Data Science in Oil and Gas 2021

For the prediction of the lithofacies cube, it is proposed to use new age full-functional Kolmogorov neural networks. These three-layer neural networks, which can be positioned as a new generation of neural networks, have a high degree of freedom comparable to deep multi-layer neural networks. For a more accurate lithofacies cube, it is suggested to perform the forecast in two stages. At the first stage, a separate forecast of each lithofacie is made in the form of a probability cube. At the second stage, the connection of such cubes into one lithofacies cube is based on the principle of maximum probability.

Today, the global oil and gas industry needs a solution to process huge data arrays due to a growing number of wells studied and emerging technologies that yield a broader range of information about the geological and technical factors of field development. Therefore, digital analytical tools are required enabling quick analysis of the data on production, well interventions, reservoir pressure, well interference, voidage replacement, and field studies. This paper describes the approaches to data processing and analysis employed during marker-based well logging at several large fields in the Russian Federation. The mentioned technology involves the use of quantum dot marker-reporters as high-precision flow indicators to obtain data on the flow profile and composition in horizontal wells for many years without well interventions. Data analysis was performed using hybrid digital models based on geological and reservoir modeling and a simplified physical reservoir model, involving machine-learning algorithms underlain by neural networks. This platform provides for structured storage of geological and engineering data and enables using dynamic production logging data in stochastic and traditional geological and reservoir modeling. A case study is described to demonstrate how the waterflooding system operation was optimized by applying complex analysis algorithms, generating a notable economic effect.

The novelty of the implemented solutions lies in the improvement of drilling technologies based on the application of modeling algorithms and finding the optimal network configuration to perform a reliable forecast based on the artificial neural network model. Without comprehensive automation, it is impossible to reduce the role of personnel, which implies the robotization of part of the drilling process and technologies of descent operations. The presented concept of a geographically distributed system of intelligent monitoring and management is easily adaptable to various technological processes when working in emergency situations due to information support of construction processes. The introduction of technologies provides a reduction in operating costs, an increase in gas and oil production of about 10% and a reduction in well downtime of at least 50 % from the classic technologies of drilling, construction and operation in remote fields.

In this work, the principal components analysis was applied along with a correlation heatmap to determine the relationship between liquid hydrocarbons samples and source rock samples in the Nile Delta Basin. The principal component analysis made it possible to display oil samples with source rock samples in a single space, and the correlation heatmap helped to determine geological factors that stand for axes of this plot. Based on the results of the study, it was possible to determine the characteristics of source rocks for hydrocarbons which are consist of kerogen type III and have an initial hydrogen Index of less than 200 mg HC/g TOC, that produced liquid hydrocarbons at the early and main stages of oil window. The Miocene source rocks are the closest to the studied oils in terms of depositional environment.

This paper presents a methodology for selecting candidate wells for hydraulic fracturing. The technique is based on well clustering. Allocation of wells into clusters is carried out according to the most coinciding technological indicators of wells operation. Further selection of wells, one cluster or another, for hydraulic fracturing is performed using well-known optimization algorithms

Currently, the information search in regulatory documentation for well construction and repairs is carried out primarily manually. This imposes restrictions on the convenience and speed of the supervisor’s work, who is controlling these activities. In this article, the development of the offline search system for the oil company internal regulatory documentation is considered and tested on real queries from «Gazprom Neft» production sites. The proposed search system is installed on the supervisor’s automated workplace (tablet) and demonstrates the best results, when using algorithm based on ElasticSearch. This enables successfully process 72,4% of queries with an average processing time of less than 0,9 second.

The paper presents an approximation neural network algorithm for solving conditionally correct coefficient inverse problems of geoelectrics in the class of media with piecewise constant electrical conductivity given on a parametrization grid. It is shown that the degree of ambiguity (error) of solutions monotonically increases with an increase in the dimension of the parametrization grid. A method is proposed for constructing an optimal parametrization grid, which has the maximum dimension provided that the a priori estimates of the ambiguity of the solutions do not exceed a given value. It is shown that the inverse problem in the considered class of media is reduced to the classical approximation-interpolation problem using neural network polynomials, the solution of which is the essence of the approximation neural network (ANN) method. The intrinsic error of the ANS method is determined, a posteriori estimates of the ambiguity (error) of the obtained approximate solutions are calculated with the achieved synthesis discrepancy. The method makes it possible to formalize and uniformly obtain solutions to the inverse problem of geoelectrics with the total number of the required parameters of the medium ∼ n 10 ^ 3.

