Research on distributed photovoltaic power prediction based on spatiotemporal information ensemble method

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