Forecasting road traffic speeds by considering area-wide spatio-temporal dependencies based on a graph convolutional neural network (GCN)

Yu, Byeonghyeop; Lee, Yongjin; Sohn, Keemin

doi:10.1016/j.trc.2020.02.013

Cited by 164 publications

(56 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Jo et al [46] proposed a convolutional neural network to deal with map images representing traffic states. Yu et al [47] proposed forecasted road traffic speeds by considering area‐wide spatio‐temporal dependencies based on a graph convolutional neural network. Gu et al [48] utilised an improved Bayesian combination model with deep learning to predict the traffic volume by assigning appropriate weights to different sub‐predictors by considering their performance during the several past time intervals.…”

Section: Related Workmentioning

confidence: 99%

Efficient deep learning based method for multi‐lane speed forecasting: a case study in Beijing

Wan

et al. 2020

IET intell. transp. syst

View full text Add to dashboard Cite

Real-time and accurate multi-lane traffic condition forecasting is of great importance to the connected and automated vehicle highway system. However, the majority of existing deep learning based traffic prediction methods focus on pursuing the precision of the methods while neglect to improve the efficiency of the methods. To achieve the high accuracy and high efficiency of multi-lane traffic flow prediction simultaneously, this study proposes a novel combination method via the integration of the clockwork recurrent neural network (CWRNN) and random forest (RF) method, which is RF-CWRNN. To the best of the authors' knowledge, this is the first time that the CWRNN is introduced to capture the temporal feature of lane-level traffic flow and make traffic speed prediction. Meanwhile, the RF method is employed to measure the temporal relevance of the traffic flow and determine the optimal input time window. To verify the performance of the RF-CWRNN method, the ground-truth data of the expressways in Beijing were utilised to carry out experiments. The results indicate that the RF-CWRNN method is superior to the baseline models in terms of accuracy and robustness. Besides, the proposed method can save plenty of training time compared with the classical long short-term memory neural network.

show abstract

Section: Related Workmentioning

confidence: 99%

Efficient deep learning based method for multi‐lane speed forecasting: a case study in Beijing

Wan

et al. 2020

IET intell. transp. syst

View full text Add to dashboard Cite

show abstract

“…All normalized weights w j,i are non-negative and the sum of them in the receptive field N k i equals to 1. The third equation is a regular extended graph convolution [18] using the generated kernel. The last equation is a shared fullyconnected (FC) layer with C output units.…”

Section: A Dynamic Graph Convolution (Dgc)mentioning

confidence: 99%

“…In most deep learning based traffic forecasting models, temporal dependencies and spatial correlations are modeled separately by different attention layers, e.g. [12], [18], [22], but this disentangled interpretation does not conform to traffic flow For simplification, we consider multistep observations but only one step prediction. The encoder encrypts observed traffic conditions in the past m − 1 time steps into the context vector C, and C is concatenated with the current input X t .…”

Section: Spatial Correlationsmentioning

confidence: 99%

Dynamic Graph Filters Networks: A Gray-box Model for Multistep Traffic Forecasting

Knoop

Lint

2020

2020 IEEE 23rd International Conference on Intelligent Transportation Systems (ITSC)

View full text Add to dashboard Cite

Short-term traffic forecasting is one of the key functions in Intelligent Transportation System (ITS). Recently, deep learning is drawing more attention in this field. However, how to develop a deep learning based traffic forecasting model that can dynamically extract explainable spatial correlations from traffic data is still a challenging issue. The difficulty mainly comes from the inconsistency between static model structures and the dynamic evolution of traffic conditions. To overcome this difficulty, we proposed a novel multistep speed forecasting model, Dynamic Graph Filters Networks (DGFN). The major contribution is that the regular pixel-wise dynamic convolution is extended to graph topology. DGFN has a simple recurrent cell structure where local area-wide graph convolutional kernels are dynamically computed from varying inputs. Experiments on ring freeways show that DGFN is able to precisely predict short-term evolution of traffic speed. Furthermore, we theoretically explain why DGFN is not a pure "black-box", but a "gray-box" model that actually reduces entangled spatial and temporal features into one component representing dynamic spatial correlations. It permits tracking real-time interactions among adjacent links. DGFN has the potential to relate trained parameters in deep learning models with physical traffic variables.

show abstract

“…Despite the superior predictive performance, machine learning models are criticized for the lack of interpretability [17]. Though spatial and temporal correlations can be captured to a certain extent by some machine learning models (e.g., deep neural network [6,18], attention network [5,19], convolutional neural network (CNN) [20,21]), it is challenging to (i) explicitly interpret these correlations, and (ii) quantify their impacts on outcomes and to make relevant inferences. Few studies have developed models considering both predictive performance and interpretability in the traffic state prediction domain when STI is concerned [12,17].…”

Section: Introductionmentioning

confidence: 99%

Online Bus Speed Prediction With Spatiotemporal Interaction: A Laplace Approximation-Based Bayesian Approach

Cui

Xie

Hu³

et al. 2021

IEEE Access

View full text Add to dashboard Cite

This study proposes a novel Bayesian hierarchical approach for online bus speed prediction by explicitly accounting for the spatiotemporal interaction (STI) of speed observations. The use of Laplace approximation can expedite the estimation of Bayesian models and enable the implementation of online prediction. Large numbers of trials are carried out to identify significant predictors and the optimal length of the look-back time window to achieve the highest prediction accuracy. The spatiotemporal interacting patterns are also explored, and results show that the Type IV model assuming the structured spatial effect interacts with the structured temporal effect can best accommodate the bus speed data. Besides, prediction errors of the Type IV model randomly distribute over time and space. The proposed model can achieve high prediction accuracy and computational efficiency without compromising the interpretability of the contributing factors and the unobserved spatiotemporal heterogeneity. The proposed model can be used to assist public transit operation and management, such as bus scheduling, congestion warning, and the development of proactive measures to mitigate bus delays.

show abstract

Forecasting road traffic speeds by considering area-wide spatio-temporal dependencies based on a graph convolutional neural network (GCN)

Cited by 164 publications

References 29 publications

Efficient deep learning based method for multi‐lane speed forecasting: a case study in Beijing

Efficient deep learning based method for multi‐lane speed forecasting: a case study in Beijing

Dynamic Graph Filters Networks: A Gray-box Model for Multistep Traffic Forecasting

Online Bus Speed Prediction With Spatiotemporal Interaction: A Laplace Approximation-Based Bayesian Approach

Contact Info

Product

Resources

About