The purpose of seismic survey is to build a depth-velocity geological model based on the joint interpretation of seismic and well data. Seismic surveys provide uniform coverage of the studied area, and well data provide more complete and accurate information about the studied geological medium at a discrete set of points (well locations). Well data in conjunction with the analysis of seismic stacks and various attributes are used within one software on the seismic interpretation stage. At the same time, the stages of seismic processing and interpretation are historically separated by different software packages. This reduces the efficiency of teamwork within the same project. Therefore, a relevant objective is the development of convenient software tools for joint work. Ideally, work should be carried out in a single software environment in order to ensure effectiveness of teamwork during a project. Thus, the purpose of the study is to create a series of software tools whic are designed to facilitate the interaction between the stages of processing and interpretation.

An example of building a velocity model of the upper part of a seismic-geological section is given. Under the model constructing the times of the first arrivals of refracted waves are used. The created model is applied in the problem of static correction, but it can also be utilized for data migration. To improve the efficiency of the proposed technological solution, the methods of machine learning, ray seismic tomography and factors decomposition are used.

The implemented approach in the software made it possible to integrate disparate data of an oil and gas producing enterprise on the basis of a single digital platform. The goal of automated preparation of a business plan for the development of an oil and gas producing enterprise with the level of downhole detail to the level of contractual terms has been achieved. As a result of the program implementation of the process approach, the calculation of net cash flow for each well and infrastructure facilities was implemented, which made it possible to improve the quality of calculations and the level of justification of the indicators laid down for the calculation of the business plan. Integration within the digital platform with automatic production forecasting based on measurement data, data on technological modes and virtual production electricity consumption depending on planned production. This software product is being implemented at oil and gas producing enterprises in Russia. In the future, it is planned to expand the functionality based on the proposed scalable ontological model. For example, the selection of optimal development options based on the specified time limits, finances, technical characteristics and target function. In addition, it is planned to expand the intellectual analysis of actual data and factor analysis of deviations.

The paper presents the results of testing the hypothesis that there is a relationship between the value of the attenuation parameter of seismic signals observed at the surface and the pressure measured in the wells. Hypothesis testing was performed on the basis of field seismic data that underwent standard processing. The attenuation parameter values ​​determined from them were used to obtain correlations between attenuation and pressure. Based on these relationships, predicted pressure values ​​were calculated. Comparison of the predicted and measured pressures showed their high accuracy. This confirmed the validity of the hypothesis under consideration.

Seismic data processing is a long, complex and iterative process implemented at Gazprom Neft. One of the stages of this process is the ground roll noise attenuation. Implementation requires the presence in the company of specialists-geophysicists with strong mathematical background, who currently solve this problem by manually selecting combinations of various predetermined filters. The company has been dealing with this task for a long time and has accumulated a representative sample of data where manual noise attenuation has been implemented. This information can be used as an automatic hypothesis test for new filtering methods. Based on this approach, we propose a method for obtaining ground roll noise masks and a method for its attenuation in the Fourier frequency domain (FK transform) based on heuristic rules.

The paper provides an approach for interpreting downhole distributed temperature sensing (DTS) and the results of its application in cases of synthetic and real production data. The outcome of such interpretation is a profile of fluid flows from reservoir layers. The given problem, however, is ambiguous, that is why the suggested approach consists of three steps: formulation of the inverse problem based on minimization of the constructed functional with the developed fast gradient optimization method, massive parallel inversions to collect a set of different interpretations and Bayesian inference of the most probable flow profiles incorporating uncertainty. All three issues are discussed in detail. Modifications of gradient optimizer making it fast and robust are described along with regularization allowing us to approach global functional minimum for synthetic data (illustration is provided) and decrease the ambiguity for real data. Explanation and example of how statistical analysis turns a set of interpretations into the most probable flow profiles and corresponding uncertainty with EM-clustering using Dirichlet distribution are included. All in all, the developed approach for effective evaluation of flow profiles and their statistical analysis can become a useful tool in oil and gas industry automating a big part of DTS interpretation process.

The prospects for the oil and gas potential of the Pre-Jurassic complex in one field in Western Siberia are associated with the weathering crust. To solve the problem of identifying highly productive zones, a complex interpretation of information on the material composition of rocks and the results of clustering of APS and gamma-ray log data was performed

The paper presents a general view of the pipeline for deploying a machine learning model to production. It is shown that today the infrastructural costs of embedding ML into the production circuit can exceed the costs of creating and training a model by almost an order of magnitude.