Load-Driven and Energy Consumption Conversion-based Enterprise Carbon Footprint Estimation Using Stacking Ensemble Learning

Abstract

Excessive carbon emissions have been established as the primary driving force behind global climate change, making the accurate prediction of carbon emissions crucial for addressing the imminent environmental crisis. The estimation of corporate carbon footprint (CCF) primarily relies on conventional annual carbon audits to determine a company’s carbon emissions. However, this approach may yield inaccurate results and inherently suffer from a one-year lag period. To address this challenge, our study presents a real-time CCF estimation method, introducing for the first time a fusion model based on Stacking ensemble learning. This model generates precise predictions regarding fossil energy consumption, subsequently calculating the corresponding direct carbon emissions. Indirect carbon emissions stem from the factory’s electricity consumption, which, when combined with direct carbon emissions, comprise the total corporate carbon emissions, ultimately enabling the estimation of the corporate carbon footprint. According to the results of empirical research, the proposed model exhibits a performance of 2.14% in Mean Absolute Percentage Error (MAPE) and 0.000513 in Root Mean Square Error (RMSE), metrics that significantly outperform other comparable predictive models.