A hybrid artificial neural network-numerical model for ground water problems.

Area of Science:

• Hydrogeology
• Computational modeling
• Artificial intelligence

Background:

• Numerical models are advanced tools for groundwater systems but face challenges with parameter uncertainty and prediction errors.
• Artificial neural networks (ANNs) offer high predictive accuracy at specific locations using field data.
• Integrating ANNs with numerical models presents an opportunity to leverage the strengths of both approaches.

Purpose of the Study:

• To present a novel hybrid modeling paradigm combining numerical models and ANNs.
• To explore methods for solving the overdetermined system created by integrating ANN predictions into numerical models.
• To assess the potential for improved numerical model accuracy through this hybrid approach.

Main Methods:

• Developed a hybrid modeling paradigm integrating ANN predictions into numerical models.
• Created an overdetermined system of equations by appending ANN models to numerical models.
• Applied mathematical techniques to solve the overdetermined system and evaluate accuracy improvements.

Main Results:

• Two of three tested methods for solving the overdetermined system showed improved numerical model accuracy.
• Accuracy improvements were observed across various levels of synthetic ANN errors.
• The hybrid approach demonstrated promise even with a limited number of constrained head values.

Conclusions:

• The hybrid approach of combining numerical models with ANNs shows potential for enhancing groundwater model accuracy.
• This method offers a promising strategy to mitigate parameter uncertainty inherent in numerical modeling.
• The presented hybrid paradigm is adaptable to other machine learning techniques beyond ANNs, such as regression or support vector machines.