A METAHEURISTICS BASED SIMULATION TOOL
TO OPTIMIZE DEMAND RESPONSIVE
TRANSPORTATION SYSTEMS
Eneko Osaba, Pablo Fernandez
Mobility Unit, Deusto Technology Center, Av. Universidades, 24, Bilbao, Spain
Roberto Carballedo, Asier Perallos
Department of Software Engineering, University of Deusto, Av. Universidades, 24, Bilbao, Spain
Keywords: Demand simulation tool, Intelligent transport, Demand responsive transport, Evolutionary computing.
Abstract: Public passenger transport is an important area that affects our quality of life. The design of routes and
frequencies of such systems is very important to ensure their economic viability. The work presented in this
paper focuses on the design of a software tool that assists in the creation of routes and schedules of
passenger transport systems. For this, we have developed an application that is based on evolutionary
computing techniques to simulate passenger demand and adjust the routes and frequency of the services to
meet those demands. The result of work done is a software tool, and a metaheuristic algorithm that can be
used for solving optimization problems.
1 INTRODUCTION
There are different kinds of public transportation
systems, each one with its own features but they all
share some disadvantages. One major disadvantage
of the conventional transportation systems is the lack
of resources to satisfy all the users’ demands. In
rural areas and places with low demographic
density, the existence of regular public
transportation systems is not profitable. There are
some areas where public transportation vehicles do
tours and stops without picking or delivering any
passenger. If this situation is constantly repeated the
route will eventually be eliminated. Whenever the
supply of public transport services in an area is
reduced, we could say that the quality of life of
people living in that region decreases. To avoid this
loss of quality of life, we should analyze the demand
of passengers, and to adapt transportation systems to
passenger demand. This is essentially the concept of
transport systems on demand.
Transportation-On-Demand (TOD) (Jorgensen,
Larsen and Bergvinsdottir, 2007) is concerned with
the transportation of passengers or goods between
specific origins and destinations at the request of
users. Most TOD problems are characterized by the
presence of three often conflicting objectives:
maximizing the number of requests served,
minimizing operating costs and minimizing user
inconvenience. As is common in many
combinatorial optimization problems, these
objectives are conflicting and it is needed to sort
them by importance.
Many of the techniques used to solve these
problems do not yield an exact solution. This is
because the types of problems to be solved are
classified as NP-hard (Garey and Johnson, 1990).
For this reason, heuristics techniques are used for
obtaining good approximations.
In this work an algorithmic solution to this
problem has been developed. This solution consists
in a hybrid algorithm combining two evolutionary
computing techniques: a Genetic Algorithms and a
Simulated Annealing algorithm.
Besides the algorithm, a simulation tool has been
developed. The tools is oriented to the bus public
transport, it allows the user to create virtual
environments with dynamic bus stations, stops and
clients requests. The tool is able to obtain the best
route through several bus stations that satisfies
384
Osaba E., Fernandez P., Carballedo R. and Perallos A..
A METAHEURISTICS BASED SIMULATION TOOL TO OPTIMIZE DEMAND RESPONSIVE TRANSPORTATION SYSTEMS.
DOI: 10.5220/0003499803840389
In Proceedings of the 6th International Conference on Software and Database Technologies (ICSOFT-2011), pages 384-389
ISBN: 978-989-8425-77-5
Copyright
c
2011 SCITEPRESS (Science and Technology Publications, Lda.)
passengers’ requests. The design of the route is
made by means of the hybrid algorithm mentioned.
The aim of the new tool is to support the
decision to create or delete a particular route. The
developed simulation tool has also been used to
validate the algorithmic solution proposed.
This paper is structures as follow: the first
section makes a brief state-of-the-art about the most
common transportation problems and the existing
traffic simulation systems. Then the simulation tool
is presented as well as the proposed algorithms. The
final solution is presented in the next section and the
development of the hybrid algorithm is explained.
Finally the conclusions and the future work is
presented.
2 CONTEXT
2.1 Transportation Problems
Most of the problems arisen in transportation on
demand topic have similar characteristics, which
means that they can be framed as instances of other
generic and well know problems. In this section, we
present the most common traditional problems in the
field of transportation on demand.
Travelling Salesman Problem (TSP)
(Applegate, Bixby, Chvátal and Cook, 2006): The
Travelling Salesman Problem (TSP) is an NP-hard
problem in combinatorial optimization studied in
operations research and theoretical computer
science. Given a list of cities and their pair-wise
distances, the task is to find a shortest possible tour
that visits each city exactly once. This type of
problem is used as a benchmark for many
optimization algorithms.
