Modeling Utility Load and Temperature Relationships for Use with Long-Lead Forecasts

Abstract

Models relating system-wide average temperature to total system load were developed for the Virginia Power and Duke Power service areas in the southeastern United States. Daily data for the 1985?91 period were used. The influence of temperature on load was at a minimum around 18°C and increased more rapidly with increasing temperatures than with decreasing ones. The response was sensitive to the day of the week, and models using separate weekdays as well as one using pooled data were created. None adequately accounted for civic holidays or for extreme temperatures. Estimates of average loads over a 3-month period, however, were accurate to within ±3%. The models were used to transform the probability distribution of 3-month average temperatures for each system, derived from the historical record, into load probabilities. These were used with the categorical temperature probabilities given by the National Weather Service long-lead forecasts to estimate the forecast load probabilities. In summer and winter the resultant change in distribution is sufficient to have an impact on the advance fuel purchase decisions of the utilities. Results in spring and fall are more ambiguous.