关键词: 光伏功率预测, 模态分解, 卷积神经网络, BiLSTM神经网络, 注意力机制

Abstract: Due to the intermittency and instability of solar radiation, photovoltaic power generation exhibits high randomness and volatility, posing challenges to the stable operation of the power grid. To improve prediction accuracy, a combined forecasting model based on dual-mode decomposition and OOA-CNN-BiLSTM-Attention is proposed. First, Complete Ensemble Empirical Mode Decomposition with Adaptive Noise is used to decompose PV power data, obtaining intrinsic mode functions at different frequencies. Next, based on sample entropy, these functions are clustered into high, medium, and low-frequency components using K-means clustering, and the high-frequency component is further refined with Variational Mode Decomposition. Finally, a composite deep learning prediction model is constructed by integrating Convolutional Neural Network, Bidirectional Long Short-Term Memory network, and Attention mechanism, with hyperparameters optimized using the Owl Optimizer Algorithm. Experimental results show that the proposed model achieves a Root Mean Square Error of 4.11, Mean Absolute Error of 2.88, Mean Absolute Percentage Error of 3.08%, and a Coefficient of Determination of 98.89%, outperforming other models. These results indicate that the method effectively captures the multi-scale features of PV power generation, demonstrating strong generalization ability and application potential.

Key words: PV power forecasting, mode decomposition, convolutional neural network, BiLSTM neural network, attention mechanism