The waterborne traffic density in the field of intelligent transportation is nonlinear, unstable and fluctuating, etc. So the accurate prediction of traffic density is faced with challenges. In addition, waterway transportation is an important way of logistics and trade in China. Compared with road transportation, waterway transportation has higher safety and lower cost, so it is particularly important for us to predict the accuracy of waterway traffic density. In this paper, a multi-feature graph convolution and long and short time memory network model is proposed to solve the spatio-temporal correlation problem of traffic flow data and realize the prediction of traffic density in water transportation. The model uses graph convolution operation to extract spatial features and short and long time memory network to extract temporal features. Three related features of ship's navigable density, ship's average speed and ship's density are introduced to achieve more accurate traffic volume prediction. Experimental verification of the model was carried out on the real AIS data set. Through multiple groups of comparative experiments, the experimental results showed that the proposed model was better than the baseline LSTM network and the single-feature graph convolution LSTM model in predicting the navigable density of water transportation, with the accuracy increased by 25% and 13% respectively.