Tyler O'Neal, Staff Editor

Skoltech scientists use ML to optimize hydraulic fracturing design for oil wells

Skoltech researchers and their industry colleagues have created a data-driven model that can forecast the production from an oil well stimulated by multistage fracturing technology. This model has high commercialization potential, and its use can boost oil production via optimized fracturing design. The research, supported by Gazprom Neft Science and Technology Center, was published in the Journal of Petroleum Science and Engineering.

Hydraulic fracturing, essentially pumping fluid at high pressures into the reservoir formation, which creates fractures and help bring hydrocarbons to the well and ultimately to the surface, is one of the most widely used techniques for stimulation of oil and gas production. Over the last decades, the technical complexity of HF has grown so much that it now requires extensive design and prior modeling with complex multi-module simulators.

"At the same time, bridging the predictions of these simulators with reality is still a major problem of calibrating, verifying and validating models on real data. Moreover, to close the loop between fracturing simulator and production data, one needs to couple fracturing design modeling with a reservoir simulator, which increases complexity and uncertainty even more. As an alternative, we decided to look right at the field data on frac design and production, which is the measure of success," explains Professor Andrei Osiptsov, Head of Multiphase Systems Lab at the Skoltech Center for Hydrocarbon Recovery and a coauthor of the study. {module INSIDE STORY} E

Researchers of the M-Phase Lab together with their colleagues at CDISE led by Professor Evgeny Burnaev, head of ADASE group, decided to see whether a data-driven approach to HF design based on machine learning can help address this challenge.

The key component of their project, which was initiated in 2018, is a digital database on fracturing jobs and oil production from some 6 thousand wells in around 20 oilfields in Western Siberia, Russia, within the perimeter of JSC Gazprom Neft. Each data point contains 92 variables on the reservoir, well and the fracturing design parameters as well as 16 oil production parameters.

"We managed to collect and clean up a very big database of completed works on hydraulic fracturing. By applying machine learning methods to this database, we can already predict hydraulic fracturing results with good accuracy, depending on the process parameters. We still have to solve the difficult task of building optimal recommendations for choosing the parameters of the hydraulic fracturing process based on this forecast," says Professor Burnaev, a coauthor of the study.

Albert Vainshtein, senior engineer and project manager at the M-Phase Lab and a coauthor of the study, notes that the project was "very challenging right from the start" due to ambiguity of real data, high uncertainty and heterogeneity.

"I think that the development of a digital database will allow us to test various hypotheses which, in turn, will clear up multiple hidden patterns of the fracturing processes. As an example, it is important to determine at which injected proppant tonnage our cumulative oil production stops increasing. Depending on the conditions, a common approach is to inject 60 tons per fracturing stage. Using the machine learning model and statistics, we can confirm or reject this hypothesis," says Anton Morozov, a Skoltech PhD student and research intern at the M-Phase Lab.

Scientists have already produced pilot well fracturing design recommendations based on their machine learning approaches, which have been delivered to the industry partner. They hope an upcoming field-testing campaign will show the potential of their technology for oil production. Still, Burnaev reiterates that there is "quite a large amount of uncertainty in the input data describing the design of a hydraulic fracturing system". In the next phase of the project, they aim to develop new methods for estimating this uncertainty.

"Working with real field data takes courage and care, as it is very sensitive and requires special handling procedures. It would have been impossible without unconditional support from our technology partner, Gazpromneft Science and Technology Center, and the largest production entity of the operator, Gazpromneft Khantos, which is our ultimate client on this project," Osiptsov says.

"Our data-driven approach opens an avenue towards a recommendation system which would advise DESC engineers on the optimum set of fracturing design parameters, or at least narrowing down the intervals, where this optimum design can be found," he concludes.

Grigory Paderin, who leads the Optimal Hydraulic Fracturing project at the Gazprom Neft Science and Technology Center, also noted that this project "is not just a unique scientific challenge aimed at optimizing hydraulic fracturing design, it is also very important for the digitization of processes at Gazprom Neft. It allows us to take a new look at the value of our data and to reconsider our attitudes towards collecting, storing and processing this data."

UTokyo-IIS scientists develop a machine-learning algorithm to help design new materials

Tyler O'Neal, Staff Editor

Researchers at the Institute of Industrial Science, The University of Tokyo (UTokyo-IIS) used artificial intelligence to rapidly infer the excited state of electrons in materials. This work can help material scientists study the structures and properties of unknown samples and assist with the design of new materials.

