Predicting power outages utilizing machine learning and sensitivity analysis

Abstract

Due to climate change and global warming, power outages have been increasing in frequency and severity. Such power outages not only cause disruptions to day-to-day life but also endanger hospital patients and people using power-reliant medical equipment at home, as well as elderly people relying on cooling and heating. With such an increase in power outages, it becomes imperative to find a solution to mitigate the impacts and frequency of such power outages. This paper analyzes the relationships between power outages and climate factors, such as temperature, precipitation, humidity, wind speed, and solar radiation to determine which are the most significant. Machine learning techniques are used to develop predictive models to identify climate conditions that are likely to cause power grid failures. In addition, this paper also investigates the use of these models in analyzing the resiliency of power grids. This study is focused on the West Coast of the United States. The climate factors and power outage data from 2017-2023 are used in this analysis. Artificial Neural Networks (ANN) were developed to perform the analysis. The results obtained during this research indicated that the ANN models were not effective in predicting power outages based on weather factors for the locations chosen. This is likely due to the highly skewed data sets, as well as existing robustness of these locations against weather-related factors. Future work should include applying such modeling on other locations that are more susceptible to weather-related power outages as well as investigating other modeling techniques.