VRP Mathematical Models - Heuristics

VRP Mathematical
Models & Heuristics
:Prepeared by
Ayoub Faisal Alzabidi
Supervised by:
Prof. Mehdi Mrad
OUTLINE
Introduction
VRP Mathematical Models:

 Vehicle flow formulation
 Set partitioning formulation
VRP Heuristics:
 Constructive heuristics
 Two-phase heuristics
 Improvement heuristics
Introduction
The vehicle routing problem (VRP) asks the following question: "What is the optimal
set of routes for a fleet of vehicles to traverse in order to deliver to a given set of
customers?“
TSP is a special case of VRP when the number
of vehicles is one and its capacity is infinity.
It is an NP-hard in combinatorial optimization.
VRP is important in industry, in which, good optimization of the problem can
reduce transportation cost of a product almost from (10%) down to (5%).

VRP Solution Techniques
Vehicle Flow
Mathematical
Formulation
Models
Set Partitioning
Formulation
These models represent the mathematical

formulation of the classical VRP (CVRP),
however, they can be used to formulate other
variants of VRP by adding and/ or modifying
constraints to satisfy each variant.
VRP- Vehicle Flow Formulation (VFF)
- Assume two nodes to represent the same single depot (0, n+1), where all routes
must start on 0 and return to n+1.
- K: max number of vehicles (available in the depot), each one has a max capacity Q.
- : demand of node i, where .
- Let : continuous decision variable corresponding to the cumulated demand on the
route that visits node j.
Minimize the total travel cost of the routes - Let
All customers are visited exactly once
Correct flow of vehicles if a vehicle arrives to a node h, then it

must depart from this node
Limits the maximum number of

routes to K
Avoid subtours (cycling

routes that do not pass
Ensure that vehicle capacity is
through the depot). It is
extension of TSP MTS 𝑞𝑖 not exceeded
formulation
Binary restrictions
VRP- Vehicle Flow Formulation (VFF)
 VFF is based on binary variables associated to arcs of a network representation of the problem.
 Different types of constraints are proposed in the literature to impose vehicle capacities and/or avoid
Irnich, S., Toth, P., & Vigo, D. (2014). Chapter 1: The family of vehicle routing problems. In Vehicle
subtours. Ref. (Click here)
Routing: Problems, Methods, and Applications, Second Edition (pp. 1-33). Society for Industrial and
Applied Mathematics.
 It is more intuitive and leads to a compact model that can be straightforwardly put on a black-box
optimization solver.
 Number of constraints becomes large in terms of the number of customers.
 Lower bound provided by the linear relaxation of this model is known to be weak in relation to other
models.
 Branch-and-cut methods based on specialized

F. Semet, P. Toth, and D.valid inequalities
Vigo. (2014). can exact
Chapter 2: Classical be devised tothesolve
algorithms for these
capacitated formulations.
vehicle routing problem. In
(Click here) P. Toth and D. Vigo, editors, Vehicle Routing: Problems, Methods, and Applications, MOS/SIAM Ser Optim, pages 37-58.
VRP- Set Partitioning Formulation (SPF)
- Let : set of routes that satisfy the problem requirements.
- : set of customers,
- K: max number of vehicles (available in the depot), each one has a max capacity Q.
- Let
- Let
Minimizes the total cost of the selected
routes
Exactly one visit to each customer node
If K is sufficiently large for Imposes the maximum number of vehicles

the problem, then this available at the depot
constraint can be dropped
from the formulation
Binary restrictions
Computed using arc costs

VRP- Set Partitioning Formulation (SPF)
 The number of constraints in SPF is much smaller with respect to VFF, but it has a huge
number of variables: one for each feasible route in the problem.
 Generating all routes of is impractical in general, as the number of routes is exponential in
terms of the numbers of customers.
 SPF require using the column generation technique for solving

Lübbecke,the linear
M. E., relaxation
& Desrosiers, J. (2005).of
(Click here) Selected topics in column generation.
Operations research, 53(6), 1007-1023.
model starting with a small subset of routes . Ref.
 SPF has stronger linear relaxations than VFF, but requires the implementation of sophisticate
(Click here)
M. Poggi and E. Uchoa. (2014). Chapter 3: New exact algorithms for the capacitated vehicle routing problem. In P. Toth
techniques
and such
D. Vigo, as column
editors, generation
Vehicle routing: Problems,and specialized
methods, branch-and-price
and applications, methods.
MOS/SIAM Ser Optim, pages 59-86. Ref.
VRP Solution Techniques
Vehicle Flow
Mathematical
Formulations
Formulation
Set Partitioning
Formulation
VRP- Heuristics
VRP Heuristics: methods that perform a relatively limited exploration

