Dual-LSTM Based State Estimation for Backstepping Control of EPS System

초록 (Abstract)

This paper proposes a dual long short-term memory (LSTM) based state estimation method for backstepping control of electric power steering (EPS) system using a second-order C-EPS dynamics. The proposed framework employs two LSTM networks in parallel, one LSTM estimates steering states, while the other estimates the lumped disturbance representing modeling errors, nonlinear effects, and unmodeled dynamics. By combining state estimation and disturbance estimation within a unified learning-based structure, the proposed method is designed to provide more effective information for backstepping control than conventional observer-based approaches. Simulation results show that the proposed dual-LSTM framework provides effective state and disturbance information for EPS torque control and achieves improved steering tracking performance compared with the extended state observer-based method. These results demonstrate the potential of learning-based parallel state and disturbance estimation for advanced EPS torque control.

사사 과제