A neurodynamic approach to model predictive control of piecewise linear systems

This paper presents a neurodynamic approach to model predictive control (MPC) of constrained piecewise linear systems. A novel procedure for estimating uncertain system parameters of piecewise linear systems is proposed. The model predictive control problem is then formulated as a quadratic optimization problem. To realize the real-time optimization in MPC, a one-layer recurrent neural network is employed for solving the quadratic optimization problem during each sampling interval. The overall MPC approach is of low computational complexity. Simulation results are included to substantiate the effectiveness and usefulness of the proposed approach.