Abstract: In recent years, with the introduction of "double carbon" strategic goal of achieving carbon peak and carbon neutralization, wind power generation has pivotal component of renewable energy power generation. Aiming to improve the accuracy of short-term wind power prediction, a short-term wind power prediction method was proposed based on density clustering and complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and the combination of convolutional neural network and long short-term memory network (CNN-LSTM). The wind power and weather characteristics were initially classified into different categories of data sets using density clustering. Through the adaptive noise complete integration empirical mode decomposition algorithm, various types of data were decomposed in frequency domain to obtain subsequence components. On this basis, the selected subsequence components and weather characteristics were input into the prediction model of convolutional neural network and long-term and short-term memory network. Finally, different prediction results were superimposed to obtain the final outcomes. Through clustering, decomposition and feature selection techniques, the accuracy of short-term wind power prediction was effectively improved.