The utilization of fossil fuels has led to a significant rise in carbon dioxide (CO2) emission. Among various sectors, the energy industry plays a substantial role in contributing to global CO2 emission. In this research, we have employed both multivariate and univariate time-series models, as well as machine learning and deep learning techniques, to analyze a dataset on CO2 emission. The dataset, obtained from the central electricity authority, encompasses attributes such as coal supply information, CO2 emission, peak demand, and peak met. To evaluate the performance of the applied models, we have utilized several performance metrics including RMSPE, MAE, RMSE, MSE, MAPE, SMAPE, and RAE. The dataset spans from 2005 to 2021, enabling us to conduct training and testing. Furthermore, we have utilized the best-performing models to forecast CO2 emission from 2022 to 2050. The results of our study demonstrate that autoregression is the top-performing model, achieving a remarkable ranking of 1.85 according to the Friedman ranking. Additionally, we have conducted a comparative analysis between multivariate and univariate approaches.
Paper
Full text
CO2 emission prediction from coal used in power plants: a machine learning-based approach
Semantic Scholar · Environmental Science · 2024
Abstract
The utilization of fossil fuels has led to a significant rise in carbon dioxide (CO2) emission. Among various sectors, the energy industry plays a substantial role in contributing to global CO2 emission. In this research, we have employed both multivariate and univariate time-series models, as well as machine learning and deep learning techniques, to analyze a dataset on CO2 emission. The dataset, obtained from the central electricity authority, encompasses attributes such as coal supply information, CO2 emission, peak demand, and peak met. To evaluate the performance of the applied models, we have utilized several performance metrics including RMSPE, MAE, RMSE, MSE, MAPE, SMAPE, and RAE. The dataset spans from 2005 to 2021, enabling us to conduct training and testing. Furthermore, we have utilized the best-performing models to forecast CO2 emission from 2022 to 2050. The results of our study demonstrate that autoregression is the top-performing model, achieving a remarkable ranking of 1.85 according to the Friedman ranking. Additionally, we have conducted a comparative analysis between multivariate and univariate approaches.