In order to guarantee the optimal dispatch of power systems and preserve grid stability, it is crucial to be able to make precise short-term predictions of wind power. Over the past few years, deep learning methods have yielded remarkable outcomes in the domain of Wind Power Prediction (WPP). However, a single prediction model has obvious limitations in performance, inspired by hybrid deep learning models, this paper puts forth a hybrid short-term WPP model founded upon Convolutional Neural Network (CNN) and Transformer. The model consists of an encoder consisting of a convolutional layer of CNN and Transformer in series, as well as a single-layer neural network decoder. First, the model employs box plots to detect outliers in the data and uses spline interpolation to handle outliers to ensure data integrity and consistency. Then, the processed data is fed into the concatenated convolutional layers and Transformer encoder. CNN employs a convolutional layer to capture the local features of WPP, while Transformer utilizes self-attention mechanisms to characterize the long-distance dependencies of WPP. Finally, we chose a single-layer neural network as the decoding layer to output the prediction results, reducing the complexity of the model. We evaluate the proposed model in comparison with existing baselines and assess the effectiveness of the sub-modules through experimentation. The experimental findings show that the CNN-Transformer prediction model, which is the subject of this research, has a quicker prediction speed and a smaller prediction error.