AI50 project to modelize Capacited Vehicle Routing Problem (CVRP) and find solutions using Multi-Agents Systems, metaheuristics and Machine Learning.

This project requires a 3.8+ python version and multiple library to be runned. Before running it for the first time please ensure that you have every python library required by running the following command:

pip install requirements.txt

Note: You can also run the setup.bat (or setup.sh dependng on your os) to install pyhton dependancies. See the next section for more details.

To be runned the application needs Redis server available (version for windows and linux are present in the redis folder).

Finally, since the multi-agent algorithm as been written in sarl you will need it in order to get a solution from this algorithm.

The installation will check for you if your python version and your os is supported by the ROAD software.

To install all dependencies for python on windows, please run the following command (or double click on it):

setup.bat

It will ask you your python command (please enter the python command that will enable you to run python 3.8+). For example python. The program will then ask you your pip command. Please enter your pip command linked to the python you have entered previously. For example pip.

Finally you will have a new batch script generated: run.bat. You can run it using the following command (or double clicking on it):

run.bat

To install all dependencies for linux os, please run the following command:

./setup.sh

It will ask you your python command (please enter the python command that will enable you to run python 3.8+). For example python3.8. The program will then ask you your pip command. Please enter your pip command linked to the python you have entered previously. For example pip3.8.

Finally you will have a new shell script generated: run.sh. You can run it using the following command:

./run.sh

Note: To run the multi-agent algorithm, you need to launch the sarl program in the sarl/vrp directory. You then need to launch the ListnerAgent.

Note: If you have generated a run.bat or run.sh you can just run this file, no need to launch python. You can then ignore this section.

You can add one or more of the followings arguments:

You may also change the configuration of the application in the file /gui/config.py. All parameters are in this file, and commented.

To launch the web application, run the main.py file. Run the following command in the root file:

python main.py

The web application runned by default on http:https://localhost:8080/.

The algorithm is based on the scientific article Agents toward Vehicle Routing Problem. It is written in sarl and bridged to python using redis to enable the communication.

Two metaheuristics have been implemented:

• Tabu search
• Genetic Algorithm
• Grey Wolf Optimizer

Some metaheuristics start from a solution provided by the Clark & Wright saving algorithm and try to improve it. Here's a small example of the result provided by the tabu search algorithm.

The Grey Wolf Optimizer will build a solution without needing one at the very start. The algorithm is based on the scientific article Using Grey Wolf Algorithm to Solve the Capacitated Vehicle Routing Problem and the wikiversity page dedicated to the Grey Wolf Algorithm.

As said previously we also use some heuristics to build a first solution for some of the meta-heuristics. Here's the list of implemented heuristics:

• Clark & Wright Saving Algorithm
• First Fit Decreasing (not available on the web gui)
• Nearest Neighbors (not available on the web gui)

A neural network model has been trained with our dataset that we created (Dataset.csv in the folder ML) to find the best setting to use for the tabu search algorithm.

Moreover a K_means algorithm have been implemented to define different clusters of customers in the graph and to compute the Davies-Bouldin index used among the inputs of our neural network model.

Neural network model used :

The Artificial Intelligence have been integrated to the software and gives directly the best settings to use for the tabu search algorithm.

The last type of algorithm implement is an implementation of the capacited K-means. Note that this algorithm is not available from the web interface, but may be used from python.