基于深度学习的电力大数据融合与异常检测方法

摘要

In order to mine existing safety threats in power grid using heterogeneous data sources in power big data,this paper maps heterogeneous data to a unified embedded vector space with deep restricted Boltzmann machine,and achieves the fusion of heterogeneous data sources.The anomaly detection is achieved by drawing a profile for embedded vector dataset using recurrent neural networks.The experiments indicate that the proposed anomaly detection approach has the highest value in our proposed mutual information index,and it is obviously better than other anomaly detection algorithms in accuracy,false positive rate and false negative rate.%为了充分利用电力大数据中的异构数据源挖掘出电网中存在的安全威胁,采用深度受限玻尔兹曼机将不同格式的异构数据映射到统一的嵌入式向量空间,实现了异构数据的融合.采用循环神经网络对得到的嵌入式向量数据建立画像,实现了数据中异常事件的检测.实验结果表明,提出的异常检测方法在提出的互信息量度量指标中具有很高的互信息量.此外提出的方法在准确率、误报率和漏报率中的结果也优于其他异常检测方法.