Photovoltaic Power Prediction Based on Hybrid Deep Learning Networks and Meteorological Data

• DOI: 10.3390/s24051593
• Publication Date: 2024-03-01T08:31:23Z
• Authors: Wei Guo; Li Xu; Tian Wang; Danyang Zhao; Xujing Tang
• Keywords: Quantile regression; Kernel (algebra); Prediction interval

Conventional point prediction methods encounter challenges in accurately capturing the inherent uncertainty associated with photovoltaic power due to its stochastic and volatile nature. To address this challenge, we developed a robust model called QRKDDN (quantile regression kernel density estimation deep learning network) by leveraging historical meteorological data conjunction data. Our aim is enhance accuracy of deterministic predictions, interval probabilistic predictions incorporating quantile (QR) (KDE) techniques. The proposed method utilizes Pearson correlation coefficient for selecting relevant factors, employs Gaussian Mixture Model (GMM) clustering similar days, constructs based on convolutional neural network (CNN) combined bidirectional gated recurrent unit (BiGRU) attention mechanism. experimental results obtained using dataset from Australian DKASC Research Centre unequivocally demonstrate exceptional performance deterministic, interval, (PV) generation. effectiveness was further validated through ablation experiments comparisons classical machine models.