基于模式识别的非参数回归在短时交通流上的预测应用

Nonparametric Regression Based on Pattern Recognition and Its Application in Short-term Traffic Flow Forecasting

针对非参数回归在短时交通流预测上的局限性,改进传统K近邻方法,加入模式识别功能(通过匹配数l实现)和变K和l搜索算法,得到最优K和l值及相应的预测结果。通过实验发现:改进的K近邻方法在误差范围为5%、9%时对应的预测准确率为84.4%、96.10%。将其与传统K近邻方法进行对比,通过计算两者预测效果的各方面指标,发现改进的K近邻方法在精度和实时性上都有了很大的提高。

Although nonparametric regression is widely used in short-term traffic flow forecasting,some questions still exist.For this reason,the ipmprovements in two aspects are made according to the traditional K-nearest neighbor method：（1） joining pattern recognition function which has the matching number l and（2） changing search algorithm for K and l to obtain optimal value and corresponding prediction results.The results of experiments show the improved K-nearest neighbor method has higher predictive accuracy and more real-time quality.