Optimal Tracking Control of Partial Unknown Continuous-Time Systems Using Integral Reinforcement Learning

An integral reinforcement learning (IRL) algorithm is used for solving the optimal tracking control problem for partial unknown continuous-time systems that try to chase a polynomial reference signal. By using IRL to get the control input of the Bellman equation derived from the problem, the approximate controller is able to get without sufficient system dynamics. Firstly, the LQT problem is formulated. An augmented vector is defined, and an algebraic Riccati equation is obtained based on the dynamic programming method. Then, employing IRL yields the iterative Bellman equation and policy updating expression. And the approximate optimal tracking control policy is finally solved, under which the reference signal with higherorder polynomials and unknown model parameters can be tracked by a linear system with partial known model parameters, meanwhile the specific performance can be minimized. Finally, by a simulation example, the efficiency of the provided method is confirmed.