Permeability Prediction of Carbonate Rock by Conventional Logging with MA-Based Adaptive Model

Models for permeability prediction of carbonate reservoirs by conventional logging curves can be classified into two major types, physical and fitting models. However, accurate physical parameters are often difficult to obtain, and it is difficult to directly predict carbonate reservoir permeability with strong heterogeneity by fitting the model. In order to optimize the physical model parameters and machine learning model parameters simultaneously, a powerful MA-based adaptive model is proposed, which is selected and modified by the mayfly algorithm (MA) to optimize hyper-parameter tuning. In this paper, taking the reservoirs of the Deng 2 members as a case, the permeability prediction capability of the MA-based adaptive model is validated by two experiments, which can be subdivided into static mode and dynamic mode. The results show that: ① The permeability prediction ability of the model can be improved by breaking the physical model and machine learning model, and the MA-based adaptive model presents better applicability on permeability prediction when the basic data of the physical model is insufficient. ② MA can more effectively optimize the parameters of the physical and machine learning models, and the formed MA-XGBoost displays better performances, which is deemed suitable for permeability prediction. ③ MA is more suitable for determining the empirical coefficient in the DRT clustering algorithm. The MA-based adaptive model provides a new idea for the accurate prediction of carbonate reservoir permeability.