A new data-driven predictor, PSO-XGBoost, used for permeability of tight sandstone reservoirs: A case study of member of chang 4+5, western Jiyuan Oilfield, Ordos Basin

Abstract Permeability is universally regarded as a critical analysis parameter for some geological work such as formation characterization and oil deposit exploitation. It can be obtained by physical models or fitting methods. Generally, the available physical models perform inefficiently when employed to solve permeability prediction of the unconventional reservoirs, since those models are much more complex, in which some parameters also are very difficult to be determined accurately. XGBoost, a machine learning model, has been widely used in many science fields as it shows to be much powerful in fitting analysis. However, due to the problem that XGBoost needs to complete its computation using many hyper-parameters, the model usually has difficulty to produce satisfactory results in most cases. In order to provide a better fitting model for permeability prediction, this paper then proposes utilizing PSO (particle swarm optimization) to improve XGBoost. The validation data for PSO-XGBoost is collected from the tight sandstone reservoirs of member of Chang 4 + 5 in western Jiyuan Oilfield, Ordos Basin. Three experiments are well designed to verify prediction capability of the proposed model. Three other models, including stepwise, SVR (support vector regression), and GBDT (gradient boosting decision tree), are introduced in the experiments to enhance the validation effect. Experiment results manifest the fitting errors generated by the proposed model are all the smallest in all experiments, well demonstrating PSO-XGBoost has stronger prediction capability compared with other validated models, further indicating its better application prospect in the study field of logging interpretation.