Research On Variable Selection And Prediction Modeling Method For Industrial Complex Data

In the industrial process,real-time monitoring of quality is a very important part of quality control,but due to economic,environmental,experimental conditions and other factors,it is difficult to directly measure some output variables or the measurement period is long.The predictive modeling technology can realize the prediction of the target variable through the process variables that are related and easy to measure,eliminating the inconvenience caused by direct measurement.Although modeling and prediction technology has been applied to some industrial processes,the current research and application are still in the static assumption stage.With the continuous popularization and update of data acquisition equipment,the characteristics of industrial process data sets have become more and more complicated The characteristics of large data volume,high nonlinearity,high dimensionality and time-varying are becoming more and more obvious.This paper starts with research on variable selection and modeling prediction of complex industrial data sets.From variable selection,a variable selection method based on mixed criteria is proposed,and then a dynamic prediction model is established based on BP neural network and adaptive moving window.Finally,the variable selection method and the modeling prediction method are combined to establish a high-precision prediction model.The specific content of this article is as follows1.In view of the high correlation and high dimensional characteristics of complex industrial process data,this paper studies the problem of variable selection under the characteristics of complex industrial process data.This paper improves a variable selection method based on mixed criteria based on analyzing and using the commonly used single criterion variable selection method.This method can solve the problem that the single criterion is not accurate enough for the selection of industrial complex data variables.The method is applied to water-treatment,LAB and other cases,and the selected variable subset is used to establish a BP neural network model comparison experiment to verify the effectiveness of the mixed criterion variable method2.Aiming at the characteristics of non-linearity,large sample and mutation of industrial complex data,this paper studies the problem of BP neural network adaptive prediction modeling under the characteristics of industrial complex data.A single static prediction model is prone to failure when dealing with unbalanced data.Based on BP neural network and PLS-AMW method,this paper proposes an improved adaptive prediction model(BP-AMW).This method is applied to the TE chemical process data set,and compared with a single BP neural network using PLS-AMW modeling method.The experimental results show that the improved BP-AMW prediction performance has a certain improvement3.In order to solve the problem of modeling complex industrial processes with high redundancy,nonlinearity and catastrophe characteristics.This paper proposes a predictive modeling method based on the combination of GA-NLP and BP-AMW.First,traverse the training set through adaptive moving windows,use the GA-NLP method to select variables,and then establish a BP neural network model set.Finally,the simulation experiment of the TE chemical process is carried out,and the prediction results are compared and analyzed.It is found that the combined prediction method proposed in this paper can effectively reduce the redundancy of complex data and further improve the prediction performance of the model.