GEO ExPro

Artificial Intelligence and Petroleum System Risk Assessment

Using artificial intelligence to assess hydrocarbon charge risks.
This article appeared in Vol. 17, No. 1 - 2020

Advertisement

AI and Petroleum System Risk Assessment

At this turning point for the oil and gas industry, with escalating competitiveness and a need to optimise productivity, petroleum companies must pick their new exploration projects carefully, making sure the returns have the potential to be as high as possible, while keeping the costs and risks low. Numerical methods such as petroleum system modelling (PSM) or forward stratigraphic modelling have proved that they can play a key role in the assessment and mitigation of exploration risks in both mature and frontier areas (e.g. Chenet et al., 2014; Bryant et al., 2014). Formerly exclusively accessible to major oil and gas companies, these techniques are now part of the exploration workflows commonly used in most E&P teams. 

  • Global workflow for an accurate and efficient risk analysis embedding geological knowledge and concepts. © Kognitus.

However, in order to keep up with the evolution of geological knowledge through a prospect’s life cycle, these powerful simulation methods can be very labour-intensive, which hampers the regular updating of the models with the most recent data and information. Additionally, as the present generation of exploration experts approaches retirement, companies are facing the problem of retaining expertise while simultaneously reducing costs and increasing effectiveness.

Taking these issues into consideration, a new solution that takes advantage of the investments made in numerical modelling has been developed by Kognitus, a technology company that specialises in the use of analytics and AI to help O&G companies gain insights from subsurface data. This system is based both on knowledge gleaned from experts, which enhances and guides the determination of the uncertainties, and on machine learning (ML) techniques, which provide maps of risks at exploration scale and in timeframes compatible with operational studies. Essentially, the solution takes the explorationist’s petroleum systems model as an input from which to infer geologically meaningful uncertainties.

The process starts with the generation of a learning base from the results of physics-based simulations, which will support the training of an ML algorithm. Once the trained algorithm achieves a good prediction capability, assessed through a set of simulations kept for testing it, the solution develops a set of statistical tools that will automatically or interactively produce risk and sensitivity analyses. This solution is a first step in making basin modelling useful throughout the full life cycle of E&P assets by continuously updating the geological model as new data is acquired.

Geological Expertise Remains Key in Risk Assessment

Risk assessment workflows in exploration still predominantly rely on subjective and qualitative expert-based evaluations which, in striving for consistency, follow standardised procedures and refer to tabulated values (Milkov, 2015). As geology must remain at the core of the process, this new solution provides a systematic way to estimate quantitative and geologically meaningful uncertainties to be used in PSM, based on its embedded geological expertise. The petroleum system, with all its elements and processes, is the cornerstone of the approach. Expert systems are designed to help the explorationist quantify uncertainty on key parameters through the consideration of geological concepts, with all the information integrated in the numerical model (Hacquard, 2018; Ducros et al., 2018).

  • Expert systems are embedded in the solution to provide geologically meaningful uncertainties on key elements and aspects of petroleum systems. © Kognitus.

The approach also offers the first effective tools to include complex uncertainties, such as source rock reactivity, which are usually discarded not due to their low impact on petroleum systems but because we lack the tools or knowledge to account for them. It strengthens the role of PSM as an integration tool by creating a consistent link between information coming from different exploration facets; for example, palaeobathymetry, source-rock thickness and horizon ages – all generic input data for PSM – can be used together to provide quantitative estimates of uncertainty on source rock richness in a consistent first guess.

Quantitative Petroleum System Risk Assessment

After defining quantitative uncertainties, the Kognitus approach uses the best physics-based simulation and ML techniques to provide quantitative risk analysis in an adapted format for petroleum system assessment (output risks provided on maps, 2D sections, well logs, etc.), in a time frame which remains consistent with operational projects yet preserves accuracy.

The software relies on an extension of the ‘proxy models’ approach (also called surface responses), initially used in meteorological modelling, which is well suited to the study of spatial outputs (Gervais et al., 2018). It combines several ML techniques, such as regression and clustering, to learn the main aspects of the petroleum system. After the learning phase has been undertaken using a small (usually less than 100) set of simulations, it is possible to perform an interactive, rigorous and accurate statistical analysis which is required for the quantitative risk assessment and sensitivity analysis (see figure overleaf). This technique leads to a significant reduction in the time required to undertake risk analysis compared to the brute force of techniques such as direct Monte Carlo sampling, or derived Markov Chain Monte Carlo approaches, which require thousands of simulations.

  • Detailed illustration of the machine learning workflow. During step 1 a learning base is generated using the physics-based simulator. A principal component analysis (PCA) then provides a set of uncorrelated maps (components of the PCA), with decreasing variance, on which to project the simulation results. A reduction of the dimension is obtained by keeping the first components of the PCA (explaining more than 95% of the variance of the simulations for instance). The machine learning takes place in the 4th step to learn the relationship between uncertain parameters and coefficients of the projection. Once the relationships are determined they can be used to perform rigorous statistical analyses on the results of interest (step 5). © Kognitus.

