J65.2 New Developments in Weather-Based Power Outage Prediction Modeling

Thursday, 16 January 2020: 10:45 AM
156A (Boston Convention and Exhibition Center)
Diego Cerrai, Univ. of Connecticut, Storrs, CT; and P. Watson, M. Koukoula, F. Yang, and E. Anagnostou

Predicting the impact of storms on electric distribution networks is a very challenging task, because interactions between weather, vegetation and infrastructure are non-linear and chaotic. Historically, utility managers have been predicting power outages by combining the information of weather predictions with their knowledge of the electric power system. However, subjective predictions and evaluations of possible impacts have been difficult to defend in public hearings after the impact of severe storms. More recently, new statistical and machine learning techniques have opened the way to the aggregation of multiple weather, vegetation, infrastructure and land cover inputs for spatially distributed outage prediction. In Connecticut, in the early 2010s, it started to be evident that a system based only on managers' intuition was not an optimal way to manage storm preparedness. The damaging consequences of prolonged power outages had been extensively experienced by the Connecticut population in 2011 and 2012, when two hurricanes (Sandy and Irene) and a strong nor’easter (2011 Halloween nor’easter) hit the State. After the impact of the first two storms, a Two Storm Panel was formed by Governor Malloy to review the prevention, planning and mitigation of impacts associated with emergencies and natural disasters that can reasonably be anticipated. A recommendation of the Two Storm Panel was the development of a collaboration between the State, utilities and a university to develop a more robust hazard assessment capability that can identify 'hot spots' for storm damage and integrate early warning with preparedness and emergency management. This and other recommendations expressed in the report were taken up by the Eversource Energy Center (EEC) at the University of Connecticut. Studying and advancing the predictability of weather related power outages has therefore become a major component of storm preparedness. In the Northeastern United States most power outages are caused by disruption of overhead lines from tree branches due to severe weather events. The amount of these disruptions depends on meteorological and environmental factors that can be quantified and modeled. This presentation aims at summarizing the work performed in the past 4 years for understanding the interactions between overhead lines, vegetation and weather phenomena, through the improvement of an existing outage prediction model (OPM) and the development of new outage prediction models to forecast the number and spatial distribution of power outages in advance of upcoming storms. The importance of storm-caused outage predictions lies in the fact that the knowledge of the expected storm damages allows electric utilities emergency managers effectively allocate crews and resources. Crews in particular may be requested to other out-of-state utilities when an impactful storm is approaching a territory. For a timely emergency response in such situations, the driving time of the crews has to be taken into consideration, and the outage prediction has to be communicated to the emergency managers with a lead-time that depends on the severity of the storm's impact. More in detail, we will describe the improvements of an existing OPM for thunderstorms and extratropical storms prediction in Connecticut, and present a comparison between model predictions using forecasts and analysis. Moreover, we will describe two new outage models for predicting mixed phase precipitation events (blizzards, ice storms, freezing rain), and a model for predicting the impact of the most severe storms.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner