Uncertainties and Challenges of Modern Well Test Interpretation

Abstract

Abstract Conceptual reservoir models are the basis for any reservoir engineering study and constitute the backbone of pressure transient analysis. The selection of the correct reservoir model is of paramount importance and cannot be overstated since the validity of the well test itself is totally dependent on the applicability of selected reservoir model. Once the proper model has been selected, various computer-aided interpretation techniques and nonlinear estimation methods can be employed to obtain quantitative descriptions of important reservoir parameters and near wellbore conditions. Automated type curve matching, together with non-linear parameter estimation, are two great modern well test analysis tools made possible by recent advances in computer technology. In addition, the use of pressure derivative data has aided in model identification to a degree not experienced before. However, the problem of non-uniqueness in model responses threatens to undermine the accuracy and usefulness of these tools in helping to characterize petroleum reservoirs. In this study, we illustrate the problem of non-unique model response with an example of a simulated pressure buildup test generated for a well near a sealing fault, in an infinitely large reservoir. We show that four different reservoir models can match this model response quite accurately. This makes model selection in field cases quite impossible without requiring additional information. Using field well test data from one slanted-horizontal and a short-radius horizontal well, we show, in one case how the uncertainty in the well model, as well as the value of the effective well length led to multiple matches, resulting in different estimates of reservoir parameters. In the other case, we show that without additional information from a previous test, automatic parameter estimation alone would have led to an erroneous interpretation, and wrong parameter estimates for this short-radius horizontal well. In the field cases presented, we provide steps taken to reduce the uncertainty and non-uniqueness inherent in the process, leading to more accurate estimates of reservoir parameters from the well test data. We conclude that in many cases, integration of additional information from geology, seismic and previous well tests conducted in the same reservoir, is necessary to reduce the non-uniqueness problem and achieve an unequivocal model match. Introduction Conventional interpretation methods utilize graphical analysis by industry standard diagnostic plots (ISDP) including semi-log, log-log and derivative plots, in addition to type curve matching. Modern well test analysis has been greatly enhanced by the use of the pressure derivative plot which was introduced by Bourdet et al1. Derivative plot analysis is emphasized throughout this study as it is proven to magnify small pressure changes, differentiate between responses of various models, define a clear recognizable pattern for various flow periods and improve the overall accuracy of test interpretation and the estimation of relevant reservoir parameters1–6. Considerable improvements in the overall accuracy of modern well test interpretation were achieved by the introduction and application of computer-aided analysis methods. Modern well test analysis utilizes the recent advancement in computer speed and efficiency to apply numerical analysis methods and estimation techniques to solve for unknown reservoir parameters. In computer-aided well test analysis, the initial stage of pattern recognition, model identification and validation, can be significantly improved by the application of automated type curve matching in combination with neural network models and artificial intelligence techniques to accelerate and enhanced the overall efficiency and accuracy of model selection process. Automated type curve matching methods can also be combined with nonlinear parameter estimation algorithm such as least squares (LS) and least absolute value (LAV), to ensure most accurate match is obtained and used to solve for unknown reservoir parameters.