SlideShare a Scribd company logo
1 of 41
Download to read offline
Algorithms for the Urban Transit Routing Problem
                                      Exact and Metaheuristic


                                          Bruno Coswig Fiss

                        1 Institut   fur Technische Informatik und Mikroelektronik
                                      ¨
                                                              ¨
                                         Technische Universitat Berlin


                                        VSP Internal Seminar




Bruno Coswig Fiss (TU Berlin)                Algorithms for the UTRP                 June 13, 2012   1 / 35
Outline

1     Introduction
         Motivation to the UTRP
         Existing Solutions
         Problem Statement

2     Our Algorithms
        Exact
        Genetic

3     Current State
        Results
        Work in Progress



    Bruno Coswig Fiss (TU Berlin)   Algorithms for the UTRP   June 13, 2012   2 / 35
Introduction


Short Intro to Myself


                                 My university in Brazil:




 Bruno Coswig Fiss (TU Berlin)       Algorithms for the UTRP   June 13, 2012   3 / 35
Introduction   Motivation to the UTRP


Outline

1     Introduction
         Motivation to the UTRP
         Existing Solutions
         Problem Statement

2     Our Algorithms
        Exact
        Genetic

3     Current State
        Results
        Work in Progress



    Bruno Coswig Fiss (TU Berlin)    Algorithms for the UTRP                June 13, 2012   4 / 35
Introduction   Motivation to the UTRP


Public Transportation




                                 Bus in Porto Alegre.


     Crowded, late.
     Low resources? Are they being well employed?
 Bruno Coswig Fiss (TU Berlin)    Algorithms for the UTRP                June 13, 2012   5 / 35
Introduction   Motivation to the UTRP


Public Transportation




                     Route network in Porto Alegre with about 230 routes.
 Bruno Coswig Fiss (TU Berlin)        Algorithms for the UTRP                June 13, 2012   6 / 35
Introduction   Motivation to the UTRP


Automatization and Computer Assistance




     Complexity of network design is enormous.
     Human planners take decisions. Is that enough?
     ”I think there is a world market for maybe five computers.” –
     allegedly Thomas Watson, chairman of IBM, 1943
     Computers can help in the process of planning.




 Bruno Coswig Fiss (TU Berlin)    Algorithms for the UTRP                June 13, 2012   7 / 35
Introduction   Motivation to the UTRP


UTNDP: UTRP and UTSP.



     This problem has been studied, and is know as the Urban Transit
     Network Design Problem (UTNDP).
     Commonly divided: Urban Transit Routing Problem and Urban
     Transit Scheduling Problem.
     Scheduling depends on previous step.
     New schedules are easier to test.
     Focus here: UTRP




 Bruno Coswig Fiss (TU Berlin)    Algorithms for the UTRP                June 13, 2012   8 / 35
Introduction   Existing Solutions


Outline

1     Introduction
         Motivation to the UTRP
         Existing Solutions
         Problem Statement

2     Our Algorithms
        Exact
        Genetic

3     Current State
        Results
        Work in Progress



    Bruno Coswig Fiss (TU Berlin)    Algorithms for the UTRP            June 13, 2012   9 / 35
Introduction   Existing Solutions


Existing Solutions




The list of existing solutions is long, including:
     Multiple step solutions.
     Metaheuristics.
     Mixed non-linear mathematical models.
     Ad-hoc solutions.




 Bruno Coswig Fiss (TU Berlin)    Algorithms for the UTRP            June 13, 2012   10 / 35
Introduction   Existing Solutions


Tool Example




                                                       ´
                                 Computational tool by Alvarez et al.



 Bruno Coswig Fiss (TU Berlin)            Algorithms for the UTRP            June 13, 2012   11 / 35
Introduction   Existing Solutions


Room for Improvement


Current issues with the existing solutions:
     Many different problem definitions.
     The quality of these solutions depends fundamentally on the
     chosen algorithms.
     Large search space (there is no free lunch).
     Comparison is necessary!
     Our Goal: develop and test appropriate algorithms and methods
     for the UTRP using a well-known problem definition (and with
     common benchmarks).




 Bruno Coswig Fiss (TU Berlin)    Algorithms for the UTRP            June 13, 2012   12 / 35
Introduction   Problem Statement


Outline

1     Introduction
         Motivation to the UTRP
         Existing Solutions
         Problem Statement

2     Our Algorithms
        Exact
        Genetic

3     Current State
        Results
        Work in Progress



    Bruno Coswig Fiss (TU Berlin)    Algorithms for the UTRP           June 13, 2012   13 / 35
Introduction   Problem Statement


Input and Route Sets
     Two inputs: graph and demand matrix.




                        Transport, route and transit networks, respectively [1].

