Implementation of data-driven coupling models at road junctions
Codes used in the paper
Data-driven Models for Traffic Flow at Junctions
by M. Herty & N. Kolbe
A preprint of the paper is available on arXiv.
Julia codes were written by N. Kolbe
Contents
The repository contains a Julia project environment, which includes the package DataDrivenroadjunctions.jl implementing the coupling models and their parameter estimation, and a directory (experiments) with scripts for various use cases. The former implements Riemann solvers, handling of trajectory and macroscopic data, fitting, and a numerical scheme for experiments on 2-to-1 networks. The code relies on other Julia packages, in particular, BlackBoxOptim.jl used for global optimization in the parameter fitting, Flux.jl implementing neural networks and CentralNetworkScheme.jl for numerical simulations on the network. The folder models contains the best fit models C1, C2, ML1, ML2 and ML3 from the paper, the folder FCD is used to store the vehicle trajectory data and the results of the scripts in the experiments folder will be stored in the folder out.
The use-cases of the included scripts are as follows:
| file | use-case |
|---|---|
experiments/dset-selection.jl |
splits the data sets into training-, test- and application data; corresponding "dshash" is used in other scripts |
experiments/dset-store.jl |
generates macroscopic data corresponding to "dshash" |
experiments/data_based_delay.jl |
estimates and visualizes the coupling delay, see Section 4.1 |
experiments/statistics.jl |
prints statistics corresponding to the datasets as shown in Table 6 |
experiments/show_FD.jl |
visualizes the fundamental diagrams, see Section 4.2 |
experiments/coupling_model_comparison.jl |
compares the best fit models in terms of model error as in Table 4 |
experiments/inout.jl |
visualizes boundary fluxes in the data, see Section 6.1 |
experiments/outflux_validation.jl |
predictions of boundary outflux by the best-fit models, see Section 6.1 |
experiments/outflux_results.jl |
processing of results from experiments/outflux_validation.jl |
experiments/prediction.jl |
model predictions in case of congestion, see Section 6.2 |
experiments/model-optimization/FlowMax-model.jl |
fit of the flow maximization models C1 and C2 |
experiments/model-optimization/LinReg-model.jl |
fit of the linear models ML1 and ML2 |
experiments/model-optimization/neural-model.jl |
fit of the neural network model ML3 |
Car trajectory data has not yet been published and is therefore not included in this repository. But it is available upon request.
Usage
Clone the repository on your local machine. To run the scripts activate the project environment, for example, by starting julia from the command line within the root folder of the repository, then changing to pkg mode typing ] and afterwards running activate .. Then the experiments can be run by including the corresponding scripts, e.g., include("experiments/prediction.jl"). You can get started by going through the scripts and modifying them as you like. Further documentation will be added in the future.
In case of questions or problems please contact the authors of the paper or file a GitHub issue in this repository.