【论文阅读】A Hybrid Model Integrating Local and Global...Traffic Prediction[一种融合局部和全局空间相关性的交通预测混合模型]（2）

原文地址：https://ieeexplore.ieee.org/abstract/document/9667362

2. Related Work（相关工作）

Traffic prediction focuses on how to predict the traffic flow state of the next moment based on historical and real- time data, such as volume, density, speed, etc. Accurate traffic prediction can provide real-time traffic information for travelers, which is important to improve road capacity. Traffic flow is complex, variable, and uncertain, and the main characteristics are: nonlinearity, stochasticity, periodicity, and spatial-temporal correlation [20]. To address these characteristics, many prediction methods have been available. The methods can be classified into traditional methods and deep learning methods according to their development process [15].
交通预测研究的是如何根据历史实时数据，如流量、密度、速度等，预测下一时刻的交通流状态。准确的交通预测可以为出行者提供实时的交通信息，对提高道路通行能力具有重要意义。交通流具有复杂、多变和不确定性，其主要特征是:非线性、随机性、周期性和时空相关性[20]。为了处理这些特征，已经有了许多预测方法。根据开发过程[15]可将方法分为传统方法和深度学习方法。

A. Traditional Methods（传统方法）

Traditional methods include classical statistical methods and machine learning methods. Classical statistical methods are more used in the early stage of prediction, such as Historical Average (HA) [21], Auto-Regressive Integrated Moving Average (ARIMA), and its variants [22], [23]. These models perform well under smooth traffic flow. However, when the traffic flow changes drastically, they will show obvious shortcomings and cannot better explore the nonlinearity and uncertainty of traffic flow. Besides, these methods only consider the temporal correlation, ignoring the spatial correlation. As the demand for prediction accuracy increases, the prediction methods gradually turn to machine learning methods such as Support Vector Regression [24], K-Nearest Neighbor [25], and Bayesian model [26], etc.
传统的方法包括经典的统计方法和机器学习方法。在预测早期多采用经典统计方法，如历史平均线(HA)[21]、自回归综合移动平均线(ARIMA)及其变体[22]、[23]。这些模型在平稳的交通流下表现良好。然而，当交通流变化剧烈时，它们会表现出明显的缺点，不能更好地探究交通流的非线性和不确定性。此外，这些方法只考虑了时间相关性，忽略了空间相关性。随着对预测精度要求的提高，预测方法逐渐转向机器学习方法，如支持向量回归[24]、K最近邻[25]、贝叶斯模型[26]等。

B. Deep Learning Methods（深度学习方法）

With the rapid development of artificial intelligence technology, more and more scholars are using deep learning methods, especially neural networks, to solve traffic prediction problems [27]. Neural networks can better fit nonlinear mapping relationships and efficiently capture the internal features of traffic flow. Commonly used networks include Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Graph Convolutional Network (GCN). CNN [28], [29] can efficiently process grid-type data. It can only handle Euclidean data since its translation invariance. RNN [30], LSTM, and GRU are suitable for processing time-series data [31,32,33]. These networks rely on the sequential temporal order of the data itself. We usually use them to model the temporal correlation of traffic flow. Graph structured data has emerged in recent years. Researchers have started to study how to build deep learning on graphs [34]. GCN [35] is an important branch of graph neural networks. It has been applied to traffic prediction and is effective in extracting the spatial correlation of traffic flow [36],[37],[38].
随着人工智能技术的快速发展，越来越多的学者正在使用深度学习方法，特别是神经网络来解决流量预测问题[27]。神经网络能够更好地拟合非线性映射关系，有效地捕捉交通流的内部特征。常用的网络包括卷积神经网络(Convolutional Neural Network, CNN)、循环神经网络(Recurrent Neural Network, RNN)和图卷积网络(Graph Convolutional Network, GCN)。CNN[28]，[29]可以有效的处理网格类型的数据。由于其平移不变性，它只能处理欧几里德数据。RNN[30]、LSTM和GRU适用于处理时间序列数据[31,32,33]。这些网络依赖于数据本身的时序时序顺序。我们通常用它们来模拟交通流的时间相关性。图结构数据是近年来出现的。研究人员已经开始研究如何在图[34]上构建深度学习。GCN[35]是图神经网络的一个重要分支。该方法已应用于交通预测，可有效提取交通流[36], [37], [38]的空间相关性。

