Electric energy theft has long been a significant challenge for power supply companies.An electricity theft detection method based on the fusion of multi-source heterogeneous time series features is proposed.First,multi-source heterogeneous data such as meteorology,calendar,and family attributes are selected and a multi-feature graph structure is constructed through feature analysis.Then a graph neural network is utilized to conduct spatiotemporal modeling of multi-source heterogeneous data,and an attention mechanism is introduced to focus on key spatiotemporal features.The experiment results show that compared with a single data source,multi-source feature fusion can significantly improve detection performance.The proposed model outperforms other comparative models,which provides a new perspective for building efficient electricity theft detection systems.
关键词
窃电/多源异构时间序列/特征融合/图神经网络/注意力机制
Key words
electricity theft/multi-source heterogeneous time series/feature fusion/graph neural network/attention mechanism
维普
万方数据