Research proposal

Object Oriented Modules for Dynamic Vehicle Routing
- Research Proposal -

Victor Pillac

´
Ecole des Mines de Nantes
IRCCyN UMR 6597
Nantes, France
Universidad de Los Andes
Industrial Engineering Department
Bogot´, Colombia
a
v.pillac63@uniandes.edu.co
vpillac@mines-nantes.fr

December 9, 2011

Outline
1 Introduction
Dynamic vehicle routing
Literature review
Motivation
2 Dynamic and deterministic routing
Parallel adaptive large neighborhood search
Application to the TRSP
Bi-objective dynamic routing
Application to the D-VRPTW
3 Dynamic and stochastic routing
Multiple scenario approach
Application to the D-VRPSD
4 Conclusions
Contributions
Plan for 3rd year
V. Pillac (EMN/Uniandes) Research Proposal 2011/12/09 2 / 44

Introduction

Outline
1 Introduction
Literature review
Motivation
4 Conclusions
Contributions
Plan for 3rd year

Introduction Dynamic vehicle routing

Outline
1 Introduction
Literature review
Motivation
4 Conclusions
Contributions
Plan for 3rd year


A taxonomy of routing problems

Information quality
Deterministic
Stochastic input
input
Input known Static and Static and
Information beforehand deterministic stochastic
evolution Input changes Dynamic and Dynamic and
over time deterministic stochastic



A taxonomy of routing problems

Information quality
Deterministic
Stochastic input
input
Input known Static and Static and
Information beforehand deterministic stochastic
evolution Input changes Dynamic and Dynamic and
over time deterministic stochastic
We focus on problems in which input changes over time
New clients
Demand realization




Environment
Vehicle Starts Finishes New customer:
is ready serving A serving A X

Next: A Next: B
Decision

Update

Decision

Update
Dispatcher

Figure: Timeline of events in dynamic vehicle routing


Introduction Literature review

Outline
1 Introduction
Literature review
Motivation
4 Conclusions
Contributions
Plan for 3rd year


Literature review
Complete review in [Pillac et al., 2011a]



Literature review
Dynamic and deterministic
Fast reoptimization
[Montemanni et al., 2005, Gambardella et al., 2003]
Adaptive memory
[Gendreau et al., 1999, Bent and Van Hentenryck, 2004,
Benyahia and Potvin, 1998]



Literature review
Dynamic and deterministic
Fast reoptimization
[Montemanni et al., 2005, Gambardella et al., 2003]
Adaptive memory
[Gendreau et al., 1999, Bent and Van Hentenryck, 2004,
Benyahia and Potvin, 1998]
Dynamic and stochastic
Stochastic modeling
[Godfrey and Powell, 2002, Powell and Topaloglu, 2005,
Simao et al., 2009, Novoa and Storer, 2009]
Sampling
[Van Hentenryck and Bent, 2006]


Introduction Motivation

Outline
1 Introduction
Literature review
Motivation
4 Conclusions
Contributions
Plan for 3rd year

Introduction Motivation

Motivation
Study the different aspects of dynamic routing
Differences with static routing
Degrees of dynamism
Algorithm performance evaluation
Important algorithm features
Develop software libraries
Extensible
Flexible
Apply findings to a case study
Routing of a crew of technicians


Dynamic and deterministic routing

Outline
1 Introduction
Literature review
Motivation
4 Conclusions
Contributions
Plan for 3rd year


Consider dynamic changes in input
No information is available on the dynamically revealed data
Optimization approaches
Use simple insertion heuristics
Split the planning horizon in decision epochs
Perform an optimization each time a change occurs
Use an adaptive memory to store previous good solutions characteristics


Dynamic and deterministic routing Parallel adaptive large neighborhood search

Outline
1 Introduction
Literature review
Motivation
4 Conclusions
Contributions
Plan for 3rd year


Parallel Adaptive Large Neighborhood Search (pALNS)
Extension of ALNS [Pisinger and Ropke, 2007]
Make use of multi-core architecture
Maintain a pool of promising solutions
Limit synchronization between threads for performance
pALNS Algorithm
Input: Π0 an initial solution
Output: The best solution found by the algorithm
1: Initialize the promising solution pool Ω ← {Π0 }
2: for N iterations do
3: Select subset Ωt of solutions
4: parallel forall Π in Ωt do
5: In a separate thread, perform n ALNS iterations starting with Π
6: Report the resulting solution to the main thread
7: end forall
8: Update the pool of promising solutions Ω
9: end for
10: return The best solution from Ω


