Accurate wind power prediction is of vital importance for the secure and stable operation and economic dispatch of new-type power systems. To address the issue that traditional point prediction fails to reflect the uncertainty of wind power, a combined model of short-term point and interval prediction was proposed. First, the rime optimization algorithm (RIME) was used to optimize the parameters of variational mode decomposition (VMD), and VMD decomposition was performed on wind power. Then, meteorological factors with high correlation to wind power were selected using the Pearson correlation coefficient method,which were used as input for the convolutional neural network-bidirectional long short-term memory network(CNN-BiLSTM) prediction model. The point prediction was obtained by summing the predictions of all decomposed components. Next, based on the point forecast, an adaptive kernel density estimation (AKDE) model was constructed for interval prediction. The AKDE method improved robustness under different confidence levels and effectively quantifies uncertainty in wind power prediction. Finally, the model was validated using measured data from a wind farm in Xinjiang. The results show that the proposed method improves the accuracy of deterministic forecasting and enhances the robustness of interval prediction.