 In order to fully analyze the complex characteristics of traffic flow such as nonlinearity, randomness, and spatial- temporal correlation, hybrid neural network structure is usually utilized instead of a single neural network structure. Zheng et al. [39] developed an attention-based Conv-LSTM module to extract spatial features and temporal features separately. Zhang et al. [19] constructed a GCN-GRU two-layer network structure to analyze the spatial-temporal correlation and achieved good results when applied to the Los Angeles freeway dataset. Zhang et al. [40] used GCN and feedforward neural networks to consider time, space, weather conditions, and the date to predict highway traffic flow.
 为了充分分析交通流非线性、随机性、时空相关性等复杂特征，通常采用混合神经网络结构代替单一神经网络结构。Zheng等人[39]开发了一种基于注意力的Conv-LSTM模块，分别提取空间特征和时间特征。Zhang等人[19]构建了GCN-GRU两层网络结构来分析时空相关性，应用于洛杉矶高速公路数据集取得了良好的效果。Zhang 等人[40]利用GCN和前馈神经网络考虑时间、空间、天气条件和日期等因素对公路交通流进行预测。

 The original GCN can only analyze the geographical correlation of sensors, which does not reflect the deep spatial correlation. Some studies have been carried out to improve the GCN. Li et al. [41] combine GCN and GRU to construct a DCRNN network. The model replaces the parameter matrix in GRU with the convolution of the parameter matrix and the Laplacian matrix. Based on DCRNN, Guo et al. [42] constructed a dynamic Laplacian matrix by summing the initial Laplacian matrix and the parameterized residual matrix. Wu et al. [43] proposed that initial graph structures do not always reflect true spatial correlation. They use adaptive matrices to learn internal spatial relationships for prediction. Li et al. [44] put aside the inherent road network structure. They proposed an adaptive graph convolution structure to automatically learn the interrelationships between nodes through training. Lv [45] et al. encoded the non-Euclidean correlation and semantic structure of the road network into multiple graphs and then constructed a multi-graph convolutional network to mine these correlations. Li [17] et al. used data-driven adjacency matrices to mine the spatial relationships of graph structures.
 原有的GCN只能分析传感器的地理相关性，不能反映更深层次的空间相关性。为改进GCN进行了一些研究。Li et al.[41]结合GCN和GRU构造了一个DCRNN网络。该模型用参数矩阵与拉普拉斯矩阵的卷积来代替GRU中的参数矩阵。Guo et al.[42]在DCRNN的基础上，将初始拉普拉斯矩阵与参数化残差矩阵相加，构造了一个动态拉普拉斯矩阵。Wu等人[43]提出初始图结构并不总是反映真实的空间相关性。他们使用自适应矩阵来学习内部空间关系来进行预测。Li et al.[44]抛开固有的路网结构。他们提出了一种自适应图卷积结构，通过训练自动学习节点之间的相互关系。Lv[45]等人将道路网络的非欧几里德相关性和语义结构编码成多个图，然后构建多图卷积网络来挖掘这些相关性。Li[17]等人使用数据驱动的邻接矩阵来挖掘图结构的空间关系。

 Compared with existing methods, our model combines the local and global spatial correlation to explore the spatial correlation of traffic flow comprehensively. In the actual road, spatial correlation does not necessarily exist between adjacent sensors. In the same traffic flow direction, spatial correlation also exists between non-first-order neighboring sensors. Through the global spatial-temporal component, we can discard the useless sensor spatial information and improve the capability of global spatial-temporal correlation analysis. Also, in the local spatial-temporal component, we separate node feature learning and neighboring nodes spatial correlation mining, which can effectively avoid the smoothing problem. By combining analysis from both global and local perspectives, our model can deeply analyze the spatial-temporal correlation of traffic flow and improve the prediction capability.
 与现有方法相比，本模型结合局部和全局空间相关性，全面探讨交通流的空间相关性。在实际道路中，相邻传感器之间不一定存在空间相关性。在同一交通流方向上，非一阶相邻传感器之间也存在空间相关性。通过全局时空分量，可以剔除无用的传感器空间信息，提高全局时空相关性分析的能力。在局部时空分量中，我们将节点特征学习和相邻节点空间相关性挖掘分离，有效地避免了平滑问题。通过结合全局和局部的分析，我们的模型可以深入分析交通流的时空相关性，提高预测能力。

参考文献

[15] X. Yin, G. Wu, J. Wei, Y. Shen, H. Qi, and B. Yin, ‘‘Deep learning on traffic prediction: Methods, analysis and future directions,’’ IEEE Trans. Intell. Transp. Syst., early access, Feb. 10, 2021, doi: 10.1109/ TITS.2021.3054840.