Computational results
Tested on Solomon benchmark [Solomon, 1987]
Factor 3 speedups
Reduced running time variance
Gap increased by 0.18%

q
q
q

q
q q

200
0.06

q q q q
q q
q
q q q
q
q
q q q
q
q q
q q q
q
q

150
0.04

q
GAP

Time
q

q
q
q
q
q

100
0.02

q
q
q
q
q

q
q
q
q
0.00

50

1 2 4 8
1 2 4 8
Threads Threads


Dynamic and deterministic routing Application to the TRSP

Outline
1 Introduction
Literature review
Motivation
4 Conclusions
Contributions
Plan for 3rd year


Technician Routing and Scheduling Problem (TRSP)
Has: 6
Has:
1 5

2
3 Pickup 4

Set of requests Crew of technicians
Location Starting/ending location
Required skills, tools, spare (home)
parts Set of skills, initial tools,
Time window and service spare parts
time Working day length
Main depot
Technicians can pickup tools and spare parts
Unlimited tools and spare parts


Static TRSP
Goal
Have a reference approach for the dynamic version

Proposed approach
Adaptive Large Neighborhood Search (ALNS)
Set covering using routes generated throughout ALNS
Presented in [Pillac et al., 2011c]
Computational experiments
Solomon benchmark (VRPTW)
38/38 Optimal solutions
14/18 Best known solution, 9 improved
Randomly generated TRSP instances based on Solomon instances



Dynamic TRSP
Generated release date for a proportion of requests
Compare to a Best Insertion (BI) heuristic

Results

ALNS BI
DOD VI (%) R VI (%) R
10 4.52 0.08 16.58 0.33
50 7.10 0.42 44.81 1.75
90 9.50 1.58 52.94 2.83
Average 7.04 0.69 38.11 1.64

VI: Value of information, R: average number of rejected requests


Dynamic and deterministic routing Bi-objective dynamic routing

Outline
1 Introduction
Literature review
Motivation
4 Conclusions
Contributions
Plan for 3rd year


Static routing generally aims at designing a set of routes
For a given day (operational)
For a longer horizon (tactical,strategic)
Minimizing cost/distance/fleet size
Dynamic routing introduces new dimensions
Changes in the routing while the vehicle is en-route
Changes in the driver/request assignments
Communication with vehicles
New problematic
How to ensure cost efficiency?
How to ensure stability in the routing?
How to optimize both objectives in limited time?
What is the trade-off between the two objectives?



Measuring stability
D-VRP case: a new request r arrives at time t
Let:
Ω the set of all solutions
Πt−1 the previous solution
Πt a solution including r
How stable is Πt with respect to Πt−1 ?



Measuring stability
Let:
Deﬁne a metric δ : Ω × Ω → R
How to deﬁne δ?



Measuring stability
Let:
Deﬁne a metric δ : Ω × Ω → R
How to deﬁne δ?
Hamming distance
Number of changed assignments
Edit (Levenshtein) distance
Insertions
Removals
Substitutions



Bi-objective Parallel ALNS
Make use of multi-core architecture
Focus on the distance during ALNS iterations
Store non-dominated solutions

pBiALNS Algorithm
Input: Π0 an initial solution
Output: The non-dominated solutions found by the algorithm
1: Initialize the non-dominated solutions Ω ← {Π0 }
2: for N iterations do
3: Select subset Ωt of non-dominated solutions
4: parallel forall Π in Ωt do
5: In a separate thread, perform n ALNS iterations starting with Π
6: Report the non-dominated solutions to the main thread
7: end forall
8: Update the non-dominated solutions Ω
9: end for
10: return The non-dominated solutions Ω


Dynamic and deterministic routing Application to the D-VRPTW

Outline
1 Introduction
Literature review
Motivation
4 Conclusions
Contributions
Plan for 3rd year


Bi-objective Dynamic VRP-TW
Simulation of a dynamic setting
Instance C101
90/100 revealed requests
Part of the routes is already executed
A new request is revealed
pBiALNS run for 25,000 iterations
100
90
80
70
60 Pareto
Edit distance

50 ALNS
solutions
40
30
20
10
0
0% 2% 4% 6% 8% 10% 12% 14% 16% 18% 20%
Gap



Bi-objective Dynamic VRP-TW
Simulation of a dynamic setting
Instance C101
90/100 revealed requests
Part of the routes is already executed
A new request is revealed
pBiALNS run for 25,000 iterations
70

60

50
Pareto
40
Edit distance

ALNS
30 solutions

20

10

0
0% 1% 2% 3% 4%
Gap


Dynamic and stochastic routing

Outline
1 Introduction
Literature review
Motivation
4 Conclusions
Contributions
Plan for 3rd year


Consider dynamic changes in input
Stochastic information is available on the dynamically revealed data
Random variables of known distribution
Historical data
Optimization approaches
Ignore stochastic information
Model stochastic information
Use sampling to capture uncertainty


Dynamic and stochastic routing Multiple scenario approach

Outline
1 Introduction
Literature review
Motivation
4 Conclusions
Contributions
Plan for 3rd year



