کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

Reinforcement Learning-Based Optimal Sliding Mode Control for Nonlinear Systems with State Constraints

نویسندگان :

Ali Sheykholeslami¹ Mohsen Montazeri² Alireza Yazdizadeh³

1- دانشگاه شهید بهشتی 2- دانشگاه شهید بهشتی 3- دانشگاه شهید بهشتی

کلمات کلیدی :

nonlinear systems،optimal control،sliding mode control،reinforcement learning،state constraints،actor-critic،adaptive control

چکیده :

This paper addresses the problem of robust optimal control for a class of uncertain nonlinear systems subject to state constraints. Toward this, a novel composite control framework is proposed by integrating Sliding Mode Control (SMC) and Actor-Critic Reinforcement Learning (RL). In the proposed scheme, the SMC-based component is designed as a robust compensator to actively reject system uncertainties and external disturbances. Concurrently, an actor-critic RL agent is developed to learn the optimal control policy for the nominal system dynamics. The key innovation is the explicit handling of state constraints, which are incorporated into the reinforcement learning problem through a barrier function in the cost function. This ensures that the learned policy inherently avoids constraint violation without manual intervention. Simulation results on a benchmark nonlinear system verify the effectiveness and performance of the proposed controller, demonstrating smooth asymptotic convergence of the system states to the origin while strictly satisfying the predefined state constraints.