公路运行车速预测模型对比分析

Comparative analysis of prediction model for vehicle operating speed

为了对公路运行车速进行准确预测,采集了国道二级公路上232处典型路段的平曲线半径和纵坡度等线形数据和小轿车车速,分别应用线性回归、多项式回归和智能学习型方法中的BP神经网络和模糊神经网络建立了小轿车第85百分位车速模型,并对4种模型预测精度进行了对比分析。结果表明:在回归模型中,多项式回归的预测精度优于线性回归;在学习型算法中,模糊神经网络的预测精度优于BP神经网络,并且模糊神经网络的预测精度优于统计回归方法;从模型使用角度来看,相比较于线性回归模型要求样本随机误差满足零均值和正态分布等假设条件,神经网络算法的限制条件较低,具有更广阔的应用前景。

In order to accurately predict vehieule operating speed, 232 samples gathered from the second-level highway, including curve radius, slope and driving speed. Linear regression, poly- nomial regression and intelligent methods such as Back-Propagation neural network and fuzzy neural network were employed to predict the 85th percentile operating speed respectively and the prediction accuracies by four models were compared and analyzed. The results show that the per- formance of polynomial regression is superior to that of the linear regression and the fuzzy neural net is superior to the BP neural net. Additionally, the results indicate that the fuzzy neural net- work can get a better prediction accuracy than that by statistical regression and its restrictive con- ditions are lower, so it is of more potential in the application than the regression models. 1 tab, 10 figs, 10 refs.