This paper proposes a hybrid interval prediction model for Photovoltaic (PV) output by integrating the maximal overlap discrete wavelet transform (MODWT)
complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN)
Bi-directional Temporal Convolutional Network (BiTCN) and Transformer architecture. Firstly
MODWT and CEEMDAN are applied to perform dual decomposition of the PV power time series
yielding intrinsic mode function components. Secondly
the IMFs are categorized into temporal and stochastic components based on their Spearman correlation coefficients. Subsequently
the temporal components are predicted using Transformer
while the stochastic components are predicted using BiTCN. The prediction results from both pathways are then reconstructed to obtain point forecasts. Finally
the point forecasts are expanded into prediction intervals using a kernel density estimation method. Validation on actual case studies demonstrates that this approach
leveraging dual signal decomposition (time and time-frequency domains) and integrating the dual-channel network architecture (BiTCN and Transformer)
achieves an average accuracy improvement of 29.36 % compared to other benchmark prediction models.
Zhang C, Hua L, Ji C L, et al.An evolutionary robust solar radiation prediction model based on WT-CEEMDAN and IASO-optimized outlier robust extreme learning machine[J]. Applied Energy, 2022, 322: 119518.
Zhang D D, Chen B A, Zhu H Y, et al.Short-term wind power prediction based on two-layer decomposition and BiTCN-BiLSTM-attention model[J]. Energy, 2023, 285: 128762.
Pan C, Tan J, Feng D D.Prediction intervals estimation of solar generation based on gated recurrent unit and kernel density estimation[J]. Neurocomputing, 2021, 453: 552-562.