arXiv:1803.01271, 2018.](https://arxiv.org/abs/1803.01271) -------------------------------------------------------------------------------- /REFIT_processed/RAW_House1_1T_processed.zip: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/joker507/Power_Load_Forecasting_by_TCN/4b71f0bc31e65bc1ba92c073872ab64ed6b024a0/REFIT_processed/RAW_House1_1T_processed.zip -------------------------------------------------------------------------------- /REFIT_source/RAW_READ_ME_081116.txt: -------------------------------------------------------------------------------- 1 | REFIT: Electrical Load Measurements 2 | 3 | THE FOLLOWING DATASET IS COMPLETELY RAW APART FROM ACCOUNTING FOR DAYLIGHT SAVINGS TIME (UK) 4 | 5 | INFORMATION 6 | Collection of this dataset was supported by the Engineering and Physical Sciences Research Council (EPSRC) via the project entitled Personalised Retrofit Decision Support Tools for UK Homes using Smart Home Technology (REFIT), under Grant Reference EP/K002368/1 to the University of Strathclyde. REFIT is a collaboration among the Universities of Strathclyde, Loughborough and East Anglia. The dataset includes data from 20 households from the Loughborough area over the period 2013 - 2014. Key findings of the study are available at http://gtr.rcuk.ac.uk/projects?ref=EP%2FK002368%2F1. 7 | 8 | LICENCING 9 | This work is licensed under the Creative Commons Attribution 4.0 International Public License. See https://creativecommons.org/licenses/by/4.0/legalcode for further details. 10 | Please cite the following paper if you use the dataset: 11 | 12 | @inbook{278e1df91d22494f9be2adfca2559f92, 13 | title = "A data management platform for personalised real-time energy feedback", 14 | keywords = "smart homes, real-time energy, smart energy meter, energy consumption, Electrical engineering. Electronics Nuclear engineering, Electrical and Electronic Engineering", 15 | author = "David Murray and Jing Liao and Lina Stankovic and Vladimir Stankovic and Richard Hauxwell-Baldwin and Charlie Wilson and Michael Coleman and Tom Kane and Steven Firth", 16 | year = "2015", 17 | booktitle = "Proceedings of the 8th International Conference on Energy Efficiency in Domestic Appliances and Lighting", 18 | } 19 | 20 | Each of the houses is labelled, House 1 - House 21 (skipping House 14), each house has 10 power sensors comprising a current clamp for the household aggregate and 9 Individual Appliance Monitors (IAMs). Only active power in Watts is collected at 8-second interval. 21 | The subset of all appliances in a household that was monitored reflects the document from DECC of the largest consumers in UK households, https://www.gov.uk/government/uploads/system/uploads/attachment_data/file/274778/9_Domestic_appliances__cooking_and_cooling_equipment.pdf 22 | 23 | FILE FORMAT 24 | The file format is csv and is laid out as follows; 25 | DATETIME, UNIX TIMESTAMP (UCT), Aggregate, Appliance1, Appliance2, Appliance3, ... , Appliance9 26 | Additionally data was only recorded when there was a change in load; this data has been filled with intermediate values where not available. The sensors are also not synchronised as our collection script polled every 6-8 seconds; the sensor may have updated anywhere in the last 6-8 seconds. 27 | The file name _Part1 refers to the first iteration of the database where sensors that were not available were set to 0. 28 | The file name _Part2 refers to the second iteration of the database where sensors that were not available were set to NaN to distinguish from 0. 29 | 30 | MISSING DATA 31 | During the course of the study there are a few periods of missing data (notably February 2014). Outages were due to a number of factors, including household internet failure, hardware failures, network routing issues, etc. 32 | 33 | Household Information 34 | House, Occupancy, Construction Year, Appliances Owned, Type, Size 35 | 1 , 2 , 1975-1980 , 35 , Detached , 4 bed 36 | 2 , 4 , - , 15 , Semi-detached , 3 bed 37 | 3 , 2 , 1988 , 27 , Detached , 3 bed 38 | 4 , 2 , 1850-1899 , 33 , Detached , 4 bed 39 | 5 , 4 , 1878 , 44 , Mid-terrace , 4 bed 40 | 6 , 2 , 2005 , 49 , Detached , 4 bed 41 | 7 , 4 , 1965-1974 , 25 , Detached , 3 bed 42 | 8 , 2 , 1966 , 35 , Detached , 2 bed 43 | 9 , 2 , 1919-1944 , 24 , Detached , 3 bed 44 | 10 , 4 , 1919-1944 , 31 , Detached , 3 bed 45 | 11 , 1 , 1945-1964 , 25 , Detached , 3 bed 46 | 12 , 3 , 1991-1995 , 26 , Detached , 3 bed 47 | 13 , 4 , post 2002 , 28 , Detached , 4 bed 48 | 15 , 1 , 1965-1974 , 19 , Semi-detached , 3 bed 49 | 16 , 6 , 1981-1990 , 48 , Detached , 5 bed 50 | 17 , 3 , mid 60s , 22 , Detached , 3 bed 51 | 18 , 2 , 1965-1974 , 34 , Detached , 3 bed 52 | 19 , 4 , 1945-1964 , 26 , Semi-detached , 3 bed 53 | 20 , 2 , 1965-1974 , 39 , Detached , 3 bed 54 | 21 , 4 , 1981-1990 , 23 , Detached , 3 bed 55 | 56 | APPLIANCE LIST 57 | The following list shows the appliances that were known to be monitored at the beginning of the study period. Although occupants were asked not to remove or switch appliances monitored by the IAMs, we cannot guarantee this to be the case. It should also be noted that Television and Computer Site may consist of multiple appliances, e.g. Television, SkyBox, DvD Player, Computer, Speakers, etc. Makes and Models specified here are gathered from pictures gathered by the installation team. 58 | 59 | House 1 60 | 0.Aggregate 61 | 1.Fridge, Hotpoint, RLA50P 62 | 2.Freezer(1),Beko, CF393APW 63 | 3.Freezer(2), Unknown, Unknown 64 | 4.Washer Dryer, Creda, T522VW 65 | 5.Washing Machine, Beko, WMC6140 66 | 6.Dishwasher, Bosch, Unknown 67 | 7.Computer, Lenovo, H520s 68 | 8.Television Site, Toshiba, 32BL502b 69 | 9.Electric Heater, GLEN, 2172 70 | 71 | House 2 72 | 0.Aggregate, 73 | 1.Fridge-Freezer, Unknown, Unknown 74 | 2.Washing Machine, LG, F1289TD 75 | 3.Dishwasher, Unknown, Unknown 76 | 4.Television Site, 77 | 5.Microwave, Unknown, Unknown 78 | 6.Toaster, Unknown, Unknown 79 | 7.Hi-Fi, Unknown, Unknown 80 | 8.Kettle, Unknown, Unknown 81 | 9.Overhead Fan 82 | 83 | House 3 84 | 0.Aggregate, 85 | 1.Toaster, Dualit, DPP2 86 | 2.Fridge-Freezer, Whirlpool, ARC7612 87 | 3.Freezer, Frigidaire, Freezer Elite 88 | 4.Tumble Dryer, Unknown, Unknown 89 | 5.Dishwasher, Bosch, Exxcel Auto Option 90 | 6.Washing Machine, Unknown, Unknown 91 | 7.Television Site, Samsung, LE46A656A1FXXU 92 | 8.Microwave, Panasoinc, NN-CT565MBPQ 93 | 9.Kettle, Dualit, JKt3 94 | 95 | House 4 96 | 0.Aggregate, 97 | 1.Fridge, Neff, K1514X0GB/31 98 | 2.Freezer, Ocean, UF 1025 99 | 3.Fridge-Freezer, Ariston, DF230 100 | 4.Washing Machine(1), Servis, 6065 101 | 5.Washing Machine(2), Zanussi, Z917 102 | 6.Desktop Computer, Unknown, Unknown 103 | 7.Television Site, Sony, KDL-32W706B 104 | 8.Microwave, Matsui, 170TC 105 | 9.Kettle, Swan, Unknown 106 | 107 | House 5 108 | 0.Aggregate, 109 | 1.Fridge-Freezer, Fisher & Paykel, Unknown 110 | 2.Tumble Dryer, Unknown, Unknown 111 | 3.Washing Machine, AEG, L99695HWD 112 | 4.Dishwasher, Unknown, Unknown 113 | 5.Desktop Computer, Unknown, Unknown 114 | 6.Television Site, Unknown, Unknown 115 | 7.Microwave, Unknown, Unknown 116 | 8.Kettle, Logik, L17SKC14 117 | 9.Toaster, Breville, TT33 118 | 119 | House 6 120 | 0.Aggregate, 121 | 1.Freezer, Whirlpool, CV128W 122 | 2.Washing Machine, Bosch, Classixx 1200 Express 123 | 3.Dishwasher, Neff, Unknown 124 | 4.MJY Computer, Unknown, Unknown 125 | 5.TV/Satellite, Samsung, UE55F6500SB 126 | 6.Microwave, Neff, H5642N0GB/02 127 | 7.Kettle, ASDA, GPK101W 128 | 8.Toaster, Breville, PT15 129 | 9.PGM Computer, Unknown, Unknown 130 | 131 | House 7 132 | 0.Aggregate, 133 | 1.Fridge, Bosch, KSR30422GB 134 | 2.Freezer(1), Whirlpool, AFG 392/H 135 | 3.Freezer(2), Unknown, Unknown 136 | 4.Tumble Dryer, White Knight, Unknown 137 | 5.Washing Machine, Bosch, Unknown 138 | 6.Dishwasher, Unknown, Unknown 139 | 7.Television Site, 140 | 8.Toaster, Unknown, Unknown 141 | 9.Kettle, Sainsburys, 121988254 142 | 143 | House 8 144 | 0.Aggregate, 145 | 1.Fridge, Liebherr, KP2620 146 | 2.Freezer, Unknown, Unknown 147 | 3.Washer Dryer, Zanussi, Unknown 148 | 4.Washing Machine, 149 | 5.Toaster, Bosch, TAT6101GB/02 150 | 6.Computer, Unknown, Unknown 151 | 7.Television Site, Sony, KDL-32V2000 152 | 8.Microwave, Panasoinc, NN-CT565MBPQ 153 | 9.Kettle, Morphy Richards, 43615 154 | 155 | House 9 156 | 0.Aggregate, 157 | 1.Fridge-Freezer, Bosch, KGH34X05GB/05 158 | 2.Washer Dryer, Hotpoint, TCM580 159 | 3.Washing Machine, Bosch, Classixx 6 1200 Express 160 | 4.Dishwasher, Bosch, Classixx 161 | 5.Television Site, LG, 32LH3000 162 | 6.Microwave, Argos, MM717CFA 163 | 7.Kettle, Russel Hobbs, Unknown 164 | 8.Hi-Fi, Unknown, Unknown 165 | 9.Electric Heater 166 | 167 | House 10 168 | 0.Aggregate, 169 | 1.Magimix(Blender), Unknown, Unknown 170 | 2.Toaster, Unknown, Unknown 171 | 3.Chest Freezer, Unknown, Unknown 172 | 4.Fridge-Freezer, Unknown, Unknown 173 | 5.Washing Machine, Beko, WI1382 174 | 6.Dishwasher, AEG, Unknown 175 | 7.Television Site, Samsung, UE40ES5500K 176 | 8.Microwave, Unknown, Unknown 177 | 9.K Mix, Unknown, Unknown 178 | 179 | House 11 180 | 0.Aggregate, 181 | 1.Fridge, Gorenje, HPI 1566 182 | 2.Fridge-Freezer, Unknown, Unknown 183 | 3.Washing Machine, Unknown, Unknown 184 | 4.Dishwasher, Unknown, Unknown 185 | 5.Computer Site, Unknown, Unknown 186 | 6.Microwave, Unknown, Unknown 187 | 7.Kettle, Unknown, Unknown 188 | 8.Router, Unknown, Unknown 189 | 9.Hi-Fi, Unknown, Unknown 190 | 191 | House 12 192 | 0.Aggregate, 193 | 1.Fridge-Freezer, Gorenje, HZS 3266 194 | 2.???, Unknown, Unknown 195 | 3.???, Unknown, Unknown 196 | 4.Computer Site, Unknown, Unknown 197 | 5.Microwave, Unknown, Unknown 198 | 6.Kettle, Unknown, Unknown 199 | 7.Toaster, Unknown, Unknown 200 | 8.Television, Unknown, Unknown 201 | 9.???, Unknown, Unknown 202 | 203 | House 13 204 | 0.Aggregate, 205 | 1.Television Site, Samsung, UE55H6400AK 206 | 2.Freezer, Unknown, Unknown 207 | 3.Washing Machine, Unknown, Unknown 208 | 4.Dishwasher, Unknown, Unknown 209 | 5.???, Unknown, Unknown 210 | 6.Network Site, Unknown, Unknown 211 | 7.Microwave, Unknown, Unknown 212 | 8.Microwave, Unknown, Unknown 213 | 9.Kettle, Unknown, Unknown 214 | 215 | House 15 216 | 0.Aggregate, 217 | 1.Fridge-Freezer, Unknown, Unknown 218 | 2.Tumble Dryer, Unknown, Unknown 219 | 3.Washing Machine, Beko, WMB91242LB 220 | 4.Dishwasher, Unknown, Unknown 221 | 5.Computer Site, Unknown, Unknown 222 | 6.Television Site, LG, 22LS4D 223 | 7.Microwave, Unknown, Unknown 224 | 8.Hi-Fi, Unknown, Unknown 225 | 9.Toaster, Unknown, Unknown 226 | 227 | House 16 228 | 0.Aggregate, 229 | 1.Fridge-Freezer(1), Bosch, KGN30VW20G/01 230 | 