Artificial neural networks use for the design of geothermal probe fields

The utilization of geothermal energy is crucial for achieving sustainable heating solutions during the transition to renewable energy sources.Geothermal probes, also known as borehole heat exchangers, are used to extract heat from the ground.These probes can operate individually or be combined into probe fields to meet larger heating demands.The design of these fields must account for potential undercooling or overheating over extended periods.G-functions are typically employed to calculate the thermal behavior of the ground with interconnected borehole heat exchangers.However, calculating numerical solutions for these functions can be very time-consuming.This study involves training artificial neural networks (\gls{ANN}) using pre-calculated g-functions.The accuracies and calculation times of the trained \gls{ANN} were then compared.The study created a dataset with over \num{\sizeOfData} g-functions for a rectangular probe field.Two different ANNs were trained using 80 \% of the dataset.The study then determined the accuracy of the g-functions generated by the ANN and the calculation speed.The results indicate that the mean absolute percentage error was below 2.3 \%.On average, the calculation time was less than 0.1 \% of the original time.Large probe fields performed better than average in terms of calculation time.The reduced calculation time enables the computation of complex tasks within a reasonable timeframe, including depth sizing and borefield design optimizations.