Model predictive control based on recurrent neural network

Model predictive control algorithm with constraints research has important significance in industrial applications. In this paper, thanks to model predictive control problem with input and output inequality constraints has been researched widely in the literature, we propose a model predictive control scheme based on hybrid constraints and described it as an quadratic programming (QP) problem with restraints, a simplified dual recurrent neural network is used to solve this problem online t to obtain the optimal control action in the future. The complexity of the neural network is reduce with the number of neurons equal to the number of inequality constraints, and shown to be global convergence to the optimal solution. The simulation results show that the method has a faster speed line optimization and broaden the application field of model predictive control.