A PV power interval forecasting based on seasonal model and nonparametric estimation algorithm. (15th May 2019)

• Record Type: Journal Article
• Authors: Han, Yutong; Wang, Ningbo; Ma, Ming; Zhou, Hai; Dai, Songyuan; Zhu, Honglu
• Is Part Of: Solar energy. Volume 184(2019)
• Page Start: 515
• Page End: 526
• Publication Date: 2019-05-15
• Subjects: PV power forecasting -- Extreme learning machine -- Kernel density estimation -- PV power fluctuation -- Seasonal model; Solar energy -- Periodicals; Solar engines -- Periodicals; 621.47
• DOI: 10.1016/j.solener.2019.04.025
• Languages: English
• ISSNs: 0038-092X

Abstract

Highlights: A new method for PV power interval forecasting based on nonparametric estimation is proposed. The seasonal multi-model method is applied on the interval forecasting. The PV power output and fluctuation are of obvious seasonal characteristic. The proposed method provides narrower interval width and better coverage probability. Abstract: With the continuous increase of grid-connected photovoltaic (PV), high-precision PV power prediction is increasingly important. Extant deterministic forecasting methods do not facilitate fully effective dispatching decisions or power grid risk analysis. Furthermore, the single model also has insufficient generalization ability and unstable forecasting performance in PV power forecasting. This paper proposes an alternative multi-model PV power interval forecasting method which takes into account the seasonal distribution of power fluctuation characteristics. PV output power, absolute power deviation, and relative variation rate are first analyzed for the seasonal distribution characteristics of PV output as they fluctuate over time. Seasonal multi-models for deterministic forecasting of PV power are then built based on an extreme learning machine (ELM). Deterministic forecasting error is fitted by kernel density estimation to complete the PV power interval forecast. The effectiveness of the proposed method is validated by comparison against other methods.