三种统计预报模型在江苏省道路低温预警中的应用

Application of Three Statistical Forecast Models in Early Warning of Low-Temperature on Road Surface in Jiangsu and Their Comparison

为了更好地开展道路交通低温灾害的预警,减轻道路结冰给车辆行驶造成的危害,本文利用2012—2016年江苏省高速公路网AWMS系统交通气象观测数据,在对路面低温发生规律进行分析的基础上,结合多元线性回归、朴素贝叶斯以及支持向量机3种统计预报方法,开展了路面低温预警的统计建模与预报试验。结果表明:(1)江苏全省高速公路网路面温度出现0℃以下、-2℃以下、-5℃以下的低温频率均呈"北高南低"分布。(2)全省高速公路网路面温度出现0℃以下的低温时次大多在15:00到次日06:00之间。(3)在对京沪高速M9308站的单站建模与预报试验中发现,路面低温预报因子组合中以13:00气温、13:00—18:00气温变温、13:00路面温度、13:00—18:00路面变温、13:00路基温度、13:00—18:00路基变温、18:00相对湿度和18:00风速U分量为自变量组合的预报方程效果最好,3种方法中以朴素贝叶斯模型的预报准确率最高;(4)就全省高速公路网而言,3种统计预报模型的路面低温预报准确率均超过75%,通过对全路网路面低温预报的试验结果对比发现,多元线性回归方法对江苏省北部路网的预报效果最好,预报准确率大多在85%以上;而支持向量机模型对江苏省南部路网的预报效果最好,大部分站点的低温预报准确率达95%以上。

In order to provide a better service for the warning of low temperature disasters on road surface and mitigate the damage caused by the frozen road against cars,in this paper,the observed data of traffic meteorological factors in the Automatic Weather Monitoring System（AWMS）on the Jiangsu expressway network from 2012 to 2016 are collected to analyze the temporal and spatial pattern of low temperature occurrence on the road surface and the statistic model establishment and the forecast experiments of the low temperature warning on road surface are carried out through three statistical forecast methods：the Multiple Linear Regression,the Naive Bayes Method,and the Support Vector Machine Model.The results are showed as follows：（1）The occurrence frequency of low temperatures below 0℃,below-2℃and below-5℃ on the road surface of the expressway network in Jiangsu Province displayed the distributions ofhigher in the north part and lower in the south part.＂（2）The road surface temperature below 0℃ on the expressway network occurs between 15：00 and 06：00 of the next day in general.（3）In the model establishment and forecast experiments of a single station for M9308 Station on the Jiangsu section of the Beijing-Shanghai Expressway,it is found that the forecast models taking the air temperature at 13：00,the variation of air temperature from 13：00 to 18：00,the road temperature at 13：00,the variation of road temperature from 13：00 to 18：00,the roadbed temperature at 13：00,the variation of roadbed temperature from 13：00 to 18：00,the relative humidity at 18：00,and the U component of wind speed at 18：00 as the forecast factors have the best efficiency in the warning of road low temperature.The naive Bayes method has the highest forecasting accuracy rate in the three methods.（4）For the whole expressway network in Jiangsu,the accuracy rates of three statistical forecast models in the warning of low temperature on road surface are higher than 75%.The comparison of the low temperature forecast experiment results of road surface on the Jiangsu expressway network indicates that the Multiple Linear Regression shows the best warning efficiency in the northern Jiangsu with an accuracy rate larger than 85% and the Support Vector Machine Model displays the best warning efficiency in the southern Jiangsu with an accuracy rate higher than 95%.