Prediction techniques for power plant failure and availability: A concise systematic review

Abstract

Electricity demand continues to exceed supply in many sub-Saharan countries like South Africa, and frequent plant failures further reduce energy availability. To address this issue, it is essential to proactively predict plant failures and inform decisions on when to plan for outages. Given a myriad of prediction techniques, this study systematically analyzed various literature to provide a collective view of prediction approaches, their use cases, and context. Following the PRISMA guideline, relevant literature was searched using the Scopus database, and retrieved from the corresponding publisher sites. The selected studies focused on predicting the unplanned capability loss factor or the availability of power plants within the electricity industry domain. A thematic analysis was performed to identify emerging patterns related to current knowledge. Results revealed that prediction studies focus more on predicting availability than failure in coal-fired plants. The prediction horizon is mainly short-term, mostly in renewable plant. Artificial neural network, Bayesian analysis, and fuzzy rules are the prevalent technique found in most studies. Scholars and researchers can benefit from this study as it provided a simplified summary of power plant prediction techniques in a consolidated view.