2.Fridge-Freezer(2), Unknown, Unknown 231 | 3.Electric Heater(1), Unknown, Unknown 232 | 4.Electric Heater(2), Unknown, Unknown 233 | 5.Washing Machine, Bosch, WAB24262GB/01 234 | 6.Dishwasher, Unknown, Unknown 235 | 7.Computer Site, Unknown, Unknown 236 | 8.Television Site, Samsung, UE55HU8500T 237 | 9.Dehumidifier, Unknown, Unknown 238 | 239 | House 17 240 | 0.Aggregate, 241 | 1.Freezer, Unknown, Unknown 242 | 2.Fridge-Freezer, Whirlpool, ARC 2990 243 | 3.Tumble Dryer, Unknown, Unknown 244 | 4.Washing Machine, Bosch, Exxcel 8 Vario Perfect 245 | 5.Computer Site, Unknown, Unknown 246 | 6.Television Site, Unknown, Unknown 247 | 7.Microwave, Matsui, M195T 248 | 8.Kettle, Russel Hobbs, 17869 249 | 9.TV Site(Bedroom), Unknown, Unknown 250 | 251 | House 18 252 | 0.Aggregate, 253 | 1.Fridge(garage), LEC, R.403W 254 | 2.Freezer(garage), Unknown, Unknown 255 | 3.Fridge-Freezer, Unknown, Unknown 256 | 4.Washer Dryer(garage), Unknown, Unknown 257 | 5.Washing Machine, Unknown, Unknown 258 | 6.Dishwasher, Unknown, Unknown 259 | 7.Desktop Computer, Unknown, Unknown 260 | 8.Television Site, Unknown, Unknown 261 | 9.Microwave, Unknown, Unknown 262 | 263 | House 19 264 | 0.Aggregate, 265 | 1.Fridge Freezer, Bosch, KGS-3272-GB/01 266 | 2.Washing Machine, Bosch, WAE24060GB/03 267 | 3.Television Site, Sony, KDL32EX703 268 | 4.Microwave, Kenwood, K20MSS10 269 | 5.Kettle, Breville, VKJ336 270 | 6.Toaster, Bellini, BET240 271 | 7.Bread-maker, Unknown, Unknown 272 | 8.Games Console, Unknown, Unknown 273 | 9.Hi-Fi, Unknown, Unknown 274 | 275 | House 20 276 | 0.Aggregate, 277 | 1.Fridge, Unknown, Unknown 278 | 2.Freezer, Unknown, Unknown 279 | 3.Tumble Dryer, Unknown, Unknown 280 | 4.Washing Machine, Unknown, Unknown 281 | 5.Dishwasher, Unknown, Unknown 282 | 6.Computer Site, Unknown, Unknown 283 | 7.Television Site, Unknown, Unknown 284 | 8.Microwave, Unknown, Unknown 285 | 9.Kettle, Unknown, Unknown 286 | 287 | House 21 288 | 0.Aggregate, 289 | 1.Fridge-Freezer, Samsung, SR-L3216B 290 | 2.Tumble Dryer, Unknown, Unknown 291 | 3.Washing Machine, Beko, WMB81241LW 292 | 4.Dishwasher, AEG, FAVORIT 293 | 5.Food Mixer, Unknown, Unknown 294 | 6.Television, Unknown, Unknown 295 | 7.Kettle, Unknown, Unknown 296 | 8.Vivarium, Unknown, Unknown 297 | 9.Pond Pump, Unknown, Unknown -------------------------------------------------------------------------------- /REFIT_source/REFIT_RAW_081116/REFIT_RAW_081116.zip: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/joker507/Power_Load_Forecasting_by_TCN/4b71f0bc31e65bc1ba92c073872ab64ed6b024a0/REFIT_source/REFIT_RAW_081116/REFIT_RAW_081116.zip -------------------------------------------------------------------------------- /TCN_w12.sh: -------------------------------------------------------------------------------- 1 | # TCN 2 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Aggregate --window_size 12 --model TCN --layer 3 --patience 10 --save tuning_w12/TCN_Aggregate --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 3 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance1 --window_size 12 --model TCN --layer 3 --patience 10 --save tuning_w12/TCN_Appliance1 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 4 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance2 --window_size 12 --model TCN --layer 3 --patience 10 --save tuning_w12/TCN_Appliance2 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 5 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance3 --window_size 12 --model TCN --layer 3 --patience 10 --save tuning_w12/TCN_Appliance3 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 6 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance4 --window_size 12 --model TCN --layer 3 --patience 10 --save tuning_w12/TCN_Appliance4 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 7 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance5 --window_size 12 --model TCN --layer 3 --patience 10 --save tuning_w12/TCN_Appliance5 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 8 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance6 --window_size 12 --model TCN --layer 3 --patience 10 --save tuning_w12/TCN_Appliance6 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 9 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance7 --window_size 12 --model TCN --layer 3 --patience 10 --save tuning_w12/TCN_Appliance7 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 10 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance8 --window_size 12 --model TCN --layer 3 --patience 10 --save tuning_w12/TCN_Appliance8 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 11 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance9 --window_size 12 --model TCN --layer 3 --patience 10 --save tuning_w12/TCN_Appliance9 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 12 | 13 | # DNN 14 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Aggregate --window_size 12 --model ANN --patience 10 --save tuning_w12/ANN_Aggregate --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 15 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance1 --window_size 12 --model ANN --patience 10 --save tuning_w12/ANN_Appliance1 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 16 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance2 --window_size 12 --model ANN --patience 10 --save tuning_w12/ANN_Appliance2 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 17 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance3 --window_size 12 --model ANN --patience 10 --save tuning_w12/ANN_Appliance3 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 18 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance4 --window_size 12 --model ANN --patience 10 --save tuning_w12/ANN_Appliance4 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 19 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance5 --window_size 12 --model ANN --patience 10 --save tuning_w12/ANN_Appliance5 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 20 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance6 --window_size 12 --model ANN --patience 10 --save tuning_w12/ANN_Appliance6 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 21 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance7 --window_size 12 --model ANN --patience 10 --save tuning_w12/ANN_Appliance7 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 22 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance8 --window_size 12 --model ANN --patience 10 --save tuning_w12/ANN_Appliance8 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 23 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance9 --window_size 12 --model ANN --patience 10 --save tuning_w12/ANN_Appliance9 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 24 | 25 | # LSTM 26 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Aggregate --window_size 12 --model LSTM --patience 10 --save tuning_w12/LSTM_Aggregate --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 27 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance1 --window_size 12 --model LSTM --patience 10 --save tuning_w12/LSTM_Appliance1 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 28 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance2 --window_size 12 --model LSTM --patience 10 --save tuning_w12/LSTM_Appliance2 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 29 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance3 --window_size 12 --model LSTM --patience 10 --save tuning_w12/LSTM_Appliance3 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 30 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance4 --window_size 12 --model LSTM --patience 10 --save tuning_w12/LSTM_Appliance4 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 31 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance5 --window_size 12 --model LSTM --patience 10 --save tuning_w12/LSTM_Appliance5 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 32 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance6 --window_size 12 --model LSTM --patience 10 --save tuning_w12/LSTM_Appliance6 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 33 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance7 --window_size 12 --model LSTM --patience 10 --save tuning_w12/LSTM_Appliance7 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 34 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance8 --window_size 12 --model LSTM --patience 10 --save tuning_w12/LSTM_Appliance8 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 35 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance9 --window_size 12 --model LSTM --patience 10 --save tuning_w12/LSTM_Appliance9 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 1 36 | -------------------------------------------------------------------------------- /TCN_w192.sh: -------------------------------------------------------------------------------- 1 | # TCN 2 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Aggregate --window_size 192 --model TCN --layer 7 --patience 10 --save tuning_w192/TCN_Aggregate --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 3 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance1 --window_size 192 --model TCN --layer 7 --patience 10 --save tuning_w192/TCN_Appliance1 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 4 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance2 --window_size 192 --model TCN --layer 7 --patience 10 --save tuning_w192/TCN_Appliance2 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 5 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance3 --window_size 192 --model TCN --layer 7 --patience 10 --save tuning_w192/TCN_Appliance3 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 6 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance4 --window_size 192 --model TCN --layer 7 --patience 10 --save tuning_w192/TCN_Appliance4 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 7 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance5 --window_size 192 --model TCN --layer 7 --patience 10 --save tuning_w192/TCN_Appliance5 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 8 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance6 --window_size 192 --model TCN --layer 7 --patience 10 --save tuning_w192/TCN_Appliance6 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 9 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance7 --window_size 192 --model TCN --layer 7 --patience 10 --save tuning_w192/TCN_Appliance7 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 10 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance8 --window_size 192 --model TCN --layer 