GA-LQR for vehicle semi-active suspension with BiLSTM inverse model of magnetic rheological damper

Abstract

This paper proposes a magnetic rheological (MR) semi-active control method based on bidirectional long short-term memory (BiLSTM) neural network, linear quadratic regulator (LQR) control algorithm, and genetic algorithm (GA). The LQR algorithm with GA optimizing the weight coefficients generates the expected damping force. Due to the nonlinear hysteresis characteristics of the magnetic rheological damper (MRD) and the fact that its input and output have certain time dependence, an inverse model of MRD is established by BiLSTM. The control current is predicted by BiLSTM and then the current is input to the MRD to obtain the damping force that is infinitely close to the expected damping force. The damping force is then applied to the suspension system to form a complete closed-loop feedback control, which realizes the damping effect and generates a real-time control. The simulation results show that the MRD inverse model can accurately predict the required control current, and the GA-optimized LQR control algorithm has a good suppression effect on the vertical vehicle acceleration, dynamic tire load, and suspension dynamic stroke.