基于多通道图聚合注意力机制的共享单车借还量预测

王福建,张泽天,陈喜群,王殿海

Usage prediction of shared bike based on multi-channel graph aggregation attention mechanism

Fujian WANG,Zetian ZHANG,Xiqun CHEN,Dianhai WANG

表 1 深圳和纽约共享单车数据集在不同预测模型下的误差

Tab.1 Errors of Shenzhen and New York bike-sharing datasets under different prediction models

模型 深圳 纽约 MAE RMSE WMAPE MAE RMSE WMAPE ARIMA 10.595 17.728 0.201 1.862 3.505 0.416 Time-GCN 5.241 10.102 0.100 1.602 2.974 0.358 LSTM 5.512 10.347 0.105 1.517 2.806 0.339 CNN-bi-LSTM 4.863 9.249 0.092 1.216 2.214 0.271 GCN-LSTM-FCN 3.100 5.841 0.059 1.175 2.209 0.262 GCN-Transformer 2.011 3.305 0.038 1.023 1.986 0.228 GraphAgg-iTransformer 1.685 2.691 0.032 0.973 1.826 0.217 ST-AGCN 1.715 2.892 0.033 0.975 1.810 0.218 MCGA 1.519* 2.578* 0.029* 0.969* 1.799* 0.216*