Research And Application Of Soft Sensing Prediction Model For Rotary Kiln Temperature

In the process of pellet roasting in grate kiln,real-time online monitoring and precise control of the temperature in the kiln can improve the quality of pellet products and further ensure the safety of production.In the actual production process,the pellet production process of grate rotary kiln is very complex.Due to the limitations of the current measurement technology,the expensive instrument used for measurement,the nonlinearity and time delay in the measurement process and other reasons,it is more difficult to control the temperature of the rotary kiln,and it is difficult to detect and control the temperature in the kiln online in real time.Aiming at the problem that the temperature in the grate kiln is difficult to be predicted on-line in real time,this paper takes the actual pellet production process of the grate kiln as the research object,on the basis of fully studying the pellet roasting reaction process,puts forward a quantum particle swarm optimization algorithm(DJQPSO)based on the dynamic dual population joint search mechanism to optimize the soft measurement model of the improved RBF neural network(SCRBF)type.Firstly,eight auxiliary variables,which are closely related to the temperature in rotary kiln,are selected to build the soft sensing model.Aiming at the problems of high dimension and information redundancy of auxiliary variable sample data,a data dimensionality reduction method(KPCA-KLPP)based on the combination of KPCA and KLPP is adopted,which keeps the objective function and The combination of global structure preserving objective function and feature extraction in high-dimensional space makes the mapped low-dimensional feature space not only retain the overall variance of the sample data to maximize,but also maintain the local neighborhood structure of the data set.While reducing the dimension of the original sample data and removing redundant information,it also keeps all the feature information of the data set to the greatest extent.The test results of the improved dimensionality reduction algorithm show that the improved data preprocessing algorithm can achieve ideal dimensionality reduction and redundancy removal effect.Then,the dynamic two population joint search mechanism of quantum particle swarm optimization(DJQPSO)is used to solve the problem of SCRBF with information feedback mechanism The parameters of weight,center of basis function and width between the hidden layer and output layer of neural network soft measurement model network are optimized for training.The internal evaluation of the temperature soft measurement prediction model in the grate kiln built by the trained structural parameters of neural network is carried out by the method of 10 fold cross validation.The evaluation result can determine the optimal conclusion of the prediction model Structure parameter combination.Finally,the evaluation index prediction error index,accuracy index,fitness of training set and prediction mean square error value are used to test the external independent sample test set,which proves that the soft sensor prediction model oftemperature in rotary kiln has good robustness and accurate prediction ability.The comprehensive experimental results show that the prediction model of temperature in grate kiln has good prediction ability and reliability.