Ultra-short-term wind power prediction method based on attention mechanism

Abstract

An ultra-short-term wind power prediction model independent of meteorological data is proposed to solve the problem that there are some wild farms do not have meteorological sensors. The model consists of a two-stream network module and an attention module, one stream is used to extract spatial and time series features, the other stream is used to extract temporal features; the features of the two streams are fused with the attention mechanism to get final feature. Finally, the dimension of the final feature is reduced by the fully connection layers to get ultra-short-time wind power prediction. The validity and practicability of the prediction model are proved by an example analysis of a wind farm in east China for one year, the accuracy of prediction is high, which also provides a strong support for making power generation plan and power dispatching in actual scenarios.