Using multiple linear regression to develop a plant damage model for a major utility company

A “plant damage model” has been developed to relate meteorological conditions to damages incurred by the physical plant of Public Service Electric and Gas (PSE&G), the largest public utility in New Jersey. The model consists of equations derived from a multiple linear regression analysis of observed damage (the predictand) and corresponding observations (the predictors) from a variety of sources including METARs and local storm reports. The analysis gives a different equation for each combination of damage type (e.g., poles down, transformers blown), location (one of four PSE&G “service territories”), and “storm mode” (e.g., winter storm, thunderstorm). Given appropriate observations from (or forecasts of) a future event, the expected amount of damage can thus be determined.

Constructing the model depended on various details, including the selection of predictors, the definition of storm mode, and the preprocessing of the predictands, all of which will be discussed. A case study of a high-impact event will be used to compare output from the new damage model to output from a previous attempt, and implications for other utility companies in other contexts will be discussed. Finally, improvements to the model will be suggested.