Demand Response from Commercial and Industrial End-uses: Application to Power System Scheduling and Stability

The aim of this thesis is to investigate the potential of flexible loads, specifically commercial and industrial end uses, to enhance power system flexibility by providing fast frequency response (FFR). By developing models of commercial and industrial end-uses and integrating them with power system models, the thesis shows that providing FFR from these end uses improves the system frequency profile and reduces the rate of change of frequency (RoCoF) after a disturbance, while minimising disruption to end-user comfort. It also discusses the challenges associated with FFR integration in power system scheduling and stability, and uses machine learning techniques to classify low-inertia power system performance for rapid assessment of power system dynamic security. It also seeks to examines the potential for industrial end-uses to adjust their energy use in sympathy with carbon intensity.
The industrial and commercial sectors are of particular interest due to their size and existing monitoring, control, and communication infrastructure. The thesis develops models for specific end-uses, such as supermarket refrigeration systems and data centres, which can predict the energy consumption of individual devices. The goal is to simulate and aggregate the behaviour of a large and diverse fleet of buildings to the system level, to gain insight into their operating characteristics and improve power system flexibility by providing FFR. The developed end-use models are integrated with power system models, including a unit commitment and economic dispatch model and a stability analysis tool. The research then investigates the impacts of demand response from the developed aggregated fleet of end-uses on power system scheduling and dynamic behaviour, based on a number of flexible load control strategies. The research shows that providing FFR from commercial and industrial end uses improves the system dynamic performance after an event, while minimising disruption to end-user comfort. During the research process, issues related to FFR integration were flagged and further investigated. To address these issues, the research delves into potential approaches for integrating FFR resources within a unit commitment and economic dispatch model. Additionally, it examines the impact of factors such as load inertia, FFR size, and FFR response time on the improvement of the system frequency profile following a contingency event. This research found that running power system stability analysis frequently is computationally expensive. To overcome this challenge, the thesis employs machine learning (ML) techniques to classify low-inertia power system performance as normal or ``at risk'' using synthetic data generated from the Irish power system for the year 2030. The ML model is robust and able to handle imbalanced data, with more instances of normal performance than ``at risk'' performance. The model provides accurate predictions with significantly reduced computation times. This thesis examines the potential for data centres to adjust their energy use in response to a carbon intensity signal, in line with the goal of decarbonising energy systems and reducing emissions. Using machine learning models, the research evaluates the flexibility of a fleet of data centres within a power system unit scheduling framework, with a focus on server temperatures. The study evaluates the impact of demand response from data centres on the system demand profile, including flexibility requirements and maximum upward ramp rates. The research finds that this approach can lead to both operational cost savings and emissions reductions.