基于长短时记忆网络的电力负荷异常检测

Power load abnormal detection based on Long Short-Term Memory

电力系统或电网可能因为一系列不确定原因产生异常,而异常数据往往蕴含了重要事件信息,挖掘异常数据意义重大.电力负荷异常是电力系统的主要异常,需要快速检测和处理.为此提出了基于长短期记忆网络的电力负荷异常检测的方法,利用纵向双层长短时记忆网络,底层负责负荷序列向量表示,顶层负责时间序列重建.同其它异常检测方法的实验对比显示,新方法又有一定的应用价值,可以为电力系统的维护检修提供了便利.

Power system or power grid may produce anomalies due to a series of uncertain reasons.Abnormal data of power system often contain important event information,so mining abnormal data is of great significance.Power load anomaly is the main abnormality of power system,which needs rapid detection and processing.For this reason,a method of power load anomaly detection based on Long Short-Term Memory(LSTM)is proposed.The bottom layer is responsible for the vector representation of load series and the top layer is responsible for time series reconstruction.Compared with other anomaly detection methods,the new method has certain application value,which can provide convenience for the maintenance and repair of power system.