Power load forecasting is influenced by many uncertain events, so accurately predicting load has always been a key research direction in the industry. In response to the problem of low accuracy of traditional methods in short-term power load forecasting, a short-term power load visualization forecasting method is proposed based on two-dimensional variational mode decomposition （2D-VMD） and convolutional long short term memory neural networks（ConvLSTM） . Firstly, the Gramian angular fields （GAF） method is used to convert the preprocessed load data into a set of Gram angular field images, and then the images are decomposed into a series of sub-modes with different center frequencies through 2D-VMD and classified according to the center frequencies. ConvLSTM neural network is used to predict the image groups with different modes. Finally, the prediction results are reconstructed and inversely operated to obtain the load prediction values. The prediction results indicate that this method improves the accuracy of short-term load forecasting and provides a new method for power load forecasting.