基于评价网络近似误差的自适应动态规划优化控制

为了求解有限时域最优控制问题，自适应动态规划(ADP)算法要求受控系统能一步控制到零。针对不能一步控制到零的非线性系统，提出一种改进的ADP算法，其初始代价函数由任意的有限时间容许序列构造。推导了算法的迭代过程并证明了算法的收敛性。当考虑评价网络的近似误差并满足假设条件时，迭代代价函数将收敛到最优代价函数的有界邻域。仿真例子验证了所提出方法的有效性。

Adaptive dynamic programming optimal control based on approximation error of critic network

In order to solve finite horizon optimal control problems, the adaptive dynamic programming(ADP) algorithm demands the system can reach zero in one step of control. For the nonlinear systems which cannot be controlled to zero in one step, an improved ADP algorithm is presented, and the initial cost is constructed by arbitrary finite horizon admissible sequence. After giving the iterative process, the convergence analysis of the improved algorithm is conducted. If the approximation error of the critic network is considered and several assumptions are satisfied, the iterative cost function will converge to a finite neighborhood of the optimal cost function. A simulation example is provided to verify the effectiveness of the presented approach.