 Bruno Coswig Fiss (TU Berlin)          Algorithms for the UTRP             June 13, 2012   14 / 35
Introduction   Problem Statement


Associated costs



     Operator cost: sum of weight of edges used.
     Passenger cost: total travel time.
     Multi-objective.
     Conditions that can be considered:
             Number of routes.
             Lenght of routes.
             Cycles and backtracks.
             Penalty for making transfers.




 Bruno Coswig Fiss (TU Berlin)    Algorithms for the UTRP           June 13, 2012   15 / 35
Introduction     Problem Statement


Output

                                                             Approximation for Pareto-optimal curves


                                              240                              GA Solutions with up to 8 routes
                                                                               GA Solutions with up to 6 routes
                                              220                              GA Solutions with up to 4 routes
                                                                         Fitting curve (58.22/(x-9.86) + 44.36)
          Total route set length in minutes




                                              200

                                              180

                                              160

                                              140

                                              120

                                              100

                                               80

                                               60
                                                 10   10.5     11          11.5        12        12.5       13        13.5
                                                                 Average travel time in minutes
 Bruno Coswig Fiss (TU Berlin)                                      Algorithms for the UTRP                       June 13, 2012   16 / 35
Our Algorithms   Exact


Outline

1     Introduction
         Motivation to the UTRP
         Existing Solutions
         Problem Statement

2     Our Algorithms
        Exact
        Genetic

3     Current State
        Results
        Work in Progress



    Bruno Coswig Fiss (TU Berlin)       Algorithms for the UTRP   June 13, 2012   17 / 35
Our Algorithms   Exact


Exact Solution Summary


     ”The problem of designing a good or efficient route set (or route
     network) for a transit system is a difficult optimization problem
     which does not lend itself readily to mathematical programming
     formulations and solutions using traditional techniques” – Dr.
     Partha Chakroborty, Transportation Engineer
     Mathematical solution has been created to test feasibility and
     correctness.
     Uses a Mixed Integer Programming formulation.
     Achieved global optimal solutions for Mandl’s Swiss road network
     (to be shown) with 2 and 3 routes.
     Very slow, but useful linear relaxation and for divide and conquer.



 Bruno Coswig Fiss (TU Berlin)       Algorithms for the UTRP   June 13, 2012   18 / 35
Our Algorithms   Genetic


Outline

1     Introduction
         Motivation to the UTRP
         Existing Solutions
         Problem Statement

2     Our Algorithms
        Exact
        Genetic

3     Current State
        Results
        Work in Progress



    Bruno Coswig Fiss (TU Berlin)       Algorithms for the UTRP   June 13, 2012   19 / 35
Our Algorithms   Genetic


Genetic Algorithm Overview




     Maintain a population of potential solutions, ie. route sets.
     Create or modify routes in a route set and, if dominating another
     route set, take its place.




 Bruno Coswig Fiss (TU Berlin)       Algorithms for the UTRP   June 13, 2012   20 / 35
Our Algorithms   Genetic


Creating New Routes




     Take it from a pool of base routes.
     Apply operators to existing solutions (route sets).




 Bruno Coswig Fiss (TU Berlin)       Algorithms for the UTRP   June 13, 2012   21 / 35
Our Algorithms   Genetic


Base routes




Base routes are intrinsic to a graph:
     Shortest path for every pair of nodes (in original network).
     Minimum Spanning Tree (!).
     Paths with highest covered demand.
     Routes with high percentages in the linear relaxation of the MIP
     solution.




 Bruno Coswig Fiss (TU Berlin)       Algorithms for the UTRP   June 13, 2012   22 / 35
Our Algorithms       Genetic


Minimum Spanning Tree Demo


                     a
                                 7

                                                                    8
                         5                        b                          c
                                                                7
                                 9                                       5
                                      15
                    d                                               e
                                 6                                       9
                                                                8
                                                                    11
                                                  f                          g



 Bruno Coswig Fiss (TU Berlin)       Algorithms for the UTRP                     June 13, 2012   23 / 35
Our Algorithms       Genetic


Minimum Spanning Tree Demo


                     a
                                 7

                                                                    8
                         5                        b                          c
                                                                7
                                 9                                       5
                                      15
                    d                                               e
                                 6                                       9
                                                                8
                                                                    11
                                                  f                          g



 Bruno Coswig Fiss (TU Berlin)       Algorithms for the UTRP                     June 13, 2012   23 / 35
Our Algorithms       Genetic


Minimum Spanning Tree Demo


                     a
                                 7

                                                                    8
                         5                        b                          c
                                                                7
                                 9                                       5
                                      15
                    d                                               e
                                 6                                       9
                                                                8
                                                                    11
                                                  f                          g



 Bruno Coswig Fiss (TU Berlin)       Algorithms for the UTRP                     June 13, 2012   23 / 35
Our Algorithms       Genetic


Minimum Spanning Tree Demo


                     a
                                 7

                                                                    8
                         5                        b                          c
                                                                7
                                 9                                       5
                                      15
                    d                                               e
                                 6                                       9
                                                                8
                                                                    11
                                                  f                          g



