核函数法与最邻近法在短时交通流预测应用中的对比研究  

Comparison of Kernel Approach and KNN Approach in Short-term Forecasting of Traffic Flow

作　　者：钱海峰[1] 陈阳舟[1] 李振龙[1] 杨玉珍[1] 

机构地区：[1]北京工业大学,北京100022

出　　处：《交通与计算机》2008年第6期18-21,34,共5页Computer and Communications

基　　金：国家自然科学基金项目(批准号:60604008);北京市自然科学基金项目(批准号:4072005)资助

摘　　要：以北京三环路的一个区间路段作为短时交通流预测的背景,利用实际检测数据将非参数回归预测模型中核函数法和最邻近法2个不同的权函数方法进行了仿真对比研究,结果表明在相同的预测精度下,最邻近法更适合时间间隔比较短的交通流预测。将传统的最邻近法加以改进,依次加入前一时刻的交通流量、当前时刻的车辆平均速度和车道平均占用率作为搜索元素。通过对仿真实验结果的研究分析得出结论:前一时刻交通流量的引入保证了预测值和真实值具有相同的切线方向,使得预测精度得到了显著提高;车辆平均速度和车道平均占用率与交通流量具有一定的对应关系,其作用和当前时刻的交通流量相似,所以不能有效的改善最邻近法的预测精度。By taking the third ring road in Beijing as the background, two basic weighted functions of Nonparametric Regression including Kernel Approach and K-nearest neighbor Approach were comparatively analyzed and simulated with the real data that was detected. The results indicate that K-nearest-neighbor Approach is more suitable for the shortterm forecasting of traffic flow in the same accuracy. Furthermore, the improvement on the traditional K-nearest-neighbor Approach had been done by adding, step by step, the volume at the previous prediction period, the current average speed and the current average occupancy as searching elements. Some conclusions can be drawn from the simulation results that adding the volume at the previous prediction period can make sure that the forecasting flow and the real flow are in the same fangent direction, therefore, it can remarkably improve the forecasting performance of the K-nearest-neighbor Approach, and that the current average speed and the current average occupancy have some relevant relationship with the volume and their roles similar to the volume, therefore, they can not obviously improve the prediction accuracy of K-nearest-neighbor Approach.

关 键 词：交通流 短时预测 非参数回归 核函数 最邻近 

分 类 号：U491.54[交通运输工程—交通运输规划与管理]