A hybrid neural network for abnormal data identification in electric systems
Abnormal losses are one of the main causes of revenue loss for grid companies,but the extremely large data size poses a challenge for abnormal data identification.In this paper,a new detection method is proposed that uses hybrid deep neural networks to self-learn features to identify anomalies in system losses.The method requires only minimal input data and knowledge scope and does not require manually building feature library.The method consists of a long and short-term memory network and a multilayer perceptron network.The former network analyzes the raw daily energy loss history,and the latter integrates non-temporal data,such as contracted electricity or geographic information.The model is trained and tested with grid data.The results show that the proposed hybrid neural network performs significantly better than other anomaly data identification methods,validating its effectiveness.
万方数据
维普
