Scholarlink

<p>To effectively manage climate change and create effective environmental policy, one needs reliable CO2 emissions forecasting in the electric power sector. Unfortunately, the accuracy of traditional forecasting schemes is often limited due to the difficulty in capturing the temporal and non-linear characteristics of the emissions data. This study proposes an integrated traditional machine learning (ML) schemes and a deep learning (DL) model based on a Long Short-Term Memory (LSTM) architecture, to develop more accurate CO₂ emissions predictions. Data pre-processing and model selection are done extensively before model training. This study shows that the LSTM model outperforms traditional ML schemes (i.e. LR (MAE: 42.97, RMSE: 48.26), RF (MAE: 14.73, RMSE: 18.42), GB (MAE: 18.85, RMSE: 22.48), and SVR (MAE: 19.51, RMSE: 24.23)) and other statistical techniques (e.g. ARIMA (MAE: 536.58, RMSE: 828.03), and Grey (MAE: 459.37, RMSE: 708.71)), with the lowest MAE (10.60) and RMSE (13.02) of all the forecasting schemes tested. These results demonstrate how well the LSTM model can identify temporal patterns in emissions data, which makes it a very reliable and accurate time-series forecasting tool. The findings have real-world applications in the power sector, supporting data-driven, flexible emission control strategies and sustainable policy planning.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Comparison of Various Machine Learning and Deep Learning Models To Forecast CO2 Emissions

  • 期刊论文

  • 发表日期：

    2025年11月27日

查看全文
  • Di Mu

摘要

To effectively manage climate change and create effective environmental policy, one needs reliable CO2 emissions forecasting in the electric power sector. Unfortunately, the accuracy of traditional forecasting schemes is often limited due to the difficulty in capturing the temporal and non-linear characteristics of the emissions data. This study proposes an integrated traditional machine learning (ML) schemes and a deep learning (DL) model based on a Long Short-Term Memory (LSTM) architecture, to develop more accurate CO₂ emissions predictions. Data pre-processing and model selection are done extensively before model training. This study shows that the LSTM model outperforms traditional ML schemes (i.e. LR (MAE: 42.97, RMSE: 48.26), RF (MAE: 14.73, RMSE: 18.42), GB (MAE: 18.85, RMSE: 22.48), and SVR (MAE: 19.51, RMSE: 24.23)) and other statistical techniques (e.g. ARIMA (MAE: 536.58, RMSE: 828.03), and Grey (MAE: 459.37, RMSE: 708.71)), with the lowest MAE (10.60) and RMSE (13.02) of all the forecasting schemes tested. These results demonstrate how well the LSTM model can identify temporal patterns in emissions data, which makes it a very reliable and accurate time-series forecasting tool. The findings have real-world applications in the power sector, supporting data-driven, flexible emission control strategies and sustainable policy planning.