Case studyGeochemical property modelling of a potential shale reservoir in the Canning Basin (Western Australia), using Artificial Neural Networks and geostatistical tools
Introduction
The identification of the distribution of organic geochemical parameters is one of the main prerequisites to a successful exploration campaign. Traditionally, drill cores are obtained from the source rock Formation and subjected to organic geochemical analysis to reveal the amounts of organic content (TOC), the source rock potential (S2) and the source rock quality (HI) form well locations. However, there is generally a dearth of core samples within these source rock intervals from older wells or often incomplete geochemical laboratory analysis (Huang and Williamson, 1996) from some cored and analysed wells. Geochemists often turn to drill cuttings for analysis as this is widely available, however, accurate depth matching and contamination of the cuttings has been a long-standing issue. In consequence, several methods have historically been proposed to obtain some geochemical parameters from well logs, most notably the (Schmoker, 1983), (Meyer and Nederlof, 1984) and (Passey et al., 1990) methods of estimating the total organic carbon (TOC) content from well logs. Yu et al. (2017) compared the TOC estimates of the Canning Basin shales, using the Schmoker and Hester (1983) as well as the Passey et al. (1990) and concluded that these methods do not work well in moderate to low TOC shales – as we have in the Canning Basin.
More recently, several authors have utilized intelligent systems such as Artificial Neural Networks, and Neuro-Fuzzy Logic to estimate TOC content from petrophysical well logs, including Kadkhodaie-Ilkhchi et al., 2009, Huang and Williamson, 1996, Kamali and Allah Mirshady, 2004 and Rezaee et al. (2007) amongst others. The Artificial Neural Networks (ANN) is an intelligent system that is used to solve nonlinear, complex problems, through pattern recognition, which mimic the biological processes of the human brain (Dowd and Sarac, 1994). With this, a supervised training is utilized, where the user “trains” the system with the desired output. An error between the output and the desired output is computed and fed back into the system (Huang and Williamson, 1996), and the weights are adjusted until the approximate desired output value is achieved. For each geochemical property prediction, an algorithm for the property is designed and trained using petrophysical data from gamma ray, sonic, resistivity and density logs as the input data, and the measured property values as the desired output. A detailed methodology for this is recorded in (Boadu, 1997, Huang and Williamson, 1996). The neural network system is particularly useful in studies such as this, due to its adaptability in learning by example and its ability to generalize.
The Canning Basin is relatively underexplored, with limited geochemical information available, partly due to limited number of wells drilled to date in the Basin, especially penetrating the Formation of interest – the Ordovician Goldwyer Formation. In this study, ANN is utilized to predict organic geochemical data in two wells with no laboratory measured geochemical data, and four (4) wells with limited laboratory measured data. The distribution of these geochemical properties within the ∼200 m–300 m thick Goldwyer 3 Shale member is subsequently modelled across the Broome Platform of the Canning Basin.
Section snippets
Geological settings
The Canning Basin developed as an extensional intracratonic sag basin between the Kimberly and Pilbara blocks in the early Ordovician. The basin contains two major north-westerly trending troughs, separated by a mid basinal arch known as the Broome Platform and Crossland Platforms (Fig. 1), and flanked by marginal shelves. The Central Arch contains series of down stepping troughs which were active at different times of the basin evolution.
As a result of fault block movements, sediment
Methodology
The methods employed in this study can be subdivided into four main steps.
- a.
Well log to geochemical data compilation
- b.
Identification of relationship between well logs and geochemical property
- c.
Network Training
- d.
Geochemical Property Model
Network training and pattern recognition
In order to establish the relationship between the well logs and laboratory derived geochemical data, cross plots were used to identify the logs with strong relationships with each geochemical property as the training set. The results from this served as an input for the supervised network training to produce continuous geochemical logs (outputs). The dataset was divided into three clusters which are 70% of the data to train the network, 15% to control the networks performance and 15% to test
Discussion
The information obtained from well log data, coupled with laboratory measured geochemical data has helped in understanding the spatial distribution of organic matter in the studied sub-basin. This study leverages on the distinct response of logging tools to disseminated organic matter in sediments.
