xLSTM-Informer 融合的多尺度风电功率预测

打开文本图片集

DOI：10.16652/j.issn.1004-373x.2026.10.007

引用格式： ， . xLSTM-Informer 融合的多尺度风电功率预测[J]. 现代电子技术， 2026， 49（10）： 44-49.

中图分类号：TN911.7-34；TP18

文献标识码：A

文章编号：1004-373X（2026）10-0044-06

Multi-scale wind power forecasting based on xLSTM-Informer fusion

Zhang Xiangyu， Chen Chunmei

（School of Automation， Qingdao University， Qingdao 266071， China）

Abstract： Wind power time series exhibit frequent short-term fluctuations coexisting with complex mid- to long-term trends， which makes it difficult for traditional forecasting methods to balance short-term sensitivity and long-term stability. On this basis， a forecasting model based on xLSTM-Informer fusion mechanism （xLSTM-Informer） is proposed. Adaptive gating is employed to dynamically weight between the recursive memory pathway and the long- sequence attention pathway for the collaborative modeling of multi-time-scale features. On the 15 min resolution dataset of the actual measured wind farm， a multi-step prediction task ranging from 1 to 4 h is constructed for the verification. The results show that the MAE of the proposed forecasting model for 1 h forecasting is 3.070， the RMSE is 4.567， and the trend hit rate can reach 0.667 5， which is significantly better than the comparison baseline models. The ablation experiment and fusion mechanism are carried out to analyze the contribution of each path and the stability of the architecture， thereby validating the rationality and interpretability of the designed structure. The results demonstrate that this model can exhibit excellent performance and application potential in modeling complex time-series features and in engineering scheduling applications.

Keywords： wind power forecasting; xLSTM; Informer; multi-scale modeling; deep learning; time series prediction; energy scheduling

0 引言

风电在能源低碳转型与能源安全中的作用日益突显。（剩余9672字）