A hybrid deep learning model for short-term PV power forecasting. (1st February 2020)

• Record Type: Journal Article
• Authors: Li, Pengtao; Zhou, Kaile; Lu, Xinhui; Yang, Shanlin
• Is Part Of: Applied energy. Volume 259(2020)
• Publication Date: 2020-02-01
• Subjects: PV power forecasting -- Deep learning -- Wavelet packet decomposition -- Long short-term memory; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2019.114216
• Languages: English
• ISSNs: 0306-2619

Abstract

Highlights: Short-term PV power forecasting with a hybrid deep learning model. Wavelet packet decomposition and long short term memory network are combined. The liner weighting method for sub-series improves the forecasting results. The proposed model outperforms some other PV power forecasting models. Abstract: The integration of PV power brings great economic and environmental benefits. However, the high penetration of PV power may challenge the planning and operation of the existing power system owing to the intermittence and randomicity of PV power generation. Achieving accurate forecasting for PV power generation is important for providing high quality electric energy for end-consumers and for enhancing the reliability of power system operation. Motivated by recent advancements in deep learning methods and their satisfactory performance in the energy sector, a hybrid deep learning model combining wavelet packet decomposition (WPD) and long short-term memory (LSTM) networks is proposed in this study. The hybrid deep learning model is utilized for one-hour-ahead PV power forecasting with five-minute intervals. WPD is first used to decompose the original PV power series into sub-series. Next, four independent LSTM networks are developed for these sub-series. Finally, the results predicted by each LSTM network are reconstructed and a linear weighting method is employed to obtain the final forecasting results. The performance of the proposed method is demonstrated with a case study using an actual dataset collected from Alice Springs, Australia. Comparisons with individual LSTM, recurrent neural network (RNN), gated recurrent (GRU), and multi-layer perceptron (MLP) models are also presented. The values of three performance evaluation indicators, MBE, MAPE, and RMSE, show that the proposed hybrid deep learning model exhibits superior performance in both forecasting accuracy and stability.