基于堆叠降噪自编码器的配电数据清洗方法

Power distribution data cleaning method based on stack denoising autoencoders

准确高效的异常数据识别与缺失数据恢复是电力网络稳定运行的基础。提出了一种配网网络状态监测异常数据清洗方法。首先,利用堆叠降噪自编码器(SDAE)学习正常数据和异常数据特征,去除噪声后获取损失函数曲线。然后,采用Bootstrap方法估计置信区间,设置异常数据识别门限,通过多分类支持向量机完成异常类型识别。最后,针对缺失数据,设计了Pearson相关系数进行插补恢复。实验结果表明,该方法能够有效识别配电网络异常数据类型,且缺失数据恢复性能优于现有方法。

Accurate and efficient abnormal data identification and missing data recovery are the basis factors for stable operation of power network.Therefore a method of cleaning abnormal data in distribution network condition monitoring is proposed.Firstly,Stacked denoising reduction auto encoder(SDAE)is used to learn normal data and abnormal data features,so as to obtain the loss function curve after removing noise.Then,bootstrap method is adopted to estimate the confidence interval and set the threshold of abnormal data recognition,which then allows Multi-class support vector machine identify the abnormal types.Finally,missing data is recovered by Pearson correlation coefficient.Overall,the experimental results show that the proposed method can effectively identify the abnormal data types of distribution network,and the recovery performance of missing data is better than the existing methods.