K-ADBiGRU-AM Short-term Wind Power Prediction Method Based on Adaptive Time Spacing

The use of traditional unsupervised learning can strengthen the correlation between data and improve the model's ability to capture temporal patterns, but at the same time, it also generates the problem of irregular temporal spacing, while ignoring the effect of temporal spacing will limit the model's temporal prediction ability to a certain extent. To address the above problems, an adaptive time-distance bidirectional recurrent gated neural network model K-means adaptive distanced bidirectional gated recurrent unit attention mechanism(K-ADBiGRU-AM) based on clustering processing and attention mechanism is proposed. Firstly, the adaptive distanced(AD) algorithm is proposed to reduce the influence of irregular time distance generated by the clustering algorithm and to adjust the parameters adaptively according to the data characteristics of different wind farms. Further, the bidirectional gated recurrent unit(BiGRU) is combined with the adaptive distanced algorithm to effectively capture the irregular spacing pattern, and finally the attention mechanism(AM) is used to reduce the probability of losing important information. The proposed model can handle the irregular spacing information adaptively and effectively improve the prediction performance of the model for irregular spacing.