Photovoltaic Power Generation Power Prediction Method Based on Digital Twins and Deep Learning

A photovoltaic power generation power prediction method based on digital twins and deep learning is proposed to address the issue of multiple influencing factors and low prediction accuracy in photovoltaic power generation. Firstly, a photovoltaic power generation prediction system was constructed based on digital twins, ensuring the reliability of data processing through a combination of virtual and real methods. Then, the dynamic time warping algorithm is used for correlation analysis to obtain the most relevant environmental factors for predicting photovoltaic power generation. Finally, parallel convolutional neural networks and bidirectional long short time memory networks are used to predict photovoltaic power generation, and real-time data is used for correction to obtain the optimal predicted power output. Based on the Matlab platform, experimental analysis was conducted on the proposed method, and the experimental results showed that its MAE and MAPE on rainy, cloudy, and sunny days were 25. S3W, 57.21 W, 49.65 W, and 17.09%, 13.85%, and 6.47%, respectively, indicating significant application value.