Research Centre: Research Centre for Systems & Control
Title: Facing the Uncertainty: Data-Driven Power System Operation
Speaker: Dr Ioannis Konstantelos, Imperial College London
The uncertainty that characterises power system operation has led to a substantial increase of possible system states to be analysed, rendering real-time dynamic stability assessment a very challenging task. Machine learning has been used in the past for the construction of surrogate models (security rules) that detect the system’s stability boundary on the basis of historical data. Nevertheless, machine learning techniques can face severe challenges when applied to large power systems.In this presentation we will discuss various techniques recently developed around such data-driven workflows. In particular, we will describe a new class of high-dimensional statistical models that combine clustering, dimension reduction and vine copulas to generate training data sets of arbitrarily high density. In addition, we will showcase the benefit of deep-learning based feature selection to improve the quality of security rules. We will then introduce the challenges related to embedding security proxies to traditional optimisation problems and show how ensemble methods can be leveraged to navigate the cost-risk trade-off that arises. Finally, we will present some results of data-driven workflows when analysing a large part of the European grid and highlight some open research questions.
Ioannis Konstantelos is Lead Optimisation Engineer at Flexciton and a Research Fellow in the Control and Power research group, Electrical and Electronic Engineering, Imperial College London. He obtained his PhD from the same institution in 2013 on the topic of large-scale stochastic planning models. His research focuses on the intersection of optimisation, machine learning and data analytics applied to highly complex energy and process systems
Share this event
When and where
1.00pm - 2.00pmTuesday 16th October 2018