A New Modeling Method for Meteorological Information of Regional Distributed Photovoltaic Power Generation Based on Multi-Source Information Fusion

With the increasingly significance of distributed photovoltaic (DPV) generation in modern energy structures, requirements for intelligent operation and accurate power forecasting have grown significantly. Precise meteorological information is the foundation for achieving these functions. However, DPV sites are typically scattered with small installed capacities and generally lack dedicated weather stations. Consequently, developing effective, reliable, and cost-efficient meteorological information computation methods for DPV has become a critical research focus. The Current challenges in DPV meteorological information fusion computation include feature engineering, the reasonable selection of input variables and preliminary establishment of mapping relationships. Additionally, incomplete DPV power data further complicate the situation. To address these challenges, this paper proposes a meteorological information computation method based on multi-source information fusion. Firstly, the paper analyzes the mapping relationships among numerical weather predictions (NWP), DPV site power, and station meteorological information. This analysis demonstrates the possibility of information fusion. Then, a geographic information-based DPV power computation method is proposed to address the low quality of DPV data. Finally, a PV site meteorological information fusion method is developed using Long Short-Term Memory (LSTM) networks, integrating NWP and site power data. Verification using actual data confirms the effectiveness of the proposed methods. The RMSE and MAE for DPV site power calculation are as low as 0.13 and 0.60, respectively. The Pearson correlation coefficient for meteorological information fusion reaches a maximum of 0.99. These metrics outperform those of convolutional neural network (CNN), back propagation (BP), support vector machines (SVM), and theoretical calculation models, demonstrating excellent seasonal adaptability and calculation accuracy.