 Bruno Coswig Fiss (TU Berlin)       Algorithms for the UTRP                     June 13, 2012   23 / 35
Our Algorithms       Genetic


Minimum Spanning Tree Demo


                     a
                                 7

                                                                    8
                         5                        b                          c
                                                                7
                                 9                                       5
                                      15
                    d                                               e
                                 6                                       9
                                                                8
                                                                    11
                                                  f                          g



 Bruno Coswig Fiss (TU Berlin)       Algorithms for the UTRP                     June 13, 2012   23 / 35
Our Algorithms       Genetic


Minimum Spanning Tree Demo


                     a
                                 7

                                                                    8
                         5                        b                          c
                                                                7
                                 9                                       5
                                      15
                    d                                               e
                                 6                                       9
                                                                8
                                                                    11
                                                  f                          g



 Bruno Coswig Fiss (TU Berlin)       Algorithms for the UTRP                     June 13, 2012   23 / 35
Our Algorithms       Genetic


Minimum Spanning Tree Demo


                     a
                                 7

                                                                    8
                         5                        b                          c
                                                                7
                                 9                                       5
                                      15
                    d                                               e
                                 6                                       9
                                                                8
                                                                    11
                                                  f                          g



 Bruno Coswig Fiss (TU Berlin)       Algorithms for the UTRP                     June 13, 2012   23 / 35
Our Algorithms   Genetic


Operators



     Mutate route
             Add a node to or remove a node from the extremity of a route.
     Simplify route set
             If the route set contains 9-3-4-5-6 and 4-5-6-12-2, we replace with
             9-3-4-5-6-12-2
     Cross-over two routes
             Join two routes at a certain intersection (cut cycles if necessary).




 Bruno Coswig Fiss (TU Berlin)       Algorithms for the UTRP        June 13, 2012   24 / 35
Current State   Results


Outline

1     Introduction
         Motivation to the UTRP
         Existing Solutions
         Problem Statement

2     Our Algorithms
        Exact
        Genetic

3     Current State
        Results
        Work in Progress



    Bruno Coswig Fiss (TU Berlin)      Algorithms for the UTRP   June 13, 2012   25 / 35
Current State   Results


Tested Networks
Two networks were used: Mandl’s and artificial British(based) city with
110 nodes and 275 links.




                                 Mandl’s Swiss road network [1].


 Bruno Coswig Fiss (TU Berlin)           Algorithms for the UTRP   June 13, 2012   26 / 35
Current State   Results


Solution Quality Quantities



All results are evaluated using the following quantities, as in previous
works:
     di is the percentage of the demand satisfied with i transfers.
     ATT is the average travel time (in minutes per passenger),
     including transfer penalties.
     CO is the cost for the operator, i.e., the total route length (in
     minutes, considering constant transport speed).
     ATTwop = ATT −              i≤TMAX tpen di i.




 Bruno Coswig Fiss (TU Berlin)         Algorithms for the UTRP   June 13, 2012   27 / 35
Current State   Results


Mandl’s Network Exact Solutions


           Best possible route sets found using the Mixed Integer formulation

                Number of routes                    2                   3
                       d0                   84.90 %              93.67 %
                       d1                   14.00 %                5.43 %
                       d2                     1.10 %               0.90 %
                     ATT                  11.33 min.           10.50 min.
                      CO                     98 min.             150 min.
               Processing time (s)              1065                78992
                  Two Routes              6-14-7-5-2-1-4-3-11-10-9-13-12
                                                 0-1-3-5-7-9-6-14-8
                    Three Routes            4-3-11-10-12-13-9-7-5-2-1-0
                                               4-3-1-2-5-14-6-9-10-11
                                                0-1-4-3-5-7-9-6-14-8



 Bruno Coswig Fiss (TU Berlin)        Algorithms for the UTRP         June 13, 2012   28 / 35
Current State    Results


Genetic Algorithm on Mandl’s Network

   Comparison between best UTRP multi-objective solutions on Mandl’s Network
                           Scenario           Qp     Best previous results    Our metaheuristic
                                                                      ([1])    approach results
                      Best for Passenger      d0                 94.54 %              98.84 %
                                              d1                  5.46 %               1.16 %
                                              d2                  0.00 %               0.00 %
                                             ATT               10.36 min.           10.10 min.
                                             CO                  283 min.             259 min.
                     Compromise Solution      d0                 93.19 %              93.61 %
                        (CO ≤ 148)            d1                  6.23 %               6.20 %
                                              d2                  0.58 %               0.19 %
                                             ATT               10.46 min.           10.43 min.
                                             CO                  148 min.             147 min.
                     Compromise Solution      d0                 90.88 %              91.23 %
                        (CO ≤ 126)            d1                  8.35 %               7.84 %
                                              d2                  0.77 %               0.93 %
                                             ATT               10.65 min.           10.59 min.
                                             CO                  126 min.             126 min.
                       Best for Operator      d0                 66.09 %              77.78 %
                                              d1                 30.38 %              21.32 %
                                              d2                  3.53 %               0.90 %
                                             ATT               13.34 min.           12.97 min.
                                             CO                   63 min.              63 min.




