基于深度长短期记忆网络的地铁进站客流预测

Forecast of Subway Inbound Passenger Flow Based on DLSTM Recurrent Network

提出利用多源数据(地铁刷卡数据、气候数据和节假日数据)进行数据特征构造,并采用深度长短期记忆网络(DLSTM)方法预测地铁进站客流量。以深圳北站地铁站为研究对象,选取该站3个月的地铁IC卡数据记录,前两个月的数据为训练集,后一个月的数据为测试集。介绍了数据预处理方法和DLSTM模型构建原理。试验结果表明:DLSTM模型的预测准确度随着DLSTM模型的深度增加而增高;与其它模型相比,DLSTM模型的预测精度更高。

The data feature structure is done based on the multi-source data,including metro card swipe data,climate data and holiday data,the deep long and short term memory(DLSTM)recurrent network is used to forecast the inbound passenger flow of subway stations.Taking Shenzhen North Station as the research target,the transit IC card data from Jan.1 to Mar.31 in 2014 are collected.The data in the first two months are the training set,and the data for the next month are the test set,the data processing method and DLSTM model construction principle are introduced.The test results show that the forecast accuracy of DLSTM model increases as the model depth increases compared with other models,DLSTM outperforms other models in terms of forecast accuracy.