基于多源异构时间序列特征融合的电力窃电检测的研究

Research on electricity theft detection based on multi-source heterogeneous time series feature fusion

长期以来,供电企业一直面临着电能被窃这一严重问题。为了有效应对这一难题,本文提出了一种基于多源异构时间序列特征融合的电力窃电检测方法。首先,通过特征分析,选择气象、日历、家庭属性等多源异构数据并构建多特征图结构;其次,利用图神经网络对多源异构数据进行时空建模,并引入注意力机制聚焦关键的时空特征。实验表明:与单一数据源相比,多源特征融合可显著提升检测性能,所提出的模型优于其他对比模型,为构建高效的电力窃电检测系统提供了新思路。

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.