Data-driven Parameter Self-correcting Modeling And Its Application In The Device Of S＿Zorb

In the industrial production process,in order to obtain some real-time parameters,we usually get real-time production data through the instrument measurement on the production line.However,this method has some limitations for the parameters that are not easy to be measured on-line or expensive.With the informatization transformation and upgrading of the whole production in the industry,a large number of production data have been accumulated,the data-driven intelligent algorithm prediction model has become one of the exploration directions of industrial process modeling.This paper establishes the theoretical framework of industrial process modeling based on datadriven,and expounds the working principle of parameter self-corrected model.Taking the S＿Zorb device as the research object,a parameter self-corrected XGBoost online prediction model that based on data-driven is established.It provides guidance for establishing an efficient and accurate data model for real-time prediction and monitoring of key parameters in the industrial production process.In view of how to conduct data-driven industrial process modeling,this paper provides theoretical guidance for the following empirical industrial process data modeling by establishing a data-driven industrial process modeling theoretical framework.In order to make the model trained from historical data maintain high prediction accuracy and ensure the timeliness of the training data set in the changing actual production process,this paper considers the optimization of the modeling process,and proposes a parameter self-correction method by analyzing the advantages and disadvantages of model recursive correction and model reconstruction correction methods.The parameter self-correction method establishes a continuously updated training data set based on the recursive principle,and uses online performance monitoring indicators to establish a confidence interval for model performance.When the model performance exceeds the confidence interval,the model is reconstructed on the new training data set to obtain Parameter self-correcting online prediction model.Finally,this paper takes the S＿Zorb device as the research object,and realizes the data modeling and parameter self-correction modeling process completely.When using XGBoost method to build a data model,the loss value of the octane number of the product that is monitored and costly in the S＿Zorb device is taken as the target variable.When selecting key variables,a joint scoring coefficient feature selection method based on maximum information coefficient and random forest importance score is proposed,and performed on 360 operating variables and 13 property variables.Then,the prediction performance of XGBoost are compared with the prediction performance of the multiple linear regression model and the back propagation artificial neural network prediction model those commonly used in relate literature,and XGBoost is finally selected as the prediction model according to the prediction effect.Finally,the parameter self-corrected XGBoost model was established,and the prediction performance of the models was compared with the uncorrected model several times.The results showed that the RMSE and MAE of the predicted results of the corrected model were reduced by 14.43% and 7.78% compared with the uncorrected model.It shows that the parameter self-calibration XGBoost model is more adaptable to the actual production process,and the model performance is better.