Identification of key enterprise sources and design of target emission reduction pathways contributing to air quality based on unsupervised learning and explainable prediction – A case study of Beijing

Abstract

Industrial enterprises are the emission sources whose operations directly affect air quality. Identification of key sources and design of emission reduction pathways for enterprises have been crucial elements of air quality prevention. However, most current studies on air quality prevention focus primarily on macro-level pollution influencing factors or pollutant sources, often failing to to trace the main entity sources contributing to air quality degradation. Therefore, this paper firstly selects Kmeans to cluster the enterprises’ emissions data based on their geographical locations, considering that the enterprises data is noise-free, exhibits clear distance relationships, and is supported by comparisons of multiple clustering effects. Subsequently, we align the industrial enterprise emission clusters with air quality monitoring data, select SVR suitable for predicting the integrated small-sample dataset for air quality forecasting, and utilize the interpretability method SHAP to identify industrial enterprise clusters that have a significant impact on air quality. Finally, the entropy weight method is utilized to determine the different pollution levels of enterprises based on their contributions to air quality and the four quadrants method is applied to classify the enterprises under each pollution level. Based on this classification, the sources of enterprise pollution emissions are investigated, and emission reduction pathways are designed for each category.