Porosity prediction of tight reservoir rock using well logging data and machine learning

The accurate quantification of porosity in tight reservoirs is crucial for optimizing oil and gas exploration and production. Traditional predictive models often face challenges such as high costs, low efficiency, and limited accuracy, hindering effective exploration activities. To address these issues, we apply advanced machine learning algorithms-gradient boosting decision tree (GBDT), random forest, XGBoost, and multilayer perceptron-using well logging data, including acoustic time (AC), well logging (CAL), compensating neutrons (CNL), density (DEN), natural gamma (GR), resistivity (RT), and spontaneous potential (SP). These models are further optimized with the particle swarm optimization (PSO) algorithm to enhance their predictive accuracy. Comparative analysis reveals that the PSO-GBDT model outperforms other models, achieving an R² exceeding 0.99. Validation on two additional wells confirms the model's robustness, showcasing its superior predictive precision and efficiency. These findings suggest that the PSO-GBDT model has strong potential for improving porosity prediction in tight reservoirs, offering significant implications for future exploration and development efforts.