Search

Distributed photovoltaic power generation can efficiently utilize idle resources and reduce carbon emissions. In order to reduce the impact of grid-connected distributed photovoltaic power fluctuations on grid operation, this paper simultaneously exploits the temporal dependence of power series and the spatial correlation of meteorological data to propose a combined prediction model with temporal characteristics and spatial relationships fused for distributed photovoltaic power plants with spatiotemporal information. First, in the study of time-dependent prediction, we propose a long and short-term memory neural network ensemble prediction model based on genetic algorithm-natural gradient boosting, which efficiently fits multiple sets of temporal characteristics of distributed photovoltaic. In the study of spatial correlation prediction, the meteorological data affecting photovoltaic power generation are selected by κ correlation coefficients, the target power plant and spatial reference power plant meteorological data are reconstructed into a two-dimensional matrix, and a two-dimensional convolutional neural network spatial feature extraction power prediction model is designed. Finally, the advantages of the two prediction models of temporal information and spatial features are fused by multiple error evaluation criteria improved information entropy, and a distributed photovoltaic power plant is constructed and implements highly accurate spatiotemporal information combination prediction model. The effect of the forecasting model in this study is validated using the photovoltaic cluster dataset in Hebei Province, China. Compared with other models, the results of this study show that the five prediction performance evaluation metrics of the proposed combined spatiotemporal information model are better. [ABSTRACT FROM AUTHOR]