Value of Forecast Accuracy and Horizon in Predictive Energy Management of An Islanded Saharan PV–Wind–Battery Microgrid With A Deferrable Water Load: A Nine-Year Simulation Study

Abstract

In an islanded renewable microgrid without a backup generator, control quality depends on the information available about the coming hours. We quantify how forecast accuracy and horizon affect the management of a Saharan PV–wind–battery system (Ouargla). The modelled load comprises a critical part (65%) and a deferrable water station (35%) with a tank. We use nine years of hourly NASA POWER weather (2016–2024) and simulated, lead-time-dependent forecast errors to compare four strategies: a perfect-information oracle, a linear model predictive controller (MPC; 48 h horizon), a rule tuned on an independent year (2015) and a strategy without flexibility. The MPC criterion exceeds that of the oracle by 8.5% (95% bootstrap interval over years: 7.7–9.4%), against 53.4% for the tuned rule and 242% without flexibility. Unserved critical energy is 0.54, 3.33 and 11.46 MWh/year, respectively. Five additional error draws give nine-year mean excesses of 8.7–9.3%. With perfect forecasts the excess is 0.5%, and horizons shorter than 24 h degrade performance in this configuration. With a naive 24 h persistence forecast the excess rises to 59%, no better than the tuned rule, although the critical load stays protected. The results quantify performance under the stated load, component and forecast-error assumptions, and rest entirely on simulation. The fair-reference protocol (tuning on an independent year, paired nine-year evaluation) can be reused to assess other controllers.