Skip to content

Latest commit

 

History

History
186 lines (107 loc) · 9.96 KB

README.md

File metadata and controls

186 lines (107 loc) · 9.96 KB

Computing Nash Flows Over Time and Thin Flows

The Nash Flow Computation Tool (NFC) allows the computation of thin flows with resetting and consequently Nash flows over time with or without spillback. NFC, being embedded in a graphical user-interface, facilitates not only quick and easy creation of networks but also the analysis of e.g. earliest arrival time functions and inflow functions in a dynamic equilibrium, which can be computed with a single click for further research. The flow models used are due to [1] in the basic case without spillback and due to [2] in the case of spillback. The algorithms to compute thin flows are based on [2] and [3]. A complete guide on Nash flow over time can be found in [4].

At the moment, NFC is only able to handle single-commodity flow networks. For a useful tool to compute multi-commodity flows (although no equilibria), please check out this repository.

NFC was developed being part of a bachelor thesis [3] (found here) and then afterwards enhanced within the TU Berlin/ECMath project Dynamic Models and Algorithms for Equilibria in Traffic Networks.

As an example we provide the breass network with spillback for which the Nash flow over time consists of infinitly many phases, as it is described in [5].

Requirements

NFC was designed to run on Linux-based OS.

Install (Windows / Mac)

For Windows or Mac we recommend to use a virtual machine VirtualBox running Ubuntu 20.04.4 Focal Fossa.

To do so, first follow the guide to set up the Ubuntu virtual machine with VirtualBox. Then follow the installation guide for Ubuntu.

VirtualBox Guide

This guide has been tested with VirtualBox 7.0.

  1. In the main windows, choose "new" to create a new virtual machine.

  2. Type in some name and choose Type: Linux and Version: Ubuntu (64bit). Keep ISO Image empty.

  3. Choose at least 4 GB of memory and 4 CPUs.

  4. Select "Use an Existing Virtual Hard Drive File" and open the "Hard Disk Selector".

  5. Add the Ubuntu 20.04.4 (64bit).vdi file downloaded from www.osboxes.org.

  6. Finish the dialog.

  7. In the main windows: press start while the newly created virtual machine is selected to start it. A graphical user interface with a login mask should open. (If only a terminal opens something went wrong with the installation of the virtual machine or VirtualBox.)

  8. The password of user osboxes is osboxes.org. This user has administration rights and the password needs to be entered for sudo commands.

Install (Ubuntu)

Warning: Newer Ubuntu version might not have Qt5 anymore and it can be tricky to install it. However, this might help. Otherwise, this easiest way is to set up a virtual machine with VirtualBox (see above) and afterwards follow the steps below within the VM.

The following has been tested in a virtual machine with VirtualBox 7.0 running Ubuntu 20.04.4 Focal Fossa 64bit.

Open a terminal and follow these steps:

  1. Install python3 (often already installed) and git:

    sudo apt update && sudo apt install python3 git
  1. Install python packages:

    sudo apt install python3-pyqt5 python3-matplotlib python3-numpy python3-networkx
  2. Clone the repository into a folder NashFlowComputation:

    git clone https://github.com/LeonSering/NashFlowComputation.git
  3. Download SCIP from the SCIP download page. Choose the newest Debian package (ends with .deb). Tested with SCIPOptSuite-8.1.0-Linux-ubuntu.deb.

  4. Install SCIP by either clicking on SCIPOptSuite-8.1.0-Linux-ubuntu.deb or by

    sudo apt install ./Downloads/SCIPOptSuite-8.1.0-Linux-ubuntu.deb 
  5. Go into the NashFlowComputation directory with:

    cd NashFlowComputation
  6. Open the GUI with:

    python3 mainControl.py
  7. Set the SCIP path by clickling "Select binary" and choose the scip binary which can often be found under /usr/bin/scip (otherwise you can find it by using the command whereis scip).

If there are any errors, check the versions of the python packages with apt-cache policy <package_name>. To install the correct version use sudo apt install <package_name>=<version> with the version given in requirements-section.

Usage

Create the network and compute a Nash Flow Over Time

  1. First choose if you want to compute a Nash Flow Over Time with Spillback (choose "Spillback") or without (choose "General").

