Industrial carbon emission forecasting considering external factors based on linear and machine learning models

Accurate forecasting of carbon emissions has become a critical task for the government to formulate effective policies and sustainable development. However, previous studies have mainly focused on large-scale carbon emissions forecasting, while urban-level carbon emission forecasting is equally important but rarely covered. In this study, we propose a novel carbon emission forecasting framework combining linear and machine learning models that considers both time dynamics and external influences. To improve the accuracy and explanatory power of the proposed model, we first introduce twelve initial influencing factors by considering the urban development, economic development, industrial energy consumption, and demographic factors. And then Lasso regression algorithm is adopted to filter out the indicators with poor predictive power. Finally, a combined prediction model by integrating Autoregressive Integrated Moving Average (ARIMA) and Support Vector Regression (SVR) models is established to capture linear and nonlinear features, respectively. The simulation results show that compared with benchmark models, the proposed model indicates stronger prediction performance with a Mean Absolute Percentage Error (MAPE) of 0.096 and a R-squared (R2) of 97.5%. In addition, six future development scenarios, including carbon emission projections for industrial growth and environmental protection factors, are also performed in this study to provide recommendations for carbon emission reduction programmers and related policy formulation. In conclusion, the forecasting framework proposed in this research can help to identify the key factors affecting carbon dioxide emissions and provide a quantitative reference for carbon dioxide emission reduction.