Contents lists available at ScienceDirect Computers and Geotechnics journal homepage: www.elsevier.com/locate/compgeo Research Paper A machine learning approach to energy pile design Nikolas Makasis a , Guillermo A. Narsilio a, ⁎ , Asal Bidarmaghz a,b a Department of Infrastructure Engineering, The University of Melbourne, Parkville, Australia b Department of Engineering, University of Cambridge, Cambridge, United Kingdom ARTICLE INFO Keywords: Energy piles Finite elements Optimisation Machine learning Regression Geothermal energy ABSTRACT Incorporating ground heat exchangers (GHEs) into building foundations allows them to also provide thermal energy for space heating and cooling. However, this introduces certain constraints to ground-source heat pump (GSHP) design, such as on the geometry, and thus a different design approach is required. One such approach, introduced in this article, uses machine learning techniques to very quickly and accurately determine the maximum amount of thermal energy that can reasonably be provided. A comprehensive validation of this methodology for energy piles is presented, using different geometries and thermal load distributions, drawing conclusions about how the approach can best be utilised. 1. Shallow geothermal systems and energy piles Ground-source heat pump (GSHP) systems can be used to efficiently provide geothermal energy for heating and cooling purposes. These shallow geothermal energy systems extract and reject heat from and to the ground within a few tens of metres below the surface. The heat pump upgrades this thermal energy and is connected to an acclimati- sation distribution circuit within the building, which transfers the heat to and from the building, as well as to a series of ground heat ex- changers (GHEs), which transfer the heat from and to the ground [1].A GHE, which traditionally can take many forms such as vertical bore- holes or horizontal trenches, contains loops (usually high-density polyethylene (HDPE) pipes) with a circulating fluid (usually water) that acts as the heat conductor in the process. These systems are known to typically be able to run at a coefficient of performance (COP) of about 4, meaning producing 4 kW of heating/cooling energy for every 1 kW of electricity consumed [2–4]. Moreover, GSHP systems are the most used amongst the different applications of direct geothermal energy [5] and have attracted much attention over the past decade for the purpose of better understanding how they can be most suitably and efficiently utilised and designed [6–10]. A promising application of GSHP systems that can minimise their capital cost is the use of energy piles, where the GHE loops are in- corporated within pile foundations as shown in Fig. 1 [11–18]. Since the most significant associated cost of these installations is drilling, by adding the loops into the piles (already needed for structural purposes) that cost is considerably minimised as drilling is already accounted for; a detailed breakdown and analysis of these costs can be found in the literature [19]. However, due to the high variability of potential pile configurations and geometries and the fact that this technology is re- latively new to the industry, there is a notable absence of available reliable and fast design tools for energy piles and limited information on not only how the design can be undertaken but also how efficient the technology can be [20]. A key difference between energy piles and typical vertical borehole GHEs is that for the former, the pile number, configuration and length are not primarily designed to fulfil the (thermal) energy needs of the building, but rather for its geo-mechanical stability. This leaves little room for optimisation of the geothermal ground loop design, as the main design parameters, such as the (energy) pile length and separa- tion, are pre-determined. Therefore, the provision of 100% of the heating and cooling energy required (thermal load) cannot be guaran- teed and instead a hybrid system must often be used, to complement the produced geothermal energy using auxiliary means [21,22]. An important challenge is to accurately determine the maximum thermal energy that the geothermal system can provide using the al- ready structurally designed energy piles, which can be either very dif- ficult and time consuming or not as reliable as required using the limited existing design approaches (detailed numerical simulations or analytical commercial software respectively) [20]. While there exist ‘geothermal’ parameters that are not necessarily fixed in energy pile design projects, such as the pipe loop diameter, flow rate and geome- trical configuration of the pipes, identifying the amount of thermal energy the energy piles can provide in the first place is extremely im- portant. A further optimisation of the above-mentioned parameters can further increase heat exchange rates with the ground and thus https://doi.org/10.1016/j.compgeo.2018.01.011 Received 2 September 2017; Received in revised form 18 January 2018; Accepted 19 January 2018 ⁎ Corresponding author at: Engineering Block B 208, Department of Infrastructure Engineering, The University of Melbourne, Parkville, VIC 3010, Australia. E-mail address: narsilio@unimelb.edu.au (G.A. Narsilio). Computers and Geotechnics 97 (2018) 189–203 0266-352X/ © 2018 Elsevier Ltd. All rights reserved. T