To address the problems of fuzzy distortion in long-term echoes and low accuracy in predicting strong echoes in existing radar echo extrapolation models, this paper designs a radar echo extrapolation method based on self-attention and dense convolution improved convolutional long short-term memory (ConvLSTM) network by using the composite reflectivity mosaic image of Doppler radar data in Anhui from May to September 2016. Based on ConvLSTM, the model incorporates self-attention mechanism into each cell and Encoder-Decoder to enhance the ability of extracting features with long-term spatial dependence. Meanwhile, the model uses dense convolution instead of common convolution to improve the feature reuse ability. The experiment uses the past 1-h radar echo image to predict the future 2-h radar echo image, and compares the resluts with the ConvLSTM before the improvement, proving that the proposed model can improve the accuracy of radar echo extrapolation.