AI for Oil and Gas Exploration
In recent years, the “hard-to-recover” oil and gas fields, whether new or old have dominated the Oil and Gas industry landscape. Many newly discovered fields are located in harsh environments, like the Arctic, or deep below seawater, or they are of complex geometry and with poor permeability. This has called practitioners for new and innovative solutions to minimize expenses, optimize efficiency, and maintain profitability margin. In this article, we will talk about the application of Artificial Intelligence (AI) in one of the most capital-intensive operations in the Oil and Gas industry: Exploration.
Context
The Oil and Gas exploration operation is a long process that maps regions in search of potential hydrocarbon reserves. It constitutes three different major activities: Seismic Surveying, Well Logging, and Core Analysis. The accumulation of the three activities’ results leads to a refined geological model of a given field. After that, reservoir engineers go ahead and design various field development schemes that include well drilling and well operation plans and run simulations to find the best schema.
Seismic Surveying
Seismic surveying is a reservoir-scale operation to model the subsurface geology. In simple words, it uses acoustic waves to see through layers of formation underneath the earth’s surface. The operation involves generating waves underneath the earth’s surface and then recording their reflection through sensors called geophones when they bounce back. The same is true for offshore exploration to look through formations underneath the water level. The recordings are then processed to result in a noisy reconstruction of the subsurface and the formation that constitutes it. The reconstruction is called a seismic cube. Expert interpreters then study and analyze the cube and recommend exploratory drilling to further enhance their knowledge of the subsurface formations. Given the size and complexity of seismic surveying, the interpretation is very time-consuming, and can take more than a year to come up with accurate surveys.
The recent advances in the domain of Artificial intelligence promise more efficient, accurate, and less time-consuming seismic surveying operations. For instance, researchers are proposing to use Convolutional Neural Networks to increase the quality and speed of the seismic reconstruction phase. On the other hand, to speed up the seismic cube interpretation activity, it is important to develop computer-aided tools to assist interpreters. Deep Learning techniques are proposed by researchers to speed up the seismic interpretation process by a factor of 10–1000. For example, a promising approach to automatically detect seismic faults from seismic cubes is proposed. A fault is a planar fracture in a rock where there has been significant displacement as a result of rock movements. Those faults are usually the main factors determining the most significant hydrocarbon traps. Automatically detecting those faults will aid interpreters and speed up the entire activity.
Well Logging
The seismic cubes are not enough for experts to make costly decisions on where to drill. This is because they are macro-scale and need to be refined through micro-scale data. To generate this data, seismic cubes interpreters propose several exploratory drilling positions to collect data from them. This data is collected using well logging operations.
The operations aim is to get more precise data related to the physical properties of the subsurface formations around drilled wells. Logging sensors are moved through the entire wellbore while recording different measurements of the surrounding formations such as natural gamma-ray intensity, electrical resistivity, and response to magnetic excitation. Those recordings are then analyzed, processed, and used to refine the seismic interpretation.
Industry practitioners are using machine learning to automatically interpret the logging recordings, to predict sediment types for example. A case study reported that a machine learning approach can achieve up to 92% agreement with the manually interpreted records while being 1000 times faster. The involved researchers in the study then asked the same interpreters to interpret the same data another time. Surprisingly, the two manual interpretations showed an agreement of 91% between each other. This hints at the potential of AI not only to accelerate the process but also to reduce the experts’ subjectiveness from the process. Acceleration and human-error reduction are considered to be among the major contributions of AI in the field. Machine Learning can also be used to directly reduce the expanses of the physical part of well logging. Internal trials have shown that machine learning algorithms can be easily used to reconstruct a major part of the logging measurements.
Core Analysis
Core analysis is similar to well logging, but it is done by directly inspecting core samples extracted from the subsurface. The core samples are long cylindrical sections of the ground extracted by drilling using special drilling equipment. The analysis results in the most accurate measurement of different well characteristics including porosity and permeability. Computer vision techniques based on the latest technologies can be used, as proposed by researchers, for automatic rock typing. Also, machine learning algorithms are used to predict the porosity and permeability of cores based on the routine core analysis data. Those algorithms follow the most employed machine learning setting which is called supervised learning. They are fed a load of labeled historical data so that they can learn a mapping between the input and the output. The richer and more diverse the training data is the better they perform.
Challenges
While there is a significant potential for using AI in exploration, the availability of large amounts of labeled data emerges as a key challenge. A large amount of labeled data is necessary for AI algorithms to achieve superior results that make them suitable for real-world applications. The stance that oil and gas players take regarding cross-company and cross-border data sharing is critical for the success of AI and its adoption. If this stance is negative, AI will perform poorly due to the lack of training data. This will yield a wrong perception of AI usability in the industry among practitioners.
Aside from exploration operations, AI has even more potential and a faster adoption rate in drilling, production, and maintenance operations. Different major players in the industry have already acquired AI startups or initiated partnerships with them.
