Accuracy and Challenges of Machine Learning in Predicting Carbon Emissions from Coal-Fired Power Plants: An Analysis Based on Load Variations

Abstract

This study explores the application of machine learning (ML) techniques in predicting carbon emissions from coal-fired power plants, with a focus on the impact of load variations. The purpose of this research is to evaluate the accuracy and challenges associated with ML models when predicting carbon emissions in response to dynamic load changes, a critical factor for emission forecasting in power generation. The study employs a combination of time-series forecasting models, including regression techniques and advanced deep learning algorithms, such as Long Short-Term Memory (LSTM) networks, to analyze historical load and emission data from a selected coal-fired plant. The results show that while ML models can offer significant improvements in prediction accuracy compared to traditional methods, the accuracy decreases during periods of rapid load fluctuation, with prediction errors increasing by 10-15% under high load variations. Furthermore, model explainability and the integration of real-time data pose considerable challenges. The study highlights the importance of robust feature engineering, including the integration of fuel composition and environmental factors, as well as the need for real-time data processing to enhance model performance. In conclusion, the research demonstrates that machine learning can be a pow-erful tool for carbon emission prediction in coal-fired power plants, but it also faces limitations in terms of accuracy under fluctuating load conditions. To overcome these challenges, future work should focus on improving model ro-bustness through hybrid approaches that combine ML techniques with physical modeling and real-time data analytics. This would enable more accurate and reliable carbon emissions forecasting, helping to optimize emissions reduction strategies and support regulatory compliance in the power sector.