of the search space and typically produce good quality solutions
within small computational times, these heuristics can be classified
into three categories:
 Constructive heuristics
 Two-phase heuristics
 Improvement heuristics
VRP- Constructive Heuristics
Constructive Heuristics (1960s – 1980s): gradually build a feasible

solution while keeping and eye on the solution cost, which in general
are used to find a good initial solution, these heuristics can be
classified into two types:
 Saving heuristics
 Insertion heuristics
Saving heuristics: merge routes step by step based on a savings

measure
 Sequential version: Clarke & Wright (1964)
 Parallel version: Gaskell (1967) and Yellow (1970)
 Version with several enhancements and acceleration

procedures: Nelson et al. (1985) and Paessens (1988)
Insertion heuristics: build a solution by inserting one customer at a

time.
 Sequential insertion heuristics: build one route at a time
 Parallel insertion heuristics: build many or all routes in parallel
 Two-phase sequential and parallel insertion heuristics .

VRP- Two-phase Heuristics
Two-phase Heuristics (1960s – 1990s): methods where the

construction of the solution is obtained by using a clustering phase
and a route construction phase, these heuristics can be classified into
two classes:
 Cluster-first Route-second
 Route-first Cluster-second
Cluster-first Route-second: perform a single clustering of the vertex

set and then determine a vehicle route on each cluster, these heuristics
can be classified into three main algorithms:
 Fisher & Jaikumar algorithm
 Sweep algorithm
 Petal algorithm
Fisher & Jaikumar algorithm: determines ’a priori’ a number of seed

customers to initialize each cluster, then it solves a generalized
assignment problem considering allocation costs, customer weights
and vehicle capacity. Finally, each cluster is turned into a route by
solving a TSP to optimality.
 Fisher and Jaikumar (1981)
Sweep algorithm: sweep the deliveries in circular order to create

clusters, for which a new cluster is initiated each time the capacity is
exceeded. Then, each cluster is turned into a route by solving a TSP to
optimality.
 Gillett & Miller (1974)
Petal algorithm: is an extension of the sweep algorithm consist of

generating several routes (petals) and make a final selection by solving
a set partitioning problem.
𝑎𝑖𝑘= {1 𝑖𝑓𝑐𝑢𝑠𝑡𝑜𝑚𝑒𝑟 𝑖 𝑏𝑒𝑙𝑜𝑛𝑔𝑠 𝑡𝑜 𝑟𝑜𝑢𝑡𝑒 𝑘 𝑑𝑘 =𝐶𝑜𝑠𝑡 𝑜𝑓 𝑟𝑜𝑢𝑡𝑒𝑘
𝑥 𝑘={ 1 𝑖𝑓 𝑟𝑜𝑢𝑡𝑒𝑘𝑏𝑒𝑙𝑜𝑛𝑔𝑠 𝑡𝑜 𝑡 h 𝑒 𝑠𝑜𝑙𝑢𝑡𝑖𝑜𝑛
 Solved as an exact algorithm: Balinski & Quandt (1964)
 Solved heuristically: Foster & Ryan (1976) and Ryan et al. (1993)
 Solved heuristically with some configurations: Renaud et al.(1996)

Route-first Cluster-second: construct in a first phase a giant TSP tour

that visits all customers, disregarding side constraints, and decompose
this tour into feasible vehicle routes in a second phase by assimilating
to a shortest path problem in a auxiliary directed acyclic graph.
 Newton & Thomas (1974)
 Bodin & Berman (1979)
 Beasley (1983)
VRP- Improvement Heuristics
Improvement Heuristics (1960s – 2000s): local search heuristics start

with initial solution and perform operations to improve the objective
of the solution, these heuristics can be classified into two classes:
 Intra-route improvement heuristics
 Inter-route improvement heuristics

Intra-route heuristics: improvement operations are performed on each

vehicle route separately, any improvement method for the TSP can be
applied, like:
 λ-opt mechanisms: Lin (1965):
 Deleting and reinserting λ arcs
 Or-opt mechanisms: Cordeau et al. (2004):
 Relocate sequences of bounded size

Inter-route heuristics: improvement operations are performed on

several routes at a time, these improvement operations can be
classified into three types:
 Relocate: removing k customers from their current route
and reinserting them elsewhere.
 Swap: swapping customers between different routes.
 Cross: exchange and or reverse two sequences differently.

