Short term traffic flow prediction of expressway service area based on STL-OMS
Short term traffic flow prediction of expressway service area based on STL-OMS
复制标题
基于STL-OMS的高速公路服务区短期交通流预测
DOI：
10.1016/j.physa.2022.126937
复制
发表时间：
2022-03-02
影响因子：
3.3
通讯作者：
Liu, Wenhui
中科院分区：
文献类型：
--
作者：
Zhao, Jiandong;Yu, Zhixin;Liu, Wenhui
为了提高高速公路服务区的管理能力，及时制定交通流变化策略，提出了一种短期交通流预测模型。首先，根据规则对提取的数据进行清理，构建四种特征(时间、空间、统计和外部因素)。在此基础上，建立了短时交通流预测模型WADNN。该模型分别利用长短期记忆神经网络(LSTM)、卷积神经网络(CNN)和自我注意机制提取不同的特征。此外，还采用基于黄土的季节趋势分解算法对交通流进行分解，使其更好地适应趋势。对分解后的3个分量进行OMS(最优模型选择)运算，将各分量的预测值相加得到最终预测值，用均方根误差(RMSE)、平均绝对误差(MAE)、平均绝对百分比误差(MAPE)和R2系数衡量模型效果。最后，以某高速公路服务区为例，将该模型与常用的几种模型进行了比较。结果表明，WADNN的预测效果较好，STL-OMS可以进一步提高预测精度。(C)2022爱思唯尔B.V.保留所有权利。
To improve the management ability of expressway service area and formulate strategies for traffic flow changes in time, a short-term traffic flow prediction model is pro-posed. Firstly, cleaning the extracted data according to the rules and constructing four kinds of features (temporal, spatial, statistical and external factors). Then, a short-term traffic flow prediction model WADNN (wide attention and deep neural networks) is constructed. In the model, LSTM (long and short-term memory neural network), CNN (convolution neural network) and self-attention mechanism are used to extract different features respectively. In addition, the STL (Seasonal-Trend decomposition procedure based on LOESS) algorithm is used to decompose the traffic flow to fit the trend better. For the three decomposed components, the OMS (optimal model selection) operation is carried out, the prediction of each component is added to obtain the final predicted value, and the model effect is measured according to the RMSE (root mean square error), MAE (mean absolute error), MAPE (Mean Absolute Percentage Error) and R2 coefficient. Finally, taking an expressway service area as an example, the proposed model is compared with some common models. The results show that the prediction effect of WADNN is better and STL-OMS can further improve the accuracy. (C) 2022 Elsevier B.V. All rights reserved.