Multi-machine learning methods for rapid and synergistic inversion of groundwater contamination source, hydrogeologic...

Area of Science:

• Environmental Science
• Hydrogeology
• Machine Learning

Background:

• Groundwater pollution inversion is a critical environmental challenge.
• Existing machine learning methods have limitations in simultaneously identifying pollution sources, hydrogeological parameters, and boundary conditions.

Purpose of the Study:

• To develop a synergistic and rapid method for identifying multiple groundwater pollution inversion variables.
• To compare the performance of different machine learning models for this task.

Main Methods:

• Utilized multilayer perceptron (MLP), kernel extremum learning machine, support vector machine (SVR), and back-propagation neural network.
• Established inverse mapping between simulation model outputs and inputs.
• Combined SVR and MLP (SVR-MLP) for synergistic identification.

Main Results:

• SVR demonstrated high accuracy in identifying hydraulic conductivity and boundary conditions.
• MLP showed good accuracy in identifying pollutant release intensity.
• The combined SVR-MLP method significantly improved overall inversion accuracy, with an average absolute percentage error below 4%.

Conclusions:

• The SVR-MLP model offers a stable and reliable approach for groundwater pollution inversion.
• This method provides a robust basis for effective groundwater pollution remediation and management strategies.