Electricity Demand Forecasting in Libya under Structural Disruptions: An Ensemble Approach with Recovery and Volatility Indicators

Abstract

Accurate electricity-demand forecasting is essential for generation planning, network investment, maintenance scheduling, and demand-side management. Libya presents a structurally unstable forecasting environment in which long-run growth is interrupted by sharp contractions, rebounds, infrastructure constraints, and political disruption. This study analyzes annual national electricity demand for 2000–2024 using Ember data processed by Our World in Data and evaluates eight transparent forecasting models through expanding-window rolling-origin validation. The naive, Theta, and drift models achieved the strongest out-of-sample performance, whereas unrestricted linear-trend and grey models substantially over-predicted demand. An equal-weight ensemble of the three best models forecasts recorded demand of 37.0 TWh in 2030 and 39.0 TWh in 2035, with a 95% residual-bootstrap interval of 32.5–43.2 TWh in 2035. Historical demand reached 37.99 TWh in 2013, while the 2024 value was 34.59 TWh, equivalent to 91.1% of the historical peak. Per-capita demand recovered to 77.8% of its maximum. The study contributes a reproducible, disruption-aware small-sample framework that integrates model competition, forecast combination, uncertainty intervals, and recovery and volatility indicators. The forecast represents recorded or served demand; latent demand remains outside the empirical scope because public annual data do not contain outages, load shedding, peak deficits, or unserved energy.