基于状态空间神经网络的短期公交调度模型

Short Term Bus Dispatching Model Based on the State Space Neural Networks

本文从公交线路状态时空变化规律的角度出发,讨论了应用状态空间神经网络模型解决短期公交调度问题的方法。采用能描述实际公交线路状态(包括客流状态以及车辆运行速度等)的网络拓扑结构,结合前一时段的公交线路状态,预测下一时段的状态并选择与其相适应的调度方案。本文以南京市某公交线路的数据作为实例进行模型应用,与BP神经网络和AMRA模型的对比结果显示状态空间神经网络模型能在短期内更好地针对客流空间、时间变化对公交发车间隔进行调整,模型预测精度高,自适应性强,值得推广应用。

This paper discussed the solution of bus dispatching in short terms using state space neural networks（SSNN） according to the spatial and temporal variation law of bus lines. Utilizing the state of bus line in the previous interval,and the SSNN＇s network topology,which is derived from the physical state of bus line,the ability to predict the state in the next interval and the corresponded optimal dispatching scheme were got. Themodel performance was tested with a set public transit data in Nanjing. And the result was compared with those from BP neural networks and ARMA model. Results of the comparison indicated that the model was better in adjusting bus departing interval based on passenger flow space and time variations,and predicting bus dispatching with higher precision.