基于K邻近算法的城市道路短时交通预测

Urban Short-Term Traffic Prediction Based on K-Nearest Neighbor Method

摘要: 为了提高城市道路交通状态预测的准确度，适应交通状态剧烈变化，提出了基于K邻近算法的实时交通预测框架。该框架以路段平均速度的时间序列构建特征向量，提出并应用差分序列考虑交通状态的幅度变化，滚动预测不同道路类型的短时交通状态。实验结果表明，增加差分序列的K邻近算法能准确地实现不同道路类型的短期交通状态预测；对比支持向量与随机森林算法，验证K邻近算法更适应交通状态变化剧烈的次干道交通预测。

Abstract: To improve the accuracy of urban traffic prediction, a framework based on K-Nearest Neighbor (KNN) methods is proposed to adapt to drastic changes in traffic conditions. The eigenvector con-sists of the time series of average speed and the difference series considering the changes on the real urban traffic. Arolling horizon approach is proposed to predict the short-term traffic of different urban road sections. The empirical results of Shenzhen presented that the KNN method with the difference series is able to achieve a short-term and real-time prediction for different urban road sections. On the comparisons with Support Vector Machine and Random Forest, the KNN is more suitable to the prediction of sub-arterial road with highly variable traffic state.