Research of multivariable GPC based on multiple Hopfield networks and its application

Multivariate general predictive control (MGPC) has a good application in the control of plant with inertia and delay. But it has some defects such as great amount of computation online and poor treatment of constraints. This paper introduces Hopfield neural network into MGPC. Firstly, the MGPC was decomposed into several multi-input and single-output systems, then it was converted to several quadratic constrained optimizing problems. Several Hopfield networks were used to solve each quadratic constrained optimizing problem respectively. Hopfield network has the merits of simple arithmetic and rapid computation. The combination of the two methods can overcome the defects of MGPC. Then the new method was applied to the control of unit load system in power plant which is a 2 /spl times/ 2 multivariable plant with coupling and constraints. Simulation proved that the new method has good performance.