Four Tank Water Neural Network Predictive Control

In the modern industrial actual control, we often encountered in the object model with unknown, most of these objects with multiple variables and the characteristics of strong coupling, and some of the characteristics of a large time delay. Quadruple-tank is a typical nonlinear, time delay and strong coupling system, which can simulate industrial complex object. In this paper, we study the control problem of Mechanism model of Quadruple-tank, and presented with a fuzzy neural network decoupling control strategy.Firstly, introduces a generalized predictive control based on neural network theory and the structure of the NGPC algorithm, use the fast convergence speed and highly accuracy algorithm-Levenberg Marquardt to train neural network, derive the derivative equation of neural network, calculate Jacobian matrix and Hessian matrix of the input u by iterative method.Secondly, introduce the basic theory of fuzzy control and fuzzy feedforward decoupling, give the specific block diagram of the decoupling strategy, present a fuzzy decoupling output coefficients and fuzzy decoupling interaction coefficients, and then base on the structural characteristics of the quadruple-tank, two equations are listed, derived the method of calculating the two coefficients to be calculated online.Finally, proceed the control experiments of mechanism model of the quadruple-tank, and the control result are analyzed contrastively. Experiment results show that generalized predictive control based on neural network combined with the fuzzy decoupling control strategy get good results in the mechanism model of the quadruple-tank, which with multiple variables,strong coupling, and large time delay. Adjustment time is short, the process no vibration and the final value is stability. Set point adjustable range is large, and set point changes without having to change any parameters of the system,So the control tactics has little human intervention and strong robustness.