Prediction method of electricity stealing behavior based on multi-dimensional features and BP neural network

Abstract

When using current methods to predict electricity theft behavior, there are problems of large sample data errors before training, high line loss rate, and low prediction effect. To this end, a power-stealing behavior prediction method based on multi-dimensional features and BP neural network is proposed. The method first extracts the abnormal power usage characteristics according to the abnormal features of the user’s power usage behavior. Then the genetic algorithm is used to optimize the BP network, and the optimized network is used to train the abnormal power consumption characteristics. Finally, a prediction model is established based on the training results to achieve the final prediction. The experimental results show that the training error test, line loss rate test and prediction effect test of the method before and after training verify the practicability and effectiveness of the method.

Keywords

• Multi-dimensional features and BP neural network
• Electricity stealing behavior prediction
• Feature extraction
• Training sample data