Research on Electricity Theft Identification Method Based on Multi-Component Feature Fusion

The huge amount of electricity consumption data in the smart grid makes data-driven electricity theft detection possible, but extracting features from electricity consumption time series that can help electricity theft detection is a challenging problem. To address this problem, a power theft user identification method based on STL decomposition and feature fusion is proposed. Firstly, the original sequence is decomposed into trend, seasonal and residual components using a robust STL decomposition method; then, according to the characteristics of each component, the features of the trend and seasonal components are extracted using an LSTM neural network with an attentional mechanism (Att-LSTM) and a one-dimensional convolutional neural network (1D-CNN), respectively; for the residual component, it is treated by a dynamics system analysis method is processed and the fractal dimension is calculated as its feature by GP algorithm; finally, the three component features are fused using a fully connected layer with attention mechanism for electricity theft user identification and compared with other methods. The results show that the feature extraction method used in this paper can effectively extract the features of electricity theft, and the method can accurately identify the electricity theft users.