实时交通流预测的并行SVR预测方法

A parallel SVR approach to real-time traffic flow forecasting

提高交通流预测的精度和实时性是智能交通系统(ITS)应用发展的一个重要问题.与广义神经网络(GNN)方法相比,支持向量回归(SVR)方法应用于交通流预测理论优势得以实现的前提是选取合适的回归参数.分析、讨论了简单而实际的直接从训练集中选取SVR参数的方法,给出了一个大规模路网交通流SVR预测模型和集群环境下的一种贪婪负载均衡并行算法(G-LB).实验结果证明了基于G-LB算法的并行SVR方法(GLB-SVR)可获得比并行的GNN方法(P-GNN)更好的预测精度和实时性.

Accurate and real-time traffic flow forecasting is a key problem to the application development of intelligent transportation systems（ITS） .Comparing with generalized neural network（GNN） method,the theoretical advantage of applying support vector regression（SVR） method to traffic flow forecasting highly depends on good parameter selection.Simple yet practical approach to SVR parameters setting directly from the training set is analyzed and discussed,and a traffic flow SVR forecasting model for large-scale road network and a greedy load balancing（G-LB） algorithm in cluster environment are proposed.Experimental results demonstrate that it can better satisfy real-time and accurate demands of traffic flow forecasting using parallel SVR approach based on G-LB（GLBSVR） algorithm than using parallel GNN（P-GNN） method.