7 --patience 10 --save tuning_w192/TCN_Appliance8 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 11 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance9 --window_size 192 --model TCN --layer 7 --patience 10 --save tuning_w192/TCN_Appliance9 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 12 | 13 | # DNN 14 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Aggregate --window_size 192 --model ANN --patience 10 --save tuning_w192/ANN_Aggregate --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 15 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance1 --window_size 192 --model ANN --patience 10 --save tuning_w192/ANN_Appliance1 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 16 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance2 --window_size 192 --model ANN --patience 10 --save tuning_w192/ANN_Appliance2 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 17 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance3 --window_size 192 --model ANN --patience 10 --save tuning_w192/ANN_Appliance3 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 18 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance4 --window_size 192 --model ANN --patience 10 --save tuning_w192/ANN_Appliance4 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 19 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance5 --window_size 192 --model ANN --patience 10 --save tuning_w192/ANN_Appliance5 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 20 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance6 --window_size 192 --model ANN --patience 10 --save tuning_w192/ANN_Appliance6 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 21 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance7 --window_size 192 --model ANN --patience 10 --save tuning_w192/ANN_Appliance7 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 22 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance8 --window_size 192 --model ANN --patience 10 --save tuning_w192/ANN_Appliance8 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 23 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance9 --window_size 192 --model ANN --patience 10 --save tuning_w192/ANN_Appliance9 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 24 | 25 | # LSTM 26 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Aggregate --window_size 192 --model LSTM --patience 10 --save tuning_w192/LSTM_Aggregate --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 27 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance1 --window_size 192 --model LSTM --patience 10 --save tuning_w192/LSTM_Appliance1 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 28 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance2 --window_size 192 --model LSTM --patience 10 --save tuning_w192/LSTM_Appliance2 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 29 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance3 --window_size 192 --model LSTM --patience 10 --save tuning_w192/LSTM_Appliance3 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 30 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance4 --window_size 192 --model LSTM --patience 10 --save tuning_w192/LSTM_Appliance4 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 31 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance5 --window_size 192 --model LSTM --patience 10 --save tuning_w192/LSTM_Appliance5 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 32 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance6 --window_size 192 --model LSTM --patience 10 --save tuning_w192/LSTM_Appliance6 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 33 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance7 --window_size 192 --model LSTM --patience 10 --save tuning_w192/LSTM_Appliance7 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 34 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance8 --window_size 192 --model LSTM --patience 10 --save tuning_w192/LSTM_Appliance8 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 35 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance9 --window_size 192 --model LSTM --patience 10 --save tuning_w192/LSTM_Appliance9 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 36 | -------------------------------------------------------------------------------- /TCN_w24.sh: -------------------------------------------------------------------------------- 1 | # TCN 2 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Aggregate --window_size 24 --model TCN --layer 4 --patience 10 --save tuning_w24/TCN_Aggregate --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 3 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance1 --window_size 24 --model TCN --layer 4 --patience 10 --save tuning_w24/TCN_Appliance1 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 4 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance2 --window_size 24 --model TCN --layer 4 --patience 10 --save tuning_w24/TCN_Appliance2 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 5 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance3 --window_size 24 --model TCN --layer 4 --patience 10 --save tuning_w24/TCN_Appliance3 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 6 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance4 --window_size 24 --model TCN --layer 4 --patience 10 --save tuning_w24/TCN_Appliance4 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 7 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance5 --window_size 24 --model TCN --layer 4 --patience 10 --save tuning_w24/TCN_Appliance5 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 8 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance6 --window_size 24 --model TCN --layer 4 --patience 10 --save tuning_w24/TCN_Appliance6 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 9 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance7 --window_size 24 --model TCN --layer 4 --patience 10 --save tuning_w24/TCN_Appliance7 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 10 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance8 --window_size 24 --model TCN --layer 4 --patience 10 --save tuning_w24/TCN_Appliance8 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 11 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance9 --window_size 24 --model TCN --layer 4 --patience 10 --save tuning_w24/TCN_Appliance9 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 12 | 13 | # DNN 14 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Aggregate --window_size 24 --model ANN --patience 10 --save tuning_w24/ANN_Aggregate --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 15 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance1 --window_size 24 --model ANN --patience 10 --save tuning_w24/ANN_Appliance1 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 16 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance2 --window_size 24 --model ANN --patience 10 --save tuning_w24/ANN_Appliance2 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 17 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance3 --window_size 24 --model ANN --patience 10 --save tuning_w24/ANN_Appliance3 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 18 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance4 --window_size 24 --model ANN --patience 10 --save tuning_w24/ANN_Appliance4 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 19 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance5 --window_size 24 --model ANN --patience 10 --save tuning_w24/ANN_Appliance5 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 20 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance6 --window_size 24 --model ANN --patience 10 --save tuning_w24/ANN_Appliance6 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 21 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance7 --window_size 24 --model ANN --patience 10 --save tuning_w24/ANN_Appliance7 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 22 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance8 --window_size 24 --model ANN --patience 10 --save tuning_w24/ANN_Appliance8 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 23 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance9 --window_size 24 --model ANN --patience 10 --save tuning_w24/ANN_Appliance9 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 24 | 25 | # LSTM 26 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Aggregate --window_size 24 --model LSTM --patience 10 --save tuning_w24/LSTM_Aggregate --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 27 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance1 --window_size 24 --model LSTM --patience 10 --save tuning_w24/LSTM_Appliance1 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 28 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance2 --window_size 24 --model LSTM --patience 10 --save tuning_w24/LSTM_Appliance2 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 29 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance3 --window_size 24 --model LSTM --patience 10 --save tuning_w24/LSTM_Appliance3 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 30 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance4 --window_size 24 --model LSTM --patience 10 --save tuning_w24/LSTM_Appliance4 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 31 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance5 --window_size 24 --model LSTM --patience 10 --save tuning_w24/LSTM_Appliance5 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 32 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance6 --window_size 24 --model LSTM --patience 10 --save tuning_w24/LSTM_Appliance6 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 33 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance7 --window_size 24 --model LSTM --patience 10 --save tuning_w24/LSTM_Appliance7 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 34 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance8 --window_size 24 --model LSTM --patience 10 --save tuning_w24/LSTM_Appliance8 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 35 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance9 --window_size 24 --model LSTM --patience 10 --save tuning_w24/LSTM_Appliance9 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 2 36 | -------------------------------------------------------------------------------- /TCN_w384.sh: -------------------------------------------------------------------------------- 1 | # TCN 2 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Aggregate --window_size 384 --model TCN --layer 8 --patience 10 --save tuning_w384/TCN_Aggregate --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 3 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance1 --window_size 384 --model TCN --layer 8 --patience 10 --save tuning_w384/TCN_Appliance1 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 4 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance2 --window_size 384 --model TCN --layer 8 --patience 10 --save tuning_w384/TCN_Appliance2 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 5 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance3 --window_size 384 --model TCN --layer 8 --patience 10 --save tuning_w384/TCN_Appliance3 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 6 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance4 --window_size 384 --model TCN --layer 8 --patience 10 --save tuning_w384/TCN_Appliance4 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 7 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance5 --window_size 384 --model TCN --layer 8 --patience 10 --save tuning_w384/TCN_Appliance5 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 8 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance6 --window_size 384 --model TCN --layer 8 --patience 10 --save tuning_w384/TCN_Appliance6 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 9 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance7 --window_size 384 --model TCN --layer 8 --patience 10 --save tuning_w384/TCN_Appliance7 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 10 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance8 --window_size 384 --model TCN --layer 8 --patience 10 --save tuning_w384/TCN_Appliance8 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 11 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance9 --window_size 384 --model TCN --layer 8 --patience 10 --save tuning_w384/TCN_Appliance9 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 12 | 13 | # DNN 14 | # python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Aggregate --window_size 384 --model ANN --patience 10 --save tuning_w384/ANN_Aggregate --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 15 | # python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance1 --window_size 384 --model ANN --patience 10 --save tuning_w384/ANN_Appliance1 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 16 | # python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance2 --window_size 384 --model ANN --patience 10 --save tuning_w384/ANN_Appliance2 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 17 | # python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance3 --window_size 384 --model ANN --patience 10 --save tuning_w384/ANN_Appliance3 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 18 | # python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance4 --window_size 384 --model ANN --patience 10 --save tuning_w384/ANN_Appliance4 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 19 | # python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance5 --window_size 384 --model ANN --patience 10 --save tuning_w384/ANN_Appliance5 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 20 | # python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance6 --window_size 384 --model ANN --patience 10 --save tuning_w384/ANN_Appliance6 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 21 | # python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance7 --window_size 384 --model ANN --patience 10 --save tuning_w384/ANN_Appliance7 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 22 | # python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance8 --window_size 384 --model ANN --patience 10 --save tuning_w384/ANN_Appliance8 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 23 | # python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance9 --window_size 384 --model ANN --patience 10 --save tuning_w384/ANN_Appliance9 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 24 | 25 | # LSTM 26 | # python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Aggregate --window_size 384 --model LSTM --patience 10 --save tuning_w384/LSTM_Aggregate --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 27 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance1 --window_size 384 --model LSTM --patience 10 --save tuning_w384/LSTM_Appliance1 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 28 | # python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance2 --window_size 384 --model LSTM --patience 10 --save tuning_w384/LSTM_Appliance2 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 29 | # python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance3 --window_size 384 --model LSTM --patience 10 --save tuning_w384/LSTM_Appliance3 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 30 | # python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance4 --window_size 384 --model LSTM --patience 10 --save tuning_w384/LSTM_Appliance4 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 31 | # python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance5 --window_size 384 --model LSTM --patience 10 --save tuning_w384/LSTM_Appliance5 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 32 | # python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance6 --window_size 384 --model LSTM --patience 10 --save tuning_w384/LSTM_Appliance6 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 33 | # python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance7 --window_size 384 --model LSTM --patience 10 --save tuning_w384/LSTM_Appliance7 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 34 | # python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance8 --window_size 384 --model LSTM --patience 10 --save tuning_w384/LSTM_Appliance8 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 35 | # python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance9 --window_size 384 --model LSTM --patience 10 --save tuning_w384/LSTM_Appliance9 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 36 | -------------------------------------------------------------------------------- /TCN_w48.sh: -------------------------------------------------------------------------------- 1 | # TCN 2 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Aggregate --window_size 48 --model TCN --layer 5 --patience 10 --save tuning_w48/TCN_Aggregate --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 3 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance1 --window_size 48 --model TCN --layer 5 --patience 10 --save tuning_w48/TCN_Appliance1 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 4 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance2 --window_size 48 --model TCN --layer 5 --patience 10 --save tuning_w48/TCN_Appliance2 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 5 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance3 --window_size 48 --model TCN --layer 5 --patience 10 --save tuning_w48/TCN_Appliance3 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 6 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance4 --window_size 48 --model TCN --layer 5 --patience 10 --save tuning_w48/TCN_Appliance4 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 7 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance5 --window_size 48 --model TCN --layer 5 --patience 10 --save tuning_w48/TCN_Appliance5 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 8 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance6 --window_size 48 --model TCN --layer 5 --patience 10 --save tuning_w48/TCN_Appliance6 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 9 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance7 --window_size 48 --model TCN --layer 5 --patience 10 --save tuning_w48/TCN_Appliance7 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 10 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance8 --window_size 48 --model TCN --layer 5 --patience 10 --save tuning_w48/TCN_Appliance8 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 11 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance9 --window_size 48 --model TCN --layer 5 --patience 10 --save tuning_w48/TCN_Appliance9 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 12 | 13 | # DNN 14 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Aggregate --window_size 48 --model ANN --patience 10 --save tuning_w48/ANN_Aggregate --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 15 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance1 --window_size 48 --model ANN --patience 10 --save tuning_w48/ANN_Appliance1 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 16 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance2 --window_size 48 --model ANN --patience 10 --save tuning_w48/ANN_Appliance2 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 17 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance3 --window_size 48 --model ANN --patience 10 --save tuning_w48/ANN_Appliance3 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 18 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance4 --window_size 48 --model ANN --patience 10 --save tuning_w48/ANN_Appliance4 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 19 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance5 --window_size 48 --model ANN --patience 10 --save tuning_w48/ANN_Appliance5 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 20 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance6 --window_size 48 --model ANN --patience 10 --save tuning_w48/ANN_Appliance6 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 21 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance7 --window_size 48 --model ANN --patience 10 --save tuning_w48/ANN_Appliance7 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 22 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance8 --window_size 48 --model ANN --patience 10 --save tuning_w48/ANN_Appliance8 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 23 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance9 --window_size 48 --model ANN --patience 10 --save tuning_w48/ANN_Appliance9 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 24 | 25 | # LSTM 26 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Aggregate --window_size 48 --model LSTM --patience 10 --save tuning_w48/LSTM_Aggregate --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 27 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance1 --window_size 48 --model LSTM --patience 10 --save tuning_w48/LSTM_Appliance1 