Secure State Estimation for Train-to-Train Communication Systems: A Neural Network-Aided Robust EKF Approach

This article studies the problem of secure state estimation for train-to-train communication systems subject to deception attacks. Different from the existing secure state estimation methods for linear systems suffering from the occasionally occurring deception attacks, a neural network (NN)-aided robust extended Kalman filter (EKF) method is proposed for nonlinear discrete train systems, where the proposed method introduces a residual saturation function to render the EKF robust against deception attacks of different magnitudes and durations, and a sufficient condition is derived to guarantee the boundedness of the estimation errors. When attacks occur, the distorted residual of the EKF is able to be limited to a finite range due to the saturation constraint, which alleviates the negative effects induced by the deception attacks on the estimation performance. To further improve the estimation accuracy, the NN is used to cascade with the EKF for predicting and correcting the estimation errors of the filter online. Experiments on a semiphysical platform validate the effectiveness and advantage of the proposed method in different scenarios.