Cardiff University | Prifysgol Caerdydd ORCA
Online Research @ Cardiff 
WelshClear Cookie - decide language by browser settings

Inferring the lithology of borehole rocks by applying neural network classifiers to downhole logs: an example from the Ocean Drilling Program

Benaouda, D., Wadge, G., Whitmarsh, R.B., Rothwell, R.G. and MacLeod, Christopher 1999. Inferring the lithology of borehole rocks by applying neural network classifiers to downhole logs: an example from the Ocean Drilling Program. Geophysical Journal International 136 (2) , pp. 477-491. 10.1046/j.1365-246X.1999.00746.x

[img]
Preview
PDF - Published Version
Download (986kB) | Preview

Abstract

In boreholes with partial or no core recovery, interpretations of lithology in the remainder of the hole are routinely attempted using data from downhole geophysical sensors. We present a practical neural net-based technique that greatly enhances lithological interpretation in holes with partial core recovery by using downhole data to train classifiers to give a global classification scheme for those parts of the borehole for which no core was retrieved. We describe the system and its underlying methods of data exploration, selection and classification, and present a typical example of the system in use. Although the technique is equally applicable to oil industry boreholes, we apply it here to an Ocean Drilling Program (ODP) borehole (Hole 792E, Izu-Bonin forearc, a mixture of volcaniclastic sandstones, conglomerates and claystones). The quantitative benefits of quality-control measures and different subsampling strategies are shown. Direct comparisons between a number of discriminant analysis methods and the use of neural networks with back-propagation of error are presented. The neural networks perform better than the discriminant analysis techniques both in terms of performance rates with test data sets (2–3 per cent better) and in qualitative correlation with non-depth-matched core. We illustrate with the Hole 792E data how vital it is to have a system that permits the number and membership of training classes to be changed as analysis proceeds. The initial classification for Hole 792E evolved from a five-class to a three-class and then to a four-class scheme with resultant classification performance rates for the back-propagation neural network method of 83, 84 and 93 per cent respectively.

Item Type: Article
Date Type: Publication
Status: Published
Schools: Earth and Ocean Sciences
Subjects: Q Science > QE Geology
Uncontrolled Keywords: Artificial intelligence; Borehole geophysics; Drill cores; Ocean drilling; Sediments; Statistical methods
Additional Information: Pdf uploaded in accordance with publisher's policy at http://www.sherpa.ac.uk/romeo/issn/0956-540X/ (accessed 19/02/2014).
Publisher: Royal Astronomical Society
ISSN: 0956-540X
Date of First Compliant Deposit: 30 March 2016
Last Modified: 04 Jun 2017 02:52
URI: http://orca.cf.ac.uk/id/eprint/13096

Citation Data

Cited 50 times in Google Scholar. View in Google Scholar

Cited 61 times in Scopus. View in Scopus. Powered By Scopus® Data

Actions (repository staff only)

Edit Item Edit Item

Downloads

Downloads per month over past year

View more statistics