Sebastian Geiger

Delft University of Technology

Deciphering fracture properties using geological well-testing and machine learning

Fractures frequently control fluid flow in geoenergy reservoirs. Well testing is a powerful tool for fracture characterisation. However, Pressure Transient Analysis is particularly challenging in the presence of fractures as they normally manifest in many different pressure derivative responses (PDRs). We propose a Machine Learning-based method for classifying fracture-induced PDRs in individual clusters and identifying characteristic patterns in the data. We used a synthetic dataset that comprised 2560 PDRs, corresponding to 10 stochastic realizations of 256 combinations of different fracture intensities and average fracture apertures, organized in two orthogonal fracture sets. This dataset was generated from numerical simulation and a fit-for-purpose Discrete Fracture Network generator. We classified the PDRs by applying Unsupervised Machine Learning, specifically the k-medoids method combined with a distance metric known as Dynamic Time Warping. Our results suggest that the algorithm is effective at recognizing similar shapes in the first pressure derivatives if the second pressure derivatives are used as the classification variable. The analysis of the Naturally Fracture Reservoirs dataset indicated that 12 clusters were appropriate to describe the full collection of PDRs. We suggest that the respective cluster medoids can be regarded as type curves for various Naturally Fractured Reservoirs and be used as a guide in the interpretation of real well tests and help to identify uncertainties in fracture network properties that need to be considered when designing static and dynamic reservoir models for naturally fractured reservoirs. The classification exercise also allowed us to identify the key geological features that influence the PDRs, namely 1) the distance from the wellbore to the closest fracture(s), 2) the local/global fracture connectivity, and 3) the local/global fracture intensity.