To deal with the real time changing of atmospheric channel with meteorological conditions in space quantum communication system, a prediction method of atmospheric attenuation coefficient of quantum signals based on deep learning is proposed. The experiment is based on meteorological data set of Xi’an, and three neural network models, namely BPNN, LSTM and GRU, are built for analysis and comparison. The results show that all the three neural network models can accomplish prediction effectively with over 80% data fitting rate, among which LSTM and GUR have better performance. Meanwhile，three network models all produce large errors at the peak value. The prediction scheme provides a basis for further research in various compensation methods and intelligent parameter optimization for atmospheric channels in space quantum communication.