Background
Deployment of next-generation monitoring equipment and upgrade of older control systems will soon allow for near-time supervision and control of power grids over wide areas. A key aspect of control is forecasting of incipient events so that they can be mitigated before they can affect the industry, public services, and the population at large.
The causality chain of incidents is not always clear, so to develop machine learning models that perform well, we must combine data from multiple sources [1]. Once models are constructed, trained, and validated, they can be asked to explain what features they are most sensitive to. Feature relevance may vary in time and be different across event types and geographic regions.
Outline of the Work
Initially, the thesis work will exploit existing data sets. You may opt to retrain models for prediction or reuse the models we have already developed. Based on these models, we will investigate the relevance of different features for different predictions. Initial focus will be on explaining why (and how) the addition of simple seasonal features appears to improve model performance. Depending on interest and progress, we may later include additional data sources, perform feature engineering, extend the data set, or consider different types of events.
Learning Outcome
You will become familiar with (a) data used in power systems monitoring, (b) supervised machine learning techniques, (c) explainable AI techniques, and (d) work with (heavily) imbalanced field data.
Prerequisites
To hit the ground running, you should have passing familiarity with (a) the Python Data Science stack, (b) supervised machine learning algorithms, and (c) understand how to evaluate models. Prior exposure to explainable AI and an understanding of power grids may be helpful, but is not required.
Contact & Questions
Volker Hoffmann (volker.hoffmann@sintef.no) // Research Scientist, SINTEF Digital, Oslo
References
[1] https://ieeexplore.ieee.org/document/9543226/ – Hoffmann, V; Klemets, J.R.A.; Torsæter, B.N.; Rosenlund, G.H.; Andresen, C.A.; “The value of multiple data sources in machine learning models for power system event prediction”, 4th International Conferences on Smart Energy Systems and Technologies (SEST), 2021