Automated Well Test Analysis Utilizing Self-Adaptive Differential Evolution Method

Abstract

AbstractWell testing is the analysis of pressure and rate data from a reservoir to estimate the reservoir parameters. It enables monitoring and optimization of reservoir performance, and helps estimating the petrophysical properties of the reservoir. Nonlinear regression was presented to well test analysis more than three decades ago, and rapidly became the typical practice in the industry since then. Despite its popularity, the nonlinear regression technique received limited development, and still has major problems in well test analysis. In nonlinear regression, wrong initial guess of regression parameters could lead to incorrect results; as the regression algorithm might not converge to the exact results when the initial parameters' estimations are far from the actual values. This is mainly due to the convergence of some algorithms, such as Levenberg–Marquardt's algorithm, to a local minimum and missing the global minimum of the objective function.The objective of this research is to develop a well test analysis technique utilizing Self-adaptive Differential Evolution (SaDE) algorithm, this technique can perform automatic reservoir parameters estimation using the pressure and its derivative responses for different reservoir models. The proposed approach is applied on different reservoir models, namely; radial flow in a radial composite reservoir model and radial flow in a circular homogeneous reservoir model. The performance of the developed method is analyzed and compared with Levenberg–Marquardt's algorithm. Sensitivity analysis is introduced to study the effect of different model parameters such as; wellbore storage, skin, mobility ratio, etc.The obtained results showed that the developed well test analysis technique, using SaDE optimization algorithm, can be used to estimate the reservoir parameters with higher accuracy compared to Levenberg–Marquardt's algorithm. Unlike Levenberg–Marquardt's algorithm, the introduced well test analysis technique has less dependency on the initial estimate of parameters, and the overall regression performance is significantly enhanced. Furthermore, for automated interpretation in which the model is already known, this method allows the elimination of the initial guess-determination step.This is the first time SaDE algorithm is implemented and applied to estimate reservoir parameters in well test analysis. Therefore, it is a step forward to, fully automate well test analysis and interpretation, where the need of initial estimate of regression parameters is eliminated. This achievement is not possible using conventional gradient-based algorithms such as Levenberg–Marquardt. Outcomes of this study would help well testing engineers and analysts to have better understanding of reservoir parameters and performance without initial guess-determination step.