The present invention relates to a short-term load forecasting method based on a cloud model. First, a three-layer classification model is established based on seasons, day types and meteorological factors, and the third-level index is extracted through the correlation coefficient method, that is, the characteristic quantity of meteorological factors affecting the load size. According to the different mechanisms of the influence of characteristic quantities on the load, the corresponding scoring standards are formulated, and the scores of each three-level index are obtained by using the membership function. The larger the score, the greater the load of the index. Then according to the importance of each index, the weight value of each index is obtained by using the AHP, and based on the cloud model, the weighted deviation degree is obtained, and the cloud map is drawn, and the load is classified through the cloud map. Finally, the score obtained by the feature quantity of the forecast day is calculated, classified according to the load, and classified into its category. Based on the bp neural network, the load data of the category to which the load belongs is used as a training sample to predict the load of the forecast day. The invention has higher classification recognition accuracy and stronger adaptability.