Work

Uncertainty-aware possibilistic load forecasting for smart grid energy conservation

Year: 2026
Type: article
Abstract: Overview efficient electricity usage management in the presence of deep uncertainty is critical for modern power systems with a high renewable share. We present a fuzzy interval regression framework that represents load observations as triangular fuzzy numbers and estimates α-cut interval forecasts without requiring a full probability distribution. The original eight-hour example is retained only as a transparent calculation: at α = 0.5, the corrected two-weight fit gives an in-sample empirical coverage of 83.3% and a mean interval width of 2.164 MW; these values are descriptive and are neither confidence levels nor predictive validation. To provide out-of-sample evidence, the method was additionally evaluated on the public Taylor half-hourly electricity-demand series for England and Wales (4,032 observations), using the first 3,360 points for estimation and the final 672 points for testing. For the one-step model, the mean absolute error (MAE) was 429.1 MW, the root mean square error (RMSE) was 574.5 MW, and the mean absolute percentage error (MAPE) was 1.45%. The residual-calibrated, time-of-day triangular spreads had an empirical coverage of 91.7% at α = 0.5 and a mean interval width of 1,804.4 MW. An example of demand-response data is given in the form of a 1 MW effort to reduce base load at four defined hours. Assuming a marginal rate of 0.7 tCO₂/MWh, this gross avoided emission would correspond to an estimate of 2.8 tCO₂ not a quantifiable system benefit measurement. The findings favour interpretability and low computational complexity, but practical deployment requires calibrated margins; energy-neutrality or rebound constraints; larger multi-zone trials; and more specific marginal-emission data.
Language: English
Open Access status: diamond
OpenAlex ID: W7226041405
FWCI:
Cites: 24
Cited by:
Related to: 10