Improved permeability prediction based on the feature engineering of petrophysics and fuzzy logic analysis in low porosity–permeability reservoir

Permeability is difficult to evaluate in reservoir petrophysics property, especially in low porosity–permeability reservoir. The conventional permeability estimation model with establishment of the regression relationship between permeability and porosity is not applicable. This regression hypothesis based on the correlation between porosity and permeability (logarithm) is not available in low porosity–permeability reservoir. It remains a challenging problem in tight and heterogeneous formations’ petrophysical interpretation. Feature engineering process, as the most significant procedure in data-driven analytics, indicates that accurate modelling should be based on the main control factor on permeability ignoring its concrete mathematical expression. To select the factors that influence the main function of the model, and use the appropriate model to carry out the model structure, fusion and optimization is the main task to permeability estimation in low porosity–permeability reservoirs. Fuzzy logic, as a widely used approach in estimation of permeability, can be used to estimate the permeability with the advantage of tolerance. Its good adaptation in objective contradictory concepts and false elements in computational processes outweighs the traditional method on permeability estimation which always lies in a wide distribution of orders of magnitude. The research takes the permeability estimation issue in Mesozoic strata, Gaoqing area as example. The area of study mainly contains reservoirs with low-to-ultra-low porosity–permeability. The relationship between porosity and permeability is somewhat certain but insufficient using the regression method to predict. The research combined specialized feature engineering process with the fuzzy logic analysis to predict permeability. First, this paper analyzes that the main control factors of permeability in the region is the homogenization by diagenetic with statistical multivariate variance analysis SNK (Student–Newman–Keuls) method. It can be characterized by Δφ\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varDelta {\varphi }$$\end{document}, the changing degrees of porosity. To characterize the permeability response in well logs, the variables standing for a comprehensive reflection of the formation hydrology, lithology, and diagenesis are selected in the result of the electrofacies, SP, LLS, AC by multivariate variable selection method. The study is trying to combine the logging principle to explain its physical meaning by the statistical results. For discrete variables like electrofacies in modelling, scale quantization should be conducted by the optimal scale analysis considering discrete variables influences on permeability instead of manual labelling by numbers. Finally, the fuzzy logic analysis is carried out to achieve the results. The study makes a comparison of results in three ways to indicate the importance of feature engineering. That is, improved results with optimized model, model without feature engineering, and ordinary regression model. The optimized model with feature engineering predicts the permeability more conformed to the core data.