CNN-BiLSTM Hybrid Model Optimized by Sparrow Search Algorithm for Photovoltaic Power Prediction

Against the backdrop of accelerating global energy transition, photovoltaic (PV) power generation technology has emerged as a core pathway to alleviate energy supply-demand imbalances and address climate change, owing to its high efficiency and low carbon emissions. However, PV output power is subject to significant non-stationarity and uncertainty due to the coupled influence of multiple factors, such as fluctuations in solar radiation intensity, dynamic cloud cover, and ambient temperature variations. These factors pose potential risks to the safe and stable operation of power systems. Accurate prediction of PV power generation is a critical step in optimizing grid dispatch and enhancing the integration capacity of renewable energy. Addressing the limitations of existing prediction models, such as insufficient adaptability under complex weather conditions and reliance on empirical settings for hyperparameter optimization, this paper proposes a hybrid prediction framework based on Sparrow Search Algorithm (SSA)-optimized CNN-BiLSTM. The model enhances global optimization capabilities through the introduction of a refraction reverse learning strategy, improves robustness by combining a dynamically adjusted sine-cosine optimization mechanism with a Cauchy mutation strategy, and employs SSA to adaptively optimize the hyperparameters of the Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) networks. This approach fully exploits the spatiotemporal nonlinear correlation features within PV power sequences. Experimental results demonstrate that the optimized model significantly reduces prediction errors across various weather scenarios, with generalization capabilities and stability markedly superior to traditional methods, validating its effectiveness in modeling complex nonlinear relationships. This study provides theoretical innovation and technical support for PV power prediction, offering significant practical value for enhancing grid resilience and promoting the large-scale application of renewable energy.