Extraction of Abnormal Points from On-line Operation Data of Intelligent Meter Based on LSTM

Aiming at the difficulty of massive data processing, the inefficiency of traditional methods and the low real-time performance, this paper proposes a method of extracting abnormal points of on-line operation data of smart meters based on LSTM. Firstly, the current, voltage and power curves of actual faulty ammeters before faults are analyzed concretely, and the sequence variance, NULL value ratio and abnormal point ratio are used as attributes to measure the abnormal degree of curves. Then, based on long short-term memory model, the abnormal points are filtered by processing sequence data, and the historical operation curve of intelligent ammeters is used as training set to operate in target time. Row curve is simulated to improve simulation accuracy by adjusting parameters such as sequence learning length. The test results verify the accuracy and validity of the model and method in this paper.