A Physics-Informed Deep Learning Paradigm for Traffic State and Fundamental Diagram Estimation

Shi, Rongye; Mo, Zhaobin; Huang, Kuang; Di, Xuan; Du, Qiang

doi:10.1109/tits.2021.3106259

Cited by 64 publications

(37 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The idea is to assess the computed traffic state of a neural network not only based on measurement data, but also on its compliance with the given physical model. Recent approaches include Huang and Agarwal (2020), who publish an approach giving promising insights into the ability to reconstruct traffic densities based on sparse measurements; Liu et al (2020), who use simulated trajectory data providing local densities, study an approach to reconstruct the traffic density in space and time and Shi et al (2021), who design a PINN with additional error term based on the LWR and a Greenshields FD in order to estimate traffic density on a road. These approaches show promising results, though, published network architectures are not applicable to the problem of traffic speed estimation with probe data: Some approach require that probe vehicles provide density information Liu et al (2020), which is rarely the case.…”

Section: Related Workmentioning

confidence: 99%

“…These approaches show promising results, though, published network architectures are not applicable to the problem of traffic speed estimation with probe data: Some approach require that probe vehicles provide density information Liu et al (2020), which is rarely the case. Additionally, the network needs to be re-trained to fit the data in each situation Shi et al (2021). This is computationally expensive, and is prone to overfitting since the network's output is optimized to match observed data.…”

Section: Related Workmentioning

confidence: 99%

“…The challenge is to design the input and expected output in such a way that the neural network is able to learn the input-output relation and be able to apply the mapping to unseen data in an optimal way. Given a well-fitting setup and neural network architecture, a neural network is able to generalize the problem without overfitting Goodfellow et al (2016). The input to the neural network needs to contain all relevant information for solving the problem.…”

Section: Feature Setup and Network Architecturementioning

confidence: 99%

“…One difference, though, is that, while image processing usually aims at the classification of images (e.g. whether there is a cat in a picture), in this case, a continuous output is required, which is called a regression network Goodfellow et al (2016). Another difference is that the given input data is sparse in time and space.…”

Section: Feature Setup and Network Architecturementioning

confidence: 99%

“…Finding the hyper parameters of a neural network that solves the reconstruction problem most accurately is the second important step Goodfellow et al (2016). After many experiments, we present the final neural network architecture that resulted in the best results (see section 6).…”

Section: Neural Network Architecturementioning

confidence: 99%

See 4 more Smart Citations

Estimating Traffic Speeds using Probe Data: A Deep Neural Network Approach

Rempe,

Franeck,

Bogenberger

2021

Preprint

View full text Add to dashboard Cite

This paper presents a dedicated Deep Neural Network (DNN) architecture that reconstructs space-time traffic speeds on freeways given sparse data. The DNN is constructed in such a way, that it learns heterogeneous congestion patterns using a large dataset of sparse speed data, in particular from probe vehicles. Input to the DNN are two equally sized input matrices: one containing raw measurement data, and the other indicates the cells occupied with data. The DNN, comprising multiple stacked convolutional layers with an encoding-decoding structure and feed-forward paths, transforms the input into a full matrix of traffic speeds.The proposed DNN architecture is evaluated with respect to its ability to accurately reconstruct heterogeneous congestion patterns under varying input data sparsity. Therefore, a large set of empirical Floating-Car Data (FCD) collected on German freeway A9 during two months is utilized. In total, 43 congestion distinct scenarios are observed which comprise moving and stationary congestion patterns. A data augmentation technique is applied to generate input-output samples of the data, which makes the DNN shift-invariant as well as capable of managing varying data sparsities. The DNN is trained and subsequently applied to sparse data of an unseen congestion scenario. The results show that the DNN is able to apply learned patterns, and reconstructs moving as well as stationary congested traffic with high accuracy; even given highly sparse input data. Reconstructed speeds are compared qualitatively and quantitatively with the results of several state-of-the-art methods such as the Adaptive Smoothing Method (ASM), the Phase-Based Smoothing Method (PSM) and a standard Convolutional Neural Network (CNN) architecture. As a result, the DNN outperforms the other methods significantly.

show abstract