Confirmed speakers
(in the order of Company)
- “Digital Twins for Drilling Fluids", Mehrdad G Shirangi*, Reza Ettehadi, Roger Aragall (Baker Hughes)
- "Challenges and Solution to AI Application in E&P Decision-making", S.Q. Sun (C&C Reservoirs), S.Y. Wu (C&C Reservoirs)
- "Augmenting seismic noise attenuation using machine learning techniques", Chengbo Li*, Yu Zhang, and Charles C. Mosher (ConocoPhillips)
- “Scalable detection of early production profile trends from public data”, Arturo Klie*, Duc Vuong, Hector Klie (DeepCast.ai)
- "Accounting for ML weaknesses to design better Neural Networks for Petrophysics", F. Basier* (Emerson)
- "Seismic Data Management in the Cloud", Allan Chatenay (Explor) from Calgary“Deep learning for salt body detection: A practical approach”, Benjamin Consolvo* (Fairfield)
- “Data Analytics (AI/ML) for Value Based Insight of Hydrocarbon Assets and Improved Operational Efficiency”, Dr. Kalyan Saikia (Halliburton)
- “Reducing the cycle time in Seismic Processing Workflows: Opportunity and Applications using Machine Learning”, Pandu Devarakota (Shell)
- “A New Machine Learning Procedure for Generating Synthetic Shear Sonic Logs in Unconventional Reservoirs”, Ilia Chaikine (Sproule) from Calgary
- “Quantitative Oil Recovery Optimization: Turning Data into Actionable Knowledge”, G. Leahy* (Tachyus)
- "Building regional seismic from machine learning (ML) predicted well logs", CenYen Ong (TGS)
- "Multicloud Digital Evolution is the next phase of transformation", Sanjay Basu (Oracle)
Abstracts
“Digital Twins for Drilling Fluids", Mehrdad G Shirangi*, Reza Ettehadi, Roger Aragall (Baker Hughes)
Abstract:
In this work, we present advances to develop and apply digital twins for drilling fluids and associated wellbore phenomena during drilling operations. A drilling fluid digital twin is a series of interconnected models that incorporate the learning from geological uncertainty, complexity of drilling fluids and operational parameters to support well planning and real-time operations. We transform data collected throughout the years from all aspects of drilling fluids to value-added applications and include real-world conditions such as operation at extended temperatures and pressures, consideration for sensitive zones with tight control of the equivalent circulating density (ECD), and hole cleaning in deviated wellbores. With the advances presented in this work, cuttings bed height along the wellbore can be determined more accurately, and hence the ‘fluid plan’ together with operational parameters (such as mud flow rate and rate of penetration (ROP)) can be optimized to a higher level. In addition to the cuttings bed, an accurate workflow for monitoring of downhole fluid rheological properties at the rigsite is developed. Through the use of a digital twin, accurate high-pressure high-temperature (HPHT) properties can be determined at the rigsite, which can more easily enable efficient and safe operations, as opposed to lab experiments that are often conducted remotely. We demonstrate the application of automated machine learning (autoML) to represent computational simulations and lab experiments. We also use test datasets and rotating cross-validation methods to ensure accurate and robust results. In both cases, very accurate models were obtained, and point the way for the inclusion of more aspects of the drilling operation including ROP optimization, fluid-loss control, drilling fluid properties management, and power transmission.
