Abstract: Accurate and dependable wind speed forecasting plays a guiding role in wind power generation. A novel wind speed forecasting method combined with Vision Transformer（ViT） and long short-term memory（LSTM） is proposed, which is applied in multidimensional time series for achieving one-step ahead and multi-step ahead wind speed forecasting. Firstly, the wind speed is decomposed into multidimensional time series by combining Spearman’s coefficient and variational mode decomposition（VMD） in order to enhance the ability of forecasting model to capture the periodicity and volatility of the wind speed sequence. And then, ViT is used to extract features and hidden information from multi-dimensional time series. Finally, while maintaining the self-attention mechanism advantage of ViT model, the LSTM component is employed to further establish the relationship between extracted features and wind speed. This approach is particularly effective in handling the time-dependent nature of the data and capturing the correlations between variables, so as to improve the generalization and forecasting accuracy of ViT-LSTM model. The experimental analysis is conducted using data recorded from a wind farm in Inner Mongolia. The proposed method achieved a reduction of 15.15%, 34.41%, 68.32%, and 81.71% in mean absolute error compared to the LSTM model at one-step ahead, six-step ahead, twelve-step ahead and twenty-four-step ahead forecasting, which indicates that the proposed model performs well in multi-step wind speed forecasting.