KIT | KIT-Bibliothek | Impressum | Datenschutz

Spatio-Temporal Short Term Photovoltaic Generation Forecasting with Uncertainty Estimates using Machine Learning Methods

Jongh, Steven de; Riedel, Tobias; Mueller, Felicitas; Yacoub, Ala Eddine; Suriyah, Michael; Leibfried, Thomas


The availability of accurate short-term forecasts of the infeed of volatile renewable power generation is of crucial importance for the safe and economic operation of future energy networks. Classical numerical weather prediction (NWP) methods offer a limited temporal and spatial resolution that is not suitable for the prediction of the infeed of individual photovoltaic (PV) systems. By combining weather data from a spatially distributed sensor network, improvements of short-term forecasts can be achieved. The suitability of a data set consisting of multiple PV sites is analyzed for this task. In this paper, spatiotemporal nowcasting methods are developed based on statistical and machine learning algorithms. The methods considered are Auto Regressive-Moving Average (ARMA) and Long Short-Term Memory (LSTM) neural networks. The increase of performance of using data from multiple sites is highlighted. Additional estimates of the forecasting uncertainty are developed using quantile regression (QR).

DOI: 10.1109/UPEC49904.2020.9209764
Zitationen: 4
Zugehörige Institution(en) am KIT Institut für Elektroenergiesysteme und Hochspannungstechnik (IEH)
Publikationstyp Proceedingsbeitrag
Publikationsmonat/-jahr 09.2020
Sprache Englisch
Identifikator ISBN: 978-1-7281-1079-0
KITopen-ID: 1000124204
HGF-Programm 37.06.01 (POF III, LK 01) Networks and Storage Integration
Erschienen in 2020 55th International Universities Power Engineering Conference (UPEC), Torino, Italy, 1-4 Sept. 2020
Verlag Institute of Electrical and Electronics Engineers (IEEE)
Seiten 1–6
Schlagwörter forecasting, Spatio Temporal, Machine, Learning, Quantile regression, LSTM
Nachgewiesen in Dimensions
KIT – Die Forschungsuniversität in der Helmholtz-Gemeinschaft
KITopen Landing Page