Vehicle Routing Problem (VRP) (Dantzig and
Ramser, 1959): The vehicle routing problem (VRP)
is a generalization of the TSP. The aim of the
problem is to service a number of customers with a
fleet of vehicles. Often the context of this type of
problem is related to deliver goods located at a
central depot to customers which have placed orders
for such goods. Implicit is the goal of minimizing
the cost of distributing the goods. Many variants of
the VRP are described in the literature (Pisinger and
Ropke, 2007). These problems include the addition
of variables and constraints. One of the most popular
variants includes time windows for deliveries. These
time windows represent the time within which the
deliveries (or visits) must be made (Repoussis,
Tarantilis and Ioannou, 2009).
Demand Responsive Transport (DRT):
Demand Responsive Transport or Demand-
Responsive Transit (DRT) or Demand Responsive
Service is an advanced, user-oriented form of public
transport. It is characterized by flexible routing and
scheduling of small/medium vehicles operating in
shared-ride mode between pick-up and drop-off
locations according to passengers needs. DRT
systems provide a public transport service in rural
areas or areas of low passenger demand, where a
regular bus service may not be economically viable.
DRT systems are characterized by the flexibility of
the planning of vehicle routes. These routes may
vary according to the passenger’ needs in real time.
This is the type of problem that we used to
benchmark the algorithm proposed in this paper.
2.2 Traffic Simulation Tools
The simulation tools help us to measure the
performance of a system in a virtual environment.
This is a usual practice before making a large scale
change in an existing platform because there are
involved a lot of costs associated with this process.
Another major feature of the simulation tool is the
ability of predicting the behaviour of one
environment in a specific context.
Talking about simulation systems in the
transportation context is talking about prediction and
traffic optimization. One of the main goals in these
simulations systems is predicting the effects and the
impact of a vehicle or road accident in the regular
traffic as stated in (Burghout, Koutsopoulos, and
Andreasson, 2010), this is a critical fact in places
where the traffic density is high because a vehicle
accident can cause several inconveniences to the
users of the road.
Another important goal of the transportation
related simulation systems is the traffic light
calibration. Thanks to the traffic simulators the
agents are able to setup the traffic lights according to
the simulated traffic and other natural factors like
the weather or the time of the day as written in
(Lopez-Mellado, 2010), in order to optimize the
light cycles.
3 SIMULATION TOOL
Nowadays there are many kinds of public
transportation systems: on demand or regular ones.
This requires the public transport to be more
demanding and sophisticated. But before performing
variations to the regular service, by adding more
A METAHEURISTICS BASED SIMULATION TOOL TO OPTIMIZE DEMAND RESPONSIVE TRANSPORTATION
SYSTEMS
385
transport units, new stops or creating new routes or
lines, it is necessary to make preliminary studies to
with the objective of improving the quality of
service and the client satisfaction without making
mistakes. To achieve this excellence level, different
techniques and tools are used and applied before
deploying the changes, in order to guarantee that the
people in charge will be able to have an idea of the
impact of the new variations.
In this work it has been developed a simulation tool,
oriented to bus public transport system (on demand
or regular), in rural or urban environments. With this
tool, different tests can be made to check the
performance improvement when adding or removing
a new stop to/from a route, or the creation of a new
line.
With this tool the user will be able to create
different environments composed of stations, that
user may place wherever he desires. The application
will calculate the optimized route to navigate
through all the placed stations using the artificial
intelligence algorithm developed that will be further
explained later. Once the route is created, the user
may make requests, which the system will manage
on an efficient way. The system will also be able to
make a simulation of how the bus would be
completing the route and managing the dynamic
requests made in real time.
It’s important point out the difference between
primary and secondary stations. The primary stations
are the ones the bus must pass through, the
secondary are the ones that will only be part of the
route if it exists enough demanding from the users.
This feature allows to create on demand transport
lines, and helps to select which stations will be
primary or secondary.
Figure 1, shows an image of main interface of
the tool developed. That image represents an
environment composed by some primary and
secondary stops.
Figure 1: Deployed Tabs.
Figure 2: Main window.
Figure 2, shows the application’s main screen.
Several deployable tabs can be seen. These tabs
contain diverse information. The top ones for
example, have information about the route of each of
the buses and the history of the active bus. The
bottom one contains a history of all the requests
made, a panel to make transportation requests and
some controls to manage the simulation of the bus
tour.
Apart from this, a mobile application has been
developed. Thanks to this, the users can kwon the
active bus route, in map or text format. Furthermore,
it also offers the possibility of making requests in the
same way as the web application.