[17] Z. Li, G. Xiong, Y. Tian, Y. Lv, Y. Chen, P. Hui, and X. Su, ‘‘A multi-stream feature fusion approach for traffic prediction,’’ IEEE Trans. Intell. Transp. Syst., early access, Oct. 7, 2020, doi: 10.1109/tits.2020.3026836.

[19] L. Zhao, Y. Song, C. Zhang, and Y. Liu, ‘‘T-GCN: A temporal graph convolutional network for traffic prediction,’’ IEEE Trans. Intell. Transp. Syst., vol. 21, no. 9, pp. 3848–3858, Sep. 2020, doi: 10.1109/TITS.2019.2935152.

[20] A. Ermagun and D. Levinson, ‘‘Spatiotemporal traffic forecasting: Review and proposed directions,’’ Transp. Rev., vol. 38, no. 6, pp. 786–814, Nov. 2018, doi: 10.1080/01441647.2018.1442887.

[21] J. Liu and W. Guan, ‘‘A summary of traffic flow forecasting methods,’’ J. Highway Transp. Res. Develop., vol. 21, no. 3, pp. 82–85, Mar. 2004.

[22] S. V. Kumar and L. Vanajakshi, ‘‘Short-term traffic flow prediction using seasonal ARIMA model with limited input data,’’ Eur. Transp. Res. Rev., vol. 7, no. 3, pp. 1–9, Sep. 2015, doi: 10.1007/s12544-015-0170-8.

[23] H. Yang, X. Li, W. Qiang, Y. Zhao, W. Zhang, and C. Tang, ‘‘A network traffic forecasting method based on SA optimized ARIMA–BP neural network,’’ Comput. Netw., vol. 193, Jul. 2021, Art. no. 108102, doi: 10.1016/j.comnet.2021.108102.

[24] S. Liu, Y. Lin, C. Luo, and W. Shi, ‘‘A novel learning method for traffic flow forecasting by seasonal SVR with chaotic simulated annealing algorithm,’’ in Proc. IEEE 6th Int. Conf. Comput. Commun. Syst. (ICCCS), Apr. 2021, pp. 205–210, doi: 10.1109/ICCCS52626.2021.9449161.

[25] F. Su, H. Dong, L. Jia, Y. Qin, and Z. Tian, ‘‘Short-term traffic forecasting using self-adjusting k-nearest neighbours,’’ IET Intell. Transp. Syst., vol. 12, no. 1, pp. 41–48, Feb. 2018, doi: 10.1049/IET-its. 2016.0263.

[26] Y. Gu, W. Lu, X. Xu, L. Qin, Z. Shao, and H. Zhang, ‘‘An improved Bayesian combination model for short-term traffic prediction with deep learning,’’ IEEE Trans. Intell. Transp. Syst., vol. 21, no. 3, pp. 1332–1342, Mar. 2019, doi: 10.1109/TITS.2019.2939290.

[27] Y. Lv, Y. Duan, W. Kang, Z. Li, and F.-Y. Wang, ‘‘Traffic flow prediction with big data: A deep learning approach,’’ IEEE Trans. Intell. Transp. Syst., vol. 16, no. 2, pp. 865–873, Apr. 2014, doi: 10.1109/tits.2014.2345663.

[28] Z. He, C. Y. Chow, and J. D. Zhang, ‘‘STCNN: A spatio-temporal convolutional neural network for long-term traffic prediction,’’ in Proc. IEEE Int. Conf. Mob. Data Manag., Jun. 2019, pp. 226–233, doi: 10.1109/MDM.2019.00-53.

[29] T. Bogaerts, A. D. Masegosa, J. S. Angarita-Zapata, E. Onieva, and P. Hellinckx, ‘‘A graph CNN-LSTM neural network for short and long- term traffic forecasting based on trajectory data,’’ Transp. Res. C, Emerg. Technol., vol. 112, pp. 62–77, Mar. 2020, doi: 10.1016/j.trc.2020. 01.010.

[30] H. Zhu, Y. Xie, W. He, C. Sun, K. Zhu, G. Zhou, and N. Ma, ‘‘A novel traffic flow forecasting method based on RNN-GCN and BRB,’’ J. Adv. Transp., vol. 2020, pp. 1–11, Oct. 2020, doi: 10.1155/2020/7586154.

[31] Z. Wang, X. Su, and Z. Ding, ‘‘Long-term traffic prediction based on LSTM encoder-decoder architecture,’’ IEEE Trans. Intell. Transp. Syst., vol. 22, no. 10, pp. 6561–6571, Oct. 2021, doi: 10.1109/ TITS.2020.2995546.

