LSTM-based Fault Prediction for Ship Power Systems

Abstract

A ship power system refers to an isolated grid consisting of power generation, electrical energy conversion, transmission, distribution, and consumption. As the heart of the ship’s route, the ship’s power station should try to avoid malfunction, and the malfunction should be quickly investigated and recovered. In this paper, a stand-alone system of a ship power station is built for simulation using the Simulink software platform, and the fault voltage and current parameters of each phase are selected as the source of the data set and input to the Long Short Term Memory (LSTM) neural network algorithm for training and diagnostic prediction of faults. By comparing the diagnostic results with those of Back Propagation (BP) neural network, the results show that LSTM has an accuracy of 98.334%, which can diagnose the failure modes of ship power stations more accurately and has higher explanatory power for the predicted data.