Screening and Optimization of Soil Remediation Strategies Assisted by Machine Learning

Abstract

A numerical approach assisted by machine learning was developed for screening and optimizing soil remediation strategies. The approach includes a reactive transport model for simulating the remediation cost and effect of applicable remediation technologies and their combinations for a target site. The simulated results were used to establish a relationship between the cost and effect using a machine learning method. The relationship was then used by an optimization method to provide optimal remediation strategies under various constraints and requirements for the target site. The approach was evaluated for a site contaminated with both arsenic and polycyclic aromatic hydrocarbons at a former shipbuilding factory in Guangzhou City, China. An optimal strategy was obtained and successfully implemented at the site, which included the partial excavation of the contaminated soils and natural attenuation of the residual contaminated soils. The advantage of the approach is that it can fully consider the natural attenuation capacity in designing remediation strategies to reduce remediation costs and can provide cost-effective remediation strategies under variable constraints for policymakers. The approach is general and can be applied for screening and optimizing remediation strategies at other remediation sites.