Application of Gradient Boosting Decision Tree Algorithm in Recoverable Reserve Assessment of Shale Gas Blocks at Different Exploration and Development Stages

The development performance of shale gas block is affected by complex couplings between geology and engineering factors, leading to significant challenges in recoverable reserves. Distinct geological and production data variability across exploration, appraisal, and early development stages necessitate differentiated predictive modeling approaches. This study introduces a multidisciplinary framework integrating gradient boosting decision tree (GBDT) with reservoir analytics to address these challenges. Methodologically, we first conduct data preprocessing followed by hybrid genesis-based qualitative analysis and distance correlation coefficient quantification. This dual approach identifies pressure coefficient, porosity, and burial depth as primary geological determinants of estimated ultimate recovery (EUR), with engineering parameters, such as fracturing stages and proppant volume per unit length are secondly influential. On this basis, the optimized GBDT algorithm incorporates stage-specific features for preliminary exploration, appraisal, and early production phases, while integrating Monte Carlo simulations for uncertainty quantification, generating P10-P50-P90 EUR distributions. A case study implemented in the Fuling shale gas field demonstrates robust predictive capability with errors constrained within 17% – 31%, which can provide result verification and technical support for the recoverable reserves calibration and listed reserves assessment in the early stage of shale gas block exploration and development.