Inter-route heuristics: include efficient move evaluations and pruning

procedures to address large-scale problem instances, like:
 Neighborhood restrictions & granular search: restrain the subset of moves to spatially
related customers:
 Johnson & McGeoch (1997)
 Toth & Vigo (2003)
 Sequential search: breaking down any pro table move into a list of arc exchanges to
prune many non-promising moves:

 Christodes & Eilon (1972)
 Irnich & Villeneuve (2003)
VRP- Meta-Heuristics
Even though we do not yet know the

optimal solutions, it is safe to say
that current metaheuristics are
capable of producing high quality
solutions for instances with up to
500 customers, and solution quality
has improved steadily over the last
decade. However, the gains in
solution quality are now becoming
marginal, an indication that the
current solution methodologies may
have reached a plateau (2).
Reference
Beresneva, E., & Avdoshin, S. (2018).
Analysis of mathematical formulations of
capacitated vehicle routing problem and
methods for their solution, 30(3).
REFERENCES
1. Munari, P., Dollevoet, T., & Spliet, R. (2016). A generalized formulation
for vehicle routing problems. arXiv preprint arXiv:1606.01935.
2. Laporte, G., Ropke, S., & Vidal, T. (2014). Chapter 4: Heuristics for the vehicle routing problem.
In Vehicle Routing: Problems, Methods, and Applications, Second Edition (pp. 87-116). Society for
Industrial and Applied Mathematics.
3. Ropke, S. (2005). Heuristic and exact algorithms for vehicle routing

problems. Unpublished PhD thesis, Computer Science Department, University of
Copenhagen.
4. Gonzalez-Feliu, J. (2008). Models and methods for the city logistics: The two-
echelon capacitated vehicle routing problem (Doctoral dissertation).
5. https://en.wikipedia.org/wiki/Vehicle_routing_problem .
Thank You

VRP Mathematical Models - Heuristics

Uploaded by

Copyright:

Available Formats

VRP Mathematical Models - Heuristics

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

VRP Mathematical Models - Heuristics

Uploaded by

Copyright:

Available Formats

VRP Mathematical

Models & Heuristics

VRP Mathematical Models:

 Set partitioning formulation

TSP is a special case of VRP when the number

of vehicles is one and its capacity is infinity.

It is an NP-hard in combinatorial optimization.

VRP is important in industry, in which, good optimization of the problem can

reduce transportation cost of a product almost from (10%) down to (5%).

These models represent the mathematical

All customers are visited exactly once

Correct flow of vehicles if a vehicle arrives to a node h, then it

Limits the maximum number of

Avoid subtours (cycling

 Number of constraints becomes large in terms of the number of customers.

 Branch-and-cut methods based on specialized

Exactly one visit to each customer node

If K is sufficiently large for Imposes the maximum number of vehicles

Computed using arc costs

number of variables: one for each feasible route in the problem.

 Generating all routes of is impractical in general, as the number of routes is exponential in

terms of the numbers of customers.

 SPF require using the column generation technique for solving

VRP Heuristics: methods that perform a relatively limited exploration

Constructive Heuristics (1960s – 1980s): gradually build a feasible

Saving heuristics: merge routes step by step based on a savings

 Parallel version: Gaskell (1967) and Yellow (1970)

 Version with several enhancements and acceleration

Insertion heuristics: build a solution by inserting one customer at a

 Parallel insertion heuristics: build many or all routes in parallel

 Two-phase sequential and parallel insertion heuristics .

Two-phase Heuristics (1960s – 1990s): methods where the

Cluster-first Route-second: perform a single clustering of the vertex

Fisher & Jaikumar algorithm: determines ’a priori’ a number of seed

Sweep algorithm: sweep the deliveries in circular order to create

Petal algorithm: is an extension of the sweep algorithm consist of

 Solved as an exact algorithm: Balinski & Quandt (1964)

 Solved heuristically with some configurations: Renaud et al.(1996)

Route-first Cluster-second: construct in a first phase a giant TSP tour

 Bodin & Berman (1979)

Improvement Heuristics (1960s – 2000s): local search heuristics start

 Inter-route improvement heuristics

Intra-route heuristics: improvement operations are performed on each

 Deleting and reinserting λ arcs

 Or-opt mechanisms: Cordeau et al. (2004):

 Relocate sequences of bounded size

Inter-route heuristics: improvement operations are performed on

and reinserting them elsewhere.

 Swap: swapping customers between different routes.

 Cross: exchange and or reverse two sequences differently.

Inter-route heuristics: include efficient move evaluations and pruning

prune many non-promising moves:

Even though we do not yet know the

3. Ropke, S. (2005). Heuristic and exact algorithms for vehicle routing

You might also like