基于CNN-BiLSTM-AM模型的交通流量预测

Traffic Volume Prediction Based on CNN-BiLSTM-AM Model

交通流量的准确预测可以为交通管理部门及个人提供更加可靠的宏观道路状况信息,为城市建设、道路规划、交通管制等问题的研究提供重要的参考。针对现有模型存在的预测准确度不理想、对特征感知能力不强等问题,结合卷积神经网络(CNN)的特征提取能力,双向长短记忆网络(BiLSTM)对于时序数据的连续性、周期性的挖掘能力以及注意力机制(Attention Mechanism,AM)对于关键信息的捕获能力,提出了一种融合多特征的CNN-BiLSTM-AM组合模型,旨在提升模型在交通流量预测准确度上的表现。采用美国明尼苏达州I-94号公路每小时西行交通流量数据进行预测实验,实验结果表明CNN-BiLSTM-AM模型具备准确预测交通流量的能力,与其他基准网络模型相比,各项误差指标均有明显下降,其中MSE降至0.00264,RMSE降至0.05135,MAE降至0.02372,判定系数R 2达到0.97001,模型预测结果与真实值拟合度较高。整体模型具有准确度高、稳定性好等优势。

Exact prediction of traffic volume enable traffic operation units and individuals to better understand roadway conditions from a macro perspective,and support important reference for the study of city construction,road planning and traffic control.Considering the problems of unsatisfactory prediction accuracy and weak feature perception ability of existing models,with the features extraction capabilities of CNN,the BiLSTM`s ability to mine periodically for time series and the critical information capturing ability of attention mechanism,a CNN-BiLSTM-AM fusion model has been proposed to improve the accuracy of prediction.Hourly westbound traffic data on Highway I-94 in Minnesota,USA,was used for predictive experiments.The experiment shows that the CNN-BiLSTM-AM model has the ability to accurately predict traffic volume.Compared with other network models,all evaluation indicators have significantly reduced.MSE dropped to 0.00264,RMSE dropped to 0.05135,MAE dropped to 0.02372,and the coefficient of determination reached 0.97001,which are well fitted to the true values.This model has better performance in terms of accuracy and stability.