 Bruno Coswig Fiss (TU Berlin)                Algorithms for the UTRP                             June 13, 2012   29 / 35
Current State    Results


Genetic Algorithm on Artificial British Network


  Comparison between best UTRP multi-objective solutions on artificial British city
                         Scenario      Qp       Best previous results   Our metaheuristic
                                                        ([1])           approach results
                        I-Passenger     d0            72.91 %               55.80 %
                                       ATT           36.28 min.            36.35 min.
                                      ATTwop         34.60 min.            34.12 min.
                                        CO           2986 min.             8406 min.
                       II-Passenger     d0            71.21 %               46.25 %
                                       ATT           37.52 min.            36.61 min.
                                      ATTwop         35.68 min.            33.77 min.
                                        CO           2378 min.             5181 min.
                         I-Operator     d0            48.62 %                9.48 %
                                       ATT           40.88 min.            55.08 min.
                                      ATTwop         37.36 min.            45.66 min.
                                        CO           1077 min.              319 min.
                        II-Operator     d0            46.97 %                8.47 %
                                       ATT           41.26 min.            55.48 min.
                                      ATTwop        37.655 min.            47.90 min.
                                        CO           1265 min.              319 min.




 Bruno Coswig Fiss (TU Berlin)              Algorithms for the UTRP                         June 13, 2012   30 / 35
Current State   Work in Progress


Outline

1     Introduction
         Motivation to the UTRP
         Existing Solutions
         Problem Statement

2     Our Algorithms
        Exact
        Genetic

3     Current State
        Results
        Work in Progress



    Bruno Coswig Fiss (TU Berlin)      Algorithms for the UTRP         June 13, 2012   31 / 35
Current State   Work in Progress


Performance




           Time used in each function. Dijkstra takes 90% of processing time.


To explore the search space faster: use GPU.
 Bruno Coswig Fiss (TU Berlin)      Algorithms for the UTRP           June 13, 2012   32 / 35
Current State   Work in Progress


Simulations




Two scenarios are being simulated:
     Porto Alegre: test effectiveness in comparison to existing network.
     Demands are artificial.
     Berlin: use MATSim as the objective function.




 Bruno Coswig Fiss (TU Berlin)      Algorithms for the UTRP         June 13, 2012   33 / 35
Thank you!


Conclusion




     New methods and algorithms for the UTRP can make public
     transport better!
     Suggestions or questions?
     Thank you!




 Bruno Coswig Fiss (TU Berlin)    Algorithms for the UTRP   June 13, 2012   34 / 35
Thank you!


References




 Lang Fan, Christine L. Mumford, and Dafydd Evans.
 A simple multi-objective optimization algorithm for the urban transit
 routing problem.
 In Proceedings of the Eleventh conference on Congress on
 Evolutionary Computation, CEC’09, pages 1–7, Piscataway, NJ, USA,
 2009. IEEE Press.




 Bruno Coswig Fiss (TU Berlin)    Algorithms for the UTRP   June 13, 2012   35 / 35

More Related Content

Similar to Seminar for verkehr

SIGEVOlution Summer 2007
SIGEVOlution Summer 2007SIGEVOlution Summer 2007
SIGEVOlution Summer 2007
Pier Luca Lanzi
 

Similar to Seminar for verkehr (20)

P-Systems for approximating NP-Complete optimization problems
P-Systems for approximating NP-Complete optimization problemsP-Systems for approximating NP-Complete optimization problems
P-Systems for approximating NP-Complete optimization problems
 
A STUDY AND IMPLEMENTATION OF THE TRANSIT ROUTE NETWORK DESIGN PROBLEM FOR A ...
A STUDY AND IMPLEMENTATION OF THE TRANSIT ROUTE NETWORK DESIGN PROBLEM FOR A ...A STUDY AND IMPLEMENTATION OF THE TRANSIT ROUTE NETWORK DESIGN PROBLEM FOR A ...
A STUDY AND IMPLEMENTATION OF THE TRANSIT ROUTE NETWORK DESIGN PROBLEM FOR A ...
 
A study and implementation of the transit route network design problem for a ...
A study and implementation of the transit route network design problem for a ...A study and implementation of the transit route network design problem for a ...
A study and implementation of the transit route network design problem for a ...
 
Natural Computing for Vehicular Networks
Natural Computing for Vehicular NetworksNatural Computing for Vehicular Networks
Natural Computing for Vehicular Networks
 
Link-wise Artificial Compressibility Method: a simple way to deal with comple...
Link-wise Artificial Compressibility Method: a simple way to deal with comple...Link-wise Artificial Compressibility Method: a simple way to deal with comple...
Link-wise Artificial Compressibility Method: a simple way to deal with comple...
 
