基于时域卷积网络和自注意力的非侵入式负荷监测方法  

Non-intrusive Load Monitoring Method Based on Temporal Convolutional Network and Self Attention

作　　者：王德文[1] 貟青青 WANG Dewen;YUN Qingqing(School of Control and Computer Engineering,North China Electric Power University,Baoding 071003,China)

机构地区：[1]华北电力大学控制与计算机工程学院,河北保定071003

出　　处：《电力科学与工程》2023年第3期42-51,共10页Electric Power Science and Engineering

基　　金：河北省自然科学基金(F2021502013)。

摘　　要：针对用电设备的运行状态多样、持续时间长等特性导致的现有负荷监测模型计算复杂度高、难以捕获设备长时间运行模式的问题,提出一种基于时域卷积网络(Temporal convolutional network,TCN)与自注意力模型的非侵入式负荷监测方法。采用序列到点学习方式,将总电能消耗输入到TCN中以提取丰富特征;同时,通过残差连接,学习用电设备不同层次的能耗模式。利用自注意力模型,计算总信号每个位置的重要性,捕捉模式间的内部相关性。利用REDD数据集开展对比实验。实验结果表明,与采用循环神经网络、去噪自编码器、卷积神经网络、生成对抗网络和TCN的负荷监测方法相比,该模型的平均绝对误差降低了约38%,F1分数提高了约17%。最后,利用UK-DALE数据集验证了该模型的泛化能力。In order to solve the problem that the existing load monitoring models have high computational complexity and are difficult to capture the long-term operation mode of equipment due to the characteristics of various operation states and long duration of electrical equipment,a non-intrusive load monitoring method based on temporal convolutional network(TCN)and self attention model is proposed.The sequence to point learning method is adopted,and the total power consumption is input into TCN to extract rich features.Meanwhile,the energy consumption mode of different levels of electric equipment is learned through residual connection.The self attention model is used to calculate the importance of each position of the total signal to capture the internal correlation between modes.The REDD dataset is used to carry out comparative experiments.The experimental results show that compared with load monitoring methods using Recurrent Neural Network,Denoising AutoEncoder,Convolutional Neural Network,Generative Adversarial Networks and TCN,the mean absolute error of the model is decreased by about 38%,and the F1 score is increased by about 17%.Finally,the generalization ability of the model is verified by UK-DALE data set.

关 键 词：非侵入 负荷监测 序列到点 时域卷积网络 残差连接 自注意力 

分 类 号：TM714[电气工程—电力系统及自动化]