Data Science is the comprehensive and integrated approach to the analysis of large volumes of information, through the application of machine learning. This technology is highly promising in terms of its potential application to the oil and gas industry. It involves a series of algorithms which analyse “raw” production data and, on that basis, “learn” to classify and categorise it, revealing hidden connections that might otherwise remain undetectable to humans, as well as extracting new and useful information from this.
Gazprom Neft developing machine-learning technologies
Specialists at the Gazprom Neft Science and Technology Centre, together with the Moscow Institute of Physics and Technology (MIPT) Engineering Centre, have begun developing algorithms based on machine-learning methodologies. This technology will improve the quality of operational data obtained from wells, as well as identifying new patterns and trends. Implementing the methodologies developed will halve the time spent on immediate analysis of operational data, will quickly integrate new trends in subsequent field development, as well as allowing optimum development methodologies to be collated, while reducing costs.
Decisions on the application of various methodologies for increasing production are taken throughout the course of field development, on the basis of operating data sourced from wells. Operational data metrics (liquid flow, oil, water encroachment, and bottomhole pressure) are sourced from all wells throughout the company, with monthly production data and technical data also generated, together with information on investigations undertaken, physical strata characteristics, and fluids and gas produced.
At the same time, the quality of this data does not always allow full analysis: data for certain time periods may be missing, and certain measurements may not always correspond with the physical model, or may be inconsistent. Incorrect data in reports can be caused both by malfunctions in the operation of measuring equipment, and by “human factors”.
Identifying mistakes by company specialists is not always possible, and incorrect information can lead to incorrect conclusions on the current state of wells and the field as a whole, as a result of which incorrect decisions might be made regarding well workovers. In most cases, well workovers involve various interventions to increase production: hydraulic fracturing of the reservoir, differing treatments for differing strata compositions, directed at increasing production, sidetracking, repair work, and so on. Tools developed through machine learning methodologies will increase the speed of managing and analysing large volumes of information sourced from the field. In addition to which, the machine learning tools used will allow the integration of disparate data, and the analysis of every megabyte of available information, leading to new conclusions able to take data quality to a new level — something which will, without a doubt, lead to better operational efficiency.
The application of Data Science methodologies makes possible the processing of huge data sets (Big Data), revealing new patterns and taking these into account in future forecasting in machine learning and integrated physical models to complete definitions in the event of missing values. As part of the project, search algorithms for identifying incorrect values and for restoring missing data have already been developed, as well as for determining well interference and classifications in terms of deviation from current and potential productivity against wells under similar geological conditions. As a result, the application of new algorithms can significantly increase speed and efficiency in the work of field development specialists, reduce the risk of making incorrect capital-intensive decisions caused by the “human factor” in development, and reduce downtime, by creating an “intelligent assistant” for specialist developers — one, moreover, who never sleeps, makes calculations instantaneously, and is almost never wrong.
The development of machine-learning algorithms is being undertaken as part of Gazprom Neft’s Electronic Asset Development (ERA) project, part of the company’s Technology Strategy, directed at developing Gazprom Neft’s IT projects in exploration and production, and covering all key areas of activity: geological exploration, geology, drilling, development, production, and field development.
Vadim Yakovlev, First Deputy CEO, Gazprom Neft, commented: "“Digital technologies are changing oil companies’ approaches in choosing options in field development and operation. Cutting-edge means of working with information make possible greater efficiency in using field-collected data, and better-informed decisions. A key aspect of our Technology Strategy is aimed at precisely this — at optimising the development of our assets with the help of cutting-edge information technologies.”
Timur Tavberidze, CEO of the MITP Engineering Centre, added: "“Analysis of Big Data, as a tool, can significantly increase the value of current information. The search for hidden but substantial inter-connections, and the full and comprehensive analysis of unstructured heterogeneous information, gives a second wind to data that would otherwise just be gathering dust. In today’s oil and gas industry, decision-making is based on data that grows exponentially over time. The ‘Big Data’ paradigm allows business strategies to be adapted in line with such ‘explosive’ data growth. Modern methods of data analysis, moreover, such as machine learning and convolutional neural networks, make possible a fundamentally new approach to solving the most urgent problems. Reworking field data is a real example of implementing this approach through the joint efforts of the Gazprom Neft Science and Technology Centre and the MIPT Engineering Centre teams.”