Carbon emission evaluation for regional public institutions operation phase based on improved LightGBM and carbon emission efficiency ratio

Regional carbon emission assessment can effectively urge public institutions to actively carry out energy conservation and emission reduction transformations. This is crucial for achieving China's ‘dual carbon’ goals. In the study, the operating carbon emissions and influencing factors for four types of public institutions in warm winter-hot summer regions from 2019 to 2022 were collected. Pandemic index, interactive characteristics and lag characteristics of carbon emissions were constructed. The study conducted structural modifications to the standard LightGBM. And carbon emission models have been created based on multiple models. The predictive performance of multiple models was compared. Based on the SHAP model, the global and local interpretability of the carbon emission prediction process was evaluated. Finally, based on the improved LightGBM, the carbon emission efficiency of public institutions was proposed, and the scale of carbon emission efficiency ratio (CEER) was formulated. By comparing the common ranking method, the evaluation method based on CEER method was verified to be effective. The results showed that the improved LightGBM exhibited excellent performance in prediction, ensuring stable prediction stability and improving training efficiency. The evaluation results of the SHAP model showed that the inherent characteristics of public institutions (building area, number of energy users and headcount) were the decisive factors of carbon emissions. Compared with the ordinary ranking quota method, CEER evaluation method not only considered more influencing factors, but also added the necessary baseline. It increased the fairness and scientificity of the carbon emission assessment of regional public institutions.