中文摘要：

为了提高短时交通流的预测精度,向交通管理部门和出行者提供更加准确可靠的交通信息,基于非参数回归与支持向量回归方法的特点,提出了一种混合预测模型（KNN-SVR）。该模型利用K近邻方法的搜索机制,重建与当前交通状态近似的历史交通流时间序列,然后利用支持向量回归原理实现短时交通流预测。针对实际的交通流数据,考虑预测路段上下游交通流的影响,对提出的KNN-SVR模型的预测精度进行了分析。研究结果表明：同时考虑预测路段和其邻近路段交通流影响的KNN-SVR模型具有更好的预测精度,其预测误差最小,平均为8.29%,而仅仅考虑预测路段交通流影响的KNN-SVR模型,其预测误差略高,平均为9.16%;KNN-SVR模型的预测精度优于传统单一的预测方法,如K-近邻非参数回归、支持向量回归以及神经网络方法。

英文摘要：

To improve the forecasting accuracy of short-term traffic flow and provide more precise and reliable traffic information to traffic management department and travelers,we proposed a hybrid prediction model（ KNN-SVR） based on the characteristics of both nonparametric regression and support vector regression.The KNN-SVR model takes the search mechanism of the K-nearest neighbor method to reconstruct the time series of historical traffic flow that is similar to the current traffic flow,then it uses the support vector regression to perform the short-term traffic flow forecast. According to the actual traffic flow data,we considered the effect of the upstream and downstream traffic flows on the target section,and analyzed the forecasting accuracy of the KNN-SVR model. The research result shows that（ 1） the KNN-SVR model considering the traffic flow influences of both the target section road and its adjacent section roads has the better performance,its forecasting error is the least and the average error value is 8. 29%,while the KNN-SVR model which only considers the target section road,its forecasting error is slightly larger and the average error value is 9. 16%;（ 2） the forecasting accuracy of the KNN-SVR model is better than those of other traditional prediction models, such as the K-nearest neighbor nonparametric regression, support vector regression,and neural networks.