In order to prevent and control power stealing behavior effectively,an abnormal detection and analysis method of power data based on(SE-CNN)power stealing behavior model is designed on the basis of the construction of power consump-tion trend decline index.Based on the CNN model,SENet is introduced to adjust the importance of the feature channels,so as to effectively improve the utilization rate of the channels.The experiem ental results show that the AUC value of SE-CNN model reaches 0.999,the detection efficiency is high,and the SE-CNN model can be applied to complex power grid environ-ment.Compared with support vector machine(SVM),XGBoost and CNN,SE-CNN model has good evaluation indexes,effec-tively reduces the experimental influence brought by invalid features,realizes feature fusion in local areas,and makes experi-mental data quickly reach the fitting state.
关键词
窃电行为/改进卷积神经网络/异常检测/评价指标
Key words
power stealing beheavior/improved convolutional neural network/abnormal detection/evaluation index
维普
万方数据