Anomaly Detection Based on Temporal Convolutional Networks in Smart Grid

Abstract Abnormal power consumption detection is an important means to detect power emergencies in time. It can effectively reduce power loss and residential losses due to electrical failures. This paper presents an anomaly detection method based on temporal convolutional networks and statistics. The predicted power consumption results of subsequent periods are compared with the actual results, and statistical methods are used to determine whether abnormal power consumption exists. If an abnormal result occurs, the system will issue a warning, search the fault location in the database established by the user, and determine the initial fault type. Simulation results show that this method can effectively detect abnormal power consumption.