Negative Binomial Regression of Electric Power Outages in Hurricanes Haibin Liu 1 ; Rachel A. Davidson, A.M.ASCE 2 ; David V. Rosowsky, M.ASCE 3 ; and Jery R. Stedinger, M.ASCE 4 Abstract: Hurricanes can cause extensive power outages, resulting in economic loss, business interruption, and secondary effects to other infrastructure systems. Currently, power companies are unable to accurately predict where outages will occur. Therefore, it is difficult for them to deploy repair personnel and materials, and make other emergency response decisions in advance of an event. This paper describes negative binomial regression models for the number of hurricane-related outages likely to occur in each one square kilometer grid cell and in each zip code in a region due to passage of a hurricane. The models are based on a large Geographic Information System database of outages in North and South Carolina from three hurricanes: Floyd 1999, Bonnie 1998, and Fran 1996. The most useful explanatory variables are the number of transformers in the area, the company affected, maximum gust wind speed, and a hurricane effect. Wind speeds were estimated using a calibrated hurricane wind speed model. Pseudo R-squared values and other diagnostic statistics are developed to facilitate model selection with generalized negative binomial models. DOI: 10.1061/ASCE1076-0342200511:4258 CE Database subject headings: Hurricanes; Electric transmission; Electric power outages; Geographic information systems; Wind; Regression models. Introduction This paper describes negative binomial regression models devel- oped to estimate the geographic distribution of electric power distribution system outages in North and South Carolina as the result of hurricane winds. Hurricanes have caused severe power interruption throughout the Atlantic and Gulf coast regions of the United States. For example, in 2003, 1.8 million of Dominion Energy’s customers 82% lost power during Hurricane Isabel Dominion Energy 2003. In 1992, Hurricane Andrew caused 1.4 million of Florida Power and Light Company’s customers 44% to lose power Larsen et al. 1996. Hurricane-related damage to the electric power system causes significant econo- mic loss, business interruption, and costly restoration efforts. In addition to considerable direct repair and restoration costs, power outages may result in interruption of security systems, finan- cial transactions, communication, health care, water distribu- tion, traffic signaling, and other lifeline systems that depend on electricity. Accurate outage occurrence predictions can help in managing preparedness and restoration efforts. When a hurricane is ap- proaching a region, power companies try to predict which areas will lose power and, based on those estimates, they deploy re- pair personnel and equipment. Underestimating repair needs can cause an undesirable delay in the restoration of service. Over- estimation can result in unnecessary expense and a less than op- timal deployment of resources. Power companies currently use outage prediction methods that are largely based on informal comparison of the incoming hurricane track with recent experi- ence, or on simple empirical models of power outages versus maximum wind speed. The model developed in this paper relates outage occurrences to explanatory variables in a consistent statis- tical framework. The resulting model should enable utilities to improve outage occurrence estimates before a hurricane makes landfall, thus potentially making response activities more effec- tive and efficient. The regression analysis relies on a large database of recent hurricane outages experienced by Progress Energy formerly Carolina Power and Light and Duke Energy, two major electric power companies that together serve much of North and South Carolina Fig. 1. Model development required integrating many large databases with a Geographic Information System GIS, then applying spatial and statistical modeling techniques. The re- sults show valuable predictive capability and promise for future application. The electric power system and typical hurricane dam- age patterns are discussed in the next section. Later sections de- scribe model development and results. 1 Graduate Student, School of Civil and Environmental Engineering, Hollister Hall, Cornell Univ., Ithaca, NY 14853-3501. E-mail: hl254@cornell.edu 2 Assistant Professor, School of Civil and Environmental Engineering, Hollister Hall, Cornell Univ., Ithaca, NY 14853-3501 corresponding author. E-mail: rad24@cornell.edu 3 Department Head and A.P. and Florence Wiley Chair Professor, Zachry Dept. of Civil Engineering, Texas A&M Univ., 3136 TAMU, College Station, TX 77843-3136. E-mail: rosowsky@tamu.edu; formerly, Professor, Dept. of Civil Engineering, Apperson Hall, Oregon State Univ. 4 Professor, School of Civil and Environmental Engineering, Hollister Hall, Cornell Univ., Ithaca, NY 14853-3501. E-mail: jrs5@cornell.edu Note. Discussion open until May 1, 2006. Separate discussions must be submitted for individual papers. To extend the closing date by one month, a written request must be filed with the ASCE Managing Editor. The manuscript for this paper was submitted for review and possible publication on February 28, 2003; approved on September 28, 2004. This paper is part of the Journal of Infrastructure Systems, Vol. 11, No. 4, December 1, 2005. ©ASCE, ISSN 1076-0342/2005/4-258–267/$25.00. 258 / JOURNAL OF INFRASTRUCTURE SYSTEMS © ASCE / DECEMBER 2005