EXPLORING UNCERTAINTY IN FLOW UNIT IDENTIFICATION AND PERMEABILITY PREDICTION

ONLINE FIRST (pre-proof)

Abstract

The study proposes a comparative uncertainty analysis of the main methods for permeability prediction or estimation, including the Cluster analysis (K-means), the Kozeny-Carman (KyC) equation for flow unit identification, and the K-nearest neighbor Density Estimate (KNN) algorithm, Kozeny-Carman equation, and One Flow Unit (OFU) for permeability prediction or estimation. The proposed analysis is applied to 13 wells in the Sacha field located in the Amazon region of Ecuador, targeting the Hollin and Napo formations, which mainly consist of sandstone, limestone, and shale. The selected wells have a sufficient number of laboratory measurements of permeability and electrical logs of porosity, permeability, natural gamma ray, medium, and deep resistivity. Initially, the K-means clustering and KyC methods are applied to identify the flow units, followed by a regression process to calculate the permeability using the KNN, KyC, and OFU methods. During the clustering process, the KyC method yielded better results, with the experimental data exhibiting uncertainties of less than ±35 mD, except in the outlier flow unit with an average porosity of 16.86% ±3.87% (Flow Unit D) whose average permeability is 407.52 mD and uncertainty of ±504.10 mD. For software simulation purposes, it is recommended to utilize the KyC method, as it employs basic concepts and equations in accordance with hydraulic principles.