Comparison of Different Machine Learning Algorithms for Predicting the Outlet Temperature of Propane Cooling Exchangers in Refinery

This study investigates the application of machine learning techniques to predict and regulate the outlet temperature of the propane cooling exchanger in the dehydration unit of the South Pars Gas Refinery. Given the critical role of this exchanger in managing gas temperature and preventing operational issues, such as excessive water vapor in molecular sieve dryers, the research focuses on evaluating the effectiveness of three machine learning algorithms: Random Forest, K-Nearest Neighbors (KNN), and Support Vector Machines (SVM). A total of ۳۱۶۸ data points were collected from the refinery's Data Collection System (DCS) for each input, and eleven input variables were used for training and testing the models. The models' performance was assessed using Mean Squared Error (MSE) and R² metrics. The results indicate that Random Forest outperforms both KNN and SVM, providing the best prediction accuracy with a low MSE of ۰.۰۰۳ and a high R² of ۰.۹۲۹. This highlights the potential of machine learning models for optimizing temperature control in industrial processes. The study concludes that Random Forest is the most suitable model for this application, though further exploration of alternative models and techniques is recommended for enhanced performance in different operational contexts.

کلیدواژه ها:

Machine Learning ، Gas Refinery ، SVM ، Random Forest ، KNN