To finish, in Figure 3 a conceptual schema of the
final system architecture is shown.
Figure 3: System architecture.
As mentioned before, there exist two ways to
use the simulation tool. One is based on a web
ICSOFT 2011 - 6th International Conference on Software and Data Technologies
386
application and the other one is an android based
mobile application. These two alternatives access to
a central server via internet, this server can be
divided into three different components. The first
one is a database where all the information about the
made requests, the active routes and the statistics are
stored. The second component is composed by some
web services that offer several access methods to the
database and business logic. The web services are
consumed by some Java Server Pages (JSP), the
third component of the server. These JSP pages
represent the presentation layer and user access to
both web application and mobile application.
4 PROPOSED ALGORITHMS
As mentioned in the preceding section, the
implemented simulation tool uses an artificial
intelligence algorithm as base. This algorithm is
used to resolve the route planning problem, in a
static way in regular transport lines, as well as in a
dynamic way, in on demand transport lines. In
application level, the algorithm will be in charge of
finding in every moment the optimized route the bus
has to go through to visit all the stations of the
environment. To implement the solution, the
problem has been treated as a DRT one, as
mentioned in the introduction of the article, and
resolved with a heuristic method that obtains good
approximations. To perform this task, we designed a
hybrid algorithm that combines simulated annealing
methods and genetic algorithms. Then we explain
the details of each technique separately.
Simulated Annealing (Rutenbar, 1989): This is
one of the most popular local search techniques. It is
based on the physical principle of cooling metal.
Using that analogy, it generates an initial solution
and the process proceeds by selecting new solutions
randomly. The new solutions are not always better
than the initial solution, but as time passes and the
temperature decreases (the metal becomes stronger),
each new solution must be better than previous
solutions.
Genetic Algorithm (Zhang, Yao and Zheng,
2009): This algorithm is inspired by the laws of
natural selection and the evolution of the animal
species. An initial population of solutions is defined.
This initial population consists of a number of
individuals (solutions of the problem.) Then, with
the combination and evolution of these individuals,
the algorithm tries to get a better solution.
Hybrid Algorithm (Kaur and Murugappan,
2008): The hybrid algorithm is the resultant from
joining the genetic and the simulated annealing
algorithms. It was decided to use this algorithm after
a test period detailed in the following section.
5 TEST AND FINAL SOLUTION
In the field of artificial intelligence, when you create
a new algorithm, or modify an existing one, results
that show that the new solution is better than other
known solutions have to be submitted. To do this,
there are sets of problems used to compare results in
computational resources and parameters related to
the quality of the solution. In our case, there is not a
set of sample problems for comparing the
performance of design algorithms for demand
responsive transport routing problems. For this
reason, we have defined our own set of testing, and
we are going to make comparisons between our
algorithm, a brute-force optimal algorithms and each
of the techniques used in our hybrid algorithm.
As indicated above, for the design of our hybrid
algorithm, separate versions of simulated annealing
and a genetic algorithm have been implemented. In
addition, we have implemented a "brute force"
algorithm, to find out the optimal solution for small
instances of the problem (with few intermediate
stops).
With these 3 algorithms, there have been a series
of tests to measure the performance of each
algorithm and the ability of each one to solve the
problem. As a result of these tests, we have obtained
several conclusions:
1. The “brute force” algorithm is optimal because
it always finds the best solution. Even so, it has
the disadvantage that the execution time is
unacceptable when the number of stations
increases to more than 9 (for a large number of
stations cannot even get a solution). This
algorithm cannot be used in a real scenario.
2. The simulated annealing algorithm only finds
the optimal solution when the first and last
station does not vary during the resolution
process. Running time is always the same
regardless of the number of stops.
3. In the case of genetic algorithm, the execution
time is constant if the number of generations is
also constant. An advantage of this algorithm is
that the probability of finding a good solution
is independent of the number of stops.
After preliminary analysis of algorithms
separately, we came to the conclusion that the results
of runtime and solution quality were not good. For
this reason we decided to combine the two
A METAHEURISTICS BASED SIMULATION TOOL TO OPTIMIZE DEMAND RESPONSIVE TRANSPORTATION
SYSTEMS
387
heuristics.
5.1 Our Hybrid Algorithm
Our hybrid algorithm came up with the aim to
combine the advantages of genetic algorithms and
simulated annealing:
Rapid and constant execution time (simulated
annealing).
Probability of finding a good solution for the
problem instances with many stops (genetic
algorithm).
The solution would avoid the main drawback of the
two algorithms:
The solution should be optimal or very close
to it.
