Enhancing Distribution Grid Failure Data by Leveraging AI-driven text classification: A Danish Case Study

Text classification techniques based on supervised machine learning or natural language models are useful tools for automatically processing text and transforming unstructured text into structured data. However, to successfully apply them to specialised tasks within the power system and distribution grid domain, they need to be contextualized. This requires suitable training data. Fortunately, over the last fifty years, Danish distribution system operators have been saving text-based information in a database on power grid failures. This study leverages the high volume of the failure comments present in the database for the training of text classification models. To illustrate the potential benefits and capabilities of the models, a use case for estimating the root causes of medium-voltage cable failures is proposed. The results indicate that text classification models can positively contribute to data imputation and quality assurance. The study ends with emphasising on how the method could enable more robust models and expand to further applications or domains such as data convergence or named entity recognition.