15:45 〜 16:00
[MGI28-02] Deep Temperature Estimation Using Neural Network in the Production and Reinjection Zones of Berlín Geothermal Field, El Salvador
キーワード：Neural Network, Simulation, Temperature, Reservoir
High temperature is one of the main characteristics of a conventional geothermal system and during its exploratory stage, data acquisition can be complicated by the lack or non-existence of exploratory wells. Geophysical methods have historically been applied to describe the physical characteristics of the subsurface, some of which characterize up to several kilometers in depth. Electrical resistivity serves as a characteristic of subsurface strata, which could be measured to depths spanning several kilometers. This property can be correlated with temperature from a geothermal system. Over the past decade, there has been a notable surge in the application of Machine Learning (ML) techniques for the analysis of diverse domains within geoscience. These applications extend to the identification and replication of behavioral patterns. Bayesian and Neural Network methods have been applied by Ishitsuka et. al (2021) in the Kakkonda Geothermal Field (KGF) to estimate temperatures in supercritical critical conditions. This study applies Neural Network approach to The Berlín Geothermal Field (BGF), a conventional liquid-dominant geothermal system located in the eastern of El Salvador, whose temperature reaches over 300°C.
Neural Network algorithms have been developed to estimate deep temperatures in geothermal reservoirs using temperature data, seismic data, and geologic boundary records. The out-temperature model is trained and calibrated using the temperature of the numerical model reservoir in a natural state until the known reservoir depth. The algorithm continues its interpolation beyond the known depth, taking resistivity data as a prior parameter. The GDB data have been processed by identifying the most suitable area in the production zone to apply the Neural Network algorithm. Due to the presence of intrusive rock in the reinjection zone, it has been of interest to apply this technique in this zone, resulting in a wide area allowing the understanding of the temperature behavior inside the geothermal reservoir.
The objective of this study is to evaluate the reliability of temperature-estimated data through the utilization of varied modeling methodologies, thereby augmenting the confidence associated with estimations. The outcomes derived from the Neural Network approach compellingly underscore their utility as pragmatic instruments for discerning resources with elevated energy potential within a geothermal system.
Neural Network algorithms have been developed to estimate deep temperatures in geothermal reservoirs using temperature data, seismic data, and geologic boundary records. The out-temperature model is trained and calibrated using the temperature of the numerical model reservoir in a natural state until the known reservoir depth. The algorithm continues its interpolation beyond the known depth, taking resistivity data as a prior parameter. The GDB data have been processed by identifying the most suitable area in the production zone to apply the Neural Network algorithm. Due to the presence of intrusive rock in the reinjection zone, it has been of interest to apply this technique in this zone, resulting in a wide area allowing the understanding of the temperature behavior inside the geothermal reservoir.
The objective of this study is to evaluate the reliability of temperature-estimated data through the utilization of varied modeling methodologies, thereby augmenting the confidence associated with estimations. The outcomes derived from the Neural Network approach compellingly underscore their utility as pragmatic instruments for discerning resources with elevated energy potential within a geothermal system.