Heat Exchanger Performance Prediction Based on Support Vector Machine, Random Forest, and BP Neural Network

This study aims to explore heat exchanger performance prediction methods based on Support Vector Machine (SVM), Random Forest (RF), and BP Neural Network (BPNN), addressing the growing need for equipment optimization and fault warning in the context of industrial automation and energy conservation. The novelty of this research lies in its comprehensive comparison of the three predictive models, which highlights the superior performance of Random Forest (RF), and in the use of large-scale data validation to enhance the reliability and practical applicability of the results. A high-quality dataset was constructed, consisting of 8 input variables and 9 output variables. After cleaning approximately 260,000 raw data entries, about 200,000 valid samples were retained. Upon completing hyperparameter optimization for all three models, the prediction performance was evaluated using three metrics: Mean Absolute Error (MAE), Mean Squared Error (MSE), and the Coefficient of Determination (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}^{\mathbf{2}}$</tex>), leading to the development of a comprehensive performance evaluation index. The experimental results show that RF achieved MAE, MSE, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}$</tex> values of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$0.0008,0.0060$</tex>, and 0.9943, respectively. Compared with BPNN, RF reduced MAE by 0.0014, decreased MSE by 0.0027, and improved <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}^{2}$</tex> by 0.0098; compared with SVM, MAE was reduced by 0.0165, MSE by 0.1410, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}$</tex> increased by 0.1165. These results clearly demonstrate that Random Forest is the best-performing prediction model in this study.