The Wavelet Neural Network Generalized Predictive Control Of Multiple-tanks System

In the modern industrial production control, it is appearing a lot ofcontrolling problems because of increasing amount and complexity of controlobjects during the practical production. The problems include the unknownmodel of controlling object, the multiple variables, misalignment, time-delay,coupling and some other in the complicated control system. The quadruple-tankssystem is a typical complicated control system which has the characteristic ofmultiple variables, great time-delay, misalignment and coupling. It imitates thecontrolled object of practical industrial production very well. In this paper, theauthor has done many related researches on the liquid level control of quadruple-tanks according to its characteristicDuring the process of study, the author applies the control strategy which isa combination of wavelet neural network and generalized predictive control. Thequadruple-tanks has the characteristic of multiple variable and misalignment. Thewavelet neural network contains good function approximation and generalizedpredictive control, and good control performance of complex process.Considering the facts above, this paper applies the wavelet neural network toidentify the controlled object model and establish a prediction model based onwavelet neural network. During the process of identification, the networktraining applies the improved BP learning algorithm. While solving the couplingproblem in system, this paper applies the fuzzy feed forward decoupling based onthe theory of feed-forward compensation decoupling. This method can takeadvantage of the universal approximation of fuzzy control, and makes somerelated design for the regulation and discourse domain in order to providespecific fuzzy feed forward decoupling strategy.In this paper, the wavelet neural network generalized predictive control is applied to simulate and analyze the two-tanks. This paper makes a controlexperiment about the quadruple-tanks liquid level control system applying thewavelet neural network generalized predictive control based on the fuzzy feedforward decoupling and analyze about the results of control experiments. Fromthe results of the analysis, it turns out that this design has the properties of shorttime adjustment, the stable final quantity and no vibration during the process ofcontrol. All the properties prove that the control strategy designed in this papercan get better control effect on the control experiment of quadruple-tanks liquidlevel control system. Meanwhile, the stabilization of the system control is alsovery good. As a result, the control strategy provides a kind of solution for thesimilar complicated control of quadruple-tanks system.