Wind speed interval prediction captures the fluctuation range of actual wind speed, effectively reflecting its randomness and uncertainty. However, due to the multi-scale fluctuation characteristics of wind speed interval sequences, a single prediction model often fails to fully represent their complex dynamics, leading to limited prediction performance. To address this issue, this paper proposes a short-term wind speed interval prediction method based on a temporal pyramid dual-layer ensemble learning architecture. The method mainly comprises three parts: data preprocessing optimization, multi-model ensemble prediction mechanism construction, and ensemble output optimization. The data preprocessing optimization employs the red-billed blue magpie optimizer to optimize the parameters of variational mode decomposition, enabling effective signal decomposition in the frequency domain. In the multi-model ensemble prediction stage, a temporal pyramid attention mechanism is introduced to integrate the advantages of multiple models and perform differentiated modeling of multi-scale feature patterns in sub-sequences. In the ensemble output optimization stage, the grey wolf optimizer is used to optimize the output weights of sub-sequence predictions, capturing the varying feature correlations between each sub-sequence and actual wind speed. Case studies demonstrate that the proposed method can significantly improve the accuracy of short-term wind speed prediction, and has strong generalization ability and stability.