Data-driven and Deep Learning-based Equivalent Modelling of HRES Plant for System Transient Stability Studies

  • Ana Radovanovic

Student thesis: Phd

Abstract

Hybrid renewable energy source (HRES) plants, a combination of renewable generation and storage technologies with a common grid connection point, have been seen as a promising option for obtaining controllable RES production, and thus, contributing to system stability maintenance. Assessing the impact of HRES plants on system dynamic behaviour requires their adequate dynamic representation in system stability studies. Dynamic equivalent models (DEMs) of HRES plants as a part of the overall system dynamic model can provide a fast reliable system stability estimation without modelling individual HRES plant components. The main contribution of the research described in this thesis is in the area of dynamic equivalent modelling of HRES plants. The thesis also looks into the concept of geographically distributed HRES plant at transmission network (TN) level from the system stability perspective, as this concept is an expected extension of the existing concept of aggregators at distribution network level. The thesis starts by discussing the techniques for dynamic equivalent modelling of plants and networks. It then provides a computationally efficient procedure for assessing the typical annual impact of HRES plant on power system stability. The outputs of this procedure are the basis for developing preliminary equivalents of HRES plant for small-disturbance, transient, frequency and long-term voltage stability studies. Two, data-driven and deep learning-based, methodologies for developing DEMs of HRES plant for transient stability studies using system-identification methods are developed. The first, data-driven, methodology is designed from the perspective of the overall transient stability assessment as the accuracy of the global transient stability status is what is of a critical importance when performing large transient stability studies. On the other hand, the deep learning-based methodology develops DEM in a conventional way, i.e., focusing on the shape of time domain HRES plant power responses. Both methodologies do not require the detailed dynamic data about physical devices in the HRES plant, provide a small set of DEMs capable of covering the most probable HRES plant dynamic behaviour in annual transient stability studies, and include a practical procedure (separate procedures were developed for each methodology) for selecting the adequate equivalent from the set of previously developed DEMs at any time of the year knowing HRES plant operating scenario only. Finally, the thesis presents an exploratory study on the potential challenges in the preservation of transient stability of the TN due to the integration of a geographically distributed HRES plant that has more than one TN connection point. The main findings of the research confirm the ability of DEMs of HRES plant, that were not derived based on the shape of time domain HRES plant power responses, to provide reliable overall transient stability assessment, and the importance of taking into account TN dynamic performance when deciding on the deployment of individual RESs within the spatially distributed HRES plant.
Date of Award1 Aug 2023
Original languageEnglish
Awarding Institution
  • The University of Manchester
SupervisorJovica Milanovic (Supervisor) & Robin Preece (Supervisor)

Keywords

  • deep learning
  • data mining
  • transient power system stability
  • hybrid renewable energy source plant
  • dynamic equivalent model

Cite this

'