Eeee2017 Conference - OR in the digital era - ICT challenges | Presentation
Eeee2017 Conference - OR in the digital era - ICT challenges | PresentationEeee2017 Conference - OR in the digital era - ICT challenges | Presentation
Eeee2017 Conference - OR in the digital era - ICT challenges | Presentation
 
Gridforum Juergen Knobloch Grids For Science 20080402
Gridforum Juergen Knobloch Grids For Science 20080402Gridforum Juergen Knobloch Grids For Science 20080402
Gridforum Juergen Knobloch Grids For Science 20080402
 
Do we measure functional size or do we count thomas fehlmann
Do we measure functional size or do we count   thomas fehlmannDo we measure functional size or do we count   thomas fehlmann
Do we measure functional size or do we count thomas fehlmann
 
“Towards Multi-Step Expert Advice for Cognitive Computing” - Dr. Achim Rettin...
“Towards Multi-Step Expert Advice for Cognitive Computing” - Dr. Achim Rettin...“Towards Multi-Step Expert Advice for Cognitive Computing” - Dr. Achim Rettin...
“Towards Multi-Step Expert Advice for Cognitive Computing” - Dr. Achim Rettin...
 
C013141723
C013141723C013141723
C013141723
 
Fast Unbalanced Optimal Transport on a Tree
Fast Unbalanced Optimal Transport on a TreeFast Unbalanced Optimal Transport on a Tree
Fast Unbalanced Optimal Transport on a Tree
 
A binary particle swarm optimization approach for power system security enha...
A binary particle swarm optimization approach for power  system security enha...A binary particle swarm optimization approach for power  system security enha...
A binary particle swarm optimization approach for power system security enha...
 
SIGEVOlution Summer 2007
SIGEVOlution Summer 2007SIGEVOlution Summer 2007
SIGEVOlution Summer 2007
 
Wavelet-based Reflection Symmetry Detection via Textural and Color Histograms
Wavelet-based Reflection Symmetry Detection via Textural and Color HistogramsWavelet-based Reflection Symmetry Detection via Textural and Color Histograms
Wavelet-based Reflection Symmetry Detection via Textural and Color Histograms
 
351 b p.3
351 b p.3351 b p.3
351 b p.3
 
351 b p.3
351 b p.3351 b p.3
351 b p.3
 
351 b p.3
351 b p.3351 b p.3
351 b p.3
 
5.3.2nd WS. News from the COGAIN Association & DTU G.Interaction
5.3.2nd WS. News from the COGAIN Association & DTU G.Interaction5.3.2nd WS. News from the COGAIN Association & DTU G.Interaction
5.3.2nd WS. News from the COGAIN Association & DTU G.Interaction
 
PhD Thesis Defense Presentation: Robust Low-rank and Sparse Decomposition for...
PhD Thesis Defense Presentation: Robust Low-rank and Sparse Decomposition for...PhD Thesis Defense Presentation: Robust Low-rank and Sparse Decomposition for...
PhD Thesis Defense Presentation: Robust Low-rank and Sparse Decomposition for...
 
Optimization of Corridor Observation Method to Solve Environmental and Econom...
Optimization of Corridor Observation Method to Solve Environmental and Econom...Optimization of Corridor Observation Method to Solve Environmental and Econom...
Optimization of Corridor Observation Method to Solve Environmental and Econom...
 

Recently uploaded

Why Teams call analytics are critical to your entire business
Why Teams call analytics are critical to your entire businessWhy Teams call analytics are critical to your entire business
Why Teams call analytics are critical to your entire business
panagenda
 
Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers:  A Deep Dive into Serverless Spatial Data and FMECloud Frontiers:  A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
Safe Software
 

Recently uploaded (20)

Apidays New York 2024 - Passkeys: Developing APIs to enable passwordless auth...
Apidays New York 2024 - Passkeys: Developing APIs to enable passwordless auth...Apidays New York 2024 - Passkeys: Developing APIs to enable passwordless auth...
Apidays New York 2024 - Passkeys: Developing APIs to enable passwordless auth...
 
Why Teams call analytics are critical to your entire business
Why Teams call analytics are critical to your entire businessWhy Teams call analytics are critical to your entire business
Why Teams call analytics are critical to your entire business
 
Elevate Developer Efficiency & build GenAI Application with Amazon Q​
Elevate Developer Efficiency & build GenAI Application with Amazon Q​Elevate Developer Efficiency & build GenAI Application with Amazon Q​
Elevate Developer Efficiency & build GenAI Application with Amazon Q​
 
Connector Corner: Accelerate revenue generation using UiPath API-centric busi...
Connector Corner: Accelerate revenue generation using UiPath API-centric busi...Connector Corner: Accelerate revenue generation using UiPath API-centric busi...
Connector Corner: Accelerate revenue generation using UiPath API-centric busi...
 
