The Study Of Air-Cushioned Headbox Control Strategy Based On Wavelet Neural Network Predictive Control

In papermaking industry,the headbox is a key component of the paper machine,whose outputs have vital influences on the produced paper quality and operation safety.The air-cushioned headbox is most widely used in China,so it is of great significance to conduct research on the control problems of air-cushioned headbox.For the control problems about air-cushioned headbox,We have deeply studied a class of self-recurrent wavelet neural networks(SRWNN).We apply the modified particle swram optimization(PSO)proposed in this thesis,to improve the SRWNN's parameters initialization.Consequently,the nonlinear predictive control(NPC)strategy about air-cushioned headbox is proposed,in which we select the SRWNN model as the prediction model.The main contents of this work are shown below:(1)To improve performance of standard PSO algorithm and avoid falling into local optimum,We have proposed a modified PSO algorithm that the total particle swarm is divided into some detective subpopulation and a exploitative subpopulation.In the proposed algorithm,all subpopulation keep up correspondences with each other based on the topological structure.The particles keep migration and substitution according to fitness during the iteration process,which can maintain population diversity,enhance ability of local search and guarantee to find the global optimum.We apply the modified PSO algorithm to wavelet neural networks(WNN)initialization,and the simulation results show that the proposed algorithm produces satisfactory performance of global optimum search,function approximation and model recognization.(2)To improve training effciency and reduce the numbers of wavelet neurons of WNN,we have deeply studied the properties of SRWNN which also takes good advantages of historical information besides current data.We make the learning rates of weights achieve adaptive adjustment.Through selecting the SRWNN as prediction model,a NPC stategy based on SRWNN is proposed.In this control strategy,the control signals are obtained through gradient descent method,and the control optimization rate condition of ensuring asymptotic convergence of control system is studied and proved.The simulation experiments verify effectiveness of the proposed control strategy.(3)The mathematical model description of air-cushioned headbox is illustrated,which is a strong nonlinear system with double input and double output.Based on this model,we have sampled some simulation data which is used as model identification with proposed PSO algorithm and SRWNN.We utilize the SRWNN model as prediction model to biuld the air-cushioned headbox predictive control structure,where there are two SRWNN models corresponding to pulp level and air-cushioned pressure.Simulation results show that the proposed control strategy achieves stable control performance for outputs,and robustness performance for interference.