Sampling based approach
Framework for online optimization
[Van Hentenryck and Bent, 2006]
Runs throughout the day
Scenario pool Scenario pool

Realizations of the random variables
Decision process Next client
Which client next?
Aggregate scenario knowledge



MSA : Scenarios

Scenario generation Example
1 Start with known data
D C

E
A

B

Goal
Capture uncertainty through sampling



MSA : Scenarios

2 Sample random variable D C
distributions
3 Solve resulting optimization X
E
problem A
Z
B Y

Goal



MSA : Scenarios

2 Sample random variable D C
distributions
3 Solve resulting optimization X
E
problem A
Z
4 Remove sample data from B Y
solution

Goal



MSA : Decisions

Decision process
Decide which request to visit
next
Aggregate information from all
scenarios
Fast process



MSA : Decisions

Decision process Example
Decide which request to visit Consensus
next Select the request that
Aggregate information from all appear ﬁrst in most
scenarios scenarios
Fast process
S1 0 4 1 6 5 0 ...
S2 0 4 1 3 0 ...
S3 0 4 1 2 3 6 0 ...
S4 0 4 1 2 3 6 5 0



MSA : Decisions

Decision process Example
Decide which request to visit Consensus
next Select the request that
Aggregate information from all appear ﬁrst in most
scenarios scenarios
Fast process
S1 0 4 1 6 5 0 ...
S2 0 4 1 3 0 ...
S3 0 4 1 2 3 6 0 ...
S4 0 4 1 2 3 6 5 0

Goal
Use scenarios to take a non-myopic decision


jMSA
Java implementation of the MSA
Highlights
Event-driven
Flexible
Extensible
Described in [Pillac et al., 2011b]

MSA
procedure Event Handler Scenario
queue manager pool

Events Handlers
Kernel

Components Scenario

Callback
Scenario Scenario Scenario
…
Problem layer generator optimizer updater

Dynamic and stochastic routing Application to the D-VRPSD

Outline
1 Introduction
Literature review
Motivation
4 Conclusions
Contributions
Plan for 3rd year


Dynamic VRP with Stochastic Demands
Extension of the Vehicle Routing Problem
Stochastic Demands (known distribution law)
Traditional approaches
Two-stage approach
Robust a-priori routing
Recourse actions in case of failure
Assumes vehicles cannot be rerouted
Why a dynamic approach?
Increasing availability of low-cost positioning systems
Real-time communication with vehicles
Allows further optimization



30 Instances from Novoa (2005)
30, 40 and 60 customers
Randomly distributed in 1x1 square
Uniform discrete demand distribution
2 vehicle capacities




Value of information
Algorithm min. max. avg.
[Secomandi, 2001] 11.1% 19.6% 13.6%
[Novoa and Storer, 2009]-1 3.5% 12.3% 5.8%
[Novoa and Storer, 2009]-2 2.8% 10.7% 4.8%
jMSA 0.9% 6.3% 3.3%




Value of information
Algorithm min. max. avg.
[Secomandi, 2001] 11.1% 19.6% 13.6%
[Novoa and Storer, 2009]-1 3.5% 12.3% 5.8%
[Novoa and Storer, 2009]-2 2.8% 10.7% 4.8%
jMSA 0.9% 6.3% 3.3%

Conclusions
Better value of information
Faster decisions (ms vs min)


Conclusions

Outline
1 Introduction
Literature review
Motivation
4 Conclusions
Contributions
Plan for 3rd year

Conclusions Contributions

Outline
1 Introduction
Literature review
Motivation
4 Conclusions
Contributions
Plan for 3rd year


Contributions I
Working papers
Pillac, V., Gendreau, M., Guret, C., Medaglia, A. L. (2011), A review of dynamic vehicle
routing problems, European Journal of Operational Research, Under review
Pillac, V., Guret, C., Medaglia, A. L. (2011), An event-driven optimization framework for
dynamic vehicle routing, Decision Support Systems, Under review
Pillac, V., Guret, C., Medaglia, A. L. (2011), On the Technician Routing and Scheduling
Problem, Optimization Letters, Under review

Technical reports
Pillac, V., Gendreau, M., Guret, C., Medaglia, A. L. (2011), A review of dynamic vehicle
routing problems, CIRRELT Research Paper, CIRRELT-2011-62
Pillac, V., Guret, C., Medaglia, A. L. (2011), An event-driven optimization framework for
dynamic vehicle routing, Technical Report 11/2/AUTO, Ecole des Mines de Nantes,
France
Pillac, V., Guret, C., Medaglia, A. L. (2010), Dynamic Vehicle Routing Problems: State of
the art and Prospects, Technical Report 10/4/AUTO, Ecole des Mines de Nantes, France



