Comprehensive prediction of soil benzo[a]pyrene content in Chinese coking enterprises from 2020-2040: an innovative full production cycle approach based on interpretable machine learning

Given the intricate and variable causes of site soil pollution, accurately predicting specific soil pollutant content over extended periods remains a challenging task. This study proposes an innovative approach to quantify the degree of enterprise full production cycle management and construct an interpretable machine learning model with dynamic optimization, thereby addressing this challenge. By incorporating 12 soil pollution influencing factors, we predicted the soil benzo[a]pyrene (BaP) content of coking enterprises in China from 2020 to 2040. Additionally, we employed SHAP and partial dependence plots techniques to conduct an in-depth analysis of the relationships between each influencing factor and soil BaP content. The random forest algorithm was identified as the optimal model for predicting soil BaP content in coking enterprises, yielding an R² value of 0.771 and an RMSE value of 2.1. Among various influencing factors, full production cycle emission standard quantification result exhibited the most significant impact with an importance score of 24.4 %. Compared with natural environmental factors (such as sunshine, rainfall, and temperature), the production activities of enterprises themselves (such as production time and output) have a more significant impact on the accumulation of soil pollutants. The highest soil BaP content among coking enterprises in China in 2020 was 231.1 mg/kg. Assuming that the unpredictable variables such as output and the number of environmental violations remain unchanged, by 2040, the average content of BaP in the soil of coking enterprises is expected to increase from 6.1 mg/kg to 7.38 mg/kg, and the over-standard rate is expected to increase by approximately 12.24 %. This study underscores the crucial role of stringent pollutant emission standards in reducing soil contamination in industrial environments.