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 28 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance2 --window_size 48 --model LSTM --patience 10 --save tuning_w48/LSTM_Appliance2 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 29 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance3 --window_size 48 --model LSTM --patience 10 --save tuning_w48/LSTM_Appliance3 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 30 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance4 --window_size 48 --model LSTM --patience 10 --save tuning_w48/LSTM_Appliance4 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 31 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance5 --window_size 48 --model LSTM --patience 10 --save tuning_w48/LSTM_Appliance5 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 32 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance6 --window_size 48 --model LSTM --patience 10 --save tuning_w48/LSTM_Appliance6 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 33 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance7 --window_size 48 --model LSTM --patience 10 --save tuning_w48/LSTM_Appliance7 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 34 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance8 --window_size 48 --model LSTM --patience 10 --save tuning_w48/LSTM_Appliance8 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 35 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance9 --window_size 48 --model LSTM --patience 10 --save tuning_w48/LSTM_Appliance9 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 3 36 | -------------------------------------------------------------------------------- /TCN_w6.sh: -------------------------------------------------------------------------------- 1 | # TCN 2 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Aggregate --window_size 6 --model TCN --layer 2 --patience 10 --save tuning_test/TCN_Aggregate --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 3 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance1 --window_size 6 --model TCN --layer 2 --patience 10 --save tuning_test/TCN_Appliance1 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 4 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance2 --window_size 6 --model TCN --layer 2 --patience 10 --save tuning_test/TCN_Appliance2 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 5 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance3 --window_size 6 --model TCN --layer 2 --patience 10 --save tuning_test/TCN_Appliance3 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 6 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance4 --window_size 6 --model TCN --layer 2 --patience 10 --save tuning_test/TCN_Appliance4 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 7 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance5 --window_size 6 --model TCN --layer 2 --patience 10 --save tuning_test/TCN_Appliance5 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 8 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance6 --window_size 6 --model TCN --layer 2 --patience 10 --save tuning_test/TCN_Appliance6 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 9 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance7 --window_size 6 --model TCN --layer 2 --patience 10 --save tuning_test/TCN_Appliance7 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 10 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance8 --window_size 6 --model TCN --layer 2 --patience 10 --save tuning_test/TCN_Appliance8 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 11 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance9 --window_size 6 --model TCN --layer 2 --patience 10 --save tuning_test/TCN_Appliance9 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 12 | 13 | # DNN 14 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Aggregate --window_size 6 --model ANN --patience 10 --save tuning_test/ANN_Aggregate --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 15 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance1 --window_size 6 --model ANN --patience 10 --save tuning_test/ANN_Appliance1 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 16 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance2 --window_size 6 --model ANN --patience 10 --save tuning_test/ANN_Appliance2 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 17 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance3 --window_size 6 --model ANN --patience 10 --save tuning_test/ANN_Appliance3 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 18 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance4 --window_size 6 --model ANN --patience 10 --save tuning_test/ANN_Appliance4 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 19 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance5 --window_size 6 --model ANN --patience 10 --save tuning_test/ANN_Appliance5 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 20 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance6 --window_size 6 --model ANN --patience 10 --save tuning_test/ANN_Appliance6 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 21 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance7 --window_size 6 --model ANN --patience 10 --save tuning_test/ANN_Appliance7 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 22 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance8 --window_size 6 --model ANN --patience 10 --save tuning_test/ANN_Appliance8 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 23 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance9 --window_size 6 --model ANN --patience 10 --save tuning_test/ANN_Appliance9 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 24 | 25 | # LSTM 26 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Aggregate --window_size 6 --model LSTM --patience 10 --save tuning_test/LSTM_Aggregate --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 27 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance1 --window_size 6 --model LSTM --patience 10 --save tuning_test/LSTM_Appliance1 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 28 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance2 --window_size 6 --model LSTM --patience 10 --save tuning_test/LSTM_Appliance2 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 29 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance3 --window_size 6 --model LSTM --patience 10 --save tuning_test/LSTM_Appliance3 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 30 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance4 --window_size 6 --model LSTM --patience 10 --save tuning_test/LSTM_Appliance4 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 31 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance5 --window_size 6 --model LSTM --patience 10 --save tuning_test/LSTM_Appliance5 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 32 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance6 --window_size 6 --model LSTM --patience 10 --save tuning_test/LSTM_Appliance6 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 33 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance7 --window_size 6 --model LSTM --patience 10 --save tuning_test/LSTM_Appliance7 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 34 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance8 --window_size 6 --model LSTM --patience 10 --save tuning_test/LSTM_Appliance8 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 35 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance9 --window_size 6 --model LSTM --patience 10 --save tuning_test/LSTM_Appliance9 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 9 36 | -------------------------------------------------------------------------------- /TCN_w96.sh: -------------------------------------------------------------------------------- 1 | # TCN 2 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Aggregate --window_size 96 --model TCN --layer 6 --patience 10 --save tuning_w96/TCN_Aggregate --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 3 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance1 --window_size 96 --model TCN --layer 6 --patience 10 --save tuning_w96/TCN_Appliance1 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 4 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance2 --window_size 96 --model TCN --layer 6 --patience 10 --save tuning_w96/TCN_Appliance2 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 5 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance3 --window_size 96 --model TCN --layer 6 --patience 10 --save tuning_w96/TCN_Appliance3 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 6 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance4 --window_size 96 --model TCN --layer 6 --patience 10 --save tuning_w96/TCN_Appliance4 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 7 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance5 --window_size 96 --model TCN --layer 6 --patience 10 --save tuning_w96/TCN_Appliance5 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 8 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance6 --window_size 96 --model TCN --layer 6 --patience 10 --save tuning_w96/TCN_Appliance6 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 9 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance7 --window_size 96 --model TCN --layer 6 --patience 10 --save tuning_w96/TCN_Appliance7 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 10 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance8 --window_size 96 --model TCN --layer 6 --patience 10 --save tuning_w96/TCN_Appliance8 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 11 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance9 --window_size 96 --model TCN --layer 6 --patience 10 --save tuning_w96/TCN_Appliance9 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 12 | 13 | # DNN 14 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Aggregate --window_size 96 --model ANN --patience 10 --save tuning_w96/ANN_Aggregate --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 15 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance1 --window_size 96 --model ANN --patience 10 --save tuning_w96/ANN_Appliance1 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 16 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance2 --window_size 96 --model ANN --patience 10 --save tuning_w96/ANN_Appliance2 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 17 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance3 --window_size 96 --model ANN --patience 10 --save