基于XGBoost融合模型的交通流量预测技术研究  

Research on Traffic Flow Prediction Technology Based on XGBoost Fusion Model

作　　者：李青 LI Qing(Hongshan Traffic Brigade,Wuhan Public Security Bureau,Wuhan 430071,China)

机构地区：[1]武汉市公安局洪山区交通大队,湖北武汉430071

出　　处：《自动化仪表》2022年第12期123-128,共6页Process Automation Instrumentation

摘　　要：为了提高城市交通信息管理能力、交互能力和处理能力,提出了基于XGBoost融合模型的交通流量预测系统。构建了梯度方向直方图(HOG)特征信息模型,采用支持向量机(SVM)分类器对城市交通信息进行识别,增强了道路信息识别能力。同时,构建了基于XGBoost融合模型的交通流量预测系统。XGBoost融合模型将分类回归树(CART)作为基分类器,应用Label Encoding和one-hot Encoding方式实现原始数据集的编码,集成多个决策树模型,共同决定样本的预测结果。通过调整模型参数,可防止出现过拟合。通过对不同数据信息进行分类,提高了数据监控能力。试验结果表明,该研究对车辆图像的识别率较高,对交通流量的预测数据的误差百分比始终低于0.1。该研究能够加强城市交通系统监控,提高交通数据信息分析和应用能力。A traffic flow prediction system based on the XGBoost fusion model is proposed in order to improve urban traffic information management,information interaction capability and processing capacity.A histogram of oriented graients(HOG)feature information model is constructed,and a support vector machine(SVM)classifier is used to identify urban traffic information,which enhances plus the road information identification capability.At the same time,a traffic flow prediction system based on XGBoost fusion model is comstructed.The XGBoost fusion model uses the elassing and regression tree(CART)as the base classifier,and the original data set is encoded by using Label Encoding and one-hot Encoding to integrate multiple decision tree models to determine the prediction results of the samples.By adjusting the model parameters to prevent overfitting,and by classifying different data information,the results improve data monitoring capabilities.The test results show that,the research method has a relatively high recognition rate of vehicle images and the percentage error of predicted data for traffic flow is always lower than 0.1.The research can enhance urban traffic system monitoring and improve traffic data information analysis and application capabilities.

关 键 词：道路信息识别 决策树 支持向量机分类器 分类回归树 XGBoost融合模型 

分 类 号：TH7[机械工程—仪器科学与技术]