Fault anticipation in distribution networks

  • Eleni Tsioumpri

Student thesis: Doctoral Thesis


This thesis is concerned with the topic of Fault Anticipation in Distribution Networks, focusing on the rapidly changing operational nature of distribution networks and outlining the anticipated data-related challenges that result from these changes, encountered in practice and from related literature. With the challenge of limited data availability in mind, a data analysis methodology for Distribution Network Operators (DNOs) is presented and demonstrated through a number of short and more detailed case studies.;The short case studies are illustrative examples of how the proposed methodology would be used by a DNO. In these, the identification of solar PV operation, phase imbalance and the detection of unusual network operation using dimensionality reduction are examined. The more detailed case studies form the main part of the thesis and focus on the following two areas: (i) Prediction of weather-related faults on minimally observed distribution networks and (ii) Impact of substation loading on the occurrence of power quality disturbances. More specifically, on the topic of weather-related fault prediction, the impact of weather conditions alone on the occurrence distribution network faults is explored, with the case study looking separately into the HV level (mainly 11kV -20kV) and LV level (0.4kV) of the distribution network.;The relationship of power quality events, mainly overcurrent and voltage swell events, with the load behaviour as observed at the LV side of secondary transformers is explored in the second detailed case study.;The contribution of the work presented in this thesis is twofold. First, the fact that distribution networks are currently minimally monitored or access to operational data is restricted for various reasons is acknowledged. This thesis attempts to overcome this challenge by exploring the potential of machine learning techniques to extract valuable information from distribution networks with minimal observation. When required, the available network information is jointly analysed with data coming from different sources that can be easily obtained, such as weather observations.;As mentioned above, this research has mainly focused on two areas which form the basis for the two more detailed case studies presented in this thesis. The weather-related fault prediction case study demonstrated that DNOs can predict the occurrence of weather-related faults in their distribution networks, using only weather observations from a nearby weather station and historic fault records. The other detailed case study which addressed the impact of distribution substation loading on power quality event occurrence identified a relation between representative load profiles and the transitions between them with the occurrence of power quality events.;Both research subjects were selected with a common final goal in mind, which was to utilise machine learning in order to develop a methodology towards the prediction of distribution network disturbances in the absence of extensive monitoring. For the second part of the contribution, the data challenges associated with the changing state of distribution networks are assessed and suggestions to deal with these issues are made. As a result of the work presented in this thesis, an overall data analysis methodology for DNOs is proposed. The main purpose of this methodology is to identify operational or environmental factors that are more likely to lead to the occurrence of certain types of disturbances and establish relations between these factors and fault occurrence, which can then be used to predict these events.;The specific case studies presented in this thesis identify relations between environmental conditions and power system faults as well as substation loading and power quality events. However, the methodology can be applied to different operating conditions and types of faults as well. Being able to establish such relations would be beneficial for DNOs as it would lead to an increased understanding of their network and allow them to act proactively in order to prevent, or minimise the impact of impending events.
Date of Award22 Jun 2020
Original languageEnglish
Awarding Institution
  • University Of Strathclyde
SponsorsEPSRC (Engineering and Physical Sciences Research Council) & University of Strathclyde
SupervisorStephen McArthur (Supervisor) & Bruce Stephen (Supervisor)

Cite this