RaphaelMeudec / Velib

This projects implements various methods for Vehicle Routing Problem, applied to a city bike repartition

Velib

This project implements various meta-heuristics for an optimization problem.

Problem

The problem tackles a bike sharing repartition. You want to equilibrate the repartition of bikes in different stations of a city. As input data, you have the number M of trucks with their bike capacity, and the number N of stations with their current count of bikes and their ideal number of bikes.

Circuit

Each station belongs to only one circuit, and a circuit is assigned to a truck. A circuit begins and ends at the same point. We note the desequilibrium of a circuit as the sum of the differences between the ideal number and the effective number of bikes in each station. We can modify the association circuit-station, and modify the order in which the stations are explored, as long as every station belongs to one circuit.

Loss function

The loss function is computed based on the desequilibrium of each circuit at the end of the tour, and the distance of each truck.
The expression is loss = 1e6*desequilibria + distances.

Run

First algorithm implemented is the steepest descent. It can be launched with:

./run --steepest2opt ../data/v0.dat

Second algorithm is a descent solver, which can explore only improving solutions, or every solution based on argument. It can be launched with:

./run --descent ../data/v0.dat --explore all
./run --descent ../data/v0.dat --explore strict

Third algorithm is the annealing solver, which takes as argument the initial temperature, and the number of iterations at each step. It can be launched with:

./run --annealing ../data/v0.dat --temperature 5000000 --itermax 40000