Strategize a Smooth Tenant-to-tenant Migration and Copilot Takeoff
Strategize a Smooth Tenant-to-tenant Migration and Copilot TakeoffStrategize a Smooth Tenant-to-tenant Migration and Copilot Takeoff
Strategize a Smooth Tenant-to-tenant Migration and Copilot Takeoff
 
Emergent Methods: Multi-lingual narrative tracking in the news - real-time ex...
Emergent Methods: Multi-lingual narrative tracking in the news - real-time ex...Emergent Methods: Multi-lingual narrative tracking in the news - real-time ex...
Emergent Methods: Multi-lingual narrative tracking in the news - real-time ex...
 
ProductAnonymous-April2024-WinProductDiscovery-MelissaKlemke
ProductAnonymous-April2024-WinProductDiscovery-MelissaKlemkeProductAnonymous-April2024-WinProductDiscovery-MelissaKlemke
ProductAnonymous-April2024-WinProductDiscovery-MelissaKlemke
 
Exploring Multimodal Embeddings with Milvus
Exploring Multimodal Embeddings with MilvusExploring Multimodal Embeddings with Milvus
Exploring Multimodal Embeddings with Milvus
 
[BuildWithAI] Introduction to Gemini.pdf
[BuildWithAI] Introduction to Gemini.pdf[BuildWithAI] Introduction to Gemini.pdf
[BuildWithAI] Introduction to Gemini.pdf
 
Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers:  A Deep Dive into Serverless Spatial Data and FMECloud Frontiers:  A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
 
Rising Above_ Dubai Floods and the Fortitude of Dubai International Airport.pdf
Rising Above_ Dubai Floods and the Fortitude of Dubai International Airport.pdfRising Above_ Dubai Floods and the Fortitude of Dubai International Airport.pdf
Rising Above_ Dubai Floods and the Fortitude of Dubai International Airport.pdf
 
How to Troubleshoot Apps for the Modern Connected Worker
How to Troubleshoot Apps for the Modern Connected WorkerHow to Troubleshoot Apps for the Modern Connected Worker
How to Troubleshoot Apps for the Modern Connected Worker
 
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, AdobeApidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
 
DEV meet-up UiPath Document Understanding May 7 2024 Amsterdam
DEV meet-up UiPath Document Understanding May 7 2024 AmsterdamDEV meet-up UiPath Document Understanding May 7 2024 Amsterdam
DEV meet-up UiPath Document Understanding May 7 2024 Amsterdam
 
Vector Search -An Introduction in Oracle Database 23ai.pptx
Vector Search -An Introduction in Oracle Database 23ai.pptxVector Search -An Introduction in Oracle Database 23ai.pptx
Vector Search -An Introduction in Oracle Database 23ai.pptx
 
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWER
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWEREMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWER
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWER
 
"I see eyes in my soup": How Delivery Hero implemented the safety system for ...
"I see eyes in my soup": How Delivery Hero implemented the safety system for ..."I see eyes in my soup": How Delivery Hero implemented the safety system for ...
"I see eyes in my soup": How Delivery Hero implemented the safety system for ...
 
DBX First Quarter 2024 Investor Presentation
DBX First Quarter 2024 Investor PresentationDBX First Quarter 2024 Investor Presentation
DBX First Quarter 2024 Investor Presentation
 
Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...
Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...
Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...
 
WSO2's API Vision: Unifying Control, Empowering Developers
WSO2's API Vision: Unifying Control, Empowering DevelopersWSO2's API Vision: Unifying Control, Empowering Developers
WSO2's API Vision: Unifying Control, Empowering Developers
 

