2017/01/23 by Jorge Ángel González Ordiano, Ordiano, Jorge Ángel González, Wolfgang Doneit +9 · 1 citation
Computer Science · Engineering · Environmental Science · #Energy Load and Power Forecasting #FOS: Electrical engineering #Hydrological Forecasting Using AI #Neural Networks and Applications #Solar Radiation and Photovoltaics #Systems and Control (eess.SY) #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.1701.06463
openalex publication_date 2017/01/23 · openalex created_date 2022/10/03 · openalex updated_date 2026/07/28
The present contribution offers a simple methodology for the obtainment of\ndata-driven interval forecasting models by combining pairs of quantile\nregressions. Those regressions are created without the usage of the\nnon-differentiable pinball-loss function, but through a k-nearest-neighbors\nbased training set transformation and traditional regression approaches. By\nleaving the underlying training algorithms of the data mining techniques\nunchanged, the presented approach simplifies the creation of quantile\nregressions with more complex techniques (e.g. artificial neural networks). The\nquality of the presented methodology is tested on the usecase of photovoltaic\npower forecasting, for which quantile regressions using polynomial models as\nwell as artificial neural networks and support vector regressions are created.\nFrom the resulting evaluation values it can be concluded that acceptable\ninterval forecasting models are created.\n