Permeability and RRT Estimation from Conventional Logs in a Middle East Carbonate Reservoir Using Neural Network Approach

Abstract

Abstract A complex carbonate reservoir of a giant on-shore structure in central Abu Dhabi consists of a porous limestone section divided by several stylolitic dense layers. The reservoir is heterogeneous. High and low permeability rock types co-exist without much variation in log characteristics. Permeability changes several orders of magnitude with no appreciable change in porosity. Reservoir rock types and permeability measurements are available from 30 cored wells. The reservoir is penetrated by more than 100 vertical wells where open hole logs of various vintages are available. Estimation of permeability using conventional logs in the intervals and wells where core data is not available has been a challenge for many years. In this study, relationships were built between core data and the open hole log suite using the Artificial Neural Network approach and were used in estimating reservoir rock types and permeability. A typical character of the reservoir is that the log signatures are similar for several dissimilar rock types. Due to limitation on the vertical resolution conventional open hole logs alone could not differentiate these rock types. A synthetic variable cementation exponent ‘m' curve, gradient curves from Gamma Ray, Bulk Density and Neutron porosity logs supplemented the open hole logs in the neural network analysis in resolving the permeability variations. Nine neural networks were constructed to handle the complexity. Robust reservoir rock type and permeability estimation was achieved in two iterations. In the initial pass, fractional facies curves were estimated using a supervised classification network. These fractional facies curves were then used in initial permeability estimation with the help of a supervised estimation network. In the next pass, this permeability curve was used to estimate the reservoir rock types. In the final step, to fine-tune the magnitude of the permeability, the log derived reservoir rock types were included in network analysis and a final permeability curve with improved magnitude was generated. The initial permeability estimation was made separately on two vertical sub divisions of the reservoir interval and the reservoir rock type estimation was made on four separate vertical sub divisions based on the occurrence of various RRTs. Examples are provided showing a satisfactory prediction of both RRTs and permeability in cored and non-cored wells. The estimated permeability and RRTs were incorporated in the 3D geological model supplementing core data.