基于深度森林算法的异常用电行为检测方法

Detection method of abnormal power consumption behavior based on deep forest algorithm

面对当前使用的逻辑回归、模糊聚类检测方法在异常用电行为数据持续增加情况下,出现的检测精准度低的问题,提出了基于深度森林算法的异常用电行为检测方法。构建基于深度森林的异常用电特征采样模型,预处理原始样本数据,消除遗漏数据。采用插值方法修复缺失值,并提取完整异常用电行为特征量。训练样本,减少拟合风险,使用深度森林算法,确定用电行为异常指数,并分类异常用电行为。由实验结果可知,该方法检测到的分时电量数据均与实际数据一致,当误检率为0.2时,检测率为0.05,具有精准检测结果。

Facing the problem of low detection accuracy of the current logical regression and fuzzy clustering detection methods when the abnormal power consumption behavior data continues to increase,an abnormal power consumption behavior detection method based on deep forest algorithm is proposed.The abnormal power consumption feature sampling model based on deep forest is constructed to preprocess the original sample data and eliminate the missing data.The missing value is repaired by interpolation method,and the complete characteristic quantity of abnormal power consumption behavior is extracted.Training samples,reducing fitting risk,using deep forest algorithm to determine abnormal power consumption behavior index and classify abnormal power consumption behavior.The experimental results show that the time-sharing electricity data detected by this method are consistent with the actual data.When the false detection rate is 0.2,the detection rate is 0.05,which has accurate detection results.