Doctoral researcher Julia Pelzer and PI Miriam Schulte from the Institute for Parallel and Distributed Systems (IPVS) have introduced a novel method for understanding the intricate interplay of geothermal heat pumps (GHWP) in shallow aquifers. Their study "Efficient two-stage modeling of heat plume interactions of geothermal heat pumps in shallow aquifers using convolutional neural networks", recently published in Geoenergy Science and Engineering (Volume 237, June 24), aims to provide more accurate and faster predictions of heat plumes, which are crucial for efficient urban planning and beyond.
In the quest for sustainable energy solutions, geothermal heat pumps hold immense promise. However, understanding their impact on groundwater temperature and the complex interactions between multiple heat plumes has been a challenge. Traditionally, analytical models like LAHM and simulations such as PFLOTRAN have been used, but they often fall short in accuracy or computational efficiency.
Pelzer and Schulte’s approach, outlined in their paper, involves a two-stage neural network methodology. The first stage rapidly predicts the shape of individual heat pump plumes, while the second stage refines these predictions by considering interactions with neighboring plumes in a so far strongly simplified setting. This innovative approach leverages convolutional neural networks (CNNs), trained on a dataset constructed using realistic subsurface flow parameters extracted from borehole measurements in the Munich region.
Importantly, the researchers' methodology enhances prediction accuracy and shifts computational costs from inference/application time to training. Through experiments conducted on various dataset sizes and input parameters, the study achieved remarkable results. With a Root Mean Square Error (RMSE) of approximately 0.1 °C in both stages, their model demonstrates exceptional accuracy in capturing the intricate dynamics of GWHP interactions.
The implications of this research are significant. By outperforming traditional analytical approximations and simulations, this two-stage model offers a practical and efficient tool for urban planning and beyond in a simplified setting. Its potential for real-time application through a web interface highlights its relevance in sustainable energy management and environmental conservation.