Effective Symbolic Regression-Based Petrophysical Model Development Workflow and Efficient Software Tool for Enabling Easy Utilization

Developing petrophysical interpretation models integrating multiphysics measurements ischallenging. Symbolic regression (SR) offers a key advantage over other machine learning methods bygenerating analytical expressions enabling petrophysicists to assess consistency with physical principles andunderstand relevancy of the input variables. Existing SR packages are not optimized for petrophysicalmodeling, which often relies on limited and error-prone core samples. To address this, we developed asoftware solution using genetic programming-based SR to produce transparent, interpretable predictionequations. Our workflow consists of five components: (1) statistical feature selection, (2) symbolicregression for multiphysics fusion, (3) ensemble modeling, (4) conditional branching for formationheterogeneity, and (5) model discrimination for optimization. To enhance usability, we integrated multipleopen-source SR packages into a user-friendly interface, providing tools for data preparation, visualization,and model evaluation. This implementation simplifies workflow execution, making SR-based petrophysicalmodeling accessible to users without advanced programming skills.