A Prediction System of Burn through Point Based on Gradient Boosting Decision Tree and Decision Rules

Abstract

According to the characteristics of sintering process, a sintering end-point prediction system based on gradient boosting decision tree (GBDT) algorithm and decision rules is proposed in this paper. The on-line parameters of the sintering machine, which can characterize the change of the properties of the sintered raw materials in real time, were selected as the input of the model. The soft measurement results of the burn-through point position and temperature were selected as the output. The problem of establishing a system model based on the data collected in the sintering process to dynamically predict the state of burn through point (BTP) was solved. With the combination of process knowledge and several feature selection methods, the important characteristic variables related to the BTP were screened out. the algorithm of GBDT was used to establish the prediction model of BTP and burn through temperature (BTT). The parameters of the ensemble algorithm were optimized by using the methods of grid search and cross-validation, and the system model based on training data was established. On this basis, the corresponding decision model was added to the output of the prediction model, and the prediction accuracy of the system was improved. The establishment process of system model is introduced in detail. The operation results show that the system has better performance.