Networked multi-agent system distributed optimization control method based on reinforcement learning

Abstract

The invention discloses a networked multi-agent system distributed optimization control method based on reinforcement learning, which belongs to the technical field of system control methods. A Bellman equation with an integrator is provided for integral reinforcement learning (IRL). According to a non-strategy (Off-poly) method, different control strategy action control systems are applied, dataexploration is added, an optimal control strategy is obtained through iterative learning, a graph game is fused with Off-poly and IRL, and the consistency of networked multi-agents is ensured under the condition that a system model is completely unknown. The system comprises the following specific steps: (1) solving an optimization control problem based on IRL and Off-poly; (2) iteratively designing a distributed approximate optimal control strategy based on an approximate strategy; and (3) adopting simulation software and a multi-manipulator control system physical experiment platform, and correspondingly adjusting a theoretical method and a control technology according to simulation and physical experiment results.

Description

technical field [0001] The invention belongs to the technical field of system control methods, and in particular relates to a distributed optimization control method for a networked multi-agent system based on reinforcement learning, combining Graphical games with Off-policy and Integral Reinforcement Learning (IRL ) fusion, given the system model is completely unknown, to ensure the consistency of the networked multi-agent and optimal performance control optimization problem solution. Background technique [0002] In the field of practical application, due to large-scale and complex processing technology and procedures, the system model is complex, the system identification accuracy is low and the cost is high, and the system cannot even be identified. When the system model is completely unknown, the existing fully model-dependent and partially model-dependent optimization control methods cannot be realized in practical applications. At present, the research on networked m...

AI Technical Summary

Problems solved by technology

[0002] In the field of practical application, due to large-scale and complex processing technology and procedures, the system model is complex, the system identification accuracy is low and the cost is high, and the system cannot even be identified

When the system model is completely unknown, the existing fully model-dependent and partially model-dependent optimization control methods cannot be realized in practical applications.

At present, the research on networked multi-agent distributed optimal control based on reinforcement learning is still in its infancy, and there is no complete theoretical system to support the development of related technologies. It is urgent to propose a networked multi-agent system distribution based on reinforcement learning. optimal control method

Images