基于数据挖掘的交通流预测模型  

Data mining using in a novel traffic flow forecasting model

作　　者：闫伟[1] 刘云岗[1] 王桂华[1] 高琦[2] 

机构地区：[1]山东大学能源与动力工程学院交通运输研究所,济南250061 [2]山东大学机械工程学院,济南250061

出　　处：《系统工程理论与实践》2010年第7期1320-1325,共6页Systems Engineering-Theory & Practice

基　　金：国家高技术研究发展计划(863计划)(2007AA04Z138);山东省优秀中青年科学家科研奖励基金(2007BS01005)

摘　　要：通过对道路交通流变化规律和影响因素的分析,提出了一种新的交通流预测模型.首先,鉴于模糊聚类方法易陷入局部最优解及运算速度慢的缺点,采用蚁群算法中Pij(t)改进模糊聚类分析;然后以最拥挤时间段的25个点交通流数据、天气数据以及天类别数据为指标,将历史数据聚分成若干簇团,并采用动量BP神经网络针对每一簇团建立相应的预测模型.对实际数据进行预测分析的结果表明:该模型不仅对普通工作日有较高的预测精度,对双休日、节假日和一些特殊情况(雨雪天气)也有较好的预测精度.A novel traffic flow forecasting model is presented in this paper. Firstly, because of fuzzy C-means algorithm （FCM） is unsatisfied in circulating precision and speed. To improve the performances of FCM algorithm, the fuzzy membership function and clustering centers are initialized by ant colony algorithm. Then the practical historical data within one year is divided into several groups by ant colony-fuzzy clustering algorithm. A separate module based on momentum BP neural networks models each group. Using data from an important road in Ji＇nan city, the satisfactory accurate results are obtained on weekends and holidays. Moreover, the model is robust, and produces accurate results in some special cases （raining or snowing）.

关 键 词：蚁群算法 模糊聚类 动量BP神经网络 交通流预测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]