A Python Multi-Algorithm Optimization Framework for Automated Parameter Calibration in OpenGeoSys

Parameter calibration remains a critical bottleneck in coupled thermo–hydro–mechanical–chemical simulations, particularly when parameters are strongly coupled and non-unique solutions exist. In OpenGeoSys (OGS), calibration is frequently performed by manual trial-and-error, resulting in workflows that are subjective, difficult to reproduce, and unsuitable for systematic comparison of calibration strategies. These limitations become especially pronounced in multiphysics settings, where equifinality can mask parameter sensitivity and bias interpretation.

This study presents a non-intrusive, reusable Python framework for automated parameter calibration in OGS that treats the simulator as a black-box forward model. The framework controls the complete calibration workflow externally, including parameter sampling within defined bounds, automated execution of OGS simulations, extraction of user-defined parameters from output files, and quantitative misfit evaluation using different metrics. A total of twelve optimization algorithms are integrated, spanning local deterministic methods, surrogate optimization, population and swarm based approaches, and hybrid strategies. All algorithms are accessed through a unified configuration interface, enabling direct and fair benchmarking under the same evaluation metrics.

The framework is evaluated using an axisymmetric hydro-mechanical borehole benchmark with prescribed pressure and stress histories. Intrinsic permeability and Young’s modulus are jointly calibrated against a reference mass-flow time series, with each optimization method limited to approximately 100 forward simulations. The results demonstrate that calibration performance is governed primarily by misfit reduction efficiency per simulation rather than algorithmic overhead. Population-based methods robustly identify favorable regions of the parameter space, local search methods exhibit rapid convergence near optimal solutions, and hybrid strategies consistently combine both strengths. The proposed framework provides a reproducible and objective basis for parameter calibration in OpenGeoSys, enabling the development of more reliable models for coupled multiphysics applications.