A Novel Vehicle Repositioning Strategy for Heterogeneous Fleets

This repository contains the implementation of a relocation algorithm for Free-Floating Vehicle Sharing Systems.

For further details concerning the theory behind our model, take a look at the paper that is provided in this repository.

Installation

Our implementation is written in python and makes use of various libraries.
To quickly install all requirements we recommend using conda.
All necessary requirements can be installed with the help of the requirements.txt file. To create a new conda environment with all requirements use the following command: conda env create --file requirements.yml
Or if you are using mamba: mamba env create --file requirements.yml Or if you are running on linux you may also use the explicit requirements, which might be faster: conda create --file requirements-linux64.txt -n VEHICLE_REPOSITION.
Make sure to select the newly created environment as your jupyter kernel.

Project Structure

The models source code is structured in folders: modules and notebooks.
The folder notebooks contains jupyter notebooks. These notebooks should be executed in the correct order, marked by their filenames prefix.
Some code is relocated into the modules folder for reusability purposes.
Before running any notebook make sure to set the PATH_DIR_TRIPS_RAW configuration variable in the config.py to the directory path of the trip data.

Configuration

The modules/config.py file allows to configure some of the models hyperparameters as well as the solver that is used to solve the linear program.

Performance

As the underlying Linear Program of our model has a large number of variables and constraints and solving Integer Linear Programs is generally NP-hard, it can take very long to solve depending on your hardware.
General tips to improve performance (and decrease model accuracy):

• Reduce H3_RESOLUTION (to 6).
• Increase PERIOD_DURATION (to 12).
• Only relocate in certain periods, e.g. only the first one. Set RELOCATION_PERIDOS_INDEX to 0.
• Reduce number of (reduced) scenarios. e.g. set N_REDUCED_SCENARIOS to 4.
• Increase number of threads used by the solver. Set SOLVER_OPTIONS to {"threads": x}
• Discard value-at-risk. Set beta to 0 when calling stochastic_program.create_model(beta=0).
• Decrease fleet size factory.set_initial_allocation(small_fleet_size).