Loading...

Fixed-Time Consensus Control of Unknown Multi-Agent Systems Based on Distributed Reinforcement Learning

Delshad, Aria | 2024

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57336 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Babazadeh, Maryam
  7. Abstract:
  8. This thesis introduces a distributed backstepping control strategy based on reinforcement learning to achieve fixed-time consensus in high-order nonlinear multi-agent systems. Each agent, whether a leader or a follower, is modeled using a strict feedback approach, incorporating uncertainties as unknown nonlinear functions and disturbance signals across all state variables. The communication network among agents can be either directed or undirected, with a fixed topology. Each follower employs a control law derived from actor-critic networks, utilizing only the outputs from neighboring agents. The proposed method combines virtual and actual backstepping control laws, formulated from two key components: the optimal policy reinforcement learning based on the Hamilton-Jacobi-Bellman equation, and fixed-time stability analysis. To streamline the adaptation of neural network weights in reinforcement learning, the gradient descent method is applied, ensuring fixed-time consensus for the multi-agent system. An approximating neural network and a disturbance estimator with fixed-time estimation capabilities are used to estimate unknown functions and disturbances. The proposed mechanism ensures that all followers successfully track the leader, with the overall consensus error confined to a specified and adjustable neighborhood of the origin within a fixed time regardless of initial conditions. By carefully selecting control tuning parameters, the consensus error can be minimized to any desired level within a fixed time frame. The effectiveness of the proposed method is validated through various simulations, which confirm all theoretical findings
  9. Keywords:
  10. Multiagent System ; Reinforcement Learning ; Backstepping Algorithm ; Fixed-Time Stability ; Uncertainty ; External Disturbance ; Optimal Backstepping Control

 Digital Object List

 Bookmark

No TOC