Accurate wind power forecasting plays a crucial role in improving wind energy utilization, optimizing power system dispatch, reducing energy waste, and promoting the stable integration of wind power into the grid. This paper proposes a novel short-term wind power forecasting method based on a combination of empirical mode decomposition (EMD) and convolutional long short-term memory networks (ConvLSTM). First, the wind power signal is decomposed into multiple sub-sequences with different frequencies using the EMD algorithm to extract multi-scale features. Then, convolutional operations are embedded into the LSTM architecture, forming a predictive model that captures the temporal-spatial characteristics of each sub-sequence. Finally, the predicted results of all sub-sequences are aggregated to yield the overall wind power forecast. Experimental results demonstrate that the proposed model outperforms traditional forecasting methods in terms of accuracy and stability, making it highly valuable for practical applications.