Nonlinear multi-agent consistency method based on state observation and experience pool

Abstract

The invention relates to a nonlinear multi-agent consistency method based on state observation and an experience pool, and belongs to the field of computers. Firstly, under the scene that a follower cannot acquire the state of a leader, a backstepping method and a dynamic surface control method are used for designing a full-dimensional observer for each agent to observe the state of the leader. Afterwards, the consistency problem is converted into the optimal tracking problem of each follower to the respective leader state observer. In the next step, a non-quadratic cost function is defined for processing aiming at asymmetric saturation constraint on control input, then an optimal control problem under a new augmented system is defined, a reinforcement learning strategy algorithm is used for iterative solution, and the stability and optimality of a solution under strategy iteration are analyzed. For the difficulty that the optimal control HJB equation is difficult to directly solve, the method utilizes the good approximation property of the neural network, and an act-critic framework is used for solving.

Description

technical field [0001] The invention belongs to the field of computers, and relates to a non-linear multi-agent consistency method based on state observation and experience pool. Background technique [0002] At present, there are more and more application scenarios of multi-agent systems, and they play an increasingly important role, such as multi-robot formation, traffic control, smart grid, and multi-robot SLAM. The primary goal of achieving multi-agent coordinated control is the multi-agent consistency problem. In addition, it is usually accompanied by reaching certain conditions, such as minimizing the energy used in the process of reaching the consensus, and the time of the consensus process being as short as possible. For single-agent systems, many optimal control methods have been proposed. In the optimal cooperative control problem in the multi-agent scenario, the cooperative optimal control is generally realized by solving the CHJB equation. However, due to the n...

AI Technical Summary

Problems solved by technology

However, in the iterative process of neural network training, the stability and convergence of the training process usually requires the continuous excitation condition to be satisfied during the training process, but this requirement is difficult to meet in practice.

Images