Parameter‐Adaptive Modulated Model Predictive Control of Induction Motors via a TD3‐Optimized Sliding‐Mode Observer

ABSTRACT This paper proposes a parameter‐adaptive modulated model predictive control (M 2 PC) scheme for induction motor (IM) drives, in which a twin delayed deep deterministic policy gradient (TD3)‐optimized sliding‐mode observer (SMO) is used instead of trial‐and‐error gain tuning. An SMO built on the full sixth‐order IM model is designed to jointly estimate the ‐axis stator currents, ‐axis rotor fluxes, and the key electrical parameters; the magnetizing inductance and rotor resistance in real time. The estimated states and parameters are fed back into the M 2 PC framework, yielding a parameter‐adaptive predictive controller. The Lyapunov adaptation gains and , which strongly affect convergence speed and robustness, are tuned offline by a TD3 agent trained on a detailed IM–SMO–M 2 PC simulation environment, rather than being selected heuristically. The optimized gains are then implemented on a low‐cost TMS320F28379D digital signal controller, where the complete TD3–SMO–based M 2 PC algorithm runs in approximately 51 μs, providing sufficient computational margin for the selected 100 μs sampling interval. Experimental results demonstrate improved speed and torque tracking, enhanced current and flux estimation, and robust performance under significant parameter variations compared with fixed‐gain SMO and nonadaptive M 2 PC schemes.