Chapter 5 - Intelligent approaches to forecast the chemical property: Case study in papermaking process

Abstract

Real-time and accurate monitoring of key physical and chemical parameters in industrial production processes is a basic work to improve product quality and reduce energy consumption. Affected by the high integration of production equipment and the working environment, sensors used to monitor key physical and chemical parameters often have problems that cannot be installed or have low measurement accuracy. As a result, the production process is not transparent enough, and the process optimization lacks data support. Establishing the prediction model of the key physical and chemical parameters can play the function of replacing hardware with software, which is not only economical and reliable, but also has a rapid dynamic response. The traditional mechanism modeling method requires an accurate theoretical foundation and is only suitable for simple process objects. For complex process objects, the model accuracy cannot meet the requirements of the production process. The data-driven modeling method based on historical data and intelligent algorithms has simple process, strong model versatility, and high precision, and is more suitable for modeling complex industrial production processes. This work takes the prediction of the beating degree of tissue paper as an example. First, the historical data of the refining process and pulp fiber parameters are collected and preprocessed. Then, based on the GBDT algorithm, a prediction model of the beating degree is established, and the k-fold cross check method is used to verify the accuracy of the model. The verification results show that the MSE of the GBDT model is 0.9637 SR° and the MAE is 0.7627 SR°, which meets the actual production requirements for the accuracy. With the same precision, compared with the SVM algorithm, the GBDT operation time is only one fiftieth of the SVM, indicating that the GBDT model has a faster operation speed. By analyzing the GBDT model, it is found that specific energy consumption and pulp flow are the variables that have higher influence on the beating degree; the twisted fiber content, fine fiber content, and brooming rate have medium influence, and the remaining variables matter less.