Data-driven soft sensors of papermaking process and its application to cleaner production with multi-objective optimization

Abstract

Papermaking industry can hardly monitor the dynamic of certain variables in the process when the test can only be conducted offline or destroy samples. Soft sensor is a predictive model that maps the measurable variables to the unknown measurements. In order to address the caused uncertainty, promote the papermaking process, a couple of soft sensors are developed with the industrial data by means of random forest (RF), gradient boosting regression (GBR), ridge regression and K-nearest neighbor (KNN) to monitor folding endurance, bursting strength, smoothness, and transverse ring compressive strength. The optimal models hold accuracy ≥0.839 (R-square) in general, and are applied to the multi-objective of cleaner papermaking production with regard to cost, energy consumption, and greenhouse gas (GHG) emission. The optimized results show that, when process is qualified with soft sensors' support, the possible reduction of cost, energy consumption and GHG emission could be up to 17.3% in total.