一种并行BP交通流预测方法  

Parallel BP Approach for Traffic Flow Forecasting

作　　者：杨际祥[1,2] 王凡[1] 谭国真[1] 王荣生[2] 

机构地区：[1]大连理工大学计算机科学与工程系,辽宁大连116024 [2]燕山大学计算机科学与工程系,河北秦皇岛066004

出　　处：《小型微型计算机系统》2009年第12期2453-2456,共4页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(60373094)资助;河北省科学研究计划项目(2007492)资助

摘　　要：BP广泛用于短时交通流预测.为了降低大规模交通流的预测时间,已提出一些并行的BP方法,但在很多情况下其并行计算的效率仍有待提高.提出一个贪婪动态负载均衡(简称GC-DLB)算法,能够提高并行计算效率和降低预测时间,并在工作站网络(NOW)系统中对该算法进行了实现.与蝶形并行BP交通流预测方法(简称DP-BP)相比较,理论和实验结果说明了DP-BP方法结合GC-DLB算法可降低预测时间.The back propagation （ BP ） is wildly used in short-term traffic flow forecasting which requires the training set size be much larger than the network size. Although a number of parallel BP approaches have been proposed for reducing the forecasting time with a large samples of traffic flow data. However, still higher performance needs to be further delivered in many cases. An load balancing strategy based on greedy algorithm considering communication cost （ GC-DLB ） is proposed to improve parallel computing efficiency and reduce predicting time and GC-DLB algorithm is implemented in network-of-workstation （ NOW ） system. Comparing with the dish parallel BP approach（ DP-BP）, our results indicate that DP-BP approach combining with GC-DLB algorithm outperforms DP-BP approach itseff, and can reduce forecasting time.

关 键 词：并行计算 交通流预测 动态负载均衡(DLB) 工作站网络(NOW) 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]