基于贝叶斯网络组合模型的公交驻站时间预测

公交驻站时间是公交行程时间的主要组成部分,其预测精度直接影响智能公交系统中公交信息发布的准确性.为了提高公交驻站时间的预测精度,提出一种基于贝叶斯网络的组合预测模型,它由反向传播神经网络和径向基函数神经网络模型组成.首先利用两种神经网络模型预测公交驻站时间；然后利用改进后的等宽数据离散方法,将两种神经网络的预测结果和观测的驻站时间数据离散后用于贝叶斯网络学习；最后通过贝叶斯网络推理得到驻站时间组合预测结果.实例分析表明,贝叶斯网络组合模型驻站时间预测结果的误差指标均优于单一模型,证明其可有效提高单一模型的预测精度.

Bus Dwell Time Prediction Based on Bayesian Network Combined Model

Dwelling time is a key component influencing the total bus travel time,and its prediction accuracy directly influences the reliability of information disseminated by intelligent public traffic system.In order to improve the accuracy of dwell time prediction,this paper puts forward a Bayesian Network(BN) combined prediction model,which is constituted by back propagation(BP) neutral network and radial basis function(RBF) neutral network.Firstly,the paper uses BP and RBF neutral network to predict the dwell time.Then an improved equal-width discrete method is adopted to disperse the predicted results and the observed dwell time for BN learning.Finally,the predicted results of combined method for dwell time are obtained through BN reasoning.The following practical analysis indicates that the performance of the index of BN combined model precedes that of single modal,and proves the effectiveness of BN combined modal in improving the accuracy of single predictor.