Research And Application Of Neural Network In CDQ System Control

Coke Dry Quenching(CDQ)is a novel coke quenching technology.Compared with Coke Wet Quenching,CDQ has higher efficiency of heat recovery,better coke quality and lower pollution.Thus,CDQ play an important role in technology upgrading of steel industry.Because CDQ is a complex nonlinear system with high time-delay,most of the time,model predictive control is used to control the CDQ.But the control accuracy of model predictive control is limited by the predictive accuracy of the model.In this paper,a nonlinear modeling method which could realize variable selection,is proposed to construct an accurate model with simple structure and high speed.Firstly,the Basic principle and the development of CDQ,artificial neural network(ANN)and variable selection are introduced.After the analysis of the working process of the CDQ system and field data of CDQ,the intake temperature of the boiler is selected as the target variable of the modeling,and 18 covariables in data set are selected as the initial alternative variables for the prediction modeling and variable selection.In order to realize variable selection and nonlinear modeling,a variable selection modeling method(Lasso-ANN)based on Lasso penalty term and neural network is proposed in this paper.Besides,the algorithm process and the key links in the algorithm are introduced detailly.There is a complex nonlinear optimization problem in the variable selection process of Lasso-ANN.Thus,a new particle swarm optimization algorithm is designed to solve this optimization problem quickly and accurately.Finally,the Lasso-ANN algorithm is implemented through MATLAB,and the simulation of the algorithm is compared with other modeling methods.The comparison results show that the prediction model obtained by this method has stronger generalization ability and higher prediction accuracy.