In order to reduce the impact of weather factors on short-term power load forecasting and improve the forecasting accuracy of the model, a short-term load forecasting model based on weather classification and convolutional neural networks is proposed. Firstly,through the preliminary classification of weather types, the model divides the original data sample set into four types such as sunny,cloudy, cloudy and rainy days. Secondly, in order to identify similar meteorological conditions, the correlation coefficient and k- means clustering method are used to find the meteorological factors that have the greatest impact on the new load output, cluster them, and select high similarity data samples. Then, according to the result of feature selection, the neural network input data set is constructed. Finally, the data set is input to the convolutional neural network for training and prediction. The proposed model has higher prediction accuracy through verification and analysis of numerical examples.