Bi-LSTM-Based Resilient Data-Driven Integral Sliding Mode Control for UMVs Under Hybrid Attacks

In this paper, based on bidirectional long short-term memory (Bi-LSTM) neural network, a resilient partial form dynamic linearization data-driven integral sliding mode control (PFDL-DDISMC) method is designed to realize the trajectory tracking for unmanned marine vehicles (UMVs) under deny-of-service (DoS) attacks and false data injection (FDI) attacks in the measurement channel. First, the generalization error bound of the Bi-LSTM, which is related to the number of samples and the bounds of the weight matrices, is proved through Rademacher complexity analysis. A Bi-LSTM-based intelligent predictor is designed to predict the position and velocity of UMVs and to estimate external disturbances. Based on this, a threshold-based attack detection and compensation scheme is proposed to determine whether the UMVs are suffering from attacks and to compensate for the compromised transmission signals. Furthermore, considering the bounds of the generalization errors of predictions, a resilient secure controller is designed using the compensation data from the proposed compensation mechanism and a data-driven discrete-time integral sliding mode control method. Compared with the data-driven control scheme, the UMVs with proposed PFDL-DDISMC can exhibit more effective trajectory tracking performance under disturbances and hybrid attacks. Theoretical analysis verifies that the proposed PFDL-DDISMC method ensures the stability of the tracking error system for UMVs subjected to hybrid attacks and external disturbances. Finally, three simulation examples are provided to demonstrate the effectiveness of the proposed resilient PFDL-DDISMC scheme and the convergence of the tracking error.