Identification of Inrush Current and Fault Current Based on Long Short-Term Memory Neural Network

Abstract

The problem of differential protection maloperation caused by inrush current during no-load closing of transformer has not been completely solved so far. To solve this problem, a method using long short-term memory (LSTM) neural network to identify inrush current and fault current is proposed. First, the simulation model of no-load closing and internal fault of transformer is built on the PSCAD software platform, and a large amount of three-phase current instantaneous sampling data is generated through simulation as the sample set to train the neural network. Then, the LSTM neural network model is built and trained by using the Keras platform. Finally, the new simulation data and fault recorder data is used to test the trained LSTM neural network. The results show that the LSTM neural network can quickly and accurately distinguish the inrush current and fault current under various conditions, which proves the effectiveness of the proposed method.

Keywords

• transformer differential protection
• long short-term memory (lstm) neural network
• inrush current identification
• fault current identification