Data-driven seasonal load uncertainty modeling for Unit Commitment and Economic Dispatch applications

Abstract

Introduction: Uncertainty characterization is an important component of Unit Commitment and Economic Dispatch (UCED) studies, where load uncertainty can influence commitment decisions, dispatch schedules, reserve allocation, and operating costs. However, many UCED-oriented studies represent load uncertainty using fixed percentage deviations, predefined uncertainty bands, or externally specified probabilistic inputs, which may not reflect the empirical behavior of the target load profile. Materials and methods: This paper proposes a Kernel Density Estimation (KDE)-assisted data-driven framework for deriving load-profile-specific and season-specific uncertainty bounds for electric and heating loads as candidate inputs to UCED-oriented uncertainty modeling. A Deep Recurrent Neural Network (DRNN) with stacked Long Short-Term Memory (LSTM) units is first used as the forecasting engine to generate four-day-ahead electric- and heating-load forecasts. The ratios of actual-to-forecast load are then analyzed using KDE, and empirical coverage intervals centered around the nominal ratio of unity are used to derive seasonal uncertainty bounds. The 90 % empirical coverage interval is used as the baseline uncertainty bound, while 80 % , 85 % , and 95 % intervals are also reported to examine sensitivity to the selected coverage level. Results: Results show clear seasonal variation at the 90 % coverage level, with spring exhibiting the largest uncertainty bounds of ± 50 % for electric load and ± 54 % for heating load, while autumn shows the smallest bounds of ± 6 % and ± 5 % , respectively. Winter and summer show intermediate bounds of ± 31 % and ± 30 % , and ± 27 % and ± 29 % , respectively. A microgrid UCED test case further shows that, relative to a fixed 20 % uncertainty bound, the empirical bounds increase operating costs in winter but reduce them in autumn. Conclusions: These findings indicate that fixed uniform deviation assumptions may be inadequate for representing empirical load uncertainty and motivate the use of data-driven uncertainty bounds in UCED input modeling. The UCED results further demonstrate the operational and economic importance of season-specific uncertainty modeling.