Implementation of seismic data quality characterisation using supervised deep learning
Joshua Thorp A C , Krista Davies B , Julien Bluteau A and Peter Hoiles BA Searcher Seismic, 15 Rheola Street, West Perth, WA 6005, Australia.
B Discover Geoscience, 15 Rheola Street, West Perth, WA 6005, Australia.
C Corresponding author. Email: j.thorp@searcherseismic.com
The APPEA Journal 60(2) 784-788 https://doi.org/10.1071/AJ19040
Accepted: 24 February 2020 Published: 15 May 2020
Abstract
Success and failure in oil and gas exploration is heavily dependent upon the quantity and quality of the subsurface data used during the analysis. Traditionally, the quality of seismic data has been characterised through the domain expertise of geoscientists and through traditional analytical methods, such as frequency decomposition, autocorrelations and signal to noise ratios. Unfortunately, due to the non-linearity and spatial heterogeneity of the data, these analytical methods can be unreliable or difficult to interpret by geoscientists. Machine learning, in particular deep learning, has been shown to analyse subsurface data in a multitude of ways by integrating domain expertise with powerful non-linear mathematical analysis. Thus, it seems that deep learning could be a natural fit for evaluating seismic data quality. In this paper we outline the issues of what is considered good- and bad-quality seismic data and introduce the labelling scheme that was implemented for the supervised learning. Examples of the labelling are shown, as well as the challenges encountered during the extensive labelling process. Neural network architecture testing and validation of the architecture are shown. Finally, applications to three-dimensional datasets on the North West Shelf, how the results can be interpreted and potential further avenues of research are discussed.
Keywords: geoscience, machine learning, subsurface.
Joshua Thorp is the Geoscience Manager at Searcher Seismic based in Perth, Australia. Joshua has a BSc in Pure Mathematics from the University of Calgary and started his career at CGG in 2007. Joshua worked as an expat in Houston, France, Angola and Brazil with CGG as a project leader on time processing and PSDM seismic processing projects. In 2012, Joshua joined Searcher to manage and QC the seismic processing projects and implement Quantitative Interpretation (QI) and Amplitude versus Offset (AVO) analysis workflows. As Geoscience Manager, Joshua has led the development of the Saismic platform, which has been purpose built for training and applying deep learning algorithms on global seismic datasets. Joshua is a member of the European Association of Geoscientists and Engineers (EAGE), the Society of Exploration Geophysicists (SEG) and PESA. |
Krista Davies is a Principal Geoscientist at Discover Geoscience. She holds a BSc(Hons) Geology (1st) from the University of Technology, Sydney and an MSc in Environmental Science specialising in Inland and Marine Aquatic Systems from Edith Cowan University, Western Australia. Krista has worked for several oil and gas companies over the past 25 years, including Woodside Energy, Shell Development Australia and Ophir Energy, in all areas of exploration and new ventures assessments. Krista has extensive experience in South-east Asian and African basins, and has a keen interest in sequence and seismic stratigraphy and deepwater geomorphology, as well as agility training her Australian Kelpie puppy. Krista is a member of PESA, the American Association of Petroleum Geologists (AAPG) and the Society of Sedimentary Geology (SEPM). |
Julien Bluteau is a geophysicist and data scientist at Searcher Seismic. Julien graduated from the French INSA de Toulouse school with an MSc in Applied Mathematics, and then joined the oil and gas industry as a geophysicist. With more than 10 years of experience, Julien has now decided to focus on enabling digitalisation for geoscience. His work involves the integration of geoscience with big data and data science pipelines. |
Peter Hoiles is a geoscientist at Discover Geoscience. Peter holds a PhD and BSc from The University of Melbourne in geology and a BSc(Hons) from James Cook University. Peter has worked in the oil and gas industry for the past 8 years in both Perth and Brisbane as a quantitative interpretation and depth imaging geophysicist, as well as his more recent role as a geoscientist. Prior to joining the oil and gas industry, Peter worked as an engineering geologist in construction and mining. Peter is a member of PESA and has recently joined the PESA WA committee as assistant treasurer. |
References
Bekara, M., and Day, A. (2019). Automatic QC of denoise processing using a machine learning classification. First Break 37, 51–58.Kleiss, E., and Wall, J. (2018). QC of marine broadband deghosting using match filters. In ‘Proceedings of the 80th EAGE Conference and Exhibition, Copenhagen, Denmark, 11–14 June 2018’. WE A10 07. (European Association of Geoscientists & Engineers: Houten, Netherlands.)
Martinez-Sansigre, A., and Ratcliffe, A. (2014). A probabilistic QC for cycle-skipping in full waveform inversion. In ‘SEG Technical Program Expanded Abstracts 2014’. pp. 1105–1109. (Society of Exploration Geophysicists.)
Paternoster, B., Lys, P. O., Crouzy, E., and Pagliccia, B. (2009). Seismic quality monitoring during processing for reservoir characterization. In ‘SEG Technical Program Expanded Abstracts 2009’. pp. 1840–1844. (Society of Exploration Geophysicists.)
Priezzhev, I. I., Veeken, P. C. H., Egorov, S. V., and Strecker, U. (2019). Direct prediction of petrophysical and petroelastic reservoir properties from seismic and well-log data using nonlinear machine learning algorithms. The Leading Edge 38, 949–958.
| Direct prediction of petrophysical and petroelastic reservoir properties from seismic and well-log data using nonlinear machine learning algorithms.Crossref | GoogleScholarGoogle Scholar |
Qian, F., Yin, M., Liu, X. Y., Wang, Y., Lu, C., and Hu, G. M. (2018). Unsupervised seismic facies analysis via deep convolutional autoencoders. Geophysics 83, A39–A43.
| Unsupervised seismic facies analysis via deep convolutional autoencoders.Crossref | GoogleScholarGoogle Scholar |
Tishchenko, I. (2016). Different methods of QC the low frequency content of broadband data for reservoir characterization. In ‘Proceedings of the 78th EAGE Conference and Exhibition, Vienna, Austria, 30 May–2 June 2016’. WS01 C01. (European Association of Geoscientists & Engineers: Houten, Netherlands.)
Yu, S., and Ma, J. (2018). Deep learning for denoising. In ‘Proceedings of the 2018 CPS/SEG International Geophysical Conference, Beijing, China, 24–27 April 2018’. pp. 461–464. (European Association of Geoscientists & Engineers: Houten, Netherlands.)
Zhao, X., Lu, P., Zhang, Y., Chen, J., and Li, X. (2019). Swell-noise attenuation: a deep learning approach. The Leading Edge 38, 934–942.
| Swell-noise attenuation: a deep learning approach.Crossref | GoogleScholarGoogle Scholar |
