|Theme||Visible||Selectable||Appearance||Zoom Range (now: 0)|
Introduction Petroleum data analytics is a solid engineering application of data science in petroleum-engineering-related problems. The engineering application of data science is defined as the use of artificial intelligence and machine learning to model physical phenomena purely based on facts (e.g., field measurements and data). The main objective of this technology is the complete avoidance of assumptions, simplifications, preconceived notions, and biases. One of the major characteristics of petroleum data analytics is its incorporation of explainable artificial intelligence (XAI). While using actual field measurements as the main building blocks of modeling physical phenomena, petroleum data analytics incorporates several types of machine-learning algorithms, including artificial neural networks, fuzzy set theory, and evolutionary computing.
This course will play a crucial role for the enthusiasts of this technology to identify the scientific realities associated with the foundation of Artificial Intelligence and Machine Learning and its true application in Petroleum Data Analytics. Want to be knowledgeable with the most up-to-date and accurate AI and Machine Learning technology? This class will get you there!
Expectations from data analytics in the upstream sector continue to evolve. Although the number and diversity of applications continue to increase, the adoption at the assetwide level faces well-known barriers and challenges. Massive data collected from field instrumentation remain unexploited to the full potential because of misalignment between people, processes, and technologies.
To begin with, a lack of consistent data-integration infrastructure (e.g., between secure field networks and corporate data repositories) makes accessing and using data in real time difficult. It is disappointing that small, missing, or broken links can interrupt the whole data flow. For example, data from a fully instrumented field may not be leveraged in the data-analytics domain because of a lack of an integrated gateway between two networks within the same company.
Furthermore, although the number of data-analytics applications continues to grow, data-assimilation algorithms are still gaining momentum in being accepted as business as usual. A disconnect exists between the available operational challenges and the applications being developed, in some cases because of data completeness and availability and in other cases because of a mismatch between the technology-driven organizations and the asset owners, operators, and decision makers.
Finally, adopting such business processes requires dedicated people with the expertise to exploit the data, with business-intelligence skills and experience with data analytics or automated-work-flow building. The competency gap does not appear to be reduced despite the avalanche of data-analytics self-training programs; it remains a highly scientific mathematical approach to solve problems, accessible to few.
The good news is that success stories are everywhere. Traditionally unsolved upstream challenges have the potential to be addressed by self-clustering and dimensionality-reduction techniques that uncover hidden patterns and trends in data from drilling, production, completion, seismic, and logs.
As an example of success, engineers can gather data from thousands of wells and automatically (without previous knowledge or target outputs) group these wells into a small subset that is similar or related through different attributes, such as rates, reservoir quality, contact area, completion parameters, or location. These applications clearly have benefited operators and service companies to focus quickly on value-added decisions, which was not possible previously unless senior experience was considered.
Assets and operators should strive to understand the value that predictive analytics can provide to their bottom line. I invite you to review the selection of key papers in this feature.
In addition, I would like to invite you to try the new SPE advance search engine (https://search.spe.org), which uses advanced analytics and artificial-intelligence techniques to offer a better experience for finding and analyzing information on SPE.org, PetroWiki, and OnePetro.
Recommended additional reading at OnePetro: www.onepetro.org.
SPE 190812 Status of Data-Driven Methods and Their Application in the Oil and Gas Industry by Karthik Balaji, University of North Dakota, et al.
SPE 187030 Logging Facies Classification and Permeability Evaluation: Multiresolution Graph-Based Clustering by Xinlei Shi, China National Offshore Oil Corporation, et al.
OTC 28002 Multiscale Ensemble-Based Data Assimilation for Reservoir Characterization and Production Forecast: Application to a Real Field by Alexandre de Lima, CGG, et al.
While many other industries have experienced tremendous benefits over the last few decades, adoption of data-driven analytics is still young in the oil and gas sectors. Benefits captured across industries involve improving the quality of decisions, improving planning and forecasting, lowering costs, and driving operational-efficiency improvements. However, many challenges for full adoption exist in our industry. In addition to the outdated data-management challenges, key gaps exist in the understanding of basic principles concerning how and when to use different data-analytics tools.