tuning_w96/ANN_Appliance3 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 18 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance4 --window_size 96 --model ANN --patience 10 --save tuning_w96/ANN_Appliance4 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 19 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance5 --window_size 96 --model ANN --patience 10 --save tuning_w96/ANN_Appliance5 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 20 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance6 --window_size 96 --model ANN --patience 10 --save tuning_w96/ANN_Appliance6 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 21 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance7 --window_size 96 --model ANN --patience 10 --save tuning_w96/ANN_Appliance7 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 22 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance8 --window_size 96 --model ANN --patience 10 --save tuning_w96/ANN_Appliance8 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 23 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance9 --window_size 96 --model ANN --patience 10 --save tuning_w96/ANN_Appliance9 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 24 | 25 | # LSTM 26 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Aggregate --window_size 96 --model LSTM --patience 10 --save tuning_w96/LSTM_Aggregate --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 27 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance1 --window_size 96 --model LSTM --patience 10 --save tuning_w96/LSTM_Appliance1 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 28 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance2 --window_size 96 --model LSTM --patience 10 --save tuning_w96/LSTM_Appliance2 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 29 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance3 --window_size 96 --model LSTM --patience 10 --save tuning_w96/LSTM_Appliance3 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 30 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance4 --window_size 96 --model LSTM --patience 10 --save tuning_w96/LSTM_Appliance4 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 31 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance5 --window_size 96 --model LSTM --patience 10 --save tuning_w96/LSTM_Appliance5 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 32 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance6 --window_size 96 --model LSTM --patience 10 --save tuning_w96/LSTM_Appliance6 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 33 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance7 --window_size 96 --model LSTM --patience 10 --save tuning_w96/LSTM_Appliance7 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 34 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance8 --window_size 96 --model LSTM --patience 10 --save tuning_w96/LSTM_Appliance8 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 35 | python train_univariate.py --data REFIT_processed/RAW_House1_1T_processed.csv --target Appliance9 --window_size 96 --model LSTM --patience 10 --save tuning_w96/LSTM_Appliance9 --epoch 1000 --batchsize 128 --lr 0.008 --gpu 4 36 | -------------------------------------------------------------------------------- /image/Cure.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/joker507/Power_Load_Forecasting_by_TCN/4b71f0bc31e65bc1ba92c073872ab64ed6b024a0/image/Cure.jpg -------------------------------------------------------------------------------- /image/Dilated_Causal_Conv.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/joker507/Power_Load_Forecasting_by_TCN/4b71f0bc31e65bc1ba92c073872ab64ed6b024a0/image/Dilated_Causal_Conv.png -------------------------------------------------------------------------------- /image/ResultTable.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/joker507/Power_Load_Forecasting_by_TCN/4b71f0bc31e65bc1ba92c073872ab64ed6b024a0/image/ResultTable.jpg -------------------------------------------------------------------------------- /image/SourceData.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/joker507/Power_Load_Forecasting_by_TCN/4b71f0bc31e65bc1ba92c073872ab64ed6b024a0/image/SourceData.png -------------------------------------------------------------------------------- /image/SourceDataTable.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/joker507/Power_Load_Forecasting_by_TCN/4b71f0bc31e65bc1ba92c073872ab64ed6b024a0/image/SourceDataTable.jpg -------------------------------------------------------------------------------- /image/TCN.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/joker507/Power_Load_Forecasting_by_TCN/4b71f0bc31e65bc1ba92c073872ab64ed6b024a0/image/TCN.png -------------------------------------------------------------------------------- /image/cure2.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/joker507/Power_Load_Forecasting_by_TCN/4b71f0bc31e65bc1ba92c073872ab64ed6b024a0/image/cure2.jpg -------------------------------------------------------------------------------- /image/line.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/joker507/Power_Load_Forecasting_by_TCN/4b71f0bc31e65bc1ba92c073872ab64ed6b024a0/image/line.jpg -------------------------------------------------------------------------------- /model.py: -------------------------------------------------------------------------------- 1 | import tensorflow as tf 2 | import tensorflow.keras as keras 3 | from keras.utils.vis_utils import plot_model 4 | 5 | # TCN残差连接块 6 | def TCN_Block(inputs, dilation_rate, in_channels, 7 | out_channels, padding, kernel_size=2,dropout_rate=0.2): 8 | 9 | init = tf.keras.initializers.RandomNormal(mean=0.0, stddev=0.1) 10 | assert padding in ["causal","same"] 11 | 12 | # block1 13 | conv1 = keras.layers.Conv1D(filters=out_channels, kernel_size=kernel_size, dilation_rate=dilation_rate, padding=padding, kernel_initializer=init) 14 | batch1 = keras.layers.BatchNormalization(axis=-1) 15 | ac1 = keras.layers.Activation("relu") 16 | drop1 = keras.layers.Dropout(rate=dropout_rate) 17 | 18 | # block2 19 | conv2 = keras.layers.Conv1D(filters=out_channels, kernel_size=kernel_size, dilation_rate=dilation_rate, padding=padding, kernel_initializer=init) 20 | batch2 = keras.layers.BatchNormalization(axis=-1) 21 | ac2 = keras.layers.Activation("relu") 22 | drop2 = keras.layers.Dropout(rate=dropout_rate) 23 | 24 | # 输出维度一致 25 | downsample = keras.layers.Conv1D(filters=out_channels, 26 | kernel_size=1, 27 | padding="same", 28 | kernel_initializer=init) 29 | 30 | ac3 = keras.layers.Activation("relu") 31 | 32 | # forward 33 | # inputs = keras.Input(shape=(window_size, in_channels)) 34 | x = conv1(inputs) 35 | x = batch1(x) 36 | x = ac1(x) 37 | x = drop1(x) 38 | x = conv2(x) 39 | x = batch2(x) 40 | x = ac2(x) 41 | y = drop2(x) 42 | if in_channels != out_channels: 43 | res = downsample(inputs) 44 | else: 45 | res = inputs 46 | 47 | outputs = ac3(keras.layers.add([y,res])) 48 | 49 | return outputs 50 | 51 | # TCN 52 | def TCN(num_inputs=10 ,num_channels=[150,150,150,150], 53 | window_size=6, kernel_size=2, dropout=0.2): 54 | 55 | # input layers 56 | inputs = keras.Input(shape=(window_size,num_inputs)) 57 | t = inputs 58 | 59 | # TCN layers 60 | num_levels = len(num_channels) # 层数 61 | for i in range(num_levels): 62 | dilation_size = 2 ** i 63 | in_channels = num_inputs if i == 0 else num_channels[i-1] 64 | out_channels = num_channels[i] 65 | # tcn clock 66 | t = TCN_Block(t, 67 | dilation_rate=dilation_size, 68 | in_channels=in_channels, 69 | out_channels=out_channels, 70 | padding='causal', 71 | kernel_size=kernel_size, 72 | dropout_rate=dropout) 73 | 74 | # output layers 75 | # outputs = keras.layers.Conv1D(filters=1, kernel_size=3)(outputs) 76 | # outputs = keras.layers.Flatten()(outputs) 77 | outputs = keras.layers.Dense(1,activation=None)(t[:,-1,:]) # (batsize,final_time_step,dim) 78 | 79 | tcn_model = keras.Model(inputs=inputs, outputs=outputs) 80 | tcn_model.compile(optimizer="adam", loss="mse", metrics=["mae"]) 81 | tcn_model.summary() 82 | return tcn_model 83 | 84 | # LSTM 85 | def LSTM(window_size,units=12,dim=10): 86 | x = keras.layers.Input(shape=(window_size, dim)) 87 | t = keras.layers.LSTM(units=units, return_sequences=False)(x) 88 | y = keras.layers.Dense(1)(t) 89 | 90 | model = keras.Model(inputs=x,outputs=y) 91 | model.summary() 92 | return model 93 | 94 | # GRU 95 | def GRU(window_size): 96 | x = keras.layers.Input(shape=(window_size, 10)) 97 | t = keras.layers.GRU(units=12, return_sequences=True)(x) 98 | t = keras.layers.Dropout(0.2)(t) 99 | t = keras.layers.GRU(units=12, return_sequences=False)(t) 100 | t = keras.layers.Dropout(0.2)(t) 101 | y = keras.layers.Dense(1)(t) 102 | 103 | model = keras.Model(inputs=x,outputs=y) 104 | # model.compile(optimizer="adam", loss="mse", metrics=["mae"]) 105 | model.summary() 106 | return model 107 | 108 | # ANN（DNN） 109 | def ANN(window_size,dim=10): 110 | x = keras.layers.Input(shape=(window_size,dim)) 111 | t = keras.layers.Flatten()(x) 112 | t = keras.layers.Dense(128, activation='sigmoid')(t) 113 | t = keras.layers.Dense(256, activation='sigmoid')(t) 114 | t = keras.layers.Dense(128, activation='sigmoid')(t) 115 | y = keras.layers.Dense(1)(t) 116 | 117 | model = keras.Model(inputs=x,outputs=y) 118 | model.compile(optimizer="adam", loss="mse", metrics=["mae"]) 119 | model.summary() 120 | return model 121 | 122 | ## 模型构建 123 | def build_model(model_name, window_size,dim=10, TCN_layers=2): 124 | model_type = model_name 125 | if model_type == "LSTM": 126 | train_model = LSTM(window_size=window_size,units=12,dim=dim) 127 | elif model_type == "ANN": 128 | train_model = ANN(window_size,dim) 129 | elif model_type == "GRU": 130 | train_model = GRU(window_size,units=12) 131 | elif model_type == "TCN": 132 | num_inputs=1 133 | num_channels=[20 for i in range(TCN_layers)] 134 | print("TCN_layers is ", TCN_layers) 135 | kernel_size=3 136 | dropout=0.2 137 | train_model = TCN(num_inputs,num_channels,window_size,kernel_size,dropout) 138 | else: 139 | assert False, "请选择正确的模型" 140 | 141 | return train_model 142 | 143 | if __name__ == "__main__": 144 | window_size = 6 145 | m = ANN(window_size,dim=1) 146 | plot_model(m, 'model.png', show_shapes=True, show_layer_names=True) 147 | 148 | 149 | -------------------------------------------------------------------------------- /requirements.txt: -------------------------------------------------------------------------------- 1 | keras==2.6.0 2 | matplotlib==3.5.2 3 | numpy==1.19.4 4 | numpy==1.19.5 5 | pandas==1.4.3 6 | tensorflow==2.6.0 7 | -------------------------------------------------------------------------------- /tools.