Search for: All records

This paper discusses how the risk of electricity grid outages is predicted using machine learning on historical data enhanced by graph embeddings of the distribution network. The process of graph creation using different embedding approaches is described. Several graph constructing strategies are used to create a graph, which is then transformed into the form acceptable for ML algorithm training. The impact of incorporating different graph embeddings on outage risk prediction is evaluated. The method used for graph embeddings is Node2Vec. The grid search is performed to find optimal hyperparameters of Node2Vec. The resulting accuracy metrics for a set of different hyperparameters are presented. The resulting metrics are compared against base scenario, where no graph embeddings were used. 

Abstract A novel method for real-time solar generation forecast using weather data, while exploiting both spatial and temporal structural dependencies is proposed. The network observed over time is projected to a lower-dimensional representation where a variety of weather measurements are used to train a structured regression model while weather forecast is used at the inference stage. Experiments were conducted at 288 locations in the San Antonio, TX area on obtained from the National Solar Radiation Database. The model predicts solar irradiance with a good accuracy (R2 0.91 for the summer, 0.85 for the winter, and 0.89 for the global model). The best accuracy was obtained by the Random Forest Regressor. Multiple experiments were conducted to characterize influence of missing data and different time horizons providing evidence that the new algorithm is robust for data missing not only completely at random but also when the mechanism is spatial, and temporal. 

A new predictive risk-based framework is proposed to increase power distribution network resiliency by improving operator understanding of the status of the grid. This paper expresses the risk assessment as the correlation between likelihood and impact. The likelihood is derived from the combination of Naive Bayes learning and Jenks natural breaks classifier. The analytics included in a GIS platform fuse together a massive amount of data from outage recordings and weather historical databases in just one semantic parameter known as failure probability. The financial impact is determined by a time series-based formulation that supports spatiotemporal data from fault management events and customer interruption cost. Results offer prediction of hourly risk levels and monthly accumulated risk for each feeder section of a distribution network allowing for timely tracking of the operating condition.