A photovoltaic power generation prediction method based on SAO-CNN-BiLSTM-Attention
Boyu Zang, He Jiang, Yan Hui Zhao, Mofan Wei · 2024
To improve the precision of photovoltaic (PV) power forecasting, this research introduces a novel hybrid model that integrates several advanced techniques: the improved algorithm for the empirical mode decomposition of adaptive noise complete ensembles (ICEEMDAN), permutation entropy (PE), the snow ablation optimization algorithm (SAO), an attention-driven convolutional neural network (CNN), and a bidirectional long short-term memory network (BiLSTM). Comparative experiments reveal that the proposed SAO-CNN-BiLSTMAttention model significantly outperforms the efficiency of both the CNN-BiLSTM and CNN-BiLSTM-Attention models, demonstrating remarkable accuracy and stability in short-term PV power predictions. This advancement illustrates the efficacy of combining these sophisticated methodologies to enhance forecasting capabilities in the renewable energy sector.