Text data mining as a tool for the asset management decision support process

Utilities are facing significant challenges, driven for example by aging grid assets and the energy transition, in budgeting their operational (OPEX) and capital (CAPEX) expenditures. Data is key in addressing these challenges and will therefore help asset managers to make well-informed decisions. Often, this information is based on structured data, including the age and geographical location of the assets, which can easily be processed by computers. Yet, more than 80% of all the information is embedded in unstructured data such as text in documents. Therefore, analysing text may significantly improve the information quality of decision-making. However, due to its subjective nature, analysing text by computers is not a trivial task. In this paper, a topic model - called LDA (Latent Dirichlet Allocation) - is proposed to automatically cluster lengthy documents. As a case study, a set of DNV GL asset failure documents on cable failures have been analysed using the LDA algorithm. Based on the LDA algorithm, a ”fingerprinting” method is presented, which provides a smart way to select the most relevant documents from a large set if only keywords are given.