Abstract: In this study, a combined prediction model based on the mutual information (MI) method, convolutional neural network (CNN), long short-term memory (LSTM) network, attention mechanism (AT), and particle swarm optimization (PSO), is proposed to successfully predict wind farm power and extract the spatiotemporal information and potential connections between the adjacent sites of the wind farm. Here, CNN is used to extract the spatial features of different sites, and LSTM is employed to obtain the time-dependent information of the wind power data from multiple sites. Furthermore, a CNN-LSTM spatiotemporal prediction model is designed and used along with deep learning algorithms, such as MI, AT, and PSO, to further improve the proposed model. The analysis of the experimental data of the two island wind farms reveals that the proposed model achieves the lowest statistical error. The CNN-LSTM model can efficiently extract the spatiotemporal information of the wind farm and perform time series forecasting; moreover, the model combined with deep learning algorithms (MI, AT, and PSO) can further improve the accuracy and stability of wind power forecasting.