Probabilistic load forecasting using post-processed weather ensemble predictions

Probabilistic forecasting of electricity demand (load) facilitates the efficient management and operations energy systems. Weather is a key determinant load. However, modelling load using weather challenging because relationship cannot be assumed to linear. Although numerous studies have focussed on forecasting, literature uncertainty in while estimating probability distribution scarce. In this study, we model for Great Britain ensemble predictions, lead times from one six days ahead. A comprises range plausible future scenarios variable. It has been shown that ensembles models tend biased underdispersed, which requires are post-processed. Surprisingly, post-processing not yet employed probabilistic forecasting. We post-process based on: (1) output statistics: correct bias dispersion errors by calibrating ensembles, (2) copula coupling: ensure remain physically consistent after calibration. The proposed approach compares favourably case when no information, raw or post-processed without coupling used during modelling.