Optimization of Buried Pipe Spacing under Seepage Conditions Based on an LSTM Algorithm

This study proposes an optimization methodology for determining the optimal spacing of buried pipes in ground-source heat pump systems under seepage conditions by utilizing a long short-term memory (LSTM) algorithm. We developed a comprehensive three-dimensional numerical model using COMSOL Multiphysics, incorporating critical factors such as permeability, porosity, thermal conductivity, and groundwater flow velocity within sandy soil layers. This model effectively simulates the heat transfer processes occurring under seepage conditions. Performance evaluations of several LSTM models with different layer configurations were conducted, and a two-layer LSTM network was selected and optimized for this specific problem. An additional multilayer perceptron (MLP) structure was integrated to enhance the model’s predictive accuracy. The model was validated using a field case at a geothermal energy demonstration site in Shaanxi Province, China, in which the LSTM predictions had a 4.6% deviation from actual monitoring data. By combining the strengths of numerical simulations with machine learning, particularly LSTM models, this study demonstrates an effective approach for predicting the temperature influence radius and optimizing the spacing of buried pipes. The results show that the optimized LSTM model provides superior accuracy compared with that of traditional exponential smoothing methods. This hybrid approach of combining COMSOL numerical simulations with LSTM predictions offers valuable insights and guidance for further research and practical applications, overcoming the limitations of traditional methods and providing innovative solutions for the design and optimization of ground-source heat pump systems.