Wind speed prediction has a significant impact on the stable and safe operation of a power system. According to the intermittent and random nature of wind speed, an integrated model of variational modal decomposition (VMD) based on grid search optimization algorithm (GS) and PSO-ELM is proposed for ultra-short-term wind speed prediction. First, the VMD is used to decompose wind speed sequence into a series of sub-sequences. By taking the orthogonality as the fitness function, the GS is used to search the key parameters of VMD adaptively, including the number of decomposed layers and a penalty factor. The purpose is to ensure information orthogonality between each component and to suppress coupling components. Then, the extreme learning machine (ELM) method is used to predict each sub-sequence. Given the unstable prediction of this model, particle swarm algorithm (PSO) is used to optimize the parameters of the initial weight and threshold, and the input dimension of each sub-sequence is determined adaptively by using the auto-regressive integrated moving average model (ARIMA). Finally, the predicted results of each sub-sequence are superimposed to obtain the final predicted wind speed. The result shows that the proposed integrated model is remarkably superior to all considered benchmark models in prediction accuracy. This work is supported by the National Natural Science Foundation of China (No. 51877161).