Short-term photovoltaic power point-interval forecasting based on double-layer decomposition and WOA-BiLSTM-Attention and considering weather classification

A reliable short-term forecast of photovoltaic power (PVPF) is essential to maintaining stable power systems and optimizing power grid dispatch. A hybrid prediction framework of PVPF considering similar day screening, signal decomposition technique, and hybrid deep learning is proposed to realize accurate point-interval prediction. First, a double similar day screening model is constructed to divided weather into the three types, which improves the quality of the training set. Second, a double-layer signal decomposition model based on improved Complementary Ensemble Empirical Mode Decomposition (ICEEMD)-Variational mode decomposition(VMD) is proposed to preprocess the data, eliminate the noise data. Third, Whale optimization algorithm(WOA) is introduced to search the best hyper parameters of BiLSTM, and the attention mechanism(AM) is adopted to focus on the influence of key information. WOA-BiLSTM-AM is constructed to predict intrinsic mode function(IMF). Fourth, Kernel density estimation (KDE) is applied to estimate the PV prediction interval at different confidence levels. Finally, compared with the benchmark models, the prediction errors of cloudy, sunny and rainy days are reduced by 56.09%, 80.10% and 71.98%, respectively. In addition, according to the coverage width criterion, KDE (Gaussian) is superior to other interval prediction models and can achieve more reliable prediction interval.