A Hybrid Predictive Model as an Emission Reduction Strategy Based on Power Plants’ Fuel Consumption Activity

Abstract

The Indonesian government has announced the initiation of carbon trading in 2023, commencing with coal-fired power plants (PLTUs). Despite this, certain PLTUs in Indonesia, including one examined in this research, still rely on manual identification of greenhouse gas (GHG) emissions and need more data to support forthcoming carbon trading endeavors. To address this gap, this study focuses on developing a predictive model using The Cross Industry Standard Process for Data Mining (CRISP-DM) methodology to forecast carbon emissions, coal fuel consumption, and gross electricity. The predictive model integrates time series models to forecast each variable, followed by a regression model for carbon emission prediction. Evaluation metrics, including MAE, RMSE, and MAPE, are utilized, and model training employs five machine learning models: Linear Regression, Support Vector Regression, Decision Tree Regression, Random Forest Regression, and LightGBM. Results reveal that for PLTU Units 3 and 7, the Linear Regression model performs optimally in time series modeling, whereas for PLTU Unit 8, Random Forest Regression is most effective. Across all units, Linear Regression emerges as the superior model in regression modeling. This study equips PLTUs with predictive insights into carbon emissions, facilitating strategic planning for emission reduction. By considering predicted outcomes of coal fuel consumption and gross electricity alongside carbon emission forecasts, PLTUs can comprehensively assess environmental and financial impacts, guiding effective mitigation strategies.

Keywords

• Coal-fired power plants
• emissions prediction
• greenhouse gas emissions
• machine learning