Development of temperature prediction method for supercritical geothermal resources using neural networks

Abstract

We propose an subsurface temperature structure prediction model using a neural network with the aim of predicting a distribution of a supercritical geothermal resources. In our proposed model, three-dimensional coordinates, specific resistance by magnetotelluric, D95, gravity anomaly value, and mineral isograds were calculated from measurement data as input features. This model training procedure was applied to the Kakkonda geothermal field, Japan. As a result of evaluation using actual measurement data, the RMSE was shown 39.3 ℃ when optimized input features.