Analyzing and optimizing CO2 geothermal energy production utilizing artificial intelligence – a deep basin approach

K. Katterbauer, A. Alhashboul, H. Chen, A. Yousef
Saudi Aramco,
Saudi Arabia

Keywords: geothermal, optimization, artificial intelligence, CO2, sustainability


Several businesses have recently developed CO2 plume geothermal technology (CPG). The concept proposes to generate geothermal energy using CO2 that is trapped in salty aquifers. Compared to traditional geothermal concepts, CPG is unique. In this instance, the feedstock makes use of CO2 as a carrier fluid to draw heat from the subterranean reservoir. Furthermore, the system may make use of ordinary sedimentary bases rather than only shallow natural hydrothermal areas. Finally, CPG may continue to produce energy in low-temperature settings where it is currently not feasible to do so using traditional geothermal methods. For the purpose of maximizing power generation from a CPG system, we introduce a novel deep learning optimization methodology. The framework employs a modified N-BEATS methodology. The method is built on an interconnected stack of backcast and forecast links, as well as ensembled feed forward networks. The framework's versatility with regard to multiple input parameters and forecastable time-series is one of its benefits. For CPG, being able to quickly record changes in the temporal dynamics and temperature responses throughout the numerous CO2 injection and production wells is very crucial. On a simulated CO2 storage reservoir based in the Taranaki basin in New Zealand, we assessed the framework. Given the existence of a sizable saline aquifer that may be ideally suited for both CO2 storage and CPG energy generation, the Taranaki basin has been extensively explored for CO2 storage. As an input to the N-BEATS framework, we generated four years' worth of CO2 generation and injection for geothermal energy production. The network demonstrated high training performance, and the model's effectiveness was assessed based on the ensuing three years' worth of energy output. In order to maximize energy output and total carbon footprint while adjusting CO2 injection and power generation from the various CPG stations, the deep learning framework is then incorporated into a global optimization framework. A novel approach to improving energy generation from CO2 storage reservoirs, the new deep learning N-BEATS optimization framework for CPG power generation offers a sustainable means to reduce carbon footprint while delivering electricity.