How can you use artificial intelligence to improve reservoir characterization?

AI transforms reservoir characterization by streamlining data processing. Techniques like deep learning enhance data quality by efficiently handling tasks like noise reduction and anomaly detection. This automation not only improves data reliability but also significantly reduces manual workload, making analysis more efficient. H2Oil leverages these AI advancements to ensure more accurate and timely reservoir insights.

An early example is using traveling salesman algorithms to sort out pressure histories in a pseudo space and then use krigging to map a continuous pressure history from discrete well measurements. That produces a 3-D pressure model that can be used to understand fluid and contact movement. Usually this is achieved by reservoir simulation, but it entails significant deviations from reality as many more parameters and calculations are involved. Overall it requires deep domain knowledge and powerful algorithms to use AI in this type of data processing. It is a laborious task, but it could be highly rewarding.

Artificial intelligence (AI) can be applied to improve reservoir characterization in the oil and gas industry by enhancing the interpretation of subsurface data, optimizing decision-making processes, and increasing the accuracy of reservoir models. This can be done by using AI in Seismic Interpretation, Well Log Analysis and interpretation, Reservoir Modeling, Multi-Disciplinary Data Integration, Uncertainty Quantification, Enhanced Imaging, and Optimization of Drilling Operations. Implementing AI in reservoir characterization requires collaboration between geoscientists, reservoir engineers, and data scientists. It's essential to validate AI models using known data and continually update them as new information becomes available.

Quite often, the data cleansing process is the most labor-intensive part of reservoir characterization, simulation and engineering. Particularly frustrating are the inevitable inaccuracies associated with public data. AI is already being utilized to generate “most likely” values for instantaneous water production data via large scale material balancing (and other techniques) for the case where regulatory agencies allow allocation. It’s being used to make sense of historical reporting that includes deliberate obfuscation of historical data (i.e., legacy FracFocus data). Flags for scenarios where unintentional mistakes are probable (i.e., wrong units of measure or misplaced decimal point) are analyzed for possible corrective action.

AI can streamline data integration from various sources, ensuring a comprehensive and holistic approach to reservoir characterization. This not only enhances the accuracy of reservoir models but also facilitates real-time monitoring and adaptive reservoir management, ultimately optimizing oil and gas recovery processes.

Krigging has been used extensively in geological mapping and that is a form of feature extraction. The reverse is more interesting, using features to Feld mapping of geological features that could be channels, dunes, sand bars, carbonate reefs, etc to map sparse well data that intercept those. To my knowledge JewelSuite (Baker Hughes) is an application that is available commercially to do that. This is another instance where deep domain knowledge (geology, stratigraphic interpretation) is required to drive computational algorithms.

Using multitudes of observations to fine-tune reservoir simulation models is a great adaptation of AI algorithms. This application is more useful when we use explainable AI vs generative AI algorithms. The key is to use flow models that are based on known processes and constrained by observations. This requires also deep domain knowledge in reservoir engineering recognizing drive mechanisms and fluid behavior to drive AI processes. Some of the key contributors to behavior overlap with other disciplines like fault mapping and seismic interpretation which makes model building very demanding computationally.

Understanding the reservoir from description and observations is the easy part. Mapping out what is needed to be done to get the best out of it, getting more revenue with the least cost is far more difficult task. Using AI in that could be helpful and insightful.