基于SVR-LightGBM模型的高速公路拥堵预测方法研究

Expressway Congestion Forecast Method Research Based on SVR-LightGBM

由于传统的高速公路拥堵预测方法大多构建的是单一的预测模型,模型的预测精度不高,并且大多以交通流的时间特征为基础,缺少对交通流空间特征的深度挖掘,往往导致预测结果不理想。本文提出了一种基于SVR与LightGBM组合的高速公路拥堵预测模型,由于高速公路上中下游路段具有时空相关性,因此首先运用SVR模型实现上下游路段对中游目标路段流量、速度和占有率的预测,进而将预测值与目标路段真实的拥堵阈值输入到LightGBM模型进行训练来预测拥堵。为验证组合模型的有效性,选用平均绝对误差(MAE)、均方根误差(RMSE)和均等系数(EC)三项评价指标对不同种预测模型进行对比结果分析。实验表明所提出的SVR-LightGBM拥堵预测模型能显著减少预测误差,预测精度更高,是一种高效快捷的拥堵预测模型。

Since most traditional highway congestion prediction methods construct a single prediction model,the prediction accuracy of the model is not high,and most of them are based on the temporal characteristics of traffic flow,lacking the deep mining of the spatial characteristics of traffic flow,which often leads to unsatisfactory prediction results.In this paper,we propose a highway congestion prediction model based on the combination of SVR and LightGBM.Since the upstream and downstream sections of the highway have spatial and temporal correlation,we first use the SVR model to achieve the prediction of traffic flow,speed and occupancy of the upstream and downstream sections of the highway on the midstream target section,and then input the predicted values of SVR and the real congestion threshold of the target section into the LightGBM model for training to predict congestion.In order to verify the effectiveness of the combined model,three evaluation indexes,mean absolute error(MAE),root mean square error(RMSE)and equal coefficient(EC),are selected to compare the results of different prediction models.The experiments show that the proposed SVR-LightGBM congestion prediction model can significantly reduce the prediction error and has higher prediction accuracy,which is an efficient and fast congestion prediction model.