--
"Challenges and Solution to AI Application in E&P Decision-making", S.Q. Sun (C&C Reservoirs), S.Y. Wu (C&C Reservoirs)
Abstract:
E&P companies strive to organize data, information and knowledge consistently to facilitate comparison, to learn lessons from the past and to better plan for the future. However, the lessons from past investments are seldom fully known or used due to lack of knowledge standards, changes in personnel, strategic priorities, cost controls and simply pressure on time. Artificial Intelligence (AI) including machine learning could be applied readily in many stages of E&P lifecycle. However, machine learning algorithms are best applied to structured and regularized data to gain meaningful results. Data preparation, regularization and standardization represent 90% of the efforts in many AI applications. To analyze and solve more complex subsurface problems at asset or portfolio level using AI, a large amount of effort would have to be made to standardize field and reservoir knowledge.We have conducted in-depth analysis and systematic documentation of the world’s most important fields and reservoirs and have established a comprehensive knowledge classification system to regularize reservoir knowledge for decision-making using AI tools. The regularized reservoir knowledge covers every known type of reservoir in all types of petroliferous basin around the world. Each documented field report details how the field was discovered followed by basin genesis and source rock, structure and trap definition, reservoir characteristics and fluid properties all the way to resources and recovery insights, including development strategy, reservoir management and improved recovery techniques applied and their outcomes. A comprehensive knowledge model, with 450 geological and reservoir engineering attributes, has been established at both reservoir and field level. Each attribute has been consistently defined and contains a set of standardized values following a pioneering classification system. Rigorous standards, consistent rules and clear guidelines have been applied to capture reservoir and field knowledge to form a global knowledge base.To facilitate translation of this knowledge base into real-time intelligence and insight, a software platform with a robust search engine and powerful set of analytics has been developed for searching, retrieving, characterizing and benchmarking E&P assets against global analogs. Our industry-leading knowledge base provides a solid foundation for the application of AI and machine learning technologies to optimize the E&P decision-making.
--
"Augmenting seismic noise attenuation using machine learning techniques", Chengbo Li*, Yu Zhang, and Charles C. Mosher (ConocoPhillips)
Abstract:
Land seismic acquisition has recently become much more efficient and economical thanks in part to the rapid adoption of advanced methods such as simultaneous sourcing with point sources and point receivers (Mosher et al., 2017). Large volumes of data can be acquired in a relatively short period of time with simultaneous vibroseis operations. In seismic processing, however, this creates a bigger challenge for noise attenuation due to additional noise sources and the lack of stack of multiple sweeps in the field. Heterogeneities in the near surface can cause back-scattering and further complicate the noise issue. Traditional noise attenuation methods often use different characteristics in frequency, wavenumber or other transform domains to separate signal from noise (e.g., Soubaras, 1995). These methods based on fixed transforms are highly efficient, but often make rigid assumptions about noise characteristics. For land data with complex noise patterns, the lack of adaptability results in an unavoidable trade-off between the preservation of signal and the amount of noise removed. Machine learning has gained increasing popularity in geophysics in recent times. The learning approach provides a promising alternative for seismic denoising by adapting to the local wavefield morphology. For example, Beckouche and Ma (2014) proposed a joint learning and denoising approach based on a variational sparse-representation model; Chen et al. (2016) combined dictionary learning with a non-adaptive basis to better handle seismic features for denoising; Li et al. (2018) introduced a weak signal recovery technique using unsupervised learning and sparse inversion. These methods offer good adaptability and are less dependent on the variance of the noise. However, most of these methods merely focus on Gaussian or unstructured noise, and become inadequate to attenuate coherent noise generated from an external noise source or a complex near surface. We investigate two dominant noise types we observed on a land seismic dataset and introduce machine learning techniques we employed to attenuate those. In both cases, we combine machine learning with traditional noise attenuation methods to achieve effective noise reduction and signal/amplitude preservation simultaneously.
--
“Scalable detection of early production profile trends from public data”, Arturo Klie*, Duc Vuong, Hector Klie (DeepCast.ai)
Abstract:
In this presentation, we introduce an automated forecasting workflow that identifies early production profile trends in both space and time. Our automated workflow is built into a platform that helps accelerate the following components: (1) data integration; (2) model building; (3) production and economic forecasting and, (4) optimization. As new data becomes available, the system reactively updates its analytics and flags any salient trends in the data. The data integration involves cleaning and normalizing public and proprietary data sources. Modeling and forecasting rely on a physics-informed AI engine to automatically extract drivers and generate faster and more accurate representation of production and economic trends. These trends and other metadata are georeferenced onto a map to an intuitive way so users can quickly identify and analyze hotspots. The optimizer uses the model and data to analyze the evolution patterns of several spatiotemporal trends hidden within the production profiles of several wells and provides the user a ranked list of prospective locations in the map. By using historical data for both conventional and unconventional assets, we are able to demonstrate an approach that provides a powerful vehicle for automatic surveillance and detection of niche prospects hidden within thousands of wells and decades of production history.
