摘要
为及时、准确地预测交通流,解决精度低、高维特征缺失等问题,建立CNN-GRU-ATT模型来对英国高速公路交通数据进行预测。模型中的卷积层用来提取特征,GRU层用于描述时间趋势,注意力层用于聚焦关键信息。输入多个路段的短期交通流信息,考虑多路段之间的相互关联以及气象因素的影响。实验发现:该模型与支持向量回归(SVR)、长短期记忆网络(LSTM)、门控循环网络(GRU)、CNN-GRU、GRU-ATT模型相比,模型精度更高,拟合优度达到了96.89%,MAPE最高降低了21.55%;将多路段与单路段数据分别进行输入,发现前者能够更好地进行预测,MAPE降低了7.56%。添加气象因素后模型精度有所提高,拟合优度达到了97.06%。
With a fast growing number of vehicles on the road,traffic congestion has become a malaise in many urban areas.The advances and applications of intelligent transportation systems(ITS)offer a promising solution to improve traffic management,effectively mitigating the congestion that plagues urban areas.ITS relies on surveillance technology to collect real-time and reliable traffic data,making the timely and accurate prediction of traffic flow a crucial focus in traffic research.To address such issues as low precision and the absence of high-dimensional features,we build a CNN-GRU-ATT model to forecast traffic data on British highways.Within this model,convolutional layers are employed to extract features,GRU layers are utilized to describe temporal trends,and attention layers are designed to focus on key information.Additionally,the model incorporates short-term traffic flow information from multiple road sections,considering the intercorrelation between these sections and the impact of meteorological factors.Our experiments reveal the CNN-GRU-ATT model achieves a higher precision,with a goodness of fit reaching 96.89%,and the mean absolute percentage error(MAPE)up by 21.55%compared with those of support vector regression(SVR),long short-term memory networks(LSTM),Gated Recurrent Units(GRU),CNN-GRU,and GRU-ATT models.When comparing the input of data from multiple road sections versus single road sections,we find the former makes better predictions,with a reduction of 7.56%in MAPE.The addition of meteorological factors further improves the model’s accuracy,with the goodness of fit reaching 97.06%.Therefore,with the input of data from multiple road sections and meteorological field,the CNN-GRU-ATT model proves to be the most effective for short-term traffic flow prediction.It excels thanks to its sophisticated architecture that takes full advantage of the strengths of convolutional neural networks for feature extraction,the temporal dynamics captured by GRU layers,and the attention mechanism that highlights the most relevant information.The integration of data from various road sections enables the model to fully consider the complex interactions and dependencies that exist between different parts of the transportation networks.Furthermore,by incorporating meteorological data,the model adjusts its predictions with full consideration of the impacts of weather conditions on traffic flow,which is a critical factor often ignored in other models.Our experimental results underscore the superiority of the CNN-GRU-ATT model in terms of accuracy and precision.The high goodness of fit indicates a strong correlation between the model’s predictions and actual traffic data while the significant reduction in MAPE demonstrates the model’s ability to make more reliable forecasts.Our comparison with other models highlights the incremental benefits of the CNN-GRU-ATT model’s design,particularly in its ability to handle high-dimensional data and complex temporal patterns.In conclusion,the CNN-GRU-ATT model,when fed with comprehensive data from multiple road sections and meteorological departments,offers a superior approach to short-term traffic flow predictions.Its ability to process and analyze large volumes of data with high accuracy makes it a valuable tool for traffic management and congestion mitigation in urban environments.
作者
周文学
赵丽雅
ZHOU Wenxue;ZHAO Liya(College of Mathematics and Physics,Lanzhou Jiaotong University,Lanzhou 730000,China)
出处
《重庆理工大学学报(自然科学)》
北大核心
2025年第1期213-218,共6页
Journal of Chongqing University of Technology:Natural Science
基金
国家自然科学基金项目(11961039,11801243)
兰州交通大学校青年科学基金项目(2017012)。
关键词
注意力机制
多路段输入
卷积神经网络
门控循环网络
attention mechanism
multi-section input
convolutional neural network
gated recurrent unit
