Integrating Machine Learning with Numerical Simulation for 3D Mineral Prospectivity Modeling in the Sanshandao-Haiyu gold belt, Eastern China

Abstract

Numerical modeling of ore-forming dynamics and 3D mineral prospectivity modeling are pivotal for deep mineral exploration, though each has inherent constraints. Commercial software such as FIDAP and FLAC2D/3D can simulate deep geodynamic processes, yet FIDAP excludes rock deformation, while FLAC2D/3D neglects chemical reactions. Meanwhile, 3D prospectivity modeling is often limited by insufficient deep data. To address these gaps, this study selects the Sanshandao-Haiyu gold belt as a case study to investigate the integrated application of these two approaches for deep mineral exploration, which remains poorly understood. First, chemical reactions were incorporated into FLAC3D via a custom-developed program to calculate the mineralization rate. Subsequently, we employed machine learning techniques to integrate simulation outcomes (i.e., volumetric strain and mineralization rate) with fault morphology in different combinations, constructing four predictive models for comparative validation. The results demonstrate that: (1) significant spatial correlations exist among zones of positive volumetric strain, negative mineralization rate, and known gold orebodies; (2) all models exhibit high predictive accuracy, with the model incorporating all considered ore-controlling features performing optimally. Based on the predictions derived from this optimal model, two prospective targets were delineated.