公路隧道交通量预测的粒子群高斯过程耦合模型  

Traffic flow prediction model of highway tunnel based on PSO-Gaussian process coupled algorithm

作　　者：万良勇[1,2] 刘开云[1] 

机构地区：[1]北京交通大学土木建筑工程学院,北京100044 [2]石家庄市轨道交通有限责任公司,石家庄050011

出　　处：《北京交通大学学报》2015年第1期33-39,共7页JOURNAL OF BEIJING JIAOTONG UNIVERSITY

基　　金：国家自然科学基金资助项目(51378052)

摘　　要：交通量的预测对公路隧道运营期通风系统的节能降耗具有重大意义,将新型小样本学习机器高斯过程引入隧道交通量预测,提出了一种组合核函数,用以改善单一核函数高斯过程的泛化性能,在网络训练过程中采用粒子群优化算法,自动搜寻泛化性能最好的高斯过程超参数,形成粒子群高斯过程耦合算法,并编写了相应的计算程序.对某公路隧道交通量进行了预测,结果表明:组合核函数高斯过程最大预测相对误差仅为4.41%,平均相对误差为1.96%;两种单一核函数高斯过程最大预测相对误差均为6.68%,平均相对误差分别为2.7%和2.67%;粒子群高斯过程耦合模型可以高精度地用于隧道交通量预测,且组合核函数可以提高单一核函数的泛化性能,并为其他类似工程提供借鉴.Traffic flow prediction of highway tunnel in operation period is of great significance to energy conservation and cost reduction for tunnel ventilation system. A new small sample learning machine-Gaussian process（GP） was introduced into the tunnel traffic flow forecasting and one combined kernel function was proposed to improve the generalization performance of single kernel function of Gaussian process. The particle swarm optimization（PSO） algorithm was applied to automatically search super parameters of Gaussian process model which possess the best generalization performance during GP network training course,thus formed the PSO-GP coupled algorithm. The corresponding computer code was programmed in Matlab. Seen from the traffic flow prediction result of one highway tunnel,the maximal relative error of combined kernel Gaussian process is only 4. 41% and the average relative prediction error is 1. 96%; as a comparison,the maximal relative error of two single kernel Gaussian process both are 6. 68% and the average relative prediction error is 2. 7% and 2. 67% respectively. It can concluded that using the combined kernel functions PSO-GP coupled model can obtain a more precise prediction result than that of single kernel GP model the tunnel traffic with high accuracy which presents reference for other similar engineering.

关 键 词：隧道 交通量预测 粒子群高斯过程耦合模型 通风系统 人工智能 

分 类 号：U457.3[建筑科学—桥梁与隧道工程] TP182[交通运输工程—道路与铁道工程]