|M.Sc Student||Canyasse Raphael|
|Subject||Supervised Learning for Stability Computation|
|Department||Department of Electrical Engineering||Supervisor||Assistant Professor Yoash Levron|
|Full Thesis text|
In this work we design and compare different supervised learning algorithms to estimate the (transient) stability of a power system. Transient stability is defined as the ability of the power system to recover a stable state following large disturbance such as sudden removal or application of load, line faults or switching operations. A main criterion for stability is that synchronous machines maintain synchronism at the end of a period of disturbances whether it is a small or a large disturbances.
We simulate different scenarios such as supply or demand change, branch switched off and perform transient stability assessment (TSA). For each such scenario we discuss its relevance for real-world application. We then constitute a dataset of power system configuration versus its corresponding stability value under the scenario of a supply change. Then we use supervised learning, a sub-field of machine learning, which allows fast prediction of the stability based on network configuration but without any domain knowledge.
The motivation for quick estimation of power system stability stems from the growing need of algorithmic-based long-term and medium-term planning methodologies in power networks.
Our method enables fast approximate estimation of stability for the case IEEE 57, achieved in run-times that are multiple orders of magnitude lower than of exact computation.