| 引用本文: | 林顺富,李毅,沈运帷,林屹峰,李东东.基于全卷积去噪自编码器与卷积块注意力模块的非侵入式居民负荷分解模型[J].电力自动化设备,2024,44(3):127-133 |
| LIN Shunfu,LI Yi,SHEN Yunwei,LIN Yifeng,LI Dongdong.Non-intrusive residential load disaggregation model based on fully convolutional denoising auto-encoder and convolutional block attention module[J].Electric Power Automation Equipment,2024,44(3):127-133 |
|
| 摘要: |
| 为了进一步提高低频居民负荷分解模型的分解精度与泛化能力，提出一种基于全卷积去噪自编码器与卷积块注意力模块的非侵入式居民负荷分解模型，该模型能够深度解析单一电器的功率曲线。基于全卷积去噪自编码器分别构建主回归子任务网络和辅助分类子任务网络；在子任务网络中，通过引入卷积块注意力模块自适应分配特征注意力权重，以减小不重要因素在模型训练过程中的影响；将辅助分类子任务网络的输出作为主回归子任务网络输出的门控单元，实现最终的负荷分解。基于公开数据集的算例结果表明，所提负荷分解模型比现有负荷分解模型具有更优的分解精度和泛化能力。 |
| 关键词: 负荷分解 全卷积去噪自编码器 注意力模块 子任务网络 门控单元 |
| DOI：10.16081/j.epae.202306004 |
| 分类号:TM714 |
| 基金项目:国家自然科学基金资助项目(51977127)；上海市科学技术委员会资助项目(19020500800)；上海市教育发展基金会和上海市教育委员会“曙光计划”项目(20SG52) |
|
| Non-intrusive residential load disaggregation model based on fully convolutional denoising auto-encoder and convolutional block attention module |
|
LIN Shunfu, LI Yi, SHEN Yunwei, LIN Yifeng, LI Dongdong
|
|
School of Electrical Engineering, Shanghai University of Electric Power, Shanghai 200090, China
|
| Abstract: |
| In order to further improve the disaggregation accuracy and generalization ability of low-frequency residential load disaggregation model, a non-intrusive residential load disaggregation model based on fully convolutional denoising auto-encoder and convolutional block attention module is proposed, which can deeply analyze the power curve of a single appliance. The main regression subtask network and auxiliary classification subtask network are respectively constructed based on fully convolutional denoising auto-encoder. In the subtask network, the convolutional block attention module is introduced to adaptively assign the feature attention weight, which reduces the influence of unimportant factors in the model training process. The output of auxiliary classification subtask network is taken as the gating unit of the output of main regression subtask network, and the final load decomposition is realized. The example results based on public datasets show that the proposed load disaggregation model has better disaggregation accuracy and generalization ability than the existing load disaggregation models. |
| Key words: load disaggregation fully convolutional denoising auto-encoder attention module subtask network gating unit |
