International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 07 | July 2024
p-ISSN: 2395-0072
www.irjet.net
DETECTION OF FRACTURES USING TRADITIONAL WELL LOG DATA AND COMBINATION OF MACHINE LEARNING TECHNIQUES MD. Ashikur Rahman1, Ishtiak Ahmed2, Mehedi Hasan3, Md. Abdullah Al Humayun4 1Graduate, Petroleum and Mining Engineering Department, Chittagong University of Engineering and Technology,
Chattogram, Bangladesh.
2Graduate, Mechanical and Production Engineering, Ahsanullah University of Science and Technology, Dhaka,
Bangladesh.
3Graduate, Electrical and Electronic Engineering, Ahsanullah University of Science and Technology, Dhaka,
Bangladesh.
4Department of Electrical and Electronic Engineering, Eastern University, Dhaka, Bangladesh
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Fractures significantly impact hydrocarbon flow
Additionally, fracture detection can minimize drilling costs and reduce environmental impacts by avoiding unnecessary drilling decreasing the risk of well bore damage.
and reservoir permeability, making their detection crucial for reservoir development, production assessment, and quality evaluation. This study identifies fracture zones using conventional well log data, consisting of two main steps. First, it analyzes well log responses to detect fracture zones, distinguishing fractured from non-fractured regions. Second, it applies a Support Vector Machine (SVM) classification to predict fracture zones based on these log responses. The well logs used include RHOB, DRHO, NPHI, DT, CALI, and PEF, which are commonly available for most wells. Current technologies like Borehole televiewer and core analysis, while effective, are expensive and time-consuming, highlighting the importance of log data for fracture detection. This study demonstrates the effectiveness of the SVM model on two wells, achieving a prediction accuracy of over 95%. The model's performance is consistent across balanced, unbalanced, scaled, and unscaled data, indicating its robustness and applicability in fracture detection.
Fractured zones can be detected through indirect or direct method. Various data sources, such as seismic sources, well logs, well tests, drilling mud histories and rock cores can be used to detect fracture zones. Formation Micro- Scanner (FMS), Borehole Televiewer (BHTV) etc. are used for high resolution images in bore hole to direct fracture detection. Conventional well log, well tests and other data are used as indirect methods to identify fracture zones [1]. Among these, one of the most common method for locating fractures in reservoir rock is to use conventional well log. As, well log which include, calliper (CAL), gamma ray (GR), sonic interval transit time (DT or AC), neutron porosity log (NPHI), density (RHOB), resistivity log etc. provide a continuous and in-situ measurements of rock properties, conventional well log data are commonly used for fracture identification. Fractures have typically some indirect and direct influence on the response of log these log data. Conventional well log measures several rock properties such as, density, resistivity, neutron porosity, sonic velocity etc. Fracture can be detected from changes in these properties, for example decreasing density, increasing sonic velocity etc. [2]. Using Machine Learning approach adding a novel step in identifying fracture zone in the reservoir. A machine learning system generates prediction models using previous data and learns from it to anticipate the results for new data. As various logs have different response to the fractured and non-fractured zones, these responses can be used for training a machine to discriminate fractured and non-fractured zones. In machine learning technique, detection of fractured zone in the formation by using conventional log data is a complex nonlinear classification problem [2]. Because in the presence of fracture responses of these conventional log are complicated. Its need to combine all log response to do classification.
Key Words: Fracture Detection, Well Log Analysis, Support Vector Machine (SVM), Hydrocarbon Flow, Reservoir Quality, Machine Learning.
1.INTRODUCTION 1.1 Background Fractures in underground formations have a significant role in fluid flow such as gas and oil. Without identifying the fracture and understanding its propagation in the reservoir, it is impossible to analyse and comprehend the behavioural traits of the reservoir and it does not lead to reliable results. In reservoir fluid flow, fracture have both positive and negative effects [1]. By making additional flow channels in the formation, fracture can help to transport hydrocarbon to the well bore. On the other hand, when fractures filled with clay or shale, it may create barriers to hydrocarbon flow and serve as a seal. They also affect the stability of engineered structures and excavations. Fracture may have also impact on the stability of excavation and designed structures.
© 2024, IRJET
|
Impact Factor value: 8.226
|
ISO 9001:2008 Certified Journal
|
Page 1248