Multi-task Learning Model of Continuous Casting Slab Temperature Based on DNNs and SHAP Analysis

In the process of continuous casting, the slab temperature is a particularly crucial production parameter. However, it is still being monitored on the surface of the slab. At present, the prediction of slab temperature using machine learning models is not feasible due to the lack of internal temperature data from actual production. In this paper, a prediction method based on DNNs algorithm integrating numerical simulation and SHAP analysis is proposed to monitor the temperature field of continuous casting slab in real-time. The dataset comprises 6 input features and 5 output features. Following data preprocessing, four distinct machine learning models were developed employing DNNs, SVM, XGBoost, RF algorithms to individually predict the temperature of the slab. The DNNs model is selected as the optimal model according to the performance comparison using performance parameters such as MAE, MSE, and R2. SHAP value is calculated for the sensitivity analysis of the influence of the characteristic parameters of DNNs prediction model on the prediction results. The experimental results indicate that the prediction model has an average success rate of 96.48 pct within a temperature accuracy of +/- 20 degrees C. Moreover, the resident time, the amount of internal side water within the slab, and the internal side heat transfer coefficient of the slab have the most significant impact on the model. This study introduces a novel method for establishing machine learning models with machine learning techniques.