Electricity Theft Identification Algorithm Based on Auto-Encoder Neural Network and Random Forest

Abstract

Non-technical losses caused by users’ electricity theft will affect the economic order for the electricity market and have a negative impact on line loss calculations and grid operations. In order to improve the power detection capability and identify the user’s power stealing method, an electricity theft identification algorithm combining auto-encoder neural network and random forest is proposed. The auto-encoder neural network can effectively extract the abstract behavior characteristics of the electricity data, and establish an early warning model of the electricity stealing behavior by the error between the actual power data and the neural network full variable reconstruction value. The random forest algorithm further derives the specific implementation mode of the warning electricity theft behavior. Based on the real data set, the proposed algorithm is verified that the proposed algorithm has a high detection sensitivity and classification accuracy.