Abstract

ABSTRACT Identifying and predicting rock unconfined compressive strength is critical to well design, wellbore stability, and hydraulic fracturing stimulation in shales. However, accurate prediction of rock mechanical properties is hard in the absence of high-resolution advanced geophysical logs (e.g., image logs) and laboratory mechanical core samples. This study focuses on rock mechanical parameter estimation using three data-driven techniques: backpropagation (BP) network, support vector machine (SVM), and extreme learning machine (ELM) models in a shale gas reservoir. Additionally, this study aimed to obtain accurate rock mechanical properties, which have proven challenging using conventional petrophysical methods in wells without downhole core data. A total of 350 samples from 22 wells with laboratory measurement data were used to train and validate the neural network. Robust algorithms were used to provide fast and accurate prediction results, which were verified by comparing them with other approaches. The trained machine-learning models were cross-validated to check their robustness. The network model was then applied to estimate rock unconfined compressive strength for the remaining wells. The predicted results had a good match well with the laboratory test conclusions. Based on the estimations, rock mechanical properties were mapped and analyzed in the target shale gas zone. This method is helpful for geomechanical modeling in shale gas reservoirs. INTRODUCTION Uniaxial compressive strength is the ultimate ability of a rock specimen to resist damage under uniaxial pressure. When a rock is subjected to external forces, damage occurs when the stress reaches a certain limit value, which is the strength of the rock. As one of the most important mechanical indexes of rock, understanding the uniaxial compressive strength is necessary for hydraulic fracturing design and wellbore stability. I To have a more accurate characterization of the reservoir geomechanical properties and the corresponding engineering parameters and to provide a basis for the later development of well pattern and stimulation scheme, well logging is generally used for the geomechanical properties prediction. However, due to the absence of several well logs and empirical parameters in the geomechanical modeling, accurate geomechanical properties estimation should be calibrated with laboratory experiments. However, in the absence of high-quality laboratory cores, it is challenging to accurately conduct experiments. Therefore, this study uses artificial intelligence techniques to predict uniaxial compressive strength by combining well logs and laboratory mechanical results.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call