基于深度学习变分自动编码器算法的电价执行稽查研究

Research on Electricity Price Inspection Implementation Based on Deep Learning Variational Autoencoder Algorithm

提出一种利用深度学习中的变分自动编码器技术进行电价执行稽查的方法,利用变分自动编码器异常检测算法模型既可以作为判别模型又可以作为生成模型的特点,有效解决了目前电价执行稽查多种异构参数的计算问题,通过重构概率来有效判定用电客户是否异常。同时,由于算法判别过程中生成模型可对数据进行同特征恢复,有效解决样本数据不完整问题。通过试验测试证明了算法有效性,解决了不同电价执行稽查工作中数据异质,提高了稽查准确性,为电力营销工作提供了有效的保障。

A method of using deep learning variational autoencoder technology for electricity price inspection implementation is put forward.The anomaly detection algorithm model of the variational autoencoder model can be used both as a discriminant model and as a feature of the generation model.It can effectively solve the calculation problems of multiple heterogeneous parameters of current electricity price execution audit and it effectively determines whether the electricity customer is abnormal by reconstructing the probability.At the same time,due to the generation of the model in the algorithm discriminating process,the same feature recovery can be performed on the data.It effectively solves the problem of incomplete sample data.The experimental results prove the effectiveness of the algorithm.It solves the heterogeneity of data in the implementation of different electricity price auditing work,improves the accuracy of auditing and provides effective protection for power marketing.