The objective of this work is to apply effective machine learning (ML) techniques in introduction potential estimation for binary cycle geothermal power generation in Japan. I utilized the dataset of the Renewable Energy Potential System (REPOS) web geographic information systems (GIS), which has been managed by the Ministry of the Environment, Government of Japan (MOE) since 2020. The REPOS database provides introduction potentials of renewable energy resources all over Japan. The database includes geothermal reserves, introduction potentials, and vertical temperature profiles at hot spring wells and geothermal wells. The resource estimation is based on the U.S. Geological Survey (USGS) volumetric methodology. The introduction potential has been estimated by a three-dimensional smoothed temperature model, featuring large scale hydrothermal system such as heat convection and conduction. Therefore, it might be less affected by pressurized hot water in faulted and fractured geothermal reservoirs at major flash steam power plants in Japan. Moreover, it is also adjusting to local land use restrictions. As a result, there is a significant discrepancy between estimated introduction capacity and installed capacity in the area close to existing geothermal power plants or active volcanoes. In this study, I focused on the investigation of the introduction potentials for binary cycle power generation. The introduction potential estimates were investigated in terms of power density. A feedforward neural network (NN) algorithm was applied for the power density prediction. The ML approach covers the shortcomings of dataset concerning lower reservoir temperature than 200 °C in Japan. The preliminary results show that the behavior of predicted power density and average reservoir temperature gives us a useful information to optimize capacity for binary cycle system and manage the influence on hot spring resources in Japan. Moreover, I applied a recurrent neural network (RNN) method with Long Short-Term Memory (LSTM) networks for prospective temperature modeling. The purpose of the RNN approach was deeper temperature prediction to mitigate geological risks for depletion of hot spring resources and geothermal production. The ML based prediction approach allows us to precisely evaluate geothermal resources using shallow borehole data without drilling additional deep wells. These data-driven ML techniques will be powerful and cost-effective tools to accelerate introduction of binary cycle system in Japan.

Press the Back button in your browser, or search again.

Copyright 2022, Stanford Geothermal Program: Readers who download papers from this site should honor the copyright of the original authors and may not copy or distribute the work further without the permission of the original publisher.