[32] D. Han, X. Yang, G. Li, S. Wang, Z. Wang, and J. Zhao, ‘‘Highway traffic speed prediction in rainy environment based on APSO-GRU,’’ J. Adv. Transp., vol. 2021, pp. 1–11, Aug. 2021, doi: 10.1155/2021/4060740.

[33] X. Chen, H. Chen, Y. Yang, H. Wu, W. Zhang, J. Zhao, and Y. Xiong, ‘‘Traffic flow prediction by an ensemble framework with data denoising and deep learning model,’’ Phys. A, Stat. Mech. Appl., vol. 565, Mar. 2021, Art. no. 125574, doi: 10.1016/j.physa.2020.125574.

[34] L. Yuting, Z. Ming, M. Chicheng, B. Bo, Z. Zhiheng, Y. Kai, W. Guanghui, and Y. Guiying, ‘‘Graph neural network,’’ SCIENTIA SINICA Math., vol. 50, no. 3, p. 367, Mar. 2020, doi: 10.1360/n012019-00133.

[35] T. N. Kipf and M. Welling, ‘‘Semi-supervised classification with graph convolutional networks,’’ in Proc. 5th Int. Conf. Learn. Represent. (ICLR) Conf. Track, 2017, pp. 1–14.

[36] C. Zheng, X. Fan, C. Wang, and J. Qi, ‘‘GMAN: A graph multi-attention network for traffic prediction,’’ in Proc. AAAI Conf. Artif. Intell., vol. 34, no. 1, Apr. 2020, pp. 1234–1241, doi: 10.1609/aaai.v34i01.5477.

[37] W. Li, X. Wang, Y. Zhang, and Q. Wu, ‘‘Traffic flow prediction over muti- sensor data correlation with graph convolution network,’’ Neurocomputing, vol. 427, pp. 50–63, Feb. 2021, doi: 10.1016/j.neucom.2020.11.032.

[38] X. Shi, H. Qi, Y. Shen, G. Wu, and B. Yin, ‘‘A spatial-temporal attention approach for traffic prediction,’’ IEEE Trans. Intell. Transport. Syst., vol. 22, no. 8, pp. 4909–4918, Apr. 2020, doi: 10.1109/ TITS.2020.2983651.

[39] H. Zheng, F. Lin, X. Feng, and Y. Chen, ‘‘A hybrid deep learning model with attention-based conv-LSTM networks for short-term traffic flow prediction,’’ IEEE Trans. Intell. Transp. Syst., vol. 22, no. 11, pp. 6910–6920, Nov. 2021, doi: 10.1109/tits.2020.2997352.

[40] T. Zhang, W. Ding, T. Chen, Z. Wang, and J. Chen, ‘‘A graph convolutional method for traffic flow prediction in highway network,’’ Wireless Commun. Mobile Comput., vol. 2021, pp. 1–8, Jul. 2021, doi: 10.1155/2021/1997212.

[41] Y. Huang, Y. Weng, S. Yu, and X. Chen, ‘‘Diffusion convolutional recurrent neural network with rank influence learning for traffic forecasting,’’ in Proc. 18th IEEE Int. Conf. Trust, Secur. Privacy Comput. Commun./13th IEEE Int. Conf. Big Data Sci. Eng. (TrustCom/BigDataSE), Aug. 2019, pp. 1–16.

[42] K. Guo, Y. Hu, Z. Qian, H. Liu, K. Zhang, Y. Sun, J. Gao, and B. Yin, ‘‘Optimized graph convolution recurrent neural network for traffic prediction,’’ IEEE Trans. Intell. Transp. Syst., vol. 22, no. 2, pp. 1138–1149, Feb. 2021, doi: 10.1109/TITS.2019.2963722.

[43] Z. Wu, S. Pan, G. Long, J. Jiang, and C. Zhang, ‘‘Graph wavenet for deep spatial-temporal graph modeling,’’ in Proc. Int. Jt. Conf. Artif. Intell. (IJCAI), Aug. 2019, pp. 1907–1913, doi: 10.24963/ijcai.2019/264.

[44] L. Bai, L. Yao, C. Li, X. Wang, and C. Wang, ‘‘Adaptive graph convolutional recurrent network for traffic forecasting,’’ in Proc. Adv. Neural Inf. Process. Syst., Dec. 2020, pp. 1–16.

[45] M. Lv, Z. Hong, L. Chen, T. Chen, and S. Ji, ‘‘Temporal multi- graph convolutional network for traffic flow prediction,’’ IEEE Trans. Intell. Transp. Syst., vol. 22, no. 6, pp. 3337–3348, Jun. 2020, doi: 10.1109/TITS.2020.2983763.