2020/12/23 by Alessandro Brusaferri, Brusaferri, Alessandro, Matteo Matteucci +5 · 2 citations
Computer Science · Decision Sciences · Engineering · #Energy Load and Power Forecasting #FOS: Computer and information sciences #FOS: Electrical engineering #Grey System Theory Applications #Image and Signal Denoising Methods #Machine Learning (cs.LG) #Signal Processing (eess.SP) #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2012.14389
openalex publication_date 2020/12/23 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Probabilistic load forecasting (PLF) is a key component in the extended\ntool-chain required for efficient management of smart energy grids. Neural\nnetworks are widely considered to achieve improved prediction performances,\nsupporting highly flexible mappings of complex relationships between the target\nand the conditioning variables set. However, obtaining comprehensive predictive\nuncertainties from such black-box models is still a challenging and unsolved\nproblem. In this work, we propose a novel PLF approach, framed on Bayesian\nMixture Density Networks. Both aleatoric and epistemic uncertainty sources are\nencompassed within the model predictions, inferring general conditional\ndensities, depending on the input features, within an end-to-end training\nframework. To achieve reliable and computationally scalable estimators of the\nposterior distributions, both Mean Field variational inference and deep\nensembles are integrated. Experiments have been performed on household\nshort-term load forecasting tasks, showing the capability of the proposed\nmethod to achieve robust performances in different operating conditions.\n