Data-Driven Machine Learning Models for Steam Turbine Efficiency Prediction
Abstract
Increased electricity consumption in Indonesia requires coal-fired power plants to operate at high efficiency. This study proposes a data-driven approach to predict the efficiency of coal-fired steam turbines by utilizing 844 historical operational data points from the XYZ coal-fired power plant. Input variables include main steam pressure and temperature, main steam flow, final feedwater temperature, and condenser vacuum pressure. Four machine learning algorithms, namely Random Forest, Extra Trees, XGBoost, and Support Vector Regression, were evaluated using R², RMSE, and MAE. Based on the results of the study, it can be concluded that predictive models using Random Forest, Extra Trees, and XGBoost can predict steam turbine efficiency with high accuracy, especially after hyperparameter adjustments that increase the R² value and reduce the RMSE and MAE values in all models. The Extra Trees model proved to be the best model with an R² value of 0.8655 and an MAE of 0.5612, demonstrating its ability to capture the complex relationship between operational variables and turbine efficiency. This approach has practical implications for continuous improvement strategies in coal-fired power plant operations.