基于天气特征的高速公路交通流预测方法研究  

Research on highway traffic flow prediction based on weather characteristics

作　　者：袁辉 谢庆 计明军 吴炜昌 曾斌 姬生忠 YUAN Hui;XIE Qing;JI Mingjun;WU Weichang;ZENG Bin;JI Shengzhong(China Railway Southern Investment Group Co.,Ltd.,Shenzhen 529566,China;College of Transportation Engineering,Dalian Maritime University,Dalian 116026,China)

机构地区：[1]中铁南方投资集团有限公司,广东深圳529566 [2]大连海事大学交通运输工程学院,辽宁大连116026

出　　处：《现代电子技术》2025年第8期164-172,共9页Modern Electronics Technique

基　　金：国家自然科学基金项目(71971035)。

摘　　要：随着高速公路网络的规模扩展和智能交通系统的不断完善,交通流预测在提高道路资源利用效率和缓解交通拥堵方面起着至关重要的作用。现有的预测方法往往忽视了天气特征动态变化对交通流的影响,故文中旨在运用集成深度学习模型来探索天气特征对高速公路交通流的影响。利用随机森林算法从历史交通流量和天气数据中提取出相关性较高的天气特征,采用粒子群优化算法对长短期记忆神经网络模型的超参数进行优化,构建一个融合天气特征数据的深度学习预测框架,将经过筛选的天气特征序列输入至预测框架模型中进行训练和预测。通过真实数据集上的实验验证了所提方法的有效性和泛化能力。实验结果表明,所提的集成深度学习方法相比现有的深度学习方法具有更好的拟合度、预测精度和稳定性,能够更准确地捕捉天气特征动态变化对交通流的影响。With the scale expansion of highway networks and the continuous improvement of intelligent transportation systems,traffic flow prediction plays a crucial role in enhancing road resource utilization and alleviating traffic congestion.Existing prediction methods often overlook the impact of dynamic weather feature changes on traffic flow.The integrated deep learning model is used to explore the influence of weather features on highway traffic flow.The random forest algorithm is used to extract highly correlated weather features from historical traffic flow and weather data.The particle swarm optimization algorithm is used to optimize the hyperparameters of long short-term memory(LSTM)neural networks.A deep learning prediction framework fusing weather feature data is constructed,and the screened weather feature sequences are input into the prediction framework model for training and prediction.The experiments on real datasets verify the effectiveness and generalization ability of the proposed method.The experimental results show that the proposed integrated deep learning method can exhibit better fitting accuracy,prediction precision,and stability compared to existing deep learning methods.It can more accurately capture the impact of dynamic weather feature changes on traffic flow.

关 键 词：智能交通系统 高速公路交通流预测 天气特征 集成深度学习 随机森林算法 粒子群优化算法 长短期记忆神经网络 超参数优化 

分 类 号：TN929.5-34[电子电信—通信与信息系统] U491.14[电子电信—信息与通信工程]