基于LSTM的轨道交通进站客流短时预测研究

Research on Short-term Prediction of Inbound Passenger Flow of Rail Transit Based on LSTM

对轨道交通进站客流进行准确的预测有助于城市交通系统更好的管理,及时做出应对措施。使用K-means聚类方法对南京地铁113个站点进行聚类,得到5个不同类别的轨道站点,分析不同类型站点进站客流的时序特征以及天气与工作日因素对客流的影响,发现是否为工作日对进站客流影响最为明显。用长短时记忆网络将前35天的数据作为训练集预测后4天的客流量,将预测结果与循环神经网络和支持向量机做比较。结果表明:类别1和类别3站点的进站客流预测精度要优于其他类别,长短时记忆网络模型对居住型轨道站点进站客流的短时预测具有很好的适用性。

The accurate forecast of the passenger flow of rail transit is helpful for the better management of urban transportation system and the timely response measures.The K-means clustering method is used to cluster 113 stations of Nanjing Metro,and five different types of rail stations are obtained.The time sequence characteristics of inbound passenger flow of different types of stations and the influence of weather and working day factors on passenger flow are analyzed,and whether working day has the most obvious influence on passenger flow is analyzed.The LSTM is used to predict the passenger flow of the next four days using the data of the first 35 days as the training set,and the prediction results are compared with the recurrent neural network and support vector machine.The prediction accuracy of inbound passenger flow of category 1 and category 3 stations is better than that of other categories.