【摘要】 针对窃电量小、窃电发生时间随机的窃电行为，提出一种基于时序偏移双残差网络(TS-Bi-ResNet)的窃电行为检测模型.将基础残差网络模型改进为双残差网络(bi-residual network,Bi-ResNet)模型，考虑到窃电行为发生时间的随机性，利用时序偏移(timing shift,TS)算法对用电数据预处理，使模型能够学习用电数据的时间因素特征，构成TS-Bi-ResNet模型.根据真实用电数据和窃电特征生成含有伪窃电数据的混合用电数据集，利用TS-Bi-ResNet模型学习其浅层特征和深层特征，进而执行窃电行为检测.仿真和实际运行结果表明，TS-Bi-ResNet模型可以有效检测窃电量小且窃电发生时间随机的窃电行为，其检测精度优于LSTM模型与残差网络(ResNet)模型.更多还原

【Abstract】 Aiming at the electricity stealing behavior with small amount of electricity stealing and the randomness of the occurrence time of electricity stealing behavior, a detection model of electricity stealing behavior based on timing shift bi-residual network(TS-Bi-ResNET) has been proposed. The basic residual network model has been improved to be bi-residual network(Bi-ResNET) model. Considering the randomness of the occurrence time of electricity stealing behavior, the timing shift(TS) algorithm has been used to preprocess the electricity consumption data, so that the model can learn the time factor characteristics of the electricity consumption data, and has formed the TS-Bi-ResNET model. According to the real electricity consumption data and electricity theft characteristics, the mixed electricity consumption data set containing pseudo electricity theft data has been generated, and the TS-Bi-ResNET model been used to learn its shallow and deep features so as to detect electricity stealing behavior. The simulation and actual operation results show that TS-Bi-ResNET model can effectively detect electricity stealing behavior with small amount of electricity stealing and the randomness of the occurrence time of electricity stealing behavior, the detection accuracy of electricity theft with is better than LSTM model and ResNET model.更多还原