基于样本卷积交互学习的窃电检测样本增强方法

Sample Enhancement Method for Electricity Theft Detection Based on Sample Convolution Interactive Learning

为了检测窃电,有多项研究和应用,但是现有的窃电检测研究未能有效地解决样本分布不平衡的问题。为了解决这一问题,提出了一种基于样本卷积交互学习的窃电检测样本增强方法。首先将数据预处理后的训练集通过样本卷积交互学习方法增加训练集中的少数样本;然后将样本增强后的训练集输入到一个3层卷积神经网络模型中进行特征提取,最后使用一个2层的全连接层输出检测结果,并使用常用的评估指标以验证样本增强机制的有效性。在中国国家电网公司(SGCC)数据集上的仿真实验结果表明,提出的TSCINet-CNN模型在70%的训练集中曲线下面积、F1-score和MAP上分别取得了0.8822、0.5445和0.5560的相对优秀结果。

In order to detect electricity theft,there are many studies and applications,but the existing research on electricity theft detection fails to effectively solve the problem of unbalanced sample distribution.In order to solve this problem,a sample enhancement method for electricity theft detection based on sample convolution interactive learning is proposed.Firstly,the training set after data preprocessing is increased by the sample convolution interactive learning method.Then,the sample enhanced training set is input into a three-layer convolutional neural network model for feature extraction.Finally,a two-layer fully connected layer is used to output the detection results,and common evaluation indicators are used to verify the effectiveness of the sample enhancement mechanism.Simulation results on the State Grid Corporation of China(SGCC)dataset show that the proposed TSCINet-CNN model achieves excellent results of 0.8822,0.5445 and 0.5560 on the area under the curve,F1-score and MAP of 70%of the training set,respectively.