عليرضا موسوي

In this dissertation, to achieve the leader-follower consensus of multi-agent systems based on the minimization of the local cost functions, a distributed Adaptive Fuzzy Sliding-Mode Control (AFSMC) method is developed based on an off-policy reinforcement learning algorithm and the graphical game theory. Since the dynamics of the multi-agent systems are unknown, the design of optimal control laws for the leader-follower consensus problem is a challenging issue. Moreover, in this study, the dynamic equations of the agents are assumed to be nonlinear and the input gain matrices are nonsquare. In the proposed method, the performance index of each agent is an explicit function of the local sliding vector and its corresponding coefficients as the optimization policies. Contrary to the conventional off-policy reinforcement learning algorithms for the consensus control of multi-agent systems, the proposed method does not require partial knowledge of the system dynamics to initialize the reinforcement learning process. Furthermore, the actor-critic fuzzy methodology is employed to approximate optimal policies using the measured input/output data. Therefore, by using the tuned sliding vector, the control input for each agent is generated which includes a fuzzy term and a robust term. In the control input designed for each agent, the fuzzy term is employed to approximate an unknown nonlinear function including the effects of system dynamics. Moreover, the robust term is considered to compensate for any possible mismatches. Therefore, using the local sliding vectors, the output singletons of the fuzzy systems and the upper bounds of the approximation errors are computed by the adaptation laws. The closed-loop asymptotic stability of the sliding variables in the presence of disturbances and uncertainties is proved using the second Lyapunov theorem and the Barbalat’s lemma. In continuation, a barrier function-based method is proposed for the consensus problem in the presence of state constraints. The effectiveness of the proposed method in the leader-follower consensus control problem of a group of REMUS Autonomous Underwater Vehicles (AUVs) is eva‎luated through numerical simulations. According to the simulation results, the proposed control structure shows acceptable performance compared to the AFSMC with ideal sliding parameters which are obtained by a model-based optimization procedure. Moreover, by using the barrier function-based AFSMC, the control objective as well as the state constraint satisfaction are achieved.

اجماع پيشرو-پيرو , كنترل لغزشي-فازي-تطبيقي , يادگيري تقويتي خارج از سياست , نظريه بازي گرافي , سيستم فازي بازيگر-منتقد

leader-follower consensus , adaptive fuzzy sliding-mode control , off-policy reinforcement learning , graphical game theory , actor-critic fuzzy system

Alireza Mousavi

Amir Hossein Davaie Markazi