Seminar for verkehr

  • 1. Algorithms for the Urban Transit Routing Problem Exact and Metaheuristic Bruno Coswig Fiss 1 Institut fur Technische Informatik und Mikroelektronik ¨ ¨ Technische Universitat Berlin VSP Internal Seminar Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 1 / 35
  • 2. Outline 1 Introduction Motivation to the UTRP Existing Solutions Problem Statement 2 Our Algorithms Exact Genetic 3 Current State Results Work in Progress Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 2 / 35
  • 3. Introduction Short Intro to Myself My university in Brazil: Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 3 / 35
  • 4. Introduction Motivation to the UTRP Outline 1 Introduction Motivation to the UTRP Existing Solutions Problem Statement 2 Our Algorithms Exact Genetic 3 Current State Results Work in Progress Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 4 / 35
  • 5. Introduction Motivation to the UTRP Public Transportation Bus in Porto Alegre. Crowded, late. Low resources? Are they being well employed? Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 5 / 35
  • 6. Introduction Motivation to the UTRP Public Transportation Route network in Porto Alegre with about 230 routes. Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 6 / 35
  • 7. Introduction Motivation to the UTRP Automatization and Computer Assistance Complexity of network design is enormous. Human planners take decisions. Is that enough? ”I think there is a world market for maybe five computers.” – allegedly Thomas Watson, chairman of IBM, 1943 Computers can help in the process of planning. Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 7 / 35
  • 8. Introduction Motivation to the UTRP UTNDP: UTRP and UTSP. This problem has been studied, and is know as the Urban Transit Network Design Problem (UTNDP). Commonly divided: Urban Transit Routing Problem and Urban Transit Scheduling Problem. Scheduling depends on previous step. New schedules are easier to test. Focus here: UTRP Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 8 / 35
  • 9. Introduction Existing Solutions Outline 1 Introduction Motivation to the UTRP Existing Solutions Problem Statement 2 Our Algorithms Exact Genetic 3 Current State Results Work in Progress Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 9 / 35
  • 10. Introduction Existing Solutions Existing Solutions The list of existing solutions is long, including: Multiple step solutions. Metaheuristics. Mixed non-linear mathematical models. Ad-hoc solutions. Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 10 / 35
  • 11. Introduction Existing Solutions Tool Example ´ Computational tool by Alvarez et al. Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 11 / 35
  • 12. Introduction Existing Solutions Room for Improvement Current issues with the existing solutions: Many different problem definitions. The quality of these solutions depends fundamentally on the chosen algorithms. Large search space (there is no free lunch). Comparison is necessary! Our Goal: develop and test appropriate algorithms and methods for the UTRP using a well-known problem definition (and with common benchmarks). Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 12 / 35
  • 13. Introduction Problem Statement Outline 1 Introduction Motivation to the UTRP Existing Solutions Problem Statement 2 Our Algorithms Exact Genetic 3 Current State Results Work in Progress Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 13 / 35
  • 14. Introduction Problem Statement Input and Route Sets Two inputs: graph and demand matrix. Transport, route and transit networks, respectively [1]. Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 14 / 35
  • 15. Introduction Problem Statement Associated costs Operator cost: sum of weight of edges used. Passenger cost: total travel time. Multi-objective. Conditions that can be considered: Number of routes. Lenght of routes. Cycles and backtracks. Penalty for making transfers. Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 15 / 35
  • 16. Introduction Problem Statement Output Approximation for Pareto-optimal curves 240 GA Solutions with up to 8 routes GA Solutions with up to 6 routes 220 GA Solutions with up to 4 routes Fitting curve (58.22/(x-9.86) + 44.36) Total route set length in minutes 200 180 160 140 120 100 80 60 10 10.5 11 11.5 12 12.5 13 13.5 Average travel time in minutes Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 16 / 35
  • 17. Our Algorithms Exact Outline 1 Introduction Motivation to the UTRP Existing Solutions Problem Statement 2 Our Algorithms Exact Genetic 3 Current State Results Work in Progress Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 17 / 35
  • 18. Our Algorithms Exact Exact Solution Summary ”The problem of designing a good or efficient route set (or route network) for a transit system is a difficult optimization problem which does not lend itself readily to mathematical programming formulations and solutions using traditional techniques” – Dr. Partha Chakroborty, Transportation Engineer Mathematical solution has been created to test feasibility and correctness. Uses a Mixed Integer Programming formulation. Achieved global optimal solutions for Mandl’s Swiss road network (to be shown) with 2 and 3 routes. Very slow, but useful linear relaxation and for divide and conquer. Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 18 / 35
  • 19. Our Algorithms Genetic Outline 1 Introduction Motivation to the UTRP Existing Solutions Problem Statement 2 Our Algorithms Exact Genetic 3 Current State Results Work in Progress Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 19 / 35
