In order to accurately predict the demand for bike-sharing in different regions of a city and solve the problem of imbalance between supply and demand，the travel distribution law of bike-sharing in Shanghai were studied based on the visualization analysis of spatio-temporal characteristics. In view of the non-strict periodicity of time travel distribution，Attention Mechanism was introduced into the Long-short Term Memory （LSTM） network to build a demand forecasting model named AM-LSTM. Spearman correlation analysis method was used to analyze characteristic influencing factors and extract model characteristic values. The prediction models of different input sequences were constructed and compared with the traditional time series prediction models. The results showed that the input sequence with a time interval of 30 min had a higher prediction accuracy. AM-LSTM model can better predict the travel demand of bike-sharing，and the prediction accuracy was higher than that of the single LSTM model. Finally，the correlation analysis of the prediction curve was conducted to verify the prediction performance of AM-LSTM model，which can provide effective information for the scheduling and distribution of urban bike-sharing.