Power Systems Dynamic Security Assessment using machine learning.

dfe7aeb8-2f4b-420a-b7ab-25d44d1fcbdb

Closed

EPSRC
EPSRC
EPSRC
EPSRC
EPSRC

No funds listed.

Sept. 30, 2018

March 31, 2022

Most countries, including the UK, are promoting the increased penetration of renewable energy sources (RES) to reduce carbon emissions. Ambitious targets are being set for the years to come, mainly because of climate change but also for economic and technical reasons. The generation mix is changing with many large, thermal power stations closing or operating with lower output and more power coming from sources connected via power electronic converters such as wind generators and Photo-Voltaics, raising questions about system security. These changes lead to significantly different dynamic behaviour of the power system that may vary in a both temporal and spatial manner due to the intermittent nature of many of the power electronic interfaced sources. Both the possible pre-disturbance operating conditions and the post-disturbance behaviour of power systems can be significantly affected. These developments, may lead to operation closer to the stability limit and increase the risk of widespread events, that might even lead to blackouts if not acted upon.

With the extensive installation of measurement devices in modern power systems, abundant data are available that hold valuable information about potential patterns of instability. Measurement data can also help in the fast prediction of imminent cascading events and enable automated control actions, all faster than with human operators.

This PhD project will deal with Dynamic Security Assessment (DSA) which focuses on the security of system dynamics in various timescales, from a few up to several seconds. DSA usually requires performing computationally intensive time domain simulations (especially for large power systems). This fact coupled with the increasing temporal and spatial variation introduced by RES as well as the huge number of possible combinations of equipment failure renders the challenge of identifying and predicting situations that might lead to cascading failures highly complex, and calls for the need of novel tools and methodologies.

Data analytics and advances in artificial intelligence (AI) and deep learning have made huge steps in recent years, providing powerful tools that can model the behaviour of complex, highly non-linear systems, such as power systems, with very high accuracy.