Thesis

Predicting power distribution network characteristics and load profiles with limited operational data

Downloadable Content

Download PDF
Creator
Rights statement
Awarding institution
  • University of Strathclyde
Date of award
  • 2026
Thesis identifier
  • T17597
Person Identifier (Local)
  • 202273153
Qualification Level
Qualification Name
Department, School or Faculty
Abstract
  • Electricity distribution systems were originally designed for unidirectional power flows and not for on-site generation. However, with the increasing integration of Distributed Energy Resources (DERs) and other Low-Carbon Technologies, the role of distribution networks is evolving. This shift introduces several challenges, including the need for more frequent network reconfiguration to identify potential voltage and thermal violations. Addressing these issues often relies on state estimation techniques, which depend on large and accurate datasets. Despite the importance of accurate monitoring, the vast number of secondary substations makes it economically impractical to install low-voltage monitoring equipment and store data over extended periods of monitoring data. Consequently, this thesis proposes three methods designed to maximise the utility of minimal data: (i) transfer learning, which utilises data from other locations; (ii) few-shot learning, which predicts outcomes with limited data from the same location. The thesis then introduces a comprehensive method for constructing a network model that leverages actual operational GIS data, ensuring the resulting representation of the network closely reflects real-world conditions. This model is subsequently utilised to perform forecasting analyses, enabling the accurate prediction of future operational scenarios. The approach is designed to align with the Distribution Future Energy Scenarios (DFES) framework used in Great Britain (GB), thereby ensuring that the forecasting results are both relevant and practical for contemporary network planning and operation. Grounded in real operational data, this thesis designs and implements a novel GIS-based model of a real-world distribution network and integrates forecasting methods based on transfer learning and few-shot learning to achieve forecast in data-sparse areas. The aim is to provide robust decision support for the GB electricity distribution sector. As a result, DNOs can achieve more cost-effective yet reliable performance across the network, reducing the need for comprehensive data sets or detailed forecast models at every substation. The significance of this research lies in its ability to extract and transfer salient information from both limited and data-rich sources. The knowledge can be applied in situ or moved to data-sparse substations, enabling cost-effective, feasible and accurate power-flow analysis and state estimation across the distribution network. When combined with a GIS-based model built from real operational data, the approach yields results that are directly actionable for DNOs
Advisor / supervisor
  • Brown, Blair
  • Stephen, Bruce
Resource Type
DOI

Relations

Items

Thumbnail Title Date Uploaded Visibility Actions
file details PDF of thesis T17597 2026-03-19 Public Download