Contributions II
Conference proceedings
Pillac, V., Guret, C., Medaglia, A. L. (2011), On the Technician Routing and Scheduling
Problem, Proceedings of the 9th Metaheuristics International Conference (MIC 2011),
675-678, Udine (Italy)
Pillac, V., Guret, C., Medaglia, A. L. (2011), A dynamic approach for the vehicle routing
problem with stochastic demands, ROADEF 2011, St Etienne (France)
Pillac, V., Guret, C., Medaglia, A. L. (2010), Solving the Vehicle Routing Problem with
Stochastic Demands with a Multiple Scenario Approach, ALIO-INFORMS 2010, Buenos
Aires (Argentina)

Software libraries
Pillac, V., Guret, C., Medaglia, A. L., VroomModeling: A general purpose modeling library
for vehicle routing problems.
Pillac, V., Guret, C., Medaglia, A. L., VroomHeuristics: A set of general heuristics for
vehicle routing problems.
Pillac, V., Guret, C., Medaglia, A. L., jMSA: An event-driven optimization framework for
dynamic vehicle routing.


Conclusions Plan for 3rd year

Outline
1 Introduction
Literature review
Motivation
4 Conclusions
Contributions
Plan for 3rd year


Plan for 3rd year
Research
1 Bi-objective approach for dynamic routing

Journal papers

1 Parallel ALNS and bi-objective dynamic routing

Conferences
1 ROADEF’12
2 ODYSSEUS’12



Questions & Answers

Thank you for your attention


References

References I
Bent, R. and Van Hentenryck, P. (2004). Scenario-based planning for partially dynamic vehicle routing with stochastic
customers. Operations Research, 52(6):977–987.

Benyahia, I. and Potvin, J. Y. (1998). Decision support for vehicle dispatching using genetic programming. IEEE
Transactions on Systems Man and Cybernetics Part A - Systems and Humans, 28(3):306–314.

Gambardella, L., Rizzoli, A., Oliverio, F., Casagrande, N., Donati, A., Montemanni, R., and Lucibello, E. (2003). Ant colony
optimization for vehicle routing in advanced logistics systems. In Proceedings of the International Workshop on Modelling
and Applied Simulation (MAS 2003), pages 3–9.

Gendreau, M., Guertin, F., Potvin, J.-Y., and Taillard, E. (1999). Parallel tabu search for real-time vehicle routing and
dispatching. Transportation Science, 33(4):381–390.

Godfrey, G. and Powell, W. B. (2002). An adaptive dynamic programming algorithm for dynamic ﬂeet management, I:
Single period travel times. Transportation Science, 36(1):21–39.

Montemanni, R., Gambardella, L. M., Rizzoli, A. E., and Donati, A. V. (2005). Ant colony system for a dynamic vehicle
routing problem. Journal of Combinatorial Optimization, 10(4):327–343.

Novoa, C. and Storer, R. (2009). An approximate dynamic programming approach for the vehicle routing problem with
stochastic demands. European Journal of Operational Research, 196(2):509–515.

Pillac, V., Gendreau, M., Gu´ret, C., and Medaglia, A. L. (2011a). A review of dynamic vehicle routing problems. Technical
e
report, CIRRELT. CIRRELT-2011-62.
Pillac, V., Gu´ret, C., and Medaglia, A. L. (2011b). An event-driven optimization framework for dynamic vehicle routing.
e
´
Technical report, Ecole des Mines de Nantes, France. Report 11/2/AUTO.

Pillac, V., Gu´ret, C., and Medaglia, A. L. (2011c). On the technician routing and scheduling problem. Optimization
e
Letters, Under review.


References

References II
Pisinger, D. and Ropke, S. (2007). A general heuristic for vehicle routing problems. Computers & Operations Research,
34(8):2403–2435.

Powell, W. B. and Topaloglu, H. (2005). Fleet management. In Wallace, S. and Ziemba, W., editors, Applications of
Stochastic Programming, volume 5 of MPS-SIAM series on Optimization, chapter 12, pages 185–215. SIAM.
Secomandi, N. (2001). A rollout policy for the vehicle routing problem with stochastic demands. Operations Research,
49(5):796–802.

Simao, H., Day, J., George, A., Giﬀord, T., Nienow, J., and Powell, W. B. (2009). An approximate dynamic programming
algorithm for large-scale ﬂeet management: A case application. Transportation Science, 43(2):178–197.

Solomon, M. M. (1987). Algorithms for the vehicle-routing and scheduling problems with time window constraints.
Operations Research, 35(2):254–265.

Van Hentenryck, P. and Bent, R. (2006). Online stochastic combinatorial optimization. MIT Press.


Research proposal

Empfohlen

Empfohlen

Weitere ähnliche Inhalte

Kürzlich hochgeladen

Kürzlich hochgeladen (20)

Empfohlen

Empfohlen (20)

Research proposal