Lithofacies Identification by a Particle Swarm–Optimized Random Forest Algorithm in a Shale Oil Reservoir

Summary Accurate lithofacies identification plays a crucial role in the exploration and development of shale oil reservoirs, while existing methods all have their own shortcomings. In this paper, focusing on the shale oil reservoirs in the Weixinan Sag of the Beibu Gulf Basin, we propose a particle swarm optimization (PSO)-random forest (RF) algorithm (PSO-RF algorithm) for lithofacies identification. First, based on the core characteristics in the study area, we classify nine lithofacies with mineral composition, grain size, and sedimentary structure as the main factors. After that, we use the principal component analysis (PCA) method to reduce the dimensionality of the logging data and eliminate redundant information among the logging curves. Finally, we use a PSO algorithm to search for the optimal hyperparameters of the RF model, which is the PSO-RF algorithm. Compared with the results of core observations, the lithofacies identification results of cored wells in the study area demonstrated the effectiveness of the PSO-RF algorithm, achieving an overall accuracy of 90% on the test set. In addition, the PSO-RF model showed excellent adaptability when applied to noncored wells, with prediction results outperforming traditional machine learning algorithms. This study provides an effective method for lithofacies identification in the Beibu Gulf Basin and similar shale oil reservoirs.