  2. Create a graph by clicking on an existing node and drag the mouse to create an arc. At the end all nodes (in particular the sink t) must be reachable from the source s.

  3. Use the middle mouse to pan the canvas and use "Del" to delete the selected node or edge.

  4. Select an edge to set the "Capacity" and the "Transit times" (and inflow and storage capacity for spillback networks). Don't forget to click "Update edge".

  5. Set the network "Inflow Rate" to the desired value.

  6. If the network is created, click on "Compute Nashflow" to start the computation.

Optional settings

  • For a larger network it is recommended to set "# Intervals" to some positive value (for example 100) to obtain a result in a reasonable amount of time. More intervals can still be computed afterward. For networks with infinite phases, the default value of -1 (compute all intervals) will crash the application.

  • You can rename and re-position nodes by clicking on them and editing the fields. Click "Update node" afterwards.

  • You can save and load graphs via the "File"-menu.

  • For no-spillback graph you can choose between three different algorithms. All should have the same result but the running time might vary drastically.

  • To store the output for each thin flow into a file, choose a Output directory and deselect the Clean-up checkbox.

Inspect the Nash Flow Over Time

  • Click on a node in the upper canvas to inspect the earliest arrival time of that node.

  • Click on an edge in the upper canvas to inspect the cumulative in- and outflow the queue size and the edge load. Above the graphs the current queue is visualized.

  • You can hide graphs by clicking on the respective line in the legend.

  • Click on the intervals to see the Thin Flow of that interval.

  • It is possible to run an animation by clicking the play button.

  • You can also set a specific time. For example if you set the time to 10:

    • The upper canvas shows the current flow state at snapshot time 10.

    • The thin flow in the lower canvas shows that thin flow for the particle/agent that starts at time 10.

    • By default active edges with no thin flow are displayed. You can show them by disabling the "Show edges w/out flow"-checkbox.

  • The graph can be exported to pdf or pgf.

  • The flow animation can be exported to an mp4-file by clicking on the black record button. It is possible to select the time window for the animation under the Option menu.

Compute only a single thin flow

  • Start the ThinFlowComputation tool by running

    python3 thinFlow_mainControl.py

    or by selecting "Open ThinFlowComputation" in the "Options" menu in the Nash Flow Computation window.

  • It is also possible to move the network from the NashFlowComputation window to the ThinFlowComputation by choosing "Move current graph to ThinFlowComputation" in the "Options" menu.

  • Create the network in the upper canvas by drag and drop.

  • Set the "Capacity" and the "Flow bound" (only for spillback thin flows) for each edge (transit times do not play any role for thin flows as long as you only include active edges). Don't forget to press "Update edge".

  • Set the status (active or resetting) of each edge. (Note that resetting edges are always active. However for the program an resetting edge that is not active is assumed to be non-active and ignored.)

  • The graph of active edges must by acyclic.

  • Again there are three different algorithms for computing a non-spillback thin flow. All of them should output the same but my vary in the running time.

  • Set the "Inflow Rate" to the desired value.

  • Click on "Compute Thin Flow" when the network is ready.

  • The bottom canvas shows the computed thin flow. Next to each node is the $\ell'$ value and next to each node the $x'$ value.

References

  • [1] Koch, Ronald, and Skutella, Martin. "Nash equilibria and the price of anarchy for flows over time." Theory of Computing Systems (2011) link
  • [2] Sering, Leon, and Vargas Koch, Laura. "Nash flows over time with spillback." Proceedings of the Thirtieth Annual ACM-SIAM Symposium on Discrete Algorithms (2019) link
  • [3] Zimmer, Max. "Nash Flows Over Time: Models and Computation." Bachelor Thesis at TU Berlin (2017) link
  • [4] Sering, Leon. "Nash flows over time." Dissertation (2020) link
  • [5] Ziemke, Theresa, and Sering. Leon, and Nagel, Kai. "Spillback Changes the Long-Term Behavior of Dynamic Equilibria in Fluid Queuing Networks." 23rd Symposium on Algorithmic Approaches for Transportation Modelling, Optimization, and Systems (2023) link