In this study, well logs with more direct relationships with geochemical data has been used to generate the input to train the network model.
Different petrophysical logs have different responses to
Conclusion
The Ordovician Canning Basin is one of the least explored Palaeozoic systems in the world, with limited information from the Ordovician shales. In this basin, well distribution and consequently data availability is one of the major exploration challenges, thereby, in cases as such, mathematical models such as artificial intelligent systems is not uncommon. Here, from the available petrophyscial data, Artificial Neural Networks have been used to achieve a quantitative and qualitative source rock
Author contribution
Lukman Johnson: Study conception, design, data analysis and interpretation, manuscript writing.
Reza Rezaee: Provided significant advice in the study, contributed to the manuscript writing and editing and revision.
Ali Kadkhodaie: Provided advise on data analysis and clean up, study conception and valuable experience in using both computer software (for the neural networks and 3D models).
Gregory Smith: Provided advise on data analysis, and valuable experience in the 3D models and interpretation.
Acknowledgements
The Authors wish to thank Schlumberger for the provision of the Petrel software. LMJ acknowledges the contribution of the Australian Government in supporting the research through the Australian Postgraduate Awards, and the Curtin University Postgraduate Scholarship. Osainaye Opeyemi and Mohammed Oloyede are greatly acknowledged for the discussions on Artificial Intelligence and Alexander Oshodi is thanked for discussions on the manuscript.
References (19)
- et al.
Hydrocarbon biomarkers from Ordovician sediments and the fossil alga Gloeocapsomorpha prisca Zalessky 1917
Geochem. Cosmochim. Acta
(1987) - et al.
Artificial neural network modelling as an aid to source rock characterization
Mar. Petrol. Geol.
(1996) - et al.
A committee neural network for prediction of normalized oil content from well log data: an example from South Pars Gas Field, Persian Gulf
J. Petrol. Sci. Eng.
(2009) - et al.
Total organic carbon content determined from well logs using ΔLogR and Neuro Fuzzy techniques
J. Petrol. Sci. Eng.
(2004) A new method for TOC estimation in tight shale gas reservoirs
Int. J. Coal Geol.
(2017)Rock properties and seismic attenuation: neural network analysis
Pure Appl. Geophys.
(1997)- et al.
- et al.
Well site geochemistry by programmed pyrolysis
- et al.
A neural network approach to geostatistical simulation
Math. Geol.
(1994)
Cited by (38)
Evaluation of shale oil and gas plays - Part I: Shale reservoir property modelling of the North Sea Kimmeridge Clay Formation
2024, Marine and Petroleum GeologyDeep sequence model-based approach to well log data imputation and petrophysical analysis: A case study on the West Natuna Basin, Indonesia
2023, Journal of Applied GeophysicsA reliable Bayesian neural network for the prediction of reservoir thickness with quantified uncertainty
2023, Computers and GeosciencesEstimating total organic carbon (TOC) of shale rocks from their mineral composition using stacking generalization approach of machine learning
2023, Upstream Oil and Gas TechnologyTotal organic carbon content logging prediction based on machine learning: A brief review
2023, Energy GeoscienceTOC prediction using a gradient boosting decision tree method: A case study of shale reservoirs in Qinshui Basin
2023, Journal of Petroleum Science and EngineeringCitation Excerpt :Lu et al. (2021) used a back propagation (BP) neural network model to predict organic carbon content in the Lunpola Basin with good fitting results. Johenson et al., 2018 used artificial neural networks and geostatistical tools to model geochemical properties in the Canning Basin of Australia and met the basic accuracy requirements. Zheng et al. (2021) utilized artificial neural networks (ANN) to evaluate hydrocarbon-rich shales in four basins of China and Canada.