A statistical forecast model of weather-related damage to a major electric utility

A model has been developed to relate meteorological conditions to damages incurred by the outdoor electrical equipment (plant) of Public Service Electric and Gas (PSE&G), the largest public utility in New Jersey. Utilizing a perfect prognosis approach, the model consists of equations derived from a backwards eliminated multiple linear regression analysis of observed damage (the predictand) and corresponding surface observations from a variety of sources including local storm reports (the predictors). The analysis gives a different equation for each combination of plant damage element (e.g., poles down, transformers blown), the four PSE&G service territories, and objectively defined storm modes (e.g., Thunderstorm, Heat Wave, None). The predictors used most often were found to be products of maximum wind gust with maximum temperature, daily liquid water equivalent precipitation, and ten day accumulated liquid equivalent precipitation, and were often found to be significant (p-value less than 0.05). The number of severe weather reports provided significant predictors for the Thunderstorm storm mode. The resulting regression equations produced coefficients of determination ranging from 0.032 to 0.697 with the lowest values for the None and Cold storm modes, and the highest values for the Thunderstorm and Mix storm modes. The appropriate model equations were applied to an independent verification dataset and the verification standard deviations were compared to the model derived standard errors which revealed heteroscedasticity (predictand error variance is proportional to the predictand itself) in the model. Both error measurements are calculated assuming independence, and they represent a lower-bound on the error estimation because the training dataset was not transformed into a normal distribution and the use of count data for damaged elements yields a non-independent dataset. Two case studies analyzed to critique model performance yielded insight into model shortcomings where lightning information and wind duration were found to be important missing predictors. The case studies were also used to develop guidelines for applying the model in an operational setting. The development of a damage model for other utility companies in other contexts is discussed.M.S.Includes abstractIncludes bibliographical referencesby Brian John Cerrut