Prediction of output temperature and fracture permeability of EGS with dynamic injection rate based on deep learning method

Abstract

The heat recovery of the enhanced geothermal system (EGS) declines noticeably as mining progresses. Therefore, it is extremely important for the regulation of the later stage thermal extraction process. The later stage thermal extraction process can be reasonably regulated if the heat recovery capacity of EGS and reservoir parameters in the late stage can be precisely predicted based on the primary production data from the same EGS. In the present work, we developed a deep learning model based on the Long Short-Term Memory (LSTM) neural network to predict the late-stage output temperature and fracture permeability of EGS with dynamic injection rate. This model could deal with time series problems. When developing this model, the numerical results for 80 years of EGS operating dynamic injection conditions are adopted as a database in which the last 20 % are set as prediction data and can be considered as later stage (the upcoming) production data. We thoroughly assess the output temperature and fracture permeability prediction performance of the LSTM network by comparing them with the numerical results. The comparisons reveal that the developed deep learning model could accurately predict the output temperature and fracture permeability of EGS under different dynamic injection rate, outperforming Gated Recurrent Unit (GRU) in prediction accuracy. This study demonstrates the potential of LSTM networks, in providing accurate, data-driven predictions for critical reservoir parameters, enabling more effective regulation of the thermal extraction process and optimizing long-term geothermal energy recovery.