With all these goals, it thought about making the
hybrid. By nature of the two algorithms, it is
appropriate to insert the execution of simulated
annealing algorithm in the execution of genetic
algorithm. That is because the first algorithm is
focused on only one solution and the second works
simultaneously with different solutions.
Having decided the model of integration, there
were two options to do the integration:
Integrate the simulated annealing in the
process of creating the initial population.
Integrate the simulated annealing in the
process of reproduction, right after generating
the new population.
After several tests, we concluded that the most
effective solution was to apply the simulated
annealing algorithm just after the reproduction
process. Below is a table showing the results of the
tests. The table shows the number of stops, the
number of generations used in the genetic algorithm,
runtime, and the percentage of times the algorithm
finds the optimal solution.
Table 1: Results of the tests.
Stations Generations T. of execution
% of optimal
solution
9 5 3 seconds 80%
9 10 5 seconds 100%
10 5 3 seconds 80%
10 10 5 seconds 100%
11 5 3 seconds 80%
11 10 5 seconds 100%
Comparing the proposed alternative with each of
the separate algorithms, we can ensure that the
execution time is right, regardless of the number of
stops. Moreover, in situations where the optimal
solution is not found, the average deviation for the
optimal solution does not exceed 3% of the value of
the optimal solution.
6 CONCLUSIONS AND FURTHER
WORK
The work presented is the result of a research project
funded by the Basque government. The project
focuses on the design of a software tool that assists
in the creation of routes and schedules of passenger
transport systems. For this, we have developed an
application that is based on evolutionary computing
techniques to simulate passenger demand and adjust
the routes and frequency of the services to meet
those demands. The result of work done is a
software tool, and a metaheuristic algorithm that can
be used for solving optimization problems.
A part of the results of this project, and as there
is no benchmark problems for this type of
transportation problem we are currently developing
a framework for algorithm validation. This new
framework will serve to validate the performance of
demand responsive routing algorithms. During the
design of this benchmark problems we will define
the basic criteria to measure algorithms
performance. Examples of such criteria are running
time, solution quality, ease of implementation,
robustness and flexibility.
REFERENCES
R M Jorgensen, J Larsen and K B Bergvinsdottir: Solving
the Dial-a-Ride problem using genetic algorithms.
Journal of the Operational Research Society, 58, pp.
1321–1331. 2007
Garey, M. R. and Johnson, D. S: Computers and
Intractability; a Guide to the Theory of Np-
Completeness. W. H. Freeman & Co. 1990.
Applegate, D. L.; Bixby, R. M.; Chvátal, V.; Cook, W. J.
The Traveling Salesman Problem. ISBN 0691129932.
2006
Dantzig, G. B.; Ramser, J. H. The Truck Dispatching
Problem. Management Science Vol. 6, No. 1, October
1959, pp. 80-91.
D Pisinger, S Ropke: A general heuristic for vehicle
routing problems. Computers and Operations
Research, Volume 34 Issue 8, August, 2007.
Repoussis, P. P.; Tarantilis, C. D.; Ioannou, G.: Arc-
guided evolutionary algorithm for the vehicle routing
problem with time windows. IEEE Transactions on
Evolutionary Computation, Volume 13 Issue 3, June
2009.
Burghout, Wilco; Koutsopoulos, Haris N; Andreasson,
Ingmar: Incident Management and Traffic Information
Tools and Methods for Simulation-Based Traffic
ICSOFT 2011 - 6th International Conference on Software and Data Technologies
388
Prediction. Transportation Research Record, Issue
2161, pp. 20-28, 2010
Lopez-Mellado Ernesto: A modeling framework for urban
traffic systems microscopic simulation. Simulation
Modelling Practice and Theory, Volume 18, Issue 8
pp. 1145-1161, Sep 2010
Rutenbar, R. A.: Simulated Annealing algorithms: an
overview. IEEE Circuits and Devices Magazine 5 (l),
pp. 19-26. 1989
L Zhang, M Yao, N Zheng: Optimization and
improvement of Genetic Algorithms solving Traveling
Salesman Problem. Image Analysis and Signal
Processing, pp. 327-332 April 2009.
Kaur, D. Murugappan, M. M: Performance enhancement
in solving Traveling Salesman Problem using hybrid
genetic algorithm. Fuzzy Information Processing
Society, pp. 1-6, May 2008.
A METAHEURISTICS BASED SIMULATION TOOL TO OPTIMIZE DEMAND RESPONSIVE TRANSPORTATION
SYSTEMS
389