Parameter identification of LPDS linear induction motors using a hybrid reinforcement learning and evolutionary strategy approach

Abstract

Abstract Accurate parameter identification of long-primary double-sided linear induction motors (LPDS-LIMs) plays a crucial role in improving control performance, thrust estimation, and overall operational efficiency. In this study, a hybrid framework based on Reinforcement Learning (RL) and Evolution Strategies (ES) is proposed for the precise estimation of key motor parameters, including stator resistance, leakage inductance, and the end-effect coefficient. The proposed approach leverages the fast convergence capability of RL together with the global search ability of ES in order to enhance estimation accuracy and improve robustness against disturbances. Simulation results demonstrate that the proposed method outperforms the standalone RL and ES approaches. Specifically, the proposed hybrid method reduces the root mean square error (RMSE) to approximately $$3.8\times {10}^{-4}$$ , achieving about 4.5 times better accuracy than RL and nearly 3 times better performance than ES. Furthermore, the proposed algorithm converges within approximately 50 episodes, indicating a significant improvement in convergence speed compared with the benchmark methods. To further validate the effectiveness of the identified parameters, a high-fidelity electromagnetic model of the LPDS-LIM was developed using the Finite Element Method (FEM). Detailed electromagnetic analyses, including magnetic flux distribution, flux density, and thrust force characteristics, were conducted. The comparison between the parameters obtained through the proposed method and the FEM-based simulation results shows strong agreement, confirming the accuracy and physical consistency of the identified parameters. Additionally, robustness analysis under noise conditions up to a level of 0.05 demonstrates the stability and reliability of the proposed approach.