Industrial Steam Quantity Prediction Model Based on Grid Search-Optimized Support Vector Regression

Support Vector Regression (SVR) is a machine learning regression method that draws on the principles of Support Vector Machines (SVM). SVR is widely used in regression analysis in several fields because of its excellent performance in dealing with nonlinear, high-dimensional, and noisy disturbed data. In this study, SVR was used as the core model, and the data collected by the boiler's sensors were utilized. Following the removal of aberrant data points through the application of a Ridge Regression model, the SVR model was optimized using a grid search algorithm and a comparison of various kernel functions, ultimately resulting in accurate prediction of steam volume. The experimental results have shown that the SVR model achieved an MSE of 0.1102 and an
$R^{2}$
value of 0.8879. When compared with the eXtreme Gradient boosting (XGBoost) regression and Random Forest regression models, the SVR model has demonstrated excellent performance in predicting steam quantity, with high accuracy and strong generalization capabilities. The contribution of this study is to provide a feasible method for steam volume prediction in the industrial field and a new idea for improving the operational efficiency of industrial boilers, which is expected to have a substantial impact in saving resources and improving industrial productivity.