Accurate and reliable wind speed prediction is crucial for efficient wind energy utilization.Considering the inherent instability of wind speed,we propose an SWD-Transformer-KAN prediction model that integrates Swarm Decomposition (SWD),Transformer architecture,and Kolmogorov-Arnold Network (KAN).First,the raw wind speed data are decomposed via the SWD method to extract key features.Subsequently,a Transformer-KAN prediction model is established for each decomposed subsequence,which fully utilizes Transformer's temporal processing capabilities and KAN's nonlinear approximation strengths.Final wind speed predictions are obtained by aggregating all subsequence forecasts.Experimental comparisons demonstrate that the proposed SWD-Transformer-KAN model exhibits superior predictive performance,achieving a coefficient of determination (R²) of 99.91%.