Design and deployment of data-driven control retrofits for energy systems in commercial buildings
File(s)
Author(s)
Bird, Max
Type
Thesis
Abstract
This thesis investigates the real-world potential of advanced control strategies, specifically model predictive control (MPC) and reinforcement learning (RL) to deliver improved energy and carbon performance across three commercial case study systems: heating, ventilation and air conditioning (HVAC), refrigeration, and PV-battery. The specific research questions look to address whether these systems can be accurately modelled using physics-based and data-driven techniques, whether advanced control methods offer meaningful operational improvements, and, whether such solutions can be deployed cost-effectively at scale.
The initial HVAC case study provided a strong foundation for a generalisable hardware-software infrastructure required to deploy such control retrofits. Building upon these findings, the novel deployment of an MPC scheme in a commercial refrigeration system was showcased. Live testing confirmed the financial attractiveness of such a retrofit, achieving a £51k net present value (NPV), 94% internal rate of return (IRR) and 2-year payback period for the case study location. Annual savings reach the millions when considering wider deployment across a large supermarket estate. RL approaches were also studied in this context, but found to underperform compared to MPC, while also requiring significantly more theoretical background and implementation effort. The suitability of MPC was further demonstrated for the simulated PV-battery case study, achieving similarly compelling financials, with a £29k NPV, 77% IRR and 2-year payback.
Despite the success of these pilot projects, several barriers to widespread adoption remain, including operational readiness of existing systems, outdated digital infrastructure at the site level, and a need for multidisciplinary skill sets that blend control theory, data engineering, and software development. Additionally, organisational inertia and risk aversion often limit the appetite for adopting non-traditional control technologies, especially when existing systems are perceived as ‘good enough’. Nevertheless, the findings of this thesis provide a clear roadmap for how predictive control can be deployed cost-effectively and at scale.
The initial HVAC case study provided a strong foundation for a generalisable hardware-software infrastructure required to deploy such control retrofits. Building upon these findings, the novel deployment of an MPC scheme in a commercial refrigeration system was showcased. Live testing confirmed the financial attractiveness of such a retrofit, achieving a £51k net present value (NPV), 94% internal rate of return (IRR) and 2-year payback period for the case study location. Annual savings reach the millions when considering wider deployment across a large supermarket estate. RL approaches were also studied in this context, but found to underperform compared to MPC, while also requiring significantly more theoretical background and implementation effort. The suitability of MPC was further demonstrated for the simulated PV-battery case study, achieving similarly compelling financials, with a £29k NPV, 77% IRR and 2-year payback.
Despite the success of these pilot projects, several barriers to widespread adoption remain, including operational readiness of existing systems, outdated digital infrastructure at the site level, and a need for multidisciplinary skill sets that blend control theory, data engineering, and software development. Additionally, organisational inertia and risk aversion often limit the appetite for adopting non-traditional control technologies, especially when existing systems are perceived as ‘good enough’. Nevertheless, the findings of this thesis provide a clear roadmap for how predictive control can be deployed cost-effectively and at scale.
Version
Open Access
Date Issued
2025-07-18
Date Awarded
2026-03-01
Copyright Statement
Attribution-NonCommercial 4.0 International Licence (CC BY-NC)
License URL
Advisor
Acha, Salvador
Shah, Nilay
Publisher Department
Department of Chemical Engineering
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)