The invention discloses a short-time weather forecast method based on wavelet denoising and Catboost, and the method comprises the following steps: S1, inputting the historical climate feature data ofa t moment, and carrying out data cleaning on input data consisting of moments t, O1-On and M1-Mm; S2, carrying out sorting on O1-On and M1-Mm, and removing the feature data of which the score is lower than the Q score; S3: carrying out one-hot coding on P sites of the climate feature sequence to be predicted; performing clock projection on the time information of the climate feature sequence tobe predicted to obtain time features; S4, performing wavelet denoising on the temperature at the height of 2 meters from the ground, the relative humidity at the height of 2 meters from the ground andthe wind speed at the height of 10 meters from the ground in the climate feature sequence to be predicted; and S5, training the Catboost model, inputting the test set into the trained Catboost model,and outputting a prediction result of the temperature at the height of 2 meters from the ground, the relative humidity at the height of 2 meters from the ground and the wind speed at the height of 10meters from the ground. Convergence time can be reduced, and prediction efficiency is improved.