AI-based back analysis of multiphysics processes in unconventional resource extraction practice

Abstract

Mulitphysics processes have been commonly identified in geotechnical engineering practice. Researchers and field engineers often carry out multiphysics simulations to understand complex engineering responses. In field practice, a back analysis is typically required along with the simulations to calibrate the most representative model parameters. This would intensify the problem as it requires further simulations to assess the parameter sensitivity. Therefore, an efficient back analysis for multiphysics processes still remains a challenge in practice due to the numerical complexity and the low computational efficiency. With recent advances in AI techniques, opportunities have opened up for meta-model development for problems involving multiphysics processes associated with a large number of properties. This study entails a meta-model developed based on Artificial Neural Networks (ANN) that intelligently learn the correlations between model parameters and the reservoir responses. This efficient meta-model is combined with Genetic Algorithm-based back analysis to report the optimal case that provides the closest output to the target time histories. The results show that the AI-based metamodel can reproduce outputs of heavy computation of the multiphysics processes and thus efficiently perform back-analysis.