Characterization of Partially Sealing Faults from Pressure Transient Data: An Artificial Neural Network Approach

Abstract

Abstract Well testing is an effective and convenient tool for characterizing the hydrocarbon reservoirs. Although, the forward solution part of the well testing theory is well advanced, the corresponding inverse solution protocols especially for the complex domains are not well established. In this paper, a new inverse solution methodology, which utilizes the artificial neural network technology in analyzing the pressure transient data, is presented. The proposed methodology is applied towards analyzing the pressure transient data collected from an anisotropic faulted reservoir. The network development begins with a simple system and the level of complexity of the system is increased as the investigation progresses. The final goals of the analysis include determination of the principal permeability values, porosity of the reservoir, distance to the fault, orientation of the fault with respect to the flow directions and the sealing characteristics of the fault. The primary goal of this work is to test the capability of the ANN methodology as an engineering tool in pressure transient analysis applications. Introduction This work proposes the utilization of soft computing protocols such as artificial networks as a toolbox in the analysis of pressure transient data collected in complex reservoir domains. The forward solution part of the well testing theory is well advanced while the corresponding inverse solution protocols are not well established. In this paper, a new inversion methodology, which utilizes the artificial neural network technology in analyzing the pressure transient data, is presented. The methodology presented in this paper is developed as a potential tool to be used in the analysis of the pressure transient data collected from an anisotropic and faulted reservoir. The goals of the analysis include determination of the principal permeability values, porosity of the reservoir, distance to the fault, orientation of the fault with respect to the flow directions and sealing characteristics of the fault. Numerous sets of drawdown pressure transient data generated using a two-dimensional, single-phase, slightly-compressible numerical simulator are shown to an artificial neural network (specifically structured for this category of applications) during the training phase of the study. After the correct identification of a competent architecture for the network, it has been possible to teach the network some of the vague and complicated relationships that exist between the reservoir specific and operational parameters. In the final stage of the development, analysis capability of the network is tested using a large set of numerically generated pressure transient data. Artificial Neural Networks Neural networks are defined as interconnected assemblies of simple processing elements, units or nodes, whose functionality are loosely based on biological neurons. The processing ability of the network is stored in the inter-unit connection strengths, or weights, obtained by a process of adaptation to, or learning from, a set of training patterns1. Neural networks can be trained in order to perform a particular function by adjusting the values of the weights (connections) between elements. After receiving training from a set of data, network prediction capability is tested via some new data sets to determine the capabilities of the network.