基于PCA-RF组合模型的隧道交通事故持续时间预测  

Prediction of Traffic Incident Duration in Tunnels Based on a PCA-RF Combined Model

作　　者：何珂 杨顺新[1] 郜勇刚 HE Ke;YANG Shunxin;GAO Yonggang(School of Transportation, Southeast University, Nanjing 210096, China;Shanxi Transportation Holdings Group Company, Taiyuan 030000, China)

机构地区：[1]东南大学交通学院,南京210096 [2]山西省交通控股有限集团,太原030000

出　　处：《交通信息与安全》2019年第5期26-32,共7页Journal of Transport Information and Safety

基　　金：山西省交通运输厅科技项目(7921000089)资助

摘　　要：使用主成分分析(PCA)和随机森林(RF)组合模型对高速公路隧道交通事故持续时间进行预测。主成分分析用来提高随机森林模型的精度与效率。此外,通过调节2个模型参数,包括决策树数目和最大树深度来提高模型精度和避免模型过拟合。参数优化的结果表明,建模时决策树数目取150、最大树深度取10可降低模型的泛化误差。用以建模的数据包括了山西省的所有高速公路隧道自2012-2017年内的2115起事故数据。每起事故数据包括16个变量,包括隧道类型,事故发生位置类型,事故类型等。结果表明,PCA-RF组合模型的平均绝对误差为12.80min,误差20min以内的准确率为89.15%,取得了良好的预测效果。并且,PCA-RF组合模型的精度高于RF模型,说明PCA-RF组合模型能够提高事故持续时间预测的精度。且PCA能够降低数据维度,提高算法的效率。与人工神经网络模型的结果表明,PCA-RF组合模型预测结果精度高且其模型更简单、效率更高。Principal Component Analysis (PCA) and Random Forests (RF) combined model is used to predict traffic incident duration in expressway tunnels. PCA is used to promote performances of a RF model. Two parameters, including number of trees and the maximum tree depth, are tuned to improve accuracy of the model and prevent it from overfitting data. The results of parameter tuning show that the tree number should be 150 and the maximum tree deep should be 10. Data used for modeling include 2 115 traffic incidents from 2012 to 2017 in all expressway tunnels in Shanxi Province, China. 16 variables including tunnel type, incident location type, incident type, etc., are used to predict duration of each incident. According to the results, the mean absolute error of the PCA-RF combined model is 12.80 min, and the accuracy rate of error within 20 min is 89.15%. The results indicate that the PCA-RF combined model outperforms RF model without PCA, proving that PCA can promote the performances of RF model. Moreover, compared with an ANN model, the PCA-RF combined model obtains better results with simpler algorithms.

关 键 词：交通安全 持续时间 预测 主成分分析 随机森林 组合模型 

分 类 号：U491.31[交通运输工程—交通运输规划与管理]