A Study of Photovoltaic Power Prediction Based on Multi-Feature Fusion and Deep Learning

Solar photovoltaic (PV) power generation has recently attracted much attention because it is a safe, clean, and renewable energy source supported by national policies. Even though PV power has quickly advanced, its unpredictability and uncontrollability hampered safety, security, and economic activity. In order to improve grid integration and maintain steady and cost-effective system operations, accurate PV energy forecasting is therefore necessary. Different meteorological factors, including solar radiation, cause variations in PV energy output. This study aims to improve PV energy by applying historical data and deep learning model modelling. Data from China and the United States were used to create three projection models: Long Short-Term Memory, Recurrent Neural Network, and Multilayer Perceptron. Overall, direct and spread radiation were jointly chosen as inputs for the model after examining factors affecting the PV era’s power. To improve forecast correctness by reducing the risk of overfitting, a multi-feature integration model was created that combines historical and meteorological data. According to the experimental results, the multi-feature fusion design based on the LSTM system model outperforms other quantitative models.