Transformer-enhanced reinforcement learning for magnetorheological semi-active suspension control under false data injection attacks

This paper investigates the control of a semi-active suspension equipped with a magnetorheological (MR) damper under nonlinear hysteresis, hard physical constraints, and unknown false data injection (FDI) attacks. A nonlinear-hysteretic dynamic model that accounts for FDI attacks is established for a semi-active suspension with an MR damper. A Transformer-enhanced Deep Deterministic Policy Gradient (DDPG-Trans) controller is proposed, in which a Transformer encoder is embedded into the Actor network to enhance feature representation and suppress the influence of FDI-corrupted sensor signals. The proposed method is applied to MR semi-active suspension control to improve ride comfort, ensure safe operation, and enhance robustness against cyberattacks. Simulation results under both attack-free and attacked conditions verify its effectiveness.