Optimal control of networks of borehole heat exchangers with machine learning
The goal of this thesis is to develop a control algorithm for networks of borehole heat exchangers based on artificial intelligence.
Background
The growing demand for sustainable energy solutions has driven significant interest in geothermal systems as efficient methods for heating and cooling. Borehole heat exchangers (BHE) are common in single family buildings, but networks of interconnected BHEs are also used in applications with more thermal demand, such as multi-family or commercial buildings. Effective management of these systems is essential for maximizing energy efficiency while minimizing operational costs and environmental impact. However, their inherent complexity, stemming from dynamic interactions at different time scales and nonlinear behavior makes it challenging to develop optimal control strategies for a given field. Commonly used methods, such as peak load estimation combined with constant or on/off operational strategies, often meet thermal demands but result in suboptimal energy efficiency. Machine learning (ML), with its ability to uncover patterns and optimize decision-making in complex systems, presents a promising avenue for addressing this challenge.
Task description
This project aims to develop and evaluate ML algorithms for optimizing control strategies in networks of borehole field exchangers. Specifically, the Julia package BoreholeNetworksSimulator.jl will be employed to train the ML model by simulating scenarios, generating training data, and evaluating different control strategies. Reinforcement learning is the primary approach to be explored, given its success in dynamic control problems. However, alternative machine learning techniques may also be considered. The task for this thesis is to develop a machine learning model capable of deriving optimal control strategies in selected case studies. These strategies will be compared against popular alternatives to evaluate improvements in energy efficiency, operational reliability, and environmental impact. By advancing control methodologies for borehole heat exchanger networks, this work aims to contribute towards the best use of geothermal energy as a sustainable and efficient solution for heating and cooling systems.
Learning outcomes
Skills in machine learning: data curation, architecture choice, training process.
Prerequisites
Familiarity with some programming language, interest in machine learning and shallow geothermal energy
Research Area
Machine learning
Duration
5 months
How to apply
Contact Marc Basquens (below) with a short introduction and motivation statement
Supervisor/Examiner
