Research on Groundwater Pollution Source Tracing of Abandoned Industrial Sites along the River Based on Bayesian Optimization

The migration of groundwater pollutants is concealed, and accurately and efficiently tracing the source of groundwater pollution is the difficulty in the remediation and control of groundwater pollution. To ensure the accuracy and efficiency of source tracing, this paper takes a contaminated site along the lower reaches of the Ganjiang River as an example and constructs a groundwater pollution source tracing framework based on Bayesian optimization. The Kepler optimization algorithm was adopted to optimize the parameters of the basic water flow model. A dynamic model was established by coupling the water level of the Ganjiang River, and the migration law of pollutants in the dynamic groundwater flow field was studied. Qualitative identification of site pollution sources is carried out through self-organizing mapping neural networks to determine the homology of pollution indicators. By using the IFM interface provided by FEFLOW and combining the Bayesian optimization algorithm with the solute transport model through the Python language, the parameters of pollution sources are inverted. The main achievements are as follows:

(1) Through the statistical analysis of groundwater quality data, it can be known that the typical pollutants in groundwater are manganese, ammonia nitrogen, iron and fluoride. There is an abnormal enrichment phenomenon caused by multi-source input and local pollution release in the field area.

(2) The pollution sources were qualitatively identified based on the self-organizing mapping neural network method. The results showed that they originated from agricultural production, livestock activities, and industrial production activities in the original factory area.

(3) Based on the site investigation data, a basic groundwater flow model was established. The model parameters were optimized through the Kepler optimization algorithm. The absolute error between the simulated water head and the actual water head at 39 water level observation points decreased from 2.62 to 0.36.

(4) By coupling the water level of the Ganjiang River to establish a dynamic water flow model, it was calculated that the influence radius of the Ganjiang River water level is approximately 350 meters. The contaminated site is precisely located at the edge of the influence radius. Through comparative experiments, it was found that the solute transport results within the site are less affected by the water level of the Ganjiang River.

(5) For the nonlinear high-dimensional optimization problem of groundwater pollution source parameter inversion, a physical constraint inversion framework based on Bayesian optimization is proposed. The Bayesian optimization algorithm can quickly identify the location with the highest possibility of pollution sources and simulate the matching parameter groups of pollution sources within a finite number of iterations. The entire process only takes 20 minutes. Subsequently, by means of combined source inversion, the locations of potential pollution sources are identified, and the causes and mechanisms of their formation are analyzed based on the integration of multi-source data.