GeoML: AI/ML for characterizing and managing geologic reservoirs

Artificial Intelligence and Machine learning (AI/ML) algorithms are currently applied to mine zettabytes of internet data. However, mining the much more limited, highly heterogeneous, multimodal, disparate, uncertain, non-uniform, and unstructured scientific data requires specialized AI/ML techniques. Developing such AI/ML methods is essential for many fundamental scientific problems. For example, we need to improve our basic understanding of the subsurface conditions and processes occurring in geologic reservoirs during various Geologic Engineering Activities (GREs), such as extraction of geothermal heat, hydrogen, and fossil fuels, sequestration of waste and carbon dioxide, and storage of energy produced by solar and wind. Geologic reservoirs are critical for the energy transition of our society and economy to renewable energy resources. The reservoirs must be characterized and managed using geologic, geochemical, and geophysical data. Our fundamental understanding of the processes occurring during GREs can be accelerated and improved by applying advanced physics/science-informed AI/ML methods and tools characterizing and managing geologic reservoirs. To address these demands, we have developed AI/ML methods and tools specifically designed to improve our fundamental understanding of complex subsurface processes impacting the flow of fluids and heat in the subsurface. We are also developing commercial software, GeoML, for (1) characterization and parameterization of subsurface conditions and processes impacting the flow of fluids and heat, (2) discovery of hidden data signatures informing the spatiotemporal characteristics of these processes, and (3) optimization of data acquisition strategies, and (4) prediction of future states and reservoir behavior under different energy injection/extraction scenarios. GeoML relies on unsupervised (self-supervised), supervised, and science-informed ML (SIML) methods. GeoML is designed to be capable of processing public and proprietary data. Our work focuses on energy extraction and storage. Many subsurface reservoirs are suitable and can be applied for both tasks. GeoML is designed to evaluate site prospectivities and select optimal energy storage and extraction locations and strategies at site, regional, and national scales. GeoML utilizes cloud computing and data management resources. We have developed a prototype of the commercial software (GeoML), providing user-friendly, fast, robust, and scientifically defensible tools for predicting energy production and storage. The prototype is available for review and comment at Geo-ML.com (http://98.60.118.144:8988). Our software relies on our novel ML methods and an advanced auto-differentiable numerical simulator for modeling subsurface conditions and energy extraction/storage prospectivity. GeoML processes existing geologic, geochemical, and geophysical public datasets. We will execute aggressive market and technological research to meet the needs of our customers and advance production and commercialization. We will expand AI/ML methods embedded in GeoML. We will expand the utilization of public and proprietary data. We will expand our collaborative efforts with industry, academia, and national research laboratories, including ongoing and upcoming DOE projects related to GRE research and applications. We will also demonstrate GeoML capabilities to address our nation's energy security. The global geothermal energy market size was valued at $6.6Bn in 2021, and is projected to reach $9.4Bn by 2027. The global thermal energy storage market size was valued at $20.8Bn in 2020 and is projected to reach $51.3Bn by 2030. Based on these market evaluations, we believe that investment in GeoML is a commercially valuable proposition. Currently, there are no similar products, either commercial or open-source. Our product will provide new research and commercial opportunities. At the moment, energy-related research is primarily limited to large organizations such as national laboratories, universities, and big companies with extensive resources to invest in detailed exploration and production analyses. With the deployment and commercial use of GeoML, smaller companies, state-level/local governments, Native American Nations, and underprivileged communities could efficiently evaluate and develop their GRE opportunities. GeoML will facilitate our nationĺs goal of making energy use more environmentally friendly, equitable, and inclusive.