With the continuous expansion of the scale of power system, the randomness, uncertainty and volatility of load become increasingly prominent, which brings huge challenges to the refined management of dispatching. Therefore, in order to forecast bus short-term load more accurately, this paper proposes a bus load forecasting model based on the fusion of maximum information coefficient (MIC), convolution neural network (CNN) and short-term memory network (LSTM). Firstly, MIC is used to analyze the correlation between multiple loads and between loads and weather factors to determine the model input. Secondly, build the CNN-LSTM network model, use the CNN network to extract features, construct feature vectors and input them into the LSTM network to complete short-term bus load forecasting. Finally, a case study is carried out with the data of electricity, cooling and heating loads of Tempe campus of Arizona State University in the United States. According to the experimental results, the accuracy of the model reaches 91.55%, which has better prediction effect than other models.