Data-Driven Prediction of Preheater Steady-State Conditions Based on Support Vector Regression

Abstract

The determination of steady-state operating conditions is a fundamental and complex task in the computation of heat exchangers. Traditional approaches often involve intricate models based on conservation laws—mass, energy, and momentum. These methods, while thorough, tend to be intricate, labor-intensive, and heavily dependent on empirical data, often making them less efficient and adaptable in diverse applications. Contrasting sharply with these conventional techniques, machine learning and data-driven methods stand out for their simplicity, user-friendliness, and swift response times. Despite their successful implementation in various domains, their integration into heat exchanger computations remains surprisingly limited. To bridge this gap, this study introduces the application of Support Vector Regression (SVR), a supervised machine learning algorithm. This research revolves around training a model with datasets generated from heat exchanger design software, aiming to accurately predict the steady-state conditions of preheaters and evaluate the robustness of the model. The results reveal that SVR is remarkably effective in forecasting the steady-state operations of preheaters. The model demonstrates an impressive accuracy level, with the maximum deviation between the predicted outcomes and actual computational results not exceeding 2%. Moreover, the model exhibits good robustness. It maintains reliability and accuracy even in the face of input fluctuations. Such robustness not only validates the effectiveness but also highlights its potential for practical deployment in real-world settings.