Three Kinds Of Optimal Consensus Control Algorithms For Nonlinear Multi-Agent Systems

Because of the high adaptability,good robustness and low maintenance cost,the multi-agent systems has become a hot topic in the scientific research,and it is also common in real life.The research on the consensus of multi-agent systems has been unable to meet the needs of the life.In order to save cost,reduce loss and improve the utilization rate of resources,the research on the optimal consistent control of multi-agent systems is extremely urgent.In this thesis three kinds of multi-agent systems optimal consensus problem have been studied.Firstly,the limited time horizon nonlinear multi-agent systems with non-quadratic performance index problem is considered.A distributed DISOPE algorithm has been designed,which mainly uses the model optimization and parameter estimation methods.And a linear system with the quadratic performance index has been employed to estimate the nonlinear system with the non-quadratic performance index,coupled with the maximum principle in the classical optimal control theory the optimal solution necessary conditions can be obtained.After several iterative times,the distributed DISOPE algorithm can get the convergence solution as the optimal solution and the follower nodes can almost track the leader node,the tracking errors converge to zero and the optimal performance index can be obtained in the finite time horizon.Secondly,the nonlinear multi-agent system with completely unknown dynamic is considered.The ADP method and PI algorithm has been employed,two neural networks are used to estimate the unknown dynamic and the performance index function respectively,then the iteration control policy can be calculated.Compared with the processing methods for such problems in existing literatures,the requirements for data sampling in the estimation of value function has been relaxed.Only the system state and control input should be sampled for calculated,the derivative of system state can be omitted,which enhance the realizability of the algorithm.Finally,a discrete time nonlinear multi-agent systems optimal control problem is considered.The local VI algorithm based on ADP is employed,which is more relaxed than PI algorithm in initial conditions.Moreover,the local VI algorithm only needs part of the data processing within the state space,reduces the calculation burden of the neural network,and speeds up the learning and operation.In this thesis the key points of the three problem contain the non-quadratic performance index,the system with unknown dynamic,and reduction of computation burden.Which are the hot topics in the current research,and also the common problems in real life.The three algorithms are all have good performance in the simulation examples,which not only make the followers track the leader,but also get the optimal control policy and optima performance index.