Recurrent neural network optimization for model predictive control

High computational burden in solving quadratic programming problem is a major obstacle when we apply model predictive control to industrial process. Recurrent neural networks offer a new quadratic programming optimization approach due to its parallel computational performance. In this paper, we present a new architecture of solving model predictive control (MPC) problem based on one layer recurrent neural network. We give algorithm of model predictive control based on recurrent neural network and prove convergence property of one layer recurrent neural network at each sample step. Two examples demonstrate the effectiveness and efficient of the proposed recurrent neural network based MPC. Simulation results show that this approach can utilize fast converge property and the parallel computation ability of recurrent neural network and apply to real-time industrial process control.