--
"Accounting for ML weaknesses to design better Neural Networks for Petrophysics", F. Basier* (Emerson)
Abstract:
In recent years, Deep Learning has proven able to tackle difficult E&P problems,but is still to become an industry standard. On real datasets, challenges such as lack of data, poor distribution or the quality of training labels are reducing prediction accuracy, while networks errors are not correlated with confidence scores. In this paper, we propose a new approach to highlighting network errors, in order to allow petrophysicists to focus on these areas.The approach is illustrated on the challenge of grain size prediction from microresistivity data. It uses a Neural Network Ensemble of networks, each dedicated to one grain size category. Each network consists of a Multiple Input Multiple Output (MIMO) Network, allowing multiple data sources (such as microresistivity and GR) to train the networks even when data is missing. Combined with a sliding window mechanism, more than 60 different predictions are realized per depth sample, allowing a statistical treatment of these NN outputs. A prediction can be realized alongside an uncertainty curve, highlighting difficult areas for study by the interpreter.This approach combines Machine Learning prediction and uncertainty analysis to reshape, using deterministic methods, the “black box” predictions of the Neural Network. This provides a more accurate methodology than conventional Machine Learning, which is also aware of its own inaccuracies. It automates most of the interpretation work while highlighting areas where machine learning is not yet able to properly perform, allowing the user to focus on the problematic areas to provide a more precise interpretation.
--
"Seismic Data Management in the Cloud", Allan Chatenay (Explor) from Calgary
Abstract:
The great seismic migration has begun. The dramatic cost savings of storage in the cloud is certainly captivating. But corporate enterprise data in the cloud is far ahead of subsurface data in the cloud. And well related data types are ahead of seismic data in moving to the cloud.This can be attributed to several factors. Two most prominent factors are that a great deal of seismic data is being held hostage in legacy storage houses, and the complexity of seismic data relationships far exceeds that of other data types, making seismic migration to the cloud particularly challenging.This talk with be an interactive discussion that walks the audience through several case study examples of seismic data migrations, the struggles and challenges faced, the solutions, and the results/outcomes. Data liberations strategies will also be discussed. Lastly, real life scenarios of actually using data in the cloud will be discussed.
--
“Deep learning for salt body detection: A practical approach”, Benjamin Consolvo* (Fairfield)
Abstract:
Interpreting salt bodies in the subsurface is a challenging manual task that can take weeks to complete. Obtaining accurate picks of salt is very important, because errors in the placement of salt can result in severe degradation of the seismic image. To meet the challenges of speeding up imaging workflows and retaining accurate salt picks, we evaluate three deep learning approaches: a 2D window-based convolutional neural network, a 3D window-based convolutional neural network, and finally a 2D “U-Net” approach. A 3D seismic volume from the deep-water field Julia in the Gulf of Mexico was used to test these approaches. The Julia field has complex salt structures with overhangs and inclusions, and the thickness of salt can reach up to 5 km. The U-Net architecture proved to be the most accurate of the three methods tested, predicting the placement of salt at 98% accuracy, as compared to the human interpretation. Beyond accuracy, U-Net also proved to be the fastest, requiring only 3.5 hours to predict salt on the 3D seismic volume. The results presented here along with other recent studies of deep learning for salt interpretation represent a clear shift in the seismic interpretation workflow.
--
“Data Analytics (AI/ML) for Value Based Insight of Hydrocarbon Assets and Improved Operational Efficiency”, Dr. Kalyan Saikia (Halliburton)
Abstract:
The expectation from data analytics in E&P industry is to make effective processes and efficient workflow with respect to maximized hydrocarbon recovery while operating under the business constraints. This is always augmented by automation of process and use of AI/ML together. In reality, E&P operations such as G&G interpretation, drilling, completion and production involving hydrocarbon recovery is very complex and all of these are interdependent. Thus, the realization of maximum recovery with minimum cost can only happen when these operations are made dynamic and analyzed through AI and ML techniques. The main objective of this talk is to show how data analytics through AI/ML in E&P industry can guide us to make accurate and fast decision and achieve maximized hydrocarbon recovery, which is the goal for every E&P companies currently. We will also touch upon how big data analytics is being blended to improve our decision-making and how cloud technology enables us to deploy and manage technology at a rapid pace.
