Adaptive Fault-tolerant Federated Filter with Fault Detection Method Based on Combination of LSTM and Chi-square Test

Abstract

In order to solve the problem that the global estimation accuracy is affected by the gradual changing fault of federated filter subsystem, the features of gradual changing fault and the advantages of a long short-term memory (LSTM) neural network classification algorithm are analyzed. On this basis, a fault detection method combining residual Chi square detection algorithm with long short-term memory neural network detection method is proposed, which can effectively detect the gradual changing fault and abrupt faults of sub filters and reduce the impact of faults on global estimation accuracy. The simulation results show that this fault detection method is better than the traditional mathematical model diagnosis methods and the convolutional neural network (CNN) detection methods when the subsystem gradual changing fault occurs.