Adaptive-Critic-Based Robust Control with Momentum Gradient Descent for Continuous-Time Nonlinear Systems
作者单位：Faculty of Information Technology Beijing University of Technology Beijing Key Laboratory of Computational Intelligence and Intelligent System Beijing University of Technology Beijing Institute of Artificial Intelligence Beijing University of Technology Beijing Laboratory of Smart Environmental Protection Beijing University of Technology
会议名称：《第43届中国控制会议》
会议日期：1000年
学科分类：0711[理学-系统科学] 07[理学] 08[工学] 081101[工学-控制理论与控制工程] 0811[工学-控制科学与工程] 071102[理学-系统分析与集成] 081103[工学-系统工程]
关 键 词：Adaptive dynamic programming Momentum-based gradient descent Online policy iteration Optimal control Robust control
摘 要：In this paper, robust control problems are investigated for nonlinear continuous-time systems. A momentum-based gradient descent(GD) approach is developed to enhance the convergence performance of parameters in adaptive dynamic programming(ADP). By introducing the idea of momentum, the oscillation in the process of GD is alleviated and the selection of the learning rate becomes more flexible. Under the framework of ADP, the robust control problem is transformed into the optimal control problem by modifying the cost function. To avoid limitations of the initial admissible condition, an additional term is employed in the computation of the current gradient. Based on the online policy iteration algorithm, the momentum-based GD approach is constructed as an improved learning algorithm to optimize the critic network weights. Finally, a simulation is conducted to verify the effectiveness of the established learning strategy.