Addressing the challenge of low prediction accuracy associated with traditional wind power forecasting methods, a novel integrated approach was proposed, which combined variational mode decomposition (VMD), the information acquisition optimizer(IAO), and the kernel extreme learning machine (KELM) for ultra-short-term wind power forecasting. The original power data, characterized by high nonlinearity, were decomposed into more stable sub-sequences utilizing the VMD technique. The information acquisition optimizer (IAO) algorithm was enhanced through the introduction of a reverse differential mutation mechanism, a fitness-distance balance (FDB) strategy, and an eagle diving strategy, culminating in the development of the improved IAO (IIAO) algorithm. The regularization parameter and the kernel function parameter of the KELM were meticulously optimized using the IIAO algorithm. An integrated prediction model, referred to as VMD-IIAO-KELM, was constructed, with the final predictions derived from the aggregation of the predicted values of each sub-sequence. Results from the case study reveal that the proposed VMD-IIAO-KELM model achieves a reduction in the root mean square error by 9.15%, the mean absolute percentage error by 41.51%, and the mean absolute error by 12.56% in comparison to the conventional KELM model. It is concluded that the application of the proposed model significantly enhances the prediction accuracy of ultra-short-term wind power.