Data-analytics benefits are being demonstrated through the efficient exploitation of data sources to derive insights and support making decisions. An exponential increase in the number of applications in recent years has been observed for enhancing data quality during/after acquisition by automatically removing noise and outliers; better assimilating new and high-frequency data into physics-based models; optimizing calendar-based inspections for preventive-maintenance tasks; increasing equipment availability of well, surface, and drilling systems; optimizing reservoir recovery on the basis of injector-to-producer allocation factors; and many others.
Machine learning is a collection of techniques, both supervised and unsupervised, that gives computers the ability to learn and adapt without being explicitly programmed. This ability to learn provides capabilities for describing past and current operating conditions, predicting, and prescribing.
Supervised learning includes regression and classification methods in which a relationship is established between the input and a known output. Unsupervised learning includes clustering, which addresses problems with no prior knowledge on the output, automatically grouping a large number of data variables into a smaller variable set.
Most data-driven projects may follow a similar approach during implementation. In the majority of these, large efforts are made in collecting and preparing the data, which typically reside in scatter sources and exist in unstructured formats. Once data are placed in proper tabular forms and relationships are established, then data are ready for analysis, which may include exploratory visualizations, model order or dimensionality reduction, clustering, regression, classification, pattern recognition, cross validation, model validation, prediction, and optimization. Insights and syntheses are derived along the analysis process.
Text mining and natural language processing (NLP) allows the possibility of efficiently extracting valuable information from text documents and reports. These methods enable an unexploited yet powerful source of insights about operational transactions (e.g., recommendations, success/failure) that are captured in unstructured text. In the drilling area, NLP has been used successfully to describe and predict nonproductive-time and invisible-lost-time causes from a massive amount of unstructured data collected from the drilling-operation reports. Major contributions will also occur in reservoir management and production optimization.
Recommended additional reading at OnePetro: www.onepetro.org.
SPE 181015 Natural Language Processing Techniques on Oil and Gas Drilling Data by M. Antoniak, Maana, et al.
OTC 27577 Assessment of Data-Driven Machine-Learning Techniques for Machinery Prognostics of Offshore Assets by Ping Lu, American Bureau of Shipping, et al.
SPE 181037 Big Data Analytics for Prognostic Foresight by Moritz von Plate, Cassantec
SPE 185695 A Novel Adaptive Nonlinear-Regression Method To Predict Shale Oil Well Performance on the Basis of Well Completions and Fracturing Data by Amol Bakshi, Chevron, et al.
One of the many challenges we face today sensor availability, and engineering models (regression, time series, and factor in the petroleum industry is the management has promoted the exponential growth analysis); pattern recognition (Markov of data and information. Data-driven models, principal components, ensemble instances, we are overwhelmed by the techniques also have diversified and averaging, classification, and regression); amount and diversity of formats, and, improved to address such incremental business intelligence (key-performanceindicator in other cases, we are blinded from the complexities. We are now referring to the dashboards, multidimensional right information to understand a process professionals who manage and find value visualization); artificial intelligence for (What has happened?), to predict the from data as "data scientists," and we planning, creativity, perception, and social immediate future (What could happen?), are calling the management of large and intelligence (knowledge representation, or to make proper decisions (What should complex data volumes "big data." The answer to these questions Data analytics, either big or small, is Bayesian inference, decision tree, naturallanguage is data analytics supporting appropriate the collection of tools that leverages data programming); machine learning engineering and management judgment collection, aggregation, processing, and (inductive logic programming, rule and the modeling of actual energy scenarios. For many decades, our technical and trends of data analytics differ from traditional A graduate degree may be required to business processes have benefited from statistics in the sense that the new master some of the techniques around the wide use of data statistics for decision data-driven predictive and prescriptive data analytics, and decades may be making.