Comparative study of machine learning methods for carbon metering in power generation enterprises

作者：Shan‐Li Wang, Bing Fang, Jiayi Zhang, Shaoxia Xing, Jiale Wu, Huanxiu Xiao · 发表于：Scientific Reports · 年份：2025 · DOI：10.1038/s41598-025-25227-6 · 被引用次数：3 · 研究领域：Energy Load and Power Forecasting、Environmental Impact and Sustainability、Integrated Energy Systems Optimization

In response to the global carbon neutrality imperative, power generation enterprises are under significant pressure to transform, with accurate carbon emission measurement and prediction being crucial to achieving this goal. While traditional carbon accounting methods face limitations in accuracy, adaptability, and cost, machine learning (ML) offers a promising alternative. This study focuses on a coal-fired power plant in Haikou, Hainan Province, utilizing operational data from January 1 to November 30, 2024, and selecting 18 feature parameters to develop and compare carbon emission prediction models.This study identified the key predictors through a hybrid feature selection strategy, including power generation, energy structure, operating time and load rate. The multiple linear regression, XGBoost and LSTM models are optimized through hyperparameter tuning. The prediction accuracy of the XGBoost model reaches 90.39%, demonstrating a strong ability to capture complex emission patterns. This research provides power generation enterprises with an accurate and efficient method for calculating carbon emissions, which can offer a scientific basis for their carbon management and energy conservation and emission reduction decisions.