This paper proposes a hybrid short-term wind speed prediction model based on Variational Mode Decomposition(VMD) and Northern Goshawk Optimization(NGO) optimized LSTM to solve the problem that wind turbines established in areas with many sudden weather changes and complex mountainous areas cannot be dispatched to the grid in a timely manner. To start with, the VMD decomposition algorithm is used to reduce the noise in the original data and generate several relatively stable subseries to improve the predictability of the wind speed series. Each subseries is then individually subjected to the NGO-LSTM's short-term prediction, and the prediction results of each subseries are fused to obtain the final wind speed prediction value. The wind tower data from the Yalong River basin is utilized as an example for the experiment, and the final experimental findings reveal that the suggested method in this research has a greater forecast accuracy.