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | 3 | ## 评估指标 4 | def rmse(true,pre): 5 | return np.sqrt(np.mean(np.square(true-pre))) 6 | 7 | def mae(true, pre): 8 | return np.mean(np.abs(true - pre)) 9 | 10 | def mse(true,pre): 11 | return np.mean(np.square(true-pre)) 12 | 13 | def mape(true, pre): 14 | return np.mean(np.abs((true - pre) / true)) * 100 #单位% 15 | 16 | ## zero-score归一化方法 17 | class Standardscale(): 18 | def __init__(self): 19 | self.mean = 0. 20 | self.std = 1. 21 | 22 | def fit(self, data): 23 | self.mean = data.mean(0) 24 | self.std = data.std(0) 25 | 26 | def transform(self, data): 27 | mean = self.mean 28 | std = self.std 29 | return (data - mean) / std 30 | 31 | def inverse_transform(self, data): 32 | mean = self.mean 33 | std = self.std 34 | 35 | # 如果参数是数组，数据是不一样的数组时，采用第一列特征进行放缩 36 | if len(mean.shape) != 0: # 参数不是数值并且与数据长度不一致时转为数值 37 | if data.shape[-1] != mean.shape[-1]: 38 | mean = mean[0] 39 | std = std[0] 40 | 41 | return (data * std) + mean 42 | 43 | ## Max-min最大最小归一化 44 | class Maxminscale(): 45 | def __init__(self): 46 | self.min = 0. 47 | self.max = 1. 48 | 49 | def fit(self, data): 50 | self.max = data.max(0) 51 | self.min = data.min(0) 52 | 53 | def transform(self, data): 54 | max = self.max 55 | min = self.min 56 | return (data - min) / (max - min) 57 | 58 | def inverse_transform(self, data): 59 | max = self.max 60 | min = self.min 61 | 62 | if len(max.shape) != 0: 63 | if data.shape[-1] != max.shape[-1]: #如果数据的维度和参数的维度不一致 采用第0个 64 | max = max[0] 65 | min = min[0] 66 | return (data * (max - min)) + min 67 | 68 | ## 无需归一化 69 | class Nonescale(): 70 | def __init__(self): 71 | pass 72 | 73 | def fit(self, data): 74 | pass 75 | 76 | def transform(self, data): 77 | return data 78 | 79 | def inverse_transform(self, data): 80 | return data -------------------------------------------------------------------------------- /train_univariate.py: -------------------------------------------------------------------------------- 1 | from numpy.random import seed 2 | seed(42) ## 设置伪随机序列 3 | from tensorflow.random import set_seed 4 | set_seed(42) 5 | 6 | import argparse 7 | import os 8 | import json 9 | import pandas as pd 10 | import numpy as np 11 | import matplotlib.pyplot as plt 12 | import tensorflow as tf 13 | import tensorflow.keras as keras 14 | from keras.utils.vis_utils import plot_model 15 | from model import build_model 16 | from tools import rmse,mae,mse,mape,Standardscale,Maxminscale,Nonescale 17 | from vmd import VMD 18 | 19 | parser = argparse.ArgumentParser(description="Electrical load curve prediction") 20 | parser.add_argument("--data", type=str, default="REFIT_processed\RAW_House1_12H_processed.csv", 21 | help="a csv file path which include some appliance load data ") 22 | parser.add_argument("--target", type=str, default='Aggregate', 23 | help="predict target cure including Aggregate or Appliancex") 24 | parser.add_argument("--test_rate", type=float, default=0.2, 25 | help="test set rate") 26 | parser.add_argument("--window_size", type=int, default=6, 27 | help="Historical reference length") 28 | parser.add_argument("--scale", type=str, default="standard", 29 | help="data scale for model inputs") 30 | parser.add_argument("--save", type=str, default=os.getcwd()+"/tuning_univariate/", 31 | help="training result folder") 32 | parser.add_argument("--model", type=str, default="CA", 33 | help="Model name") 34 | parser.add_argument("--layer", type=int, default=2, 35 | help="TCN layers") 36 | parser.add_argument("--lr", type=float, default=0.001, 37 | help="learning rate") 38 | parser.add_argument("--vmd", action='store_true', 39 | help="sue vmd to extract feature") 40 | parser.add_argument("--patience",type=int, default=10, 41 | help="early stop") 42 | parser.add_argument("--epoch",type=int, default=1000, 43 | help="training epoch") 44 | parser.add_argument("--batchsize",type=int, default=128, 45 | help="batchsize") 46 | parser.add_argument("--gpu",type=int, default=0, 47 | help="gpu device id") 48 | 49 | args = parser.parse_args() 50 | 51 | gpus = tf.config.list_physical_devices('GPU') 52 | if gpus: 53 | # 设置 TensorFlow 只使用第一块 GPU 54 | try: 55 | tf.config.set_visible_devices(gpus[args.gpu], 'GPU') 56 | logical_gpus = tf.config.experimental.list_logical_devices('GPU') 57 | print(len(gpus), "Physical GPUs,", len(logical_gpus), "Logical GPUs") 58 | except RuntimeError as e: 59 | # 异常处理 60 | print(e) 61 | 62 | if not os.path.exists(args.save): 63 | os.makedirs(args.save) 64 | log_dic = {"data":{}, "model":{}, "traing":{}, "testing":{}} 65 | plot = False 66 | 67 | #################################################################################### 68 | # Load Data 69 | #################################################################################### 70 | def make_data(file_path, test_rate, window_size,scale, target): 71 | 72 | df = pd.read_csv(file_path, parse_dates=["Time"], index_col ="Time") 73 | df = df[target] 74 | 75 | # 切分数据集 76 | test_num = int(len(df) * test_rate) 77 | train_data = df[:-test_num].copy().values 78 | test_data = df[-test_num:].copy().values 79 | print("dataset shape:",df.shape) 80 | print("train_data shape:",train_data.shape) 81 | print("test_data shape:",test_data.shape) 82 | 83 | scale.fit(train_data) 84 | train_data = scale.transform(train_data) 85 | 86 | # 测试集数据归一化 87 | test_data = scale.transform(test_data) 88 | 89 | # 窗口滑动制作数据样本 90 | train_X = [] 91 | train_y = [] 92 | train_num = train_data.shape[0] 93 | for k in range(train_num-window_size): 94 | train_X.append(train_data[k:k+window_size]) #x: [k:k+window_size) 95 | train_y.append(train_data[k+window_size]) #y: k+windowsize 96 | 97 | test_X = [] 98 | test_y = [] 99 | test_num = test_data.shape[0] 100 | for k in range(test_num-window_size): 101 | test_X.append(test_data[k:k+window_size]) #x: [k: k+window_size) 102 | test_y.append(test_data[k+window_size]) #y: k+windowsize 103 | 104 | train_X, train_y = np.array(train_X), np.reshape(train_y,(len(train_y),1)) 105 | test_X, test_y = np.array(test_X), np.reshape(test_y,(len(test_y),1)) 106 | 107 | print("train_x shape: ", train_X.shape) 108 | print("train_y shape: ", train_y.shape) 109 | print("test_x shape: ", test_X.shape) 110 | print("test_y shape: ", test_y.shape) 111 | 112 | return train_X,train_y,test_X,test_y,scale 113 | 114 | scale = Standardscale() 115 | train_X,train_y,test_X,test_y,scale = make_data(args.data, args.test_rate, args.window_size, scale, args.target) 116 | log_dic["data"]["file"] = args.data 117 | log_dic["data"]["target"] = args.target 118 | log_dic["data"]["windows"] = args.window_size 119 | 120 | 121 | ## shuufle 122 | indexes = np.arange(len(train_X)) 123 | np.random.shuffle(indexes) 124 | train_X = train_X[indexes] 125 | train_y = train_y[indexes] 126 | log_dic["data"]["seed"] = 42 127 | 128 | #################################################################################### 129 | # extract vmd feature 130 | #################################################################################### 131 | 132 | if args.vmd: # 增加vmd特征提取 133 | 134 | alpha = 2000 # moderate bandwidth constraint 135 | tau = 0. # noise-tolerance (no strict fidelity enforcement) 136 | K = 3 # 3 modes 137 | DC = 0 # no DC part imposed 138 | init = 1 # initialize omegas uniformly 139 | tol = 1e-7 140 | 141 | vmd_train_X = [] # 存储vmd数据 142 | vmd_test_X = [] 143 | for i in range(len(train_X)): # for every one example 144 | vmd_compose_matrix , u_hat, omega = VMD(train_X[i], alpha, tau, K, DC, init, tol) 145 | vmd_train_X.append(vmd_compose_matrix.T) # append(shape=(W,3)) 146 | 147 | for i in range(len(test_X)): # for every one example 148 | vmd_compose_matrix , u_hat, omega = VMD(test_X[i], alpha, tau, K, DC, init, tol) 149 | vmd_test_X.append(vmd_compose_matrix.T) # append(shape=(W,3)) 150 | 151 | train_X = np.expand_dims(train_X, axis=-1) 152 | test_X = np.expand_dims(test_X, axis=-1) 153 | train_X = np.concatenate((train_X,np.array(vmd_train_X)), axis=-1) # (N, W, 1) (N, W, 3) -> (B, W, 4) 154 | test_X = np.concatenate((test_X,np.array(vmd_test_X)), axis=-1) 155 | print(train_X.shape) 156 | print(test_X.shape) 157 | log_dic["data"]["vmd"] = 3 158 | dim = 4 ## 输入维度大小 159 | 160 | else: 161 | train_X = np.expand_dims(train_X, axis=-1) 162 | test_X = np.expand_dims(test_X, axis=-1) 163 | dim = 1 ## 输入维度大小 164 | 165 | log_dic['data']["train_x"] = str(train_X.shape) 166 | log_dic['data']["train_y"] = str(train_y.shape) 167 | log_dic['data']["test_x"] = str(test_X.shape) 168 | log_dic['data']["test_y"] = str(test_y.shape) 169 | 170 | #################################################################################### 171 | # Build Model 172 | #################################################################################### 173 | 174 | train_model = build_model(args.model, args.window_size, dim, TCN_layers=args.layer) 175 | log_dic["model"]["name"] = args.model 176 | 177 | optimizer = keras.optimizers.Adam(learning_rate=args.lr) 178 | log_dic["model"]["lr"] = args.lr 179 | 180 | train_model.compile(optimizer=optimizer, loss="mse", metrics=["mae"]) 181 | log_dic["model"]["loss"] = "mse" 182 | 183 | # 绘制模型图 184 | # plot_model(train_model, to_file=os.path.join(args.save, args.model+"_model.png"), 185 | # show_shapes=True, 186 | # show_dtype=True, 187 | # show_layer_names=True) 188 | 189 | 190 | #################################################################################### 191 | # Training Code 192 | #################################################################################### 193 | 194 | callback = keras.callbacks.EarlyStopping(monitor='val_loss', patience=args.patience) 195 | history = train_model.fit(x=train_X, y=train_y, 196 | # train_dataset, 197 | epochs=args.epoch, batch_size=args.batchsize, shuffle=True, 198 | validation_split = 0.2,callbacks=[callback]) 199 | 200 | log_dic["traing"]["epoch"] = args.epoch 201 | log_dic["traing"]["batchsize"] = args.batchsize 202 | log_dic["traing"]["early_stop"] = len(history.history["val_loss"]) 203 | 204 | epoch = len(history.history["val_loss"]) 205 | 206 | # 保存训练好的模型 207 | train_model.save(os.path.join(args.save, args.model+".h5")) 208 | 209 | # 最佳验证集结果：用来挑选超参 210 | val_loss = min(history.history['val_loss']) 211 | val_epoch = history.history['val_loss'].index(val_loss) + 1 212 | val_mae = history.history['val_mae'][val_epoch-1] 213 | train_loss = history.history['loss'][-1] 214 | train_mae = history.history['mae'][-1] 215 | log_dic["traing"]["best_val_mse"] = round(val_loss,3) 216 | log_dic["traing"]["best_val_mae"] = round(val_loss,3) 217 | log_dic["traing"]["best_epoch"] = val_epoch 218 | log_dic["traing"]["final_train_loss"] = round(train_loss,3) 219 | log_dic["traing"]["final_train_mae"] = round(train_mae,3) 220 | log_dic["traing"]["final_val_loss"] = round(history.history['val_loss'][-1],3) 221 | log_dic["traing"]["final_val_mae"] = round(history.history['val_mae'][-1],3) 222 | 223 | ## display 224 | # loss 225 | plt.figure(figsize=(20,10)) 226 | plt.plot(history.history['loss'], label="train loss") 227 | plt.plot(history.history['val_loss'], label="val loss") 228 | plt.title(args.model + 'train loss') 229 | plt.ylabel('loss') 230 | plt.xlabel('epoch') 231 | plt.legend(loc='upper right') 232 | plt.savefig(os.path.join(args.save, 'train_loss.jpg')) 233 | if plot: 234 | plt.show() 235 | 236 | #################################################################################### 237 | # Testing Model 238 | #################################################################################### 239 | 240 | train_model = keras.models.load_model(os.path.join(args.save, args.model+".h5")) 241 | print("sucessfully load the model") 242 | 243 | # 测试集评估模型 244 | evaluate_result = train_model.evaluate(x=test_X, y=test_y) 245 | print("Test loss:{}, mae:{}".format(evaluate_result[0],evaluate_result[1])) 246 | log_dic["testing"]["evaluate"] = {"loss":evaluate_result[0], "mae":evaluate_result[1]} 247 | 248 | # 分析预测结果 249 | predicts = train_model.predict(test_X) 250 | predicts_inv = scale.inverse_transform(predicts) 251 | test_predicts = np.reshape(predicts, len(predicts)) 252 | test_predicts_inv = np.reshape(predicts_inv, len(predicts)) 253 | 254 | trues = test_y 255 | trues_inv = scale.inverse_transform(test_y) 256 | test_trues = np.reshape(trues, (len(trues))) 257 | test_trues_inv = np.reshape(trues_inv, (len(trues_inv))) 258 | 259 | r = rmse(test_trues,test_predicts) 260 | m = mae(test_trues,test_predicts) 261 | ms = mse(test_trues,test_predicts) 262 | mape_t = mape(test_trues,test_predicts) 263 | r_inv = rmse(test_trues_inv,test_predicts_inv) 264 | m_inv = mae(test_trues_inv,test_predicts_inv) 265 | mse_inv = mse(test_trues_inv,test_predicts_inv) 266 | mape_inv = mape(test_trues_inv,test_predicts_inv) 267 | predict_df = pd.DataFrame({ 268 | "test_trues":test_trues, 269 | "test_predicts":test_predicts, 270 | "test_trues_inv":test_trues_inv, 271 | "test_predicts_inv":test_predicts_inv 272 | }) 273 | predict_df.to_csv(os.path.join(args.save , 'predicts.csv')) 274 | 275 | print("Test set:"+">"*5+args.model+"<"*5) 276 | print("standard result") 277 | print("rmse:",r) 278 | print("mae:",m) 279 | print("mse:",ms) 280 | print("mape:",mape_t) 281 | print("inverter result") 282 | print("rmse:",r_inv) 283 | print("mae:",m_inv) 284 | print("mse:",mse_inv) 285 | print("mape:",mape_inv) 286 | 287 | metrics_df = pd.DataFrame({ 288 | "method":["rmse","mae","mse","mape"], 289 | "standard":[r,m,ms,mape_t], 290 | "inverse":[r_inv,m_inv,mse_inv,mape_inv], 291 | }) 292 | log_dic["testing"]["standard"] = {"rmse":r, "mse": ms, "mae": m, "mape": mape_t} 293 | log_dic["testing"]["inverse"] = {"rmse":r_inv, "mse": mse_inv, "mae": m_inv, "mape": mape_inv} 294 | 295 | metrics_df.to_csv(os.path.join(args.save,'test_metrics.csv')) 296 | 297 | # predict 298 | plt.figure(figsize=(20,10)) 299 | plt.title("standard predict: rmse:{:.2f} mse:{:.2f} mae:{:.2f} mape:{:.2f}".format(r,ms,m,mape_t)) 300 | plt.plot(test_trues,'b', label="true") 301 | plt.plot(test_trues,'bo', markersize=3, label="true") 302 | plt.plot(test_predicts,'r', label='predict') 303 | plt.plot(test_predicts,'rx', markersize=5, label='predict') 304 | plt.legend(loc='best') 305 | plt.savefig(os.path.join(args.save, 'standard_predict.jpg')) 306 | if plot: 307 | plt.show() 308 | 309 | plt.figure(figsize=(20,10)) 310 | plt.title("inverse predict: rmse:{:.2f} mse:{:.2f} mae:{:.2f} mape:{:.2f}".format(r_inv,mse_inv,m_inv,mape_inv)) 311 | plt.plot(test_trues_inv,'b', label="true") 312 | plt.plot(test_trues_inv,'bo', markersize=3, label="true") 313 | plt.plot(test_predicts_inv,'r', label='predict') 314 | plt.plot(test_predicts_inv,'rx', markersize=5, label='predict') 315 | plt.legend(loc='best') 316 | plt.savefig(os.path.join(args.save, 'inverse_predict.jpg')) 317 | if plot: 318 | plt.show() 319 | 320 | ## 保存参数与结果 321 | with open(os.path.join(args.save, 'log.json'),'w') as f: 322 | json.dump(log_dic,f,indent=4) 323 | 324 | print("result dir in: ",os.path.join(args.save)) 325 | -------------------------------------------------------------------------------- /vmd.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import pandas as pd 3 | 4 | def VMD(f, alpha, tau, K, DC, init, tol): 5 | """ 6 | u,u_hat,omega = VMD(f, alpha, tau, K, DC, init, tol) 7 | Variational mode decomposition 8 | Python implementation by Vinícius Rezende Carvalho - vrcarva@gmail.com 9 | code based on Dominique Zosso's MATLAB code, available at: 10 | https://www.mathworks.com/matlabcentral/fileexchange/44765-variational-mode-decomposition 11 | Original paper: 12 | Dragomiretskiy, K. and Zosso, D. (2014) ‘Variational Mode Decomposition’, 13 | IEEE Transactions on Signal Processing, 62(3), pp. 531–544. doi: 10.1109/TSP.2013.2288675. 14 | 15 | 16 | Input and Parameters: 17 | --------------------- 18 | f - the time domain signal (1D) to be decomposed 19 | alpha - the balancing parameter of the data-fidelity constraint 20 | tau - time-step of the dual ascent ( pick 0 for noise-slack ) 21 | K - the number of modes to be recovered 22 | DC - true if the first mode is put and kept at DC (0-freq) 23 | init - 0 = all omegas start at 0 24 | 1 = all omegas start uniformly distributed 25 | 2 = all omegas initialized randomly 26 | tol - tolerance of convergence criterion; typically around 1e-6 27 | 28 | Output: 29 | ------- 30 | u - the collection of decomposed modes 31 | u_hat - spectra of the modes 32 | omega - estimated mode center-frequencies 33 | """ 34 | 35 | if len(f) % 2: 36 | f = f[:-1] 37 | 38 | # Period and sampling frequency of input signal 39 | fs = 1. / len(f) 40 | 41 | ltemp = len(f) // 2 42 | fMirr = np.append(np.flip(f[:ltemp], axis=0), f) 43 | fMirr = np.append(fMirr, np.flip(f[-ltemp:], axis=0)) 44 | 45 | # Time Domain 0 to T (of mirrored signal) 46 | T = len(fMirr) 47 | t = np.arange(1, T + 1) / T 48 | 49 | # Spectral Domain discretization 50 | freqs = t - 0.5 - (1 / T) 51 | 52 | # Maximum number of iterations (if not converged yet, then it won't anyway) 53 | Niter = 500 54 | # For future generalizations: individual alpha for each mode 55 | Alpha = alpha * np.ones(K) 56 | 57 | # Construct and center f_hat 58 | f_hat = np.fft.fftshift((np.fft.fft(fMirr))) 59 | f_hat_plus = np.copy(f_hat) # copy f_hat 60 | f_hat_plus[:T // 2] = 0 61 | 62 | # Initialization of omega_k 63 | omega_plus = np.zeros([Niter, K]) 64 | 65 | if init == 1: 66 | for i in range(K): 67 | omega_plus[0, i] = (0.5 / K) * (i) 68 | elif init == 2: 69 | omega_plus[0, :] = np.sort(np.exp(np.log(fs) + (np.log(0.5) - np.log(fs)) * np.random.rand(1, K))) 70 | else: 71 | omega_plus[0, :] = 0 72 | 73 | # if DC mode imposed, set its omega to 0 74 | if DC: 75 | omega_plus[0, 0] = 0 76 | 77 | # start with empty dual variables 78 | lambda_hat = np.zeros([Niter, len(freqs)], dtype=complex) 79 | 80 | # other inits 81 | uDiff = tol + np.spacing(1) # update step 82 | n = 0 # loop counter 83 | sum_uk = 0 # accumulator 84 | # matrix keeping track of every iterant // could be discarded for mem 85 | u_hat_plus = np.zeros([Niter, len(freqs), K], dtype=complex) 86 | 87 | # *** Main loop for iterative updates*** 88 | 89 | while (uDiff > tol and n < Niter - 1): # not converged and below iterations limit 90 | # update first mode accumulator 91 | k = 0 92 | sum_uk = u_hat_plus[n, :, K - 1] + sum_uk - u_hat_plus[n, :, 0] 93 | 94 | # update spectrum of first mode through Wiener filter of residuals 95 | u_hat_plus[n + 1, :, k] = (f_hat_plus - sum_uk - lambda_hat[n, :] / 2) / ( 96 | 1. + Alpha[k] * (freqs - omega_plus[n, k]) ** 2) 97 | 98 | # update first omega if not held at 0 99 | if not (DC): 100 | omega_plus[n + 1, k] = np.dot(freqs[T // 2:T], (abs(u_hat_plus[n + 1, T // 2:T, k]) ** 2)) / np.sum( 101 | abs(u_hat_plus[n + 1, T // 2:T, k]) ** 2) 102 | 103 | # update of any other mode 104 | for k in np.arange(1, K): 105 | # accumulator 106 | sum_uk = u_hat_plus[n + 1, :, k - 1] + sum_uk - u_hat_plus[n, :, k] 107 | # mode spectrum 108 | u_hat_plus[n + 1, :, k] = (f_hat_plus - sum_uk - lambda_hat[n, :] / 2) / ( 109 | 1 + Alpha[k] * (freqs - omega_plus[n, k]) ** 2) 110 | # center frequencies 111 | omega_plus[n + 1, k] = np.dot(freqs[T // 2:T], (abs(u_hat_plus[n + 1, T // 2:T, k]) ** 2)) / np.sum( 112 | abs(u_hat_plus[n + 1, T // 2:T, k]) ** 2) 113 | 114 | # Dual ascent 115 | lambda_hat[n + 1, :] = lambda_hat[n, :] + tau * (np.sum(u_hat_plus[n + 1, :, :], axis=1) - f_hat_plus) 116 | 117 | # loop counter 118 | n = n + 1 119 | 120 | # converged yet? 121 | uDiff = np.spacing(1) 122 | for i in range(K): 123 | uDiff = uDiff + (1 / T) * np.dot((u_hat_plus[n, :, i] - u_hat_plus[n - 1, :, i]), 124 | np.conj((u_hat_plus[n, :, i] - u_hat_plus[n - 1, :, i]))) 125 | 126 | uDiff = np.abs(uDiff) 127 | 128 | # Postprocessing and cleanup 129 | 130 | # discard empty space if converged early 131 | Niter = np.min([Niter, n]) 132 | omega = omega_plus[:Niter, :] 133 | 134 | idxs = np.flip(np.arange(1, T // 2 + 1), axis=0) 135 | # Signal reconstruction 136 | u_hat = np.zeros([T, K], dtype=complex) 137 | u_hat[T // 2:T, :] = u_hat_plus[Niter - 1, T // 2:T, :] 138 | u_hat[idxs, :] = np.conj(u_hat_plus[Niter - 1, T // 2:T, :]) 139 | u_hat[0, :] = np.conj(u_hat[-1, :]) 140 | 141 | u = np.zeros([K, len(t)]) 142 | for k in range(K): 143 | u[k, :] = np.real(np.fft.ifft(np.fft.ifftshift(u_hat[:, k]))) 144 | 145 | # remove mirror part 146 | u = u[:, T // 4:3 * T // 4] 147 | 148 | # recompute spectrum 149 | u_hat = np.zeros([u.shape[1], K], dtype=complex) 150 | for k in range(K): 151 | u_hat[:, k] = np.fft.fftshift(np.fft.fft(u[k, :])) 152 | 153 | return u, u_hat, omega 154 | 155 | 156 | if __name__ == "__main__": 157 | # some sample parameters for VMD 158 | alpha = 2000 # moderate bandwidth constraint 159 | tau = 0. # noise-tolerance (no strict fidelity enforcement) 160 | K = 3 # 3 modes 161 | DC = 0 # no DC part imposed 162 | init = 1 # initialize omegas uniformly 163 | tol = 1e-7 164 | 165 | x_data = np.sin(np.arange(0, 2*np.pi, 2*np.pi/100)) 166 | print(x_data.shape) 167 | vmd_compose_matrix , u_hat, omega = VMD(x_data,alpha, tau, K, DC, init, tol) 168 | print(vmd_compose_matrix.shape) 169 | --------------------------------------------------------------------------------