Application of machine learning technique to predict the energy performance of energy tunnels

This study investigates the application of a machine learning technique in the field of energy tunnels (i.e. tunnels equipped as ground heat exchangers), as a method to efficiently predict and optimise the thermal exchange during the operational phase and reduce the related computational time and resources. Artificial Neural Networks (ANN) models were trained based on short-term (i.e. some years) numerical simulation data. Then, the accuracy of the trained model was assessed by comparing its results for longer operational periods under the same or different thermal operation patterns with respect to those used for training. Numerical simulations were run for long-term periods, and the obtained numerical results were assumed to be the target to verify the accuracy of the prediction. The analyses under different scenarios indicate that the trained models are able to accurately and rapidly predict the thermal exchange of energy tunnels, not only for future years under the same operational pattern, but also for other thermal operation patterns. In this study, using the same computational machine, a 20-year numerical forecast of thermal output takes hours or days depending on complexity of the numerical model, but it can be reduced to minutes by using the proposed ANN model. This represents a first attempt demonstrating that with further development, machine learning techniques could be used in the practice, in combination with short-term monitoring data, as a tool for the assessment of long-term energy potential and optimisation of energy tunnels during operation.