An ontological approach to exploit Natural Language Processing for blackout analysis in power systems

The objective of this thesis is to propose a framework for the automatic collection and analysis of data about blackouts, based on an ontological approach and Natural Language Processing (NLP), which extracts information from human-written text, allowing researchers to better focus on the entities that play a major role in power outages and the relations among them. After automatically retrieving news articles from Google News, an ontology suitable for blackout analysis, the real core of this work, was designed and different experiments were conducted to study its impact on the NLP model for Named Entity Recognition (NER) and Relationship Extraction (RE). The comparative analysis shows that by using the proposed ontology, even with a very small dataset, the results are promising. However, a larger dataset, especially if composed of technical reports, would be beneficial to use the ontology in practice, thus finally covering the gap between the exploitation of NLP for blackout analysis and its application in other fields.