Lastly, these results, when used in a Bayesian framework, can assist in updating the petroleum system model during the whole exploration life cycle. This is set to be a major game changer, as it would speed up the process of reevaluating risks and resources and help identify satellite fields while acquiring new data.

Applying the Machine Learning Quantitative Risk Analysis

The results of this new method can be provided in any format used by explorationists. It is therefore possible to get access to maps of risk (P10, P50 and P90) on every simulated property, such as source-rock thermal maturity, overpressure or volume of generated petroleum. Since the method also uses statistical techniques to compute sensitivity analysis, it gives a better understanding of the key controlling factors of the petroleum systems that contribute uncertainty to the risk. It is also possible to use the approach for pore pressure prediction through logs using P10/P90 estimates of pressure at the planned position of a well.

  • Risk analysis results are computed on different classical formats used by explorationists. © Kognitus.

The complete methodology was applied to assess the deep petroleum system of the frontier Levantine Basin in the Eastern Mediterranean (Ducros, 2019). Significant gas discoveries have been made there since 2006 in the Oligo-Miocene section, where there is a proven biogenic system. While Cretaceous source rocks are known to outcrop in Lebanon, their presence in the deep basin is still speculative, so the objective was to study the potential for a deeper thermogenic system and to look at where hydrocarbons could have possibly migrated towards the Oligo-Miocene reservoirs. Uncertainty on ten elements was taken into account: three on source rock characteristics (reactivity, depth and petroleum potential); two on the thermal system (nature and thickness of the crust); three on the hydrodynamic system (permeability of the regional Eocene seal, extension of the Oligo-Miocene turbiditic system and time-to-depth conversion); and two on the Messinian Salinity Crisis (timing and duration).

The results showed that there is no doubt on the maturity level of the Cretaceous source rocks in the deep offshore basin and onshore Lebanon, which were found to be over-mature and immature respectively, but that there is a high risk along the margins (Cyprus, Lebanon and Israel) and in the basin offshore Israel. Sensitivity analysis demonstrated that the nature and reactivity of the source rocks would be the key elements to be studied in order to de-risk this aspect of the Levantine Basin petroleum system.

The Kognitus System for Assessing Petroleum System Risk

The Kognitus system for assessing risk has proved to be both time-effective and robust. It provided highly valuable results for the exploration of the still poorly known but promising Levantine Basin by giving access to the identification of sweet spots and presenting clues on how to further reduce risks. The approach can therefore be used in early basin exploration to evaluate the source rock maturity, or to assess the whole petroleum system, providing maps of risks compatible with common risk segment mapping methods, or for continuous model update.

This methodology could pave the way to a standardisation of quantitative risk analysis within a portfolio, as it provides a framework in which risks can be reviewed using quantitative and standardised uncertainties. It could also be used to enhance the role of PSM by re-evaluating risks and resources as new data is acquired. Thus, this technique could lead to a major increase in the use of PSM during all phases of basin exploration and production.

New approaches like this showcase the importance of human geological expertise, which remains at the heart of any successful AI approach.

Further Reading on Artificial Intelligence and Machine Learning in Oil and Gas Exploration

Advancing Geophysical Interpretation in Oil and Gas Exploration
Gehrig Schultz, Chris Tucker and Kirsty Simpson; EPI
Geophysics must change – but it must also remain meaningful.
This article appeared in Vol. 16, No. 3 - 2019

Artificial Intelligence and Seismic Interpretation
James Lowell, Peter Szafian and Nicola Tessen; GeoTeric
The key to all seismic interpretation is the interpreter’s experience and knowledge, so why should artificial intelligence change that? The reality is, it shouldn't.
This article appeared in Vol. 16, No. 2 - 2019

Using AI to Estimate Hydrocarbon Prospect Size
Halfdan Carstens
With the recent advances in technology and adoption of machine learning in geosciences, is it time for AI to take a hand in estimating hydrocarbon prospect sizes?
This article appeared in Vol. 15, No. 6 - 2018

Artificial Intelligence in Oil & Gas Production
Bjørn-Erik Dale & Vidar Uglane; Solution Seeker AS
Solution Seeker, a Norwegian tech start-up and spin-off from the Norwegian University of Science and Technology, is developing the world’s first artificial intelligence for real-time production optimisation.
This article appeared in Vol. 15, No. 3 - 2018

Advertisement

Related Articles

Reservoir Management North America

The Marcellus Shale

The gas potential of the Marcellus Shale of the Appalachian Basin has made it an attractive investment opportunity. Horizontal wells and hydraulic fracturing have boosted production rates and recoverable reserves in these low permeability rocks.
Intro464 thumb

Reservoir Management Worldwide

Leveraging Complex Geologies

The ability of reservoir models to leverage seismic and accommodate complex geologies ensures robust and reliable models, improved decision-making, more justifiable investments, and improved field performance.
Thumb thumb