考虑可变环境因素的公共自行车短期需求预测模型.

Existing short-term demand prediction models for public bikes ignored the difference in the nature of the impacts of different environmental factors on user demand as well as the temporal dependency of variable environmental factors. this paper distinguished environmental factors into the invariable factors that have been internalized into user demand and the variable factors that need to be considered separately, and then proposed a model called GCNN-LSTM-E in this paper. In the model, this paper used graph convolutional neural network(GCNN) to capture the non-euclidean spatial dependency of user demand, used long short-term memory (LSTM) network to capture the temporal dependencies of user demand and variable environmental factors, and applied vector concatenation and fully connected network to impose the influence of variable environmental factors on user demand. Experimental results showed that the GCNN-LSTM-E model has the best prediction performance and outperforms all benchmark models under 1 h time granularity. It indicates that the design of the model is reasonable and effective, and 1 h is the most appropriate time granularity. [ABSTRACT FROM AUTHOR]