Affiliation:
1. CSIR National Geophysical Research Institute, Hyderabad 500007, Telangana, India
2. Wadia Institute of Himalayan Geology, Dehradun 248001, Uttarakhand, India
Abstract
SUMMARY
We use the unsupervised and supervised neural network methods together to predict lithology of a gas hydrate reservoir from downhole data in the Krishna–Godavari (KG) offshore basin, India. In this study, we successfully identify the host litho-units of gas hydrate and show its effects in the identification of lithology using neural network techniques, which is not reported earlier. We use well log data acquired at three holes (10A, 03A and 04A) in 2006 during the first expedition of the Indian National Gas Hydrate Program (NGHP-01). Five different logging while drilling data (e.g. density, neutron porosity, gamma ray, resistivity and sonic) are considered for the mapping of lithology and gas hydrate. In the presence of gas hydrate, the resistivity and sonic velocity of the host sediments increase significantly, whereas density, neutron porosity and gamma ray are negligibly affected. Therefore, we calculate resistivity and sonic velocity for water-saturated sediment (without gas hydrate) theoretically to remove the effects of gas hydrate. At first, we apply the seven unsupervised classification methods (i.e. elbow, dendrogram, K-means, 3-D clustering, principal component analysis, Devies–Bouldin index and self-organizing map) to the data with gas hydrate (e.g. observed) and without gas hydrate (i.e. water-saturated/theoretical) to assess the data dimensionality and the number of clusters/litho-units. Each of the unsupervised schemes has its own pros and cons, and may provide different number of cluster/litho-units; sometimes, it is difficult to interpret from only one method. However, all seven methods provide same number of clusters in our study. Then, we apply the supervised classification method (i.e. Bayesian neural networks optimized by hybrid Monte Carlo searching technique) to the training data to refine the defined litho-units and map them with depth. Our approach identifies four types of litho-units and illustrates that the lithology in this area is dominated by clay (∼64 per cent) with some amount of silty clay, silt and minor sand. Gas hydrate is found in clay, silty clay and silt and not in sand. Results also show that, if gas hydrate is not considered as a separate unit, it is distributed as lithology in its hosts (i.e. clay, silty clay and silt). The method is very stable up to ∼15 per cent of random noise added to the data and results are well matched with the analysis of recovered core data. Identified lithologies at three wells correlate very well with seismic section crossing the wells. Very low permeability (<0.1 mD) estimated at three wells also indicates the clay-dominated lithology in our study area.
Funder
Department of Science & Technology
SERB
NNPDF
Publisher
Oxford University Press (OUP)
Subject
Geochemistry and Petrology,Geophysics
Reference88 articles.
1. Nuclear magnetic resonance logging while drilling, Oilfield Review;Alvarado,2003
2. The electrical resistivity log as an aid in determining some reservoir characteristics;Archie;J. Pet. Technol.,1942
3. Inversion of nuclear well-logging data using neural networks;Aristodemou;Geophys. Prospect.,2005
4. The effect of temperature on the density and electrical resistivity of sodium chloride solutions;Arps;J. Pet. Technol.,1953
5. Comparison of self-organizing maps classification approach with cluster and principal components analysis for large environmental data sets;Astel;Water Res.,2007
Cited by
21 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献