Hygese.jl

This package is under active development. It can introduce breaking changes anytime. Please use it at your own risk.

A solver for the Capacitated Vehicle Routing Problem (CVRP)

This package provides a simple Julia wrapper for the Hybrid Genetic Search solver for Capacitated Vehicle Routing Problems (HGS-CVRP).

Install:

] add Hygese

Use:

using Hygese, CVRPLIB
ap = AlgorithmParameters(timeLimit=1.3, seed=3) # `timeLimit` in seconds, `seed` is the seed for random values.
cvrp = CVRPLIB.readCVRP(<path to .vrp file>)
result = solve_cvrp(cvrp, ap; verbose=true) # verbose=false to turn off all outputs

• result.cost = the total cost of routes
• result.time = the computational time taken by HGS
• result.routes = the list of visited customers by each route, not including the depot (index 1). In the CVRPLIB instances, the node numbering starts from 1, and the depot is typically node 1. However, the solution reported in CVRPLIB uses numbering starts from 0.

For example, P-n19-k2 instance has the following nodes:

1 30 40
2 37 52
3 49 43
4 52 64
5 31 62
6 52 33
7 42 41
8 52 41
9 57 58
10 62 42
11 42 57
12 27 68
13 43 67
14 58 27
15 37 69
16 61 33
17 62 63
18 63 69
19 45 35

and the depot is node 1. But the solution reported is:

Route #1: 4 11 14 12 3 17 16 8 6 
Route #2: 18 5 13 15 9 7 2 10 1 
Cost 212

The last element 1 in Route #2 above represents the node number 2 with coordinate (37, 52).

This package returns visited_customers with the original node numbering. For the above example,

using Hygese, CVRPLIB
cvrp, cvrp_file, cvrp_sol_file = CVRPLIB.readCVRPLIB("P-n19-k2")
result = solve_cvrp(cvrp)

returns

julia> result.routes
2-element Vector{Vector{Int64}}:
 [19, 6, 14, 16, 10, 8, 3, 11, 2]
 [7, 9, 17, 18, 4, 13, 15, 12, 5]

To retrieve the CVRPLIB solution reporting format:

julia> reporting(result.routes)
2-element Vector{Vector{Int64}}:
 [18, 5, 13, 15, 9, 7, 2, 10, 1]
 [6, 8, 16, 17, 3, 12, 14, 11, 4]

In all data the first element is for the depot.

• x = x coordinates of nodes, size of n
• y = x coordinates of nodes, size of n
• dist_mtx = the distance matrix, size of n by n.
• service_times = service time in each node
• demands = the demand in each node
• vehicle_capacity = the capacity of the vehicles
• duration_limit = the duration limit for each vehicle
• n_vehicles = the maximum number of available vehicles

Three possibilities:

• Only by the x, y coordinates. The Euclidean distances are used.

ap = AlgorithmParameters(timeLimit=3.2) # seconds
result = solve_cvrp(x, y, demands, vehicle_capacity, n_vehicles, ap; service_times=service_times, duration_limit=duration_limit, verbose=true)

• Only by the distance matrix.

ap = AlgorithmParameters(timeLimit=3.2) # seconds
result = solve_cvrp(dist_mtx, demand, vehicle_capacity, n_vehicles, ap; service_times=service_times, duration_limit=duration_limit, verbose=true)

• Using the distance matrix, with optional x, y coordinate information. The objective function is calculated based on the distance matrix, but the x, y coordinates just provide some helpful information. The distance matrix may not be consistent with the coordinates.

ap = AlgorithmParameters(timeLimit=3.2) # seconds
result = solve_cvrp(dist_mtx, demand, vehicle_capacity, n_vehicles, ap; x_coordinates=x, y_coordinates=y, service_times=service_times, duration_limit=duration_limit, verbose=true)

As TSP is a special case of CVRP, the same solver can be used for solving TSP.

The following interfaces are provided:

• Reading .tsp or .atsp files via TSPLIB.jl:

tsp = TSPLIB.readTSP("br17.atsp")
ap = AlgorithmParameters(timeLimit=1.2)
result = solve_tsp(tsp, ap; use_dist_mtx=true)

• By the coordinates, by the distance matrix, or by both:

result1 = solve_tsp(x, y, ap)
result2 = solve_tsp(dist_mtx, ap)
result3 = solve_tsp(dist_mtx, ap; x_coordinates=x, y_coordinates=y)

The paramters for the HGS algorithm with default values are:

Base.@kwdef mutable struct AlgorithmParameters
    nbGranular :: Int32 = 20
    mu :: Int32 = 25
    lambda :: Int32 = 40
    nbElite :: Int32 = 4
    nbClose :: Int32 = 5
    targetFeasible :: Float64 = 0.2
    seed :: Int32 = 0
    nbIter :: Int32 = 20000
    timeLimit :: Float64 = 0.0
    useSwapStar :: Int32 = 1 # 1 = true, 0 = false
end

• CVRPLIB.jl

• TSPLIB.jl

• LKH.jl

• Concorde.jl

• PyHygese: A Python wrapper for HGS-CVRP