--
“Reducing the cycle time in Seismic Processing Workflows: Opportunity and Applications using Machine Learning”, Pandu Devarakota (Shell)
--
“A New Machine Learning Procedure for Generating Synthetic Shear Sonic Logs in Unconventional Reservoirs”, Ilia Chaikine (Sproule) from Calgary
Abstract:
During hydraulic fracture treatment, the rock mechanical properties surrounding a horizontal wellbore dictate how fractures propagate though the formation which heavily influences post-fracture production performance. Yet, despite the huge number of horizontal wells drilled and completed to date there is still a poor understanding of the relationship between rock mechanics and production performance. The objective is to determine this relationship from data analytics, more specifically, convolutional-recurrent neural networks, by modeling the DTS log in a tight rock formations.
Mechanical rock properties are estimated with three sets of log measurements: shear sonic travel time (DTS), compressional sonic travel time (DTP) and bulk density (RHOB). Due to cost and time to process. DTS logs are often missing. In this study, a hybrid convolutional-recurrent neural network (c-RNN) was developed to predict synthetic DTS curves. C-RNN have the advantage that they learn sequential data unlike traditional neural network (ANN) which do not have this capability. The synthetic DTS curve was generated by using five inputs: x, y and z coordinates, RHOB and DTP for every point along the wellbore.
This study focuses on the Montney formation in Alberta, Canada. Out of the 180 vertical and deviated wells in the study area, only 14 wells had DTS measurement available, thus the only suitable method for determining experiment accuracy was the “leave-one-out” cross-validation method. In this method, 13 wells were used as training data with one well as a test, the experiment was run a total of 14 times (one for each test well) and the results of all 14 experiments were examined and compared. Using the c-RNN the average mean absolute percent error (MAPE) along the entire Montney formation for the 14 wells came out to 1.2%. This result was superior when compared to that of a basic ANN, simple baselines, and empirical correlations. The results demonstrated that the new c-RNN method is a cost effective and fast alternative to running DTS logs for new wells and can be applied to any formation as along as there is a sufficient number of existing wells, typically < 20, with DTS logs for verification.
The novelty of the research reported is on the use of hybrid convolutional-recurrent neural networks for modelling the DTS log in a tight rock formation. The approach is widely applicable to other tight rock resources and is relatively easy to implement.
--
“Quantitative Oil Recovery Optimization: Turning Data into Actionable Knowledge”, G. Leahy* (Tachyus)
Abstract:
Reading the headlines, one might be forgiven for preparing for an impending invasion of machines - Artificial Intelligence and Machine Learning are starting to creep out of specialized domains and into the public psyche. Here, I provide a basic overview of different AI and ML approaches, and the kinds of applications where professionals in Oil and Gas may soon run into an AI-assisted application. Then, we’ll dive deep into the reservoir and explore how Tachyus is using AI to optimize producing assets, and where we expect the field will head in the next few years.
--
"Building regional seismic from machine learning (ML) predicted well logs", CenYen Ong (TGS)
Abstract:
Accurate characterization of the geometry and composition of rocks and fluids in the subsurface layers of the earth is important for making drilling decisions and planning production strategies in the energy industry. Different methods are used to measure the properties of the subsurface including seismic data acquisition and borehole sensing tools. Acquisition of seismic data enables geoscientists to build a 3D model of the subsurface, however, the costs and effort can be prohibitive for onshore prospects.A cost-effective alternative is to build a lithostratigraphic model from lithofacies that have been identified from well logs. The motivation to build a lithostratigraphic model necessitates the identification of formation tops for a set of densely sampled well logs in a large geographical region. Identifying lithofacies from well logs is a time-consuming task for geologists. Moreover, most well recordings are incomplete due to operator errors, missing tools or tool failure. Gaps in well recordings complicate the task of formation tops identification.We propose a machine learning workflow for well log curve completion. Cross sections of completed well log curves can be used to build initial velocity models which can be corroborated with sparsely acquired seismic data.
--
"Multicloud Digital Evolution is the next phase of transformation", Sanjay Basu (Oracle)
Abstract:
This presentation will delve into the positives and the negatives of digital transformation. As the name “digital transformation” suggests, sometime the enterprise replaces a complex manual process with another complex digital processes which takes months if not years to implement and resulting in newer issues. In this paper I would explore how we can use modern technologies and multicloud solutions to help an enterprise evolve according to market requirements and ever changing business needs.