A selective memory attention mechanism for chaotic wind speed time series prediction with auxiliary variable
Wind speed prediction is crucial for enhancing the utilization rate of wind energy and optimizing the grid connection of wind power. Its chaotic characteristics and the lack of relevant variables make accurate prediction difficult. Most studies rely solely on historical wind speed, which limits the improvement of accuracy.
Although there are many studies related to wind speed prediction and deep learning models have been widely applied. However, deep learning models are suitable for mining patterns in multivariable and nonlinear data, but the challenge of wind speed prediction often becomes tricky due to the lack of relevant predictor variables. As shown in Figure 1, it is the correlation analysis of wind speed and wind power data at different frequency intervals.
From another perspective, although wind power has a reverse causal relationship with wind speed, the long-term trend of wind power is a reflection of the nonlinear transformation of wind speed. If this trend can be explored and utilized, it will expand the research on wind speed prediction problems.
To address this issue, especially to explore the potential trends of auxiliary variables, the research team innovatively constructed an attention mechanism model for selective memory of auxiliary variables, as shown in Figure 2. The innovation of this method lies in the development of an adaptive frequency-domain selection attention weight operator to adaptively parse the effective information in different frequency domain intervals.
Taking the MAE of 15 minutes and 1 hour as the standard, the actual wind speed prediction error of the proposed method is reduced by 68% and 49% respectively compared with the classical LSTM algorithm. The feasibility of mining reverse causal relationships to improve prediction accuracy was verified. Some of the comparative verification results are shown in Figure 3.
The relevant research results are titled “A selective memory attention mechanism for chaotic wind speed time series prediction with auxiliary variable”. Published in Applied Soft Computing on July 23, 2025.
