Research On Prediction Model Of Cement Raw Material Fineness Based On Convolution Neural Network

The raw meal fineness in the cement vertical mill system is an important indicator in cement production.The accurate prediction of cement raw meal fineness is the prerequisite for energy conservation,emission reduction and intelligent control in the cement industry.However,there are problems of nonlinearity,time-varying delay and coupling among the data variables of the vertical mill system.The traditional cement raw meal fineness prediction method is difficult to solve the above problems.They make it difficult to establish an accurate cement raw meal fineness prediction model.Aiming at the above-mentioned problems in the prediction of raw meal fineness,this paper proposes a two-dimensional convolutional neural network(2D-CNN)raw meal fineness prediction model,and applies Bayesian optimization algorithm to optimize the important hyperparameters of the 2D-CNN model to realize the accurate prediction of cement raw meal fineness and the automatic optimization of model hyperparameters.The specific research work is as follows:First,based on the analysis of the raw material grinding process of the vertical mill system,the candidate input variables are initially determined.The k-nearest neighbor mutual information algorithm is used to analyze the degree of association between candidate variables and predicted variable,and then several variables that have a greater impact on the raw meal fineness are selected.The selection of input variables can reduce the computational complexity of the prediction model,reduce the training and prediction time of the model,and prepare for establishing the raw meal fineness prediction model.Secondly,in view of the problems of non-linearity,time-varying delay and coupling in the vertical mill system,the sliding window method is used to construct the data matrix with the above characteristics as the input layer of the 2D-CNN model,and the two-dimensional volume is used in the convolution process.Two-dimensional convolution kernels are used to extracts the above features.This process can effectively learn the potential rules in the data.In addition,the use of predictive model monitoring strategies can effectively reduce the overfitting phenomenon of the 2D-CNN model.The above method can effectively solve the problems in the prediction of raw meal fineness,and realize the accurate prediction of raw meal fineness.Finally,in view of the problem that the hyperparameters related to the 2D-CNN model structure are difficult to choose,the 2D-CNN model’s prediction error is taken as the optimization goal,and a 2D-CNN hyperparameter optimization model based on Bayesian optimization is established,which avoids the limitation of relying on manual experience to select hyperparameters.The actual data of a cement company is used to simulate the above method.The experimental results show that the above method can realize the accurate prediction of raw meal fineness and the automatic optimization of model hyperparameters.