this paper proposes a combined forecasting model for PV power based on CEEMDAN and a serial CNN-GRU network. Firstly
considering the impact of PV power fluctuations on the forecasting results
CEEMDAN is used to decompose the original PV power into several subsequences to reduce the non-stationarity of the sequence. Each subsequence’s sample entropy(SE) is then calculated to measure its complexity
and subsequences with similar SE values are regrouped to reduce computational load. Secondly
to overcome the limitations of a single neural network in learning the historical characteristics of PV power
a serial CNN-GRU hybrid neural network is proposed to explore the spatiotemporal features of PV powerfully. Each subsequence is input into the serial CNN-GRU network to obtain the forecasting results
and the predicted values of the subsequences are summed to yield the final PV power forecasting results. Finally
case studies on PV power plants in two regions are conducted
with comparative validation against LSTM
GRU
CEEMDAN-LSTM
CEEMDAN-GRU
and the serial CNN-GRU models. The results show that the proposed model achieves superior forecasting accuracy and generalization capability.
TORRES M E, COLOMINAS M A, SCHLOTTHAUER G, et al.A complete ensemble empirical mode decomposition with adaptive noise[C]//2011 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Prague, Czech Republic, 2011: 4144-4147.
RICHMAN J S, MOORMAN J R.Physiological time-series analysis using approximate entropy and sample entropy[J]. American journal of physiology heart and circulatory physiology, 2000, 278(6): H2039-H2049.
TAO Q, LIU F, LI Y, et al.Air pollution forecasting using a deep learning model based on 1D convnets and bidirectional GRU[J]. IEEE access, 2019, 7: 76690-76698.
GAO M Y, ZHANG N, SHEN S L, et al.Real-time dynamic earth-pressure regulation model for shield tunneling by integrating GRU deep learning method with GA optimization[J]. IEEE access, 2020, 8: 64310-64323.