[S-EM14] Electric, magnetic and electromagnetic survey technologies and scientific achievements

[SEM14-P22] Estimation of subsurface temperature distribution from resistivity and seismic wave velocity using a neural network

*Souma Yamamoto¹, Tada-nori Goto¹ (1.University of Hyogo)

Keywords:Neural networks, Resistivity, Temperature estimation

Subsurface temperature distributions are necessary in geothermal development. It is fundamental information not only to confirm geothermal reservoirs, but also to discuss hydrothermal circulation patterns before and after geothermal development. Temperature measurement requires the drilling. In some cases, they are more than 3,000 m deep. However, the number of wells is limited due to the high cost and environmental impact by the drilling. In recent years, neural networks have been used to estimate temperature. In this study, we try to estimate the subsurface temperature distribution by applying neural networks to the known temperature. The temperature obtained by a numerical simulation is used. The resistivity and seismic velocity distribution based from this temperature distribution are also used additionally, which are based on the results of rock physical property measurements. First, a learning model was created to estimate temperature using only location data, and then a learning model was created to estimate temperature using location data and resistivity data. At the same time, a learning model was also created for location data and seismic wave velocity. Here, for the neural network, the number of neurons layers in the middle layer and the number of neurons per layer were varied, and the cost function and test data were verified to see how the accuracy of learning changes. The results showed that temperature estimation using resistivity data in addition to location data was more accurate in estimating temperature distribution. In the future, we plan to apply the actual data to the neural network.