Answers are grounded in this paper and cite their sources.
Accurate forecasting of power generation from photovoltaic (PV) power plant clusters can support the safe and stable operation of the power grid. Individual PV power plants cover relatively small areas, and weather conditions within each plant are often relatively similar. Forecasting methods based on weather-pattern classification can help account for differences in input–output relationships across weather conditions. However, clusters of PV power plants typically span hundreds to thousands of square kilometers, where weather types across individual plants within the region are inconsistent. This inconsistency makes it difficult to directly apply existing classification modeling methods to the entire area. Therefore, this paper proposes a method for day-ahead power forecasting of PV power plant clusters based on sub-region extended weather patterns and dynamic graph attention networks. First, based on the total historical PV generation of the region, a Gaussian mixture clustering algorithm is employed to obtain three types of composite weather patterns and their corresponding benchmark weather scores. Second, considering the spatial distribution of each site, multiple subregions are delineated. Based on forecasted irradiance, the weather pattern for each subregion on the target day is first determined. Then, by integrating weather scores and indicator-specific historical contribution weights, the generalized weather pattern for the entire region on the target day is derived. Finally, considering the variations in PV output characteristics under different generalized weather patterns, a regional PV power forecasting model tailored to each pattern is constructed using dynamic graph attention and kernel density estimation. Case-study experiments using data from a PV plant cluster in northern China are used to assess the forecasting performance of the proposed method.
No takes yet. Share an insight, caveat, or question.
Su et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: