A novel bimodal feature fusion network-based deep learning model with intelligent fusion gate mechanism for short-term photovoltaic power point-interval forecasting

Abstract

Under the goals of carbon neutrality and peak carbon emissions, photovoltaic (PV) power generation is widely valued for its clean and green characteristics. However, the uncertainty and randomness of PV power pose challenges to energy management. Therefore, this study proposed a novel bimodal feature fusion network-based deep learning model with an intelligent fusion gate mechanism for short-term photovoltaic power point-interval forecasting. First, a threshold-guided iNNE-based outlier detection and repair method is designed for preprocessing PV data. Second, a bimodal feature fusion network was proposed to extract global and local features from PV power sequences, and the environmental factors-based rime optimization algorithm with growth mutation strategy and humidity perception mechanism was devised to optimize model's hyperparameters. Additionally, a photovoltaic power interval prediction model with a volatility segmentation strategy was introduced. Finally, the effectiveness of the proposed model, algorithm, and strategies was validated using measured datasets. The results demonstrated that under various weather conditions, the proposed model achieved point prediction evaluation metrics with an R2 exceeding 98 % and a prediction interval evaluation metric with a Prediction Interval Coverage Probability of 85.07 %. The obtained outcomes contribute to providing a basis for decision-making in the scientific scheduling and management of PV power systems.