Ask any chemist, and they will tell you that the structures and properties of materials are primarily determined by the electrons orbiting around the molecules that make it up. To be specific, the outermost electrons, which are most accessible for participating in bonding and chemical reactions, are the most critical. These electrons can rest in their lowest energy "ground state," or be temporarily kicked into a higher orbit called an excited state. Having the ability to predict excited states from ground states would go a long way to helping researchers understand the structures and properties of material samples, and even design new ones. CAPTION Scientists at The University of Tokyo use machine learning to predict the excited electronic states of materials--research that can accelerate both the characterization of materials as well as the formulation of new useful compounds{module INSIDE STORY}

Now, scientists at UTokyo-IIS have developed a machine-learning algorithm to do just that. Using the power of artificial neural networks--which have already proven themselves useful for deciding if your latest credit card transaction was fraudulent or which movie to recommend streaming--the team showed how artificial intelligence can be trained to infer the excited state spectrum by knowing the ground states of the material.

"Excited states usually have atomic or electronic configurations that are different from their corresponding ground states," says first author Shin Kiyohara. To perform the training, scientists used data from core-electron absorption spectroscopy. In this method, a high energy X-ray or electron is used to knock out a core electron orbiting close to the atomic nucleus. Then, the core electron excites to unoccupied orbitals, absorbing the energy of the high energy X-ray/electron. Measuring this energy absorption reveals information about the atomic structures, chemical bonding, and properties of materials.

The artificial neural network took as input the ground state partial density of states, which can be easily computed, and was trained to predict the corresponding excited state spectra. One of the main benefits of using neural networks, as opposed to conventional computational methods, is the ability to apply the results from the training set to completely new situations.

"The patterns we discovered for one material showed excellent transferability to others," says senior author Teruyasu Mizoguchi. "This research in excited states can help scientists better understand chemical reactivity and material function in new or existing compounds."

Dutch astronomers predict bombardment from asteroids, comets in another planetary system

Tyler O'Neal, Staff Editor

The planetary system around star HR8799 is remarkably similar to our Solar System. A research team led by astronomers from the University of Groningen and SRON Netherlands Institute for Space Research has used this similarity to model the delivery of materials by asteroids, comets, and other minor bodies within the system. Their simulation shows that the four gas planets receive material delivered by minor bodies, just like in our Solar System. The results were published by the journal Astronomy & Astrophysics on 29 May.

Counting outwards from the Sun, our Solar System consists of four rocky planets, an asteroid belt, four gas giants, and another asteroid belt. The inner planets are rich in refractory materials such as metals and silicates, the outer planets are rich in volatiles such as water and methane. While forming, the inner planets had a hard time collecting a volatile atmosphere because the strong solar wind kept blowing the gas away. At the same time, the heat from the Sun evaporated any ice clumps, so it was harder to retain water. In the outer regions, there was less solar heat and wind, so the eventual gas giants could collect water ice and also gather large atmospheres filled with volatiles. CAPTION This is a cartoon to accompany the article 'Astronomers predict bombardment from asteroids and comets in another planetary system' CREDIT Anastasia Kruchevska{module INSIDE STORY}

Simulation

Minor bodies, including asteroids, comets, and dust, fine-tuned this outcome later on by delivering refractories from the inner belt and both volatiles and refractories from the outer belt. A research team led by astronomers from the University of Groningen and SRON Netherlands Institute for Space Research wondered if the same delivery system applies to planetary systems around other stars. They created a simulation for the system around HR8799, which is similar to our Solar System with four gas giants plus an inner and outer belt, and possibly rocky planets inside the inner belt. Therefore the team could take some unknowns about HR8799 from our own Solar System.

Terrestrial planets

The simulation shows that just like in our Solar System, the four gas planets receive material delivered by minor bodies. The team predicts a total delivery of both material types of around half a millionth of the planets' masses. Future observations, for example by NASA's James Webb Space Telescope, will be able to measure the number of refractories in the volatile-rich gas giants. 'If telescopes detect the predicted amount of refractories, it means that these can be explained by a delivery from the belts as shown in the model', explains Kateryna Frantseva, first author of the paper. 'However, if they detect more refractories than predicted, the delivery process is more active than was assumed in the model, for example, because HR8799 is much younger than the Solar System. The HR8799 system may contain terrestrial planets, for which volatile delivery from the asteroid belts may be of astrobiological relevance.'