Enhanced Geothermal System Performance Prediction Based on Deep Learning Neural Networks

Abstract

The rapid development of the global economy and the sharp rise in population has increased human demand for energy. Hot dry rock (HDR) geothermal energy has attracted much attention because of its wide distribution, huge reserves, and high stability. Enhanced geothermal systems (EGS) are used for the extraction of HDR, whose recovery performance has a highly non-linear relationship with actual production constraints. Therefore, accurately predicting geothermal productivity is an important task for managing sustainable geothermal systems. In this paper, we use a convolutional neural network (CNN), a long short-term memory network (LSTM), and hybrid models based on a convolutional neural network and a long short-term memory network model (CNN-LSTM) for prediction. The dataset is obtained from numerical simulations on the dynamic economic performance of EGS extraction with different well model parameters and fracture parameters. The performance of different neural networks for geothermal capacity prediction is evaluated comprehensively. The results show that the CNN-LSTM neural network can predict geothermal energy production accurately and stably. Compared with the original LSTM and CNN neural networks, the combined network has the best geothermal capacity prediction accuracy, stability, and generalization ability. This study provides a highly accurate and efficient prediction method for geothermal capacity prediction.