Design of predictive controllers by dynamic programming and neural networks

This paper proposes a method for the design of predictive controllers for nonlinear systems, The method consists of two phases, a solution phase and a learning phase. In the solution phase, dynamic programming is applied to obtain a closed-loop control law. In the learning phase, neural networks are used to simulate the control law. This phase overcomes the "curse of dimensionality" problem that has often hindered the implementation of control laws generated by dynamic programming. Experimental results demonstrate the effectiveness of the method.