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

Abstract

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