基于双线性Koopman算子的非线性多智能体系统的分布式自触发一致性控制OA

This paper proposes a distributed self-triggered predictive control method based on bilinear Koopman oper-ators for the consensus control problem of a class of model-unknown discrete-time nonlinear multi-agent systems.Address-ing the challenges of modeling complexity and limited control performance in traditional model-based approaches for com-plex nonlinear systems,this paper constructs a fully data-driven bilinear Koopman modeling framework.By designing a deep neural network architecture comprising a lifting network,a bilinear layer,and a reconstruction network,it achieves a fi-nite-dimensional approximation of the bilinear dynamics of the unknown nonlinear system in the lifting space.To overcome the limitations in model accuracy and generalization caused by hyperparameter dependence on manual expertise in deep Koopman models,Bayesian optimization is further introduced for adaptive optimization of the network architecture and training hyperparameters.This significantly enhances predictive performance with reduced training costs,establishing a Bayesian-optimized deep bilinear Koopman model.Building upon this foundation,considering the practical constraints of computational and communication resources in multi-agent systems,a distributed model predictive control strategy based on a self-triggering mechanism is proposed.Each agent solves the optimization problem and predicts the next trigger time only at the trigger moment based on local neighbor information,effectively reducing communication frequency and compu-tational burden.Furthermore,the input-to-state stability of the consensus error system under the designed self-triggering strategy is proven,ensuring the eventual boundedness of the consensus error.Finally,a simulation experiment validates the effectiveness of the proposed method in the consensus control of nonlinear multi-agent systems,demonstrating lower trigger frequencies compared to existing control methods based on linear Koopman models.