Log-based estimation of magnitude, azimuth and causes of anisotropy using a committee machine-based model.

A medium is called anisotropic when one or more of its physical properties change with the direction of measurement. The main anisotropy causes include presence of fracture, shale layers and induced stresses. Some conventional methods of anisotropy evaluation, are determination of Thomsens coefficients and the empirical relations. The high cost of coring and the lack of generalizability of empirical relations have led to turning to other methods of anisotropy estimation. In this research, considering the characteristic of shear wave splitting into two components in an anisotropic environment, the difference in the arrival time of shear wave components (fast and slow) is considered as a magnitude of anisotropy. Then, based on the anisotropy values obtained from the shear wave splitting, intelligent methods were used to estimate the anisotropy. By selecting the effective inputs with high correlation coefficient to anisotropy, the anisotropy values were predicted and compared by each of the stand-alone methods. Considering the high errors of each individual method, an ensemble of intelligent methods and meta-heuristic algorithms was used because of their high precision and generalizability. Also, the azimuth of the fast shear wave was used to determine the direction of the fractures and the principal stresses in the study area. A total of 42 data samples from one of the southwestern Iran oilfields was used to construct the Committee Machine with Intelligent System. Afterward, anisotropy was estimated utilizing four intelligent systems including: Artificial Neural-Network (ANN), Fuzzy Inference System (FIS), Adaptive Neuro-Fuzzy Inference System (ANFIS), Radial-basis Function (RBF) and two meta-heuristic algorithms including: Genetic Algorithm (GA) and Ant-Colony Optimization Routing (ACOR). The designed Committee Machine with Intelligent System (CMIS) includes four input parameters of porosity (φ)\documentclass[12pt]{minimal}\usepackage{amsmath}\usepackage{wasysym}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage{mathrsfs}\usepackage{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$(\mathrm{\varphi })$$\end{document}, clay volume (VClay)\documentclass[12pt]{minimal}\usepackage{amsmath}\usepackage{wasysym}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage{mathrsfs}\usepackage{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$({{\text{V}}}_{{\text{Clay}}})$$\end{document}, compressional wave velocity (VP)\documentclass[12pt]{minimal}\usepackage{amsmath}\usepackage{wasysym}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage{mathrsfs}\usepackage{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$({{\text{V}}}_{{\text{P}}})$$\end{document} and shear velocity (VS)\documentclass[12pt]{minimal}\usepackage{amsmath}\usepackage{wasysym}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage{mathrsfs}\usepackage{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$({{\text{V}}}_{{\text{S}}})$$\end{document}. The results show that among the estimator models, the ANN model has the highest correlation coefficient and the RBF model has the lowest Mean-Squared Error (MSE). Moreover, among the meta-heuristic algorithms, the GA model has the best optimization coefficient according to the values of correlation coefficient (≅\documentclass[12pt]{minimal}\usepackage{amsmath}\usepackage{wasysym}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage{mathrsfs}\usepackage{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$\cong$$\end{document} 0.93) and minimum mean squared error (≅\documentclass[12pt]{minimal}\usepackage{amsmath}\usepackage{wasysym}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage{mathrsfs}\usepackage{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$\cong$$\end{document} 0.0001). Finally, having the azimuth of the fast shear velocity (30°-50o North-East), the direction of the fracture and the principal-horizontal stress were identified. [ABSTRACT FROM AUTHOR]