Long-term probabilistic forecast of peak electricity demand

24/12/2017 |  Suhoy Tanya
All Press Releases In Subject:
The Economy and Economic Activity


This paper provides probabilistic forecasts through 2050 for the levels and sequences of hours of peak electricity demand. Methodologically, we apply the forecasting model suggested by Hyndman and Fan (2015) (MEFM) with adjustments to the Israeli electricity sector. The estimation of future load curves is conditional on macroeconomic scenarios (population growth, GDP growth, the relative price of electricity, and structural and technological changes), as they affect the trend (seasonal demand). This low-frequency component has been evaluated through dynamic decomposition since 1975, and its parameters provide evidence of a structural change in the late 1990s: the massive penetration of air conditioners increased the seasonal amplitude and the (negative) elasticity of consumption by the relative price of electricity. The high-frequency (hourly) component was estimated relative to the seasonal demand, based on the hourly time series of electricity production in Israel between 2002 and 2005 and the hourly temperatures recorded at Israel Meteorological Service stations. The parameters of the hot-season model provide evidence of a decreasing intraday ratio between the maximum and minimum load (by about 6 percent), as well as decreasing withinseason sensitivity to high temperatures. The flattening of the load curve may be explained by energy efficiency, massive use of charging devices during the night and early morning hours, and a transition to continuous use of air conditioners, including at night. Simulations of future load profiles allow us to evaluate peak load levels and sequences of hours with peak demand that may be exceeded once in 10, 20 or 100 years (Probability of Exceedance Values). The current maximum hourly demand is expected to double by 2050, assuming that a rise in electricity prices does not exceed the inflation rate by more than 10 percent. However, the simulations do not reveal rising sequences of hours with extreme loads.