A deep learning-based method for ultra-short-term PV power prediction

Abstract With the rapid development of photovoltaic industry, application accuracy power prediction technology in field system control, dispatching and operation is more precise. Due to volatility randomness ultra short term, high-precision has theoretical practical significance. This paper presents an short-term model based on dual attention mechanism GRU, which realizes term. Firstly, introduced realize extraction temporal spatial features; Then, predicted by combining extracted features with characteristics GRU long-term memory ability fast calculation. The time series independently extract information historical key moments improve stability long-time effect; feature effectively calculates correlation each meteorological quantity, alters weight. Through comparison experiment baseline model, it verified that proposed higher accuracy, better generalization robustness.