Nonparametric Tree‐Based Predictive Modeling of Storm Outages on an Electric Distribution Network

Abstract

This article compares two nonparametric tree-based models, quantile regression forests (QRF) and Bayesian additive regression trees (BART), for predicting storm outages on an electric distribution network in Connecticut, USA. We evaluated point estimates and prediction intervals of outage predictions for both models using high-resolution weather, infrastructure, and land use data for 89 storm events (including hurricanes, blizzards, and thunderstorms). We found that spatially BART predicted more accurate point estimates than QRF. However, QRF produced better prediction intervals for high spatial resolutions (2-km grid cells and towns), while BART predictions aggregated to coarser resolutions (divisions and service territory) more effectively. We also found that the predictive accuracy was dependent on the season (e.g., tree-leaf condition, storm characteristics), and that the predictions were most accurate for winter storms. Given the merits of each individual model, we suggest that BART and QRF be implemented together to show the complete picture of a storm's potential impact on the electric distribution network, which would allow for a utility to make better decisions about allocating prestorm resources.

Synergistic Use of Nighttime Satellite Data, Electric Utility Infrastructure, and Ambient Population to Improve Power Outage Detections in Urban Areas

“…Outages were defined as individual locations that require a two-man restoration crew to manually intervene and restore power, which are recorded at the nearest upstream isolating device ("asset") to a fault (i.e., downed powerline, broken pole, fuses, reclosers, switches, and transformers). The count of isolating devices per 1 km grid cell were included in the model to represent the amount of infrastructure in a given area, which has been shown to be an important offset in recent studies (e.g., the count of outages per grid cell cannot exceed the count of isolating devices within a grid cell) [3,4]. Furthermore, no dynamics of the actual power grid infrastructure have been included in our study, as each grid cell is treated as spatially independent.…”

“…We selected this model for its ability to discern nonlinear patterns [38]; a feature that has proven useful in a variety of remote sensing studies for change detection and extraction of damaged areas, where the objective is to assess substantial changes between single pixel-based elements [39][40][41]. Other types of machine learning models such as random forest [3,4] and Bayesian additive regression trees [2,4] have been used to relate weather and geographic data to outages, but we believe that this is the first attempt at synergistically combining electrical infrastructure, population, and satellite observations of NTL to quantitatively estimate outages needing repair. To prevent overfitting and ensure model accuracy, we performed a 20-fold cross-validation such that 95% of the data was used for training and 5% of the data was used as an independent validation (without replacement).…”

Spatial Risk Analysis of Power Systems Resilience During Extreme Events

“…This model is able to provide accurate estimates and can be used over a wide area. The accuracy of different types of models for predicting power outages in Connecticut using data from 89 storms of different types and from different seasons to calibrate their models was examined by Wanik, Anagnostou, Hartman, Frediani, and Astitha (2015) and He et al (2016) In these works, which predict the power outages before the landfall of an extreme weather event, historical data about power outages of previous extreme weather events are required in order to train their models while their effectiveness could be limited due to data nonavailability.…”

Agent Based Model to Estimate Time to Restoration of Storm-Induced Power Outages

“…The model could be utilized prior to a storm based on outage predictions [21][22][23][24][25][26] or in real-time as outages are discovered.…”

“…In the future, this novel technique could be incorporated with outage predictions before storms hit [23][24][25][26] to give emergency managers a powerful tool to decrease restoration times in Connecticut and elsewhere. Cost of restoration and mutual assistance crews can be easily added to the model.…”