Some Case Studies of Power Outages with Possible Machine Learning Strategies for their Predictions

Abstract

Power outage has serious consequences for communities and the energy sector encompassing all aspects of life. In spite of continuous surveillance by personnel, power outages do occur in many parts of the world. This year alone, there have been power outages in Norway, Singapore and USA (California). This paper presents some analysis of historical data from power outages with an overview of the main causes for these outages based on case studies from selected regions. Very often, power outages can be to a certain extent predicted based on analysis of historical data. Using historical data and Time Delay Neural Network (TDNN), weather conditions can be predicted. The outage scenarios from historical data indicate that birds/animals, falling trees, thunderstorm and wind are the main causes of outages. In this paper, a regression model classifying outages with an accuracy of 74% due to animals and storms. Interestingly, for the historical data used in the analytics, the cause of outages due to birds is more prevalent when temperatures exceed 12°C. With surveillance drones increasingly used in many sectors, also in the energy sectors in the recent past, these findings indicate possible strategies for implementing preventive actions in avoiding power outages. As an industrially relevant problem, this project facilitated cross-disciplinary thinking and PBL mode in the teaching and learning processes.