This repo includes all the benchmarking code and hyper-parameters for the paper "A GPU-accelerated Large-scale Simulator for Transportation System Optimization Benchmarking".

Prerequisites

• Any linux distribution
• NVIDIA GPU with CUDA 12
• Python 3.9

Setup

1. Download the dataset archive data.tar.gz and unpack it.

   tar xvf data.tar.gz

   If everything is correct, you should see the following structure:

   data
   ├── congestion
   ├── dynamic
   ├── perf
   ├── road
   ├── tidal
   └── tsc
   

2. Install the required packages.

   pip install -r requirements.txt

   In case of 'GLIBCXX_3.4.30' not found error, also install:

   conda install -c conda-forge libstdcxx-ng=12

Benchmarks

Performance

Switch to the folder:

cd ./Perf

Install baseline simulators:

• Install sumo

  sudo add-apt-repository ppa:sumo/stable
  sudo apt-get update
  sudo apt-get install sumo sumo-tools sumo-doc
  pip install traci

• Install cityflow

  git clone --depth 1 https://github.com/cityflow-project/CityFlow
  pip install ./CityFlow

• Install CBLab

  git clone --depth 1 https://github.com/caradryanl/CityBrainLab
  pip install ./CityBrainLab/CBEngine/core

Run all experiments:

bash run.sh

The results will be saved under ./log/${city}/${algo}_${n}.json, which contains a list that records the duration of each simulation step.

Traffic Signal Control

Switch to the folder:

cd ./TSC

Run all experiments:

bash run.sh

The results will be saved under ./log/${algo}/${city}/${date}/info.log, where each line records ATT TP time.

Dynamic Lane Assignment

Switch to the folder:

cd ./DynamicLane

Run all experiments:

bash run.sh

The results will be saved under ./log/${algo}/${city}/${date}/info.log, where each line records ATT TP time.

Tidal Lane Control

Switch to the folder:

cd ./TidalLane

Run all experiments:

bash run.sh

The results will be saved under ./log/${algo}/${city}/${date}/info.log, where each line records ATT TP time.

Congestion Pricing

Switch to the folder:

cd ./CongestionPricing

Run all experiments:

bash run.sh

The results will be saved under ./log/${algo}/${city}/${date}/info.log, where each line records ATT TP time.

Road Planning

Switch to the folder:

cd ./CongestionPricing

Step 1. Data Preparation

• Fetch the road net from OSM data of 2019.

  python preparation/fetch_2019_geojsons.py

• Find the difference between road network of 2019 and road network of 2024, mark those roads not in road network of 2019 as constructed in the past five years.

  for city in beijing shanghai newyork paris; do
      python preparation/fetch_candidate_way_ids.py --candidate_way_path ./data/${city}_candidate_ways.pkl --city $city
  done

  • --candidate_way_path: output path for all constructed ways from 2019 to 2024.
  • --city: name of the city to process.

• Select 50 roads out of constructed roads with the highest vehicle count during the simulation of morning peak and evening peak as the optimization set.

  for city in beijing shanghai newyork paris; do
      for condition in smooth normal congested; do
          python exp/select_50_optimize_way_ids.py \
              --opt_way_path ./data/${condition}_${city}_opt_ways.pkl \
              --city $city \
              --map_path ./data/moss.map_china_${city}.pb \
              --trip_path ./data/${condition}_${city}_trip.pb \
              --candidate_way_path ./data/${city}_candidate_ways.pkl
      done
  done

  • --opt_way_path: output path for 50 ways for algorithms to optimize.
  • --city: name of the city to process.
  • --map_path: map in protobuf format.
  • --trip_path: trip in protobuf format.
  • --candidate_way_path: output path of fetch_candidate_way_ids.py

Step 2. Run Optimization Methods

Run all experiments:

bash run.sh

The results will be saved under ./log/${algo}/${city}/${date}/info.log, where each line records ATT TP time.

Ours vs CityFlow

Switch to the folder:

cd ./OptPref

Step 1. Apply the patch provided in folder data.

git clone --depth 1 https://github.com/cityflow-project/CityFlow`
git apply ./diff_data/diff_cityflow.patch`
pip install ./CityFlow

Step 2. Run Optimization Methods

Run all experiments:

bash run.sh

The results will be saved under ./log/${algo}/${city}/${date}/info.log, where each line records ATT TP time simulation_time_cost optimization_time_cost.