基于BP神经网络的公交周转时间预测研究  

Bus turnaround time prediction research based on the BP neural network

作　　者：史同广[1] 袁腾飞[2] 时柏营[2] 

机构地区：[1]山东建筑大学土木工程学院,山东济南250101 [2]山东建筑大学交通工程学院,山东济南250101

出　　处：《山东建筑大学学报》2015年第3期205-210,共6页Journal of Shandong Jianzhu University

基　　金：国家"十二五"科技支撑计划项目(2012BAJ15B05)

摘　　要：公交周转时间是制定公交行车作业计划的关键参数,公交周转时间的精确预测也是提高公交服务水平的重要保证。文章基于公交运行的动态随机性,运用数据挖掘有序样本聚类算法,研究了对公交周转时间影响较大的动态因素,分析了全天时段每个时间窗内周转时间的分布特征,建立了基于BP神经网络的公交周转时间预测模型;通过实例分析验证了模型的有效性和精度。结果表明:公交周转时间的分布特征主要受时段、天气状况、节假日或大型活动、星期、季节月份等动态因素的影响,通过划分时间窗可知,同一时间窗内的公交转时间近似呈正态分布;BP神经网络预测模型可有效地揭示公交周转时间和各动态因素之间的非线性关系,通过实例分析,预测公交周转时间的平均绝对百分比误差为5.69%,具有较高的预测精度。Bus turnaround time is the key parameter of bus driving operation plan, and the accurate prediction of the bus turnaround time is also important guarantee to improve the service level of public transportation. Based on the bus running dynamic randomness and by using the ordered sample clustering data mining algorithm, the paper analyzes the distribution features of the bus turnaround time in each time window to get the main dynamic factors, establishes the bus turnaround time prediction model, and through example analysis verifies the effectiveness and accuracy of the model. The results show using GPS data can obtain the accurate bus running bus turnaround time, and by partitioning time window, the bus turnaround time approximately normally distributed at the same time window. Through the establishment of the BP neural network prediction model can effectively reveal the bus turnaround time and the nonlinear relationship between the dynamic factors, and the example analysis results show that the forecasting bus turnaround time mean absolute percentage error is 5.69%, which has higher prediction accuracy.

关 键 词：公交周转时间 时间窗 分布特征 动态因素 BP神经网络 

分 类 号：U491.17[交通运输工程—交通运输规划与管理]