Evaluating the applicability of neural network to determine the extractable temperature from a shallow reservoir of Puga geothermal field

Abstract

The developmental works to set up a geothermal power plant by Oil Natural Gas Corporation (ONGC) in Ladakh are in niche stages. Existing studies addressing the pre-drilling power estimates of the geothermal field in Ladakh using coupled simulations explicitly correspond to specific operating conditions. Though simulating the reservoir response under unexplored operating conditions would help to analyze the optimal scenarios and devise strategies, the involved computational effort is a major barrier. In these circumstances, adopting neural network models to predict the response for unstimulated operating conditions is a compelling solution. However, studies focused on analyzing the feasibility of using neural network models are limited. Building on this research gap, this study investigates if Convolutional Neural Networks (CNN), Recurring Neural Networks (RNN), and Deep Neural Networks (DNN) can be used to estimate extractable temperature from a geothermal reservoir. Accuracy metrics reveal that the developed network models can estimate extractable temperature for a chosen operating condition under a doublet extraction scheme without compromising accuracy and with just one-tenth of computational effort involved in conducting a simulation studies. The maximum deviation between estimated and simulated temperature fields is 1.3 K, 0.8 K, and 1.1 K for CNN, RNN, and DNN models, respectively. Results suggest that RNN architecture is preferred over CNN and DNN. The developed model serves as a benchmark and helps planners to estimate the extractable power from Puga geothermal field under various operating conditions with the least computation effort while ensuring the physics captured.

Keywords

• Puga geothermal field
• Renewable energy
• Neural network models
• Extractable temperature