Impact of Forecasting Uncertainty on Hybrid Renewable Energy System Sizing in Water Utilities

Abstract

Water utilities are energy-intensive municipal systems, yet high-resolution operational data for planning on-site renewable generation remain scarce. This study assesses the influence of short-term load forecasting accuracy on the sizing of hybrid renewable energy systems that integrate photovoltaics, wind turbines, and battery energy storage. Using one year of hourly demand data from nine sublocations of a Polish water utility and meteorological reanalysis data, four one-hour-ahead models were evaluated: Long Short-Term Memory, Gated Recurrent Unit, Random Forest, and Extreme Gradient Boosting. Tree-based ensemble models demonstrated the most consistent aggregate performance. However, the best-performing model varied by site, with coefficients of determination reaching 0.95. Forecast residuals from each location’s best model were converted into a peak-conditional demand buffer using the 95th percentile of peak-hour errors during peak-demand hours. Hybrid-system sizing was then optimized under a net-zero annual energy constraint for battery storage durations of 0, 2 and 4 h. Forecast uncertainty had little effect on self-consumption but increased installed capacity by up to 58%. These results show that forecasting errors affect renewable energy planning for water utilities primarily through capital oversizing rather than operational inefficiency, highlighting the importance of site-specific forecasting for robust system design.