Electric Power Systems Research 77 (2007) 973–979 Storm modeling for prediction of power distribution system outages Dan Zhu a,∗ , Danling Cheng a , Robert P. Broadwater a , Charlie Scirbona b a Department of Electrical and Computer Engineering, Virginia Tech, Blacksburg, VA 24060, USA b Distribution Engineering, Orange and Rockland Utilities, One Blue Hill Plaza, Pearl River, NY 10965, USA Received 9 June 2006; accepted 23 August 2006 Available online 11 October 2006 Abstract Utility companies spend tremendous amounts of money on storm outage restoration. A storm model which can accurately predict time-varying outages for approaching or ongoing storms will help with outage management. With such a model both customer down time and outage costs can be reduced. This paper proposes a two-stage prediction method to forecast storm related outages. In the first stage, historical outage and weather data are employed to create empirical models for different types of storms. In the second stage, observers are employed for tracking storm outages in real time. The effectiveness of the prediction model is evaluated with simulations applying actual storm data. Lightning is a primary cause of outages during summer storms. Results are presented on a possible relationship between flash density and outages. © 2006 Elsevier B.V. All rights reserved. Keywords: Storm outages; Storm classification; Lightning caused outages; Outage prediction; Circuit corridor; Flash density 1. Introduction Major impacts of severe weather conditions on electric util- ity operations are equipment failures and power outages. It is estimated that an average storm costs an average utility about $ 100,000 h -1 including materials and labor. For large utilities, storm costs can reach $ 1,000,000 h -1 . Manpower is a significant part of power restoration cost. A utility usually needs to send out a large number of crews to restore services during a storm. Some of the crews may be borrowed from neighboring companies or are contract crews that are hired during large emergencies. It is a challenging task to manage crews effectively during a major storm. If the number of outages that are going to occur as a function of time can be accurately predicted, then the management of crews and the power restoration can be more effectively planned. The objective of this paper is to investigate storm models that may be used to predict outages due to approaching or ongoing storms. These models are functions of storm conditions. Accu- rate storm models hold the potential for lowering operational costs and reducing customer downtime. ∗ Corresponding author. Tel.: +1 540 951 5236. E-mail address: dzhu@vt.edu (D. Zhu). Previous research has been performed on storm related power outages. Early in the 1980s, Ref. [1] proposed using a database of customer no-light calls to generate outage patterns. However, the outage patterns were not successful for use in identify- ing faulted sections. Ref. [2] uses a two-stage Monte Carlo simulation to evaluate the impact of high wind storms. This work quantifies storm interruptions in a power distribution sys- tem. In order to improve outage restoration, Ref. [3] suggests a rule-based prediction for determining outage locations from the available call patterns and telemetered data. None of the ‘pre- diction’ models in Refs. [1–3] provide an outage forecast ahead of time. Ref. [4] proposes a predictor which applies a data grouping method and neural networks to estimate the amount of dam- age due to typhoons. Similar to Ref. [4], the work here uses a two-stage process. Instead of typhoons, typical storms in the northeastern section of the United States are analyzed and an observer [5] is used instead of neural networks in the second stage. Ref. [6] presents a general procedure for storm outage man- agement. It utilizes the information from weather forecasts and historical data to predict the storm damage in advance, and uses the prediction to manage the crews. However, Ref. [6] did not provide any specific storm outage models, and furthermore did not provide the details of how storm related information may be used to produce storm outage models. 0378-7796/$ – see front matter © 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.epsr.2006.08.020