基于双向长短期记忆网络的共享单车流量预测

Sharing Bicycle Flow Forecasting Based on Bi-directional Long Short-term Memory

近年来,共享单车逐渐成为流行于城市的交通出行手段,过量投放是其目前面临的最大问题,准确预测共享单车流量能有效调节共享单车投放,且能维护城市的交通秩序和形象.考虑到共享单车流量是一种时间序列,当前流量与过去和将来的流量具有密切的联系,本文提出一种基于双向长短期记忆的深度网络模型以预测未来的共享单车流量.该模型的时间步长设置为12,即以过去12个小时的数据作为输入,预测未来一个小时的共享单车流量数据,以此类推,每次向后推移一个小时,从而预测下一个数据.为了验证模型的性能,本文选取人工神经网络,循环神经网络以及长短期记忆网络作为对比模型.实验结果显示,所提出的模型在预测未来的共享单车流量的性能最佳.

In recent years,bike sharing has become a popular way of transportation in cities. Excessive investment is the biggest problem of bike sharing industry at present. Accurate prediction of sharing bicycle flow can not only effectively regulate the investment of sharing bicycles,but also maintain the traffic order and the image of the city. Considering that the sharing bicycle is a time series,the cuorrent traffic is closely related to the past and future flow. Therefore,this paper proposes a deep network model based on bidirectional long short-term memory to predict the future sharing bicycle flow. The time step of the model is set to 12,that is psing the data of the past twelve houors as input,to predict the sharing bicycle flow data in the next hour,and so on,shifting the time back one hour at a time,to predict the next data. In order to verify the performance of the proposed model Jthis paper selects artificial neural network,recuorrent neural network and long short-tenm memory as the contrast model. The experimental results show that the proposed model has the best performance in predicting the future shared single vehicle traffic.