Data-driven Bipartite Formation Control For Nonlinear Multi-agent Systems

Distributed formation control of multi-agent systems,a fundamental coordination issue,is one of the key research directions in artificial intelligence.Due to their flexibility and scal-ability,multi-agent systems have been widely used in satellite communications,warehousing and logistics,and intelligent reconnaissance surveys.With the development of science and technology,controlled systems have become increasingly complex.It is a huge challenge to employ traditional system identification methods to obtain the dynamics models of the con-trolled systems.Hence,researching how to design data-driven control algorithms for nonlinear multi-agent systems with unknown dynamics models to realize bipartite formation control is of practical significance.This thesis designs the corresponding control strategies for model-free multi-agent sys-tems from simple to complex practical scenarios,mainly using model-free adaptive control theory combined with the distributed control method,iterative learning control method,neural network prediction method,and reinforcement learning method.Moreover,the corresponding data-driven bipartite formation control algorithms are proposed,the related algorithms’ conver-gence proof is given,and a systematic theoretical framework for data-driven bipartite formation control is established.The main research contents of this thesis are summarized as follows:1.For nonlinear multi-agent systems with unknown dynamic models running aperiod-ically,a dynamic linearization data model is established along the time axis only using the input and output data.A distributed model-free adaptive bipartite formation control framework combined with the distributed control method is proposed.A saturated data-driven bipartite for-mation control algorithm and a data-driven bipartite formation control algorithm for preventing data packet loss are proposed based on the proposed framework to solve the problem of output data saturation and data dropout of controlled plants.2.For nonlinear multi-agent systems with unknown dynamic models running periodically,the concept of iterative learning is introduced,a dynamic linearization data model is established along the iterative axis according to their input and output data,and a distributed model-free adaptive iterative learning bipartite formation control framework is proposed.Then,to effec-tively solve the control system’s signal quantization problems and channel attenuation problems,a quantization data-driven bipartite formation control algorithm and a data-driven bipartite for-mation control algorithm against the fading of channels are proposed.3.For large-scale,networked nonlinear multi-agent systems with unknown dynamic mod-els,an enhanced dynamic linearization data model is established using the input and output data and the communication topology information.Then,combining the established data model with the neural network prediction algorithm,a fully distributed model-free adaptive neural network bipartite formation control algorithm framework is proposed.In addition,to effectively solve the corresponding cyberattack problems,the corresponding defense mechanisms are designed for the main cyberattack issues in the controlled plants,and the related data-driven bipartite formation control algorithms are proposed.4.For highly complex and networked nonlinear multi-agent systems with unknown dy-namic models,the concept of an ideal controller is introduced.An enhanced dynamic lineariza-tion ideal controller model is established according to the input and output data and the com-munication topology information.Then,an online learning fully distributed model-free adap-tive reinforcement learning bipartite formation control framework is proposed by combining the established data model with the reinforcement learning control algorithm.Moreover,to effectively solve the input and output constraints of the controlled systems,an adaptive event-triggered model-free reinforcement learning data-driven bipartite formation control algorithm and model-free reinforcement learning data-driven bipartite formation control algorithm under the consideration of input constraints are proposed.