Natural and anthropogenic hazards are frequently responsible for disaster events, leading to damaged physical infrastructure, which can result in loss of electrical power for affected locations. Remotely-sensed, nighttime satellite imagery from the Suomi National Polar-orbiting Partnership (Suomi-NPP) Visible Infrared Imaging Radiometer Suite (VIIRS) Day/Night Band (DNB) can monitor power outages in disaster-affected areas through the identification of missing city lights. When combined with locally-relevant geospatial information, these observations can be used to estimate power outages, defined as geographic locations requiring manual intervention to restore power. In this study, we produced a power outage product based on Suomi-NPP VIIRS DNB observations to estimate power outages following Hurricane Sandy in 2012. This product, combined with known power outage data and ambient population estimates, was then used to predict power outages in a layered, feedforward neural network model. We believe this is the first attempt to synergistically combine such data sources to quantitatively estimate power outages. The VIIRS DNB power outage product was able to identify initial loss of light following Hurricane Sandy, as well as the gradual restoration of electrical power. The neural network model predicted power outages with reasonable spatial accuracy, achieving Pearson coefficients (r) between 0.48 and 0.58 across all folds. Our results show promise for producing a continental United States (CONUS)- or global-scale power outage monitoring network using satellite imagery and locally-relevant geospatial data.

A growing number of electricity utilities use machine learning-based outage prediction models (OPMs) to predict the impact of storms on their networks for sustainable management. The accuracy of OPM predictions is sensitive to sample size and event severity representativeness in the training dataset, the extent of which has not yet been quantified. This study devised a randomized and out-of-sample validation experiment to quantify an OPM’s prediction uncertainty to different training sample sizes and event severity representativeness. The study showed random error decreasing by more than 100% for sample sizes ranging from 10 to 80 extratropical events, and by 32% for sample sizes from 10 to 40 thunderstorms. This study quantified the minimum number of sample size for the OPM attaining an acceptable prediction performance. The results demonstrated that conditioning the training of the OPM to a subset of events representative of the predicted event’s severity reduced the underestimation bias exhibited in high-impact events and the overestimation bias in low-impact ones. We used cross entropy (CE) to quantify the relatedness of weather variable distribution between the training dataset and the forecasted event.

Thunderstorms are complex weather phenomena that cause substantial power outages in a short period. This makes thunderstorm outage prediction challenging using eventwise outage prediction models (OPMs), which summarize the storm dynamics over the entire course of the storm into a limited number of parameters. We developed a new, temporally sensitive outage prediction framework designed for models to learn the hourly dynamics of thunderstorm-caused outages directly from weather forecasts. Validation of several models built on this hour-by-hour prediction framework and comparison with a baseline model show abilities to accurately report temporal and storm-wide outage characteristics, which are vital for planning utility responses to storm-caused power grid damage.

Severe weather is a leading cause of electric distribution network failures and customer outages. Predictive modelling of customer outages can mitigate the economic and personal impact of adverse weather but is challenging due to the diverse causes and nonstationarity of customer outage events. Moreover, customer outages may occur during both blue skies or storm weather and only a small proportion of weather events leads to significant outage events. Current approaches for estimating customer outages either assume stationarity and thus underestimate outages in storms or rely on expensive hand labelling of historical and prospective storms, which may or may not be associated with significant outages. In this study, we develop probabilistic storm labelling and outage prediction methods guided by storm probability estimation to improve customer outage prediction in New York State. We partition blue skies and storm weather with outlier detection algorithms, and model customer outages in blue skies weather using an autoregressive neural network and time series models – with model selection based on cross validation and the Akaike information criterion. For nonstationary phenomena like severe weather, we build several machine learning regressors (random forest, k-nearest neighbors, and Poisson) to predict customer outages. Finally, blue skies and storm outage predictions are combined in a single architecture guided by the estimated storm event probability. We demonstrate that, by carefully considering the distinct signatures of storms and blue skies regions, our approach more accurately forecasts customer outages under different storm event scenarios across New York State counties when compared with competing methods.

Power outage restoration following extreme storms is a complicated process that couples engineering processes and human decisions. Emergency managers typically rely on past experiences and have limited access to computer simulations to aid in decision-making. Climate scientists predict that although hurricane frequency may decrease, the intensity of storms may increase. Increased damage from hurricanes will result in new restoration challenges that emergency managers may not have experience solving. Our study uses agent-based modeling (ABM) to determine how restoration might have been impacted for 30 different scenarios of Hurricane Sandy for a climate in 2112 (Sandy2112). These Sandy2112 scenarios were obtained from a previous study that modeled how outages from Hurricane Sandy in 2012 might have been affected in the future as climate change intensified both wind and precipitation hazards. As the number of outages increases, so does the expected estimated time to restoration for each storm. The impact of increasing crews is also studied to determine the relationship between the number of crews and outage durations (or restoration curves). Both the number of outages and the number of crews impact the variability in time to restoration. Our results can help emergency managers and policy makers plan for future hurricanes that are likely to become stronger and more impactful to critical infrastructure.