Event-triggered Control Based On Particle Swarm Optimized Adaptive Dynamic Programming

Optimal control,as one of the branch of control theories,plays an important role in modern industrial production including people's daily life.With the development of computer hardware technology,plenty of adaptive learning algorithms have attracted much attention and are applied to solve the optimal control problems.One of these algorithms is named as adaptive dynamic programming(ADP),overcoming “the curse of dimension” from the perspective of reinforcement learning.It is an effective method of solving the optimal control problems of nonlinear systems.With the demands of the production efficiency raising,the structures of actual systems become more and more complex,and the scale of the systems are also gradually increasing.The researches on single controller systems can not meet the actual needs.As a result,some novel,resource saving and efficient theories and algorithms need to be studied urgently.Recently,the differential games based on event-triggered control have attracted much attention,which shows its potential in the research of efficient control on complex systems.In order to further save the computation and the communication resources and improve the efficiency of the learning algorithms,this thesis study the event-triggered control on differential games of nonlinear systems.In addition,the swarm intelligence algorithm is utilized to improve the success rate of the ADP method.This thesis contains the following two aspects:1.To deal with the zero-sum games of nonlinear systems,this thesis studies the ADP based event-triggered control method.First,an infinite-horizon-based value function is introduced and the optimal control strategy of both sides is formulated,and then,the zero-sum games problem is transformed into the optimal control problem.With the help of the criticonly network structure,the complexity of the algorithm is reduced.In addition,the network is updated via particle swarm optimization(PSO)rather than the traditional gradient descent algorithm,which overcomes the disadvantage of determining the initial weight by only expert experience.Thus,the learning efficiency is increased.The theoretical analysis and simulation experiment illustrate that the proposed event-triggered control scheme can guarantee the stability of the system.2.To deal with the tracking control problems of the nonzero-sum games of continuoustimes multiplayer nonlinear systems,this thesis develops an novel event-triggered control scheme.First,for the error system,the optimal value function and the optimal tracking control strategy of each player are designed,and then,the coupled Hamilton-Jacobi equation(HJE)is obtained.In order to overcome the difficulty of solving the coupled HJE,the ADP method is utilized.The synchronous policy iteration framework is employed to solve the problems approximately.Then,by using the Lyapunov method,an novel event-triggering condition is designed,which can ensure the stability of the closed-loop system.In addition,the weights of the neural network are updated via the PSO algorithm,which improves the efficiency of solving the HJE and alleviates the difficulty of selecting the initial weight.Two simulation results demonstrate the effectiveness of the developed event-triggered tracking control scheme.Finally,the conclusions of this thesis are summarized based on the experience in the study process.Moreover,some worthy improvements and research directions in the further work are expounded.