Unlocking grid flexibility of distributed energy resources

Abstract

Power systems are increasingly integrating distributed energy resources (DERs), such as distributed renewable energy sources (RES), heat pumps (HPs), and data centres. While DERs can strain grid capacity, they can also offer power flexibility to support stable system operation. To unlock this opportunity, this thesis focuses on tackling unresolved challenges in realising DER flexibility.

From a technical perspective, fully unlocking the DER flexibility requires optimal scheduling. Although extensive work has studied DER scheduling, a key challenge remains in handling the inherent uncertainty of DERs, which arises from weather dependence and human behaviour and poses a challenge to reliable power system operation. Joint chance constraints (JCC) represent a promising approach by explicitly guaranteeing the probability of joint constraint satisfaction. This method has been widely applied both in standard power system scheduling and in specific frameworks designed for efficient large-scale scheduling, such as the hierarchical scheduling regime known as DER aggregation. However, existing integrations of JCC are computationally demanding, which limits their scalability. This challenge leads to the first research question: Q1: How can flexible DERs be scheduled in a reliable and computationally scalable manner considering uncertainty?

Beyond the technical challenge of scheduling DERs, existing work has investigated the technical, economic and social impacts of many DERs. These studies also represent a crucial dimension that supports the design of effective and efficient roadmaps to realise DER flexibility. However, two important types of DERs, namely HPs and data centres, remain under-investigated. Given the pivotal role that HPs and data centres are expected to play in future power systems, this knowledge gap poses a significant challenge in unlocking the full value of DER flexibility. In particular, the impact of HPs on fuel poverty remains insufficiently investigated. Fuel poverty refers to a status where households struggle to heat their homes to a comfortable standard, representing a critical economic and social issue. The approximately 300% running efficiency of HPs makes them an especially attractive alternative for households reliant on expensive off-gas heating sources, such as heating oil and electric resistive heating. These households are often at high risk of fuel poverty, further increasing the need for deploying lower-cost, efficient heating solutions such as HPs. However, the HP impact on fuel poverty remains unclear. Furthermore, the large-scale deployment of HPs may also threaten the distribution network, and this network impact, which is interlinked with the fuel poverty impact, has also been under-explored. These gap motivate the second research question: Q2: What are the fuel poverty and distribution network impacts of replacing domestic off-gas heating systems with HPs?

For data centres, there is limited understanding in whether they can bring a new and significant source grid flexibility, particularly for the emerging data centres dedicated to Artificial Intelligence (AI). The recent expansion of the AI sector is driving the deployment of AI-focused data centres, which are more energy-intensive than traditional general-purpose data centres due to the heavy reliance on GPUs. However, it remains unclear whether AI-focused data centres represent a new and significant source of power system flexibility. Furthermore, the lack of quantification of the associated cost means that it is also uncertain whether data centre operators could be effectively incentivised to offer this flexibility. These gaps motivate the third research question: Q3: Do AI-focused data centres have a different potential for power system flexibility compared to general-purpose data centres, and what is their cost of flexibility provision?

To address Q1, this thesis develops novel methods for the efficient integration of JCC in both standard DER scheduling problems and DER aggregation. For standard DER scheduling problems, the focus is placed on a particular JCC structure: the Wasserstein distributionally robust JCC (WDRJCC) with right-hand-side (RHS) uncertainty. This formulation can be applied to ensure power system reserve sufficiency and to ensure network thermal limits with enhanced robustness. This thesis proposes a strengthened and faster linear approximation (SFLA) for the RHS-WDRJCC, demonstrating up to a 100x speedup over traditional methods, such as worst-case conditional value-at-risk (CVaR). For DER aggregation, this thesis develops a tractable method to integrate WDRJCC and classical JCC with a two-stage structure. The proposed method is shown to achieve a better balance between conservativeness and reliability compared to existing approaches. In addition, the computational complexity of the proposed method remains low and does not increase with the number of DERs nor the size of the network, thereby facilitating its applicability to real-world, industry-scale problems.

To tackle Q2, this thesis evaluates the fuel poverty and distribution network impact of HP replacing off-gas heating in the UK. A novel evaluation method is proposed to support this evaluation by integrating publicly available datasets from a wide range of sources, and the method can be extended to other countries. The results show that replacing off-gas heating with HPs could lift up to 20% of households out of fuel poverty in specific UK regions, providing both social and economic incentives for HP deployment. However, this replacement would incur significant distribution network upgrade costs in certain areas, underscoring the need to realise HP flexibility.

Finally, to address Q3, this thesis proposes a novel method to evaluate both the maximum flexibility and the corresponding cost for data centres providing power system services, ranging from short-duration, low-frequency services to long-duration, high-frequency services. To ensure robust conclusions, the proposed method is applied to datasets from 14 real-world data centres and three mainstream cloud computing platforms. The results demonstrate that AI-focused, GPU-intensive data centres can deliver greater flexibility at up to 50% lower cost compared to traditional CPU-intensive data centres. This insight provides motivation for both power grid operators and AI data centre operators to collaborate in future schemes aimed at realising data centre flexibility.

Overall, this thesis contributes to the understanding of how to unlock DER flexibility in future power systems. The proposed methods for efficient JCC integration and the evaluation of HP and data centre impacts provide valuable insights for both researchers and practitioners. The findings highlight the potential of DERs to support reliable and efficient power system operation, while also addressing critical social issues such as fuel poverty.