Research On Optimal Control Based On Iterative Neural Dynamic Programming

The optimal control problem is one of the hotspots in modern control theory,which is aim to find an admissible control law to control the dynamic characteristics of the controlled object and optimize the performance index.The adaptive dynamic programming can approximate the cost function and control law of the system by the function approximate structure,which can avoid the "dimension disaster" problem of the general dynamic programming method,and is one of the most effective method to solve the optimal control problem of complex nonlinear system.Therefore,the combination of adaptive dynamic programming and optimal control theory is of great significance,which can solve many problems in system control and greatly improve control performance.Firstly,the problem of control constraints of general discrete-time nonlinear systems is considered.In order to overcome the constrained-input problem,the utility function is designed as a non-quadratic form.The neural network structure is designed to approximate the cost function and its partial derivative,and the parameters of the neural network are updated iteratively to obtain the approximate optimal control law.Simulations compare the control performances of the iterative algorithm with and without control constraints,and prove the effectiveness of the proposed method.Secondly,in order to effectively utilize computing resources and reduce resource waste,an adaptive event-triggered constrained control method for discrete-time nonlinear systems is proposed.A trigger condition is designed and the system is proved to be asymptotically stable under this condition.The neural network is iteratively updated online to obtain the approximate optimal control strategy under the eventtriggered control.The simulations verify the excellent performance of the proposed event-triggered constraint control method.Finally,for a class of discrete-time nonlinear switched systems,an adaptive approximate optimal control method is proposed.A novel model network is designed to estimate the state input of the system to obtain the system dynamics.The off-line iterative training is performed on the neural network structure to obtain the approximate optimal control strategy of the switched system.The simulations verify the effectiveness of the optimal control method of the switched system based on adaptive dynamic programming.