  • 20. Our Algorithms Genetic Genetic Algorithm Overview Maintain a population of potential solutions, ie. route sets. Create or modify routes in a route set and, if dominating another route set, take its place. Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 20 / 35
  • 21. Our Algorithms Genetic Creating New Routes Take it from a pool of base routes. Apply operators to existing solutions (route sets). Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 21 / 35
  • 22. Our Algorithms Genetic Base routes Base routes are intrinsic to a graph: Shortest path for every pair of nodes (in original network). Minimum Spanning Tree (!). Paths with highest covered demand. Routes with high percentages in the linear relaxation of the MIP solution. Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 22 / 35
  • 23. Our Algorithms Genetic Minimum Spanning Tree Demo a 7 8 5 b c 7 9 5 15 d e 6 9 8 11 f g Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 23 / 35
  • 24. Our Algorithms Genetic Minimum Spanning Tree Demo a 7 8 5 b c 7 9 5 15 d e 6 9 8 11 f g Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 23 / 35
  • 25. Our Algorithms Genetic Minimum Spanning Tree Demo a 7 8 5 b c 7 9 5 15 d e 6 9 8 11 f g Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 23 / 35
  • 26. Our Algorithms Genetic Minimum Spanning Tree Demo a 7 8 5 b c 7 9 5 15 d e 6 9 8 11 f g Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 23 / 35
  • 27. Our Algorithms Genetic Minimum Spanning Tree Demo a 7 8 5 b c 7 9 5 15 d e 6 9 8 11 f g Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 23 / 35
  • 28. Our Algorithms Genetic Minimum Spanning Tree Demo a 7 8 5 b c 7 9 5 15 d e 6 9 8 11 f g Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 23 / 35
  • 29. Our Algorithms Genetic Minimum Spanning Tree Demo a 7 8 5 b c 7 9 5 15 d e 6 9 8 11 f g Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 23 / 35
  • 30. Our Algorithms Genetic Operators Mutate route Add a node to or remove a node from the extremity of a route. Simplify route set If the route set contains 9-3-4-5-6 and 4-5-6-12-2, we replace with 9-3-4-5-6-12-2 Cross-over two routes Join two routes at a certain intersection (cut cycles if necessary). Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 24 / 35
  • 31. Current State Results Outline 1 Introduction Motivation to the UTRP Existing Solutions Problem Statement 2 Our Algorithms Exact Genetic 3 Current State Results Work in Progress Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 25 / 35
  • 32. Current State Results Tested Networks Two networks were used: Mandl’s and artificial British(based) city with 110 nodes and 275 links. Mandl’s Swiss road network [1]. Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 26 / 35
  • 33. Current State Results Solution Quality Quantities All results are evaluated using the following quantities, as in previous works: di is the percentage of the demand satisfied with i transfers. ATT is the average travel time (in minutes per passenger), including transfer penalties. CO is the cost for the operator, i.e., the total route length (in minutes, considering constant transport speed). ATTwop = ATT − i≤TMAX tpen di i. Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 27 / 35
  • 34. Current State Results Mandl’s Network Exact Solutions Best possible route sets found using the Mixed Integer formulation Number of routes 2 3 d0 84.90 % 93.67 % d1 14.00 % 5.43 % d2 1.10 % 0.90 % ATT 11.33 min. 10.50 min. CO 98 min. 150 min. Processing time (s) 1065 78992 Two Routes 6-14-7-5-2-1-4-3-11-10-9-13-12 0-1-3-5-7-9-6-14-8 Three Routes 4-3-11-10-12-13-9-7-5-2-1-0 4-3-1-2-5-14-6-9-10-11 0-1-4-3-5-7-9-6-14-8 Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 28 / 35
  • 35. Current State Results Genetic Algorithm on Mandl’s Network Comparison between best UTRP multi-objective solutions on Mandl’s Network Scenario Qp Best previous results Our metaheuristic ([1]) approach results Best for Passenger d0 94.54 % 98.84 % d1 5.46 % 1.16 % d2 0.00 % 0.00 % ATT 10.36 min. 10.10 min. CO 283 min. 259 min. Compromise Solution d0 93.19 % 93.61 % (CO ≤ 148) d1 6.23 % 6.20 % d2 0.58 % 0.19 % ATT 10.46 min. 10.43 min. CO 148 min. 147 min. Compromise Solution d0 90.88 % 91.23 % (CO ≤ 126) d1 8.35 % 7.84 % d2 0.77 % 0.93 % ATT 10.65 min. 10.59 min. CO 126 min. 126 min. Best for Operator d0 66.09 % 77.78 % d1 30.38 % 21.32 % d2 3.53 % 0.90 % ATT 13.34 min. 12.97 min. CO 63 min. 63 min. Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 29 / 35
  • 36. Current State Results Genetic Algorithm on Artificial British Network Comparison between best UTRP multi-objective solutions on artificial British city Scenario Qp Best previous results Our metaheuristic ([1]) approach results I-Passenger d0 72.91 % 55.80 % ATT 36.28 min. 36.35 min. ATTwop 34.60 min. 34.12 min. CO 2986 min. 8406 min. II-Passenger d0 71.21 % 46.25 % ATT 37.52 min. 36.61 min. ATTwop 35.68 min. 33.77 min. CO 2378 min. 5181 min. I-Operator d0 48.62 % 9.48 % ATT 40.88 min. 55.08 min. ATTwop 37.36 min. 45.66 min. CO 1077 min. 319 min. II-Operator d0 46.97 % 8.47 % ATT 41.26 min. 55.48 min. ATTwop 37.655 min. 47.90 min. CO 1265 min. 319 min. Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 30 / 35
  • 37. Current State Work in Progress Outline 1 Introduction Motivation to the UTRP Existing Solutions Problem Statement 2 Our Algorithms Exact Genetic 3 Current State Results Work in Progress Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 31 / 35
  • 38. Current State Work in Progress Performance Time used in each function. Dijkstra takes 90% of processing time. To explore the search space faster: use GPU. Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 32 / 35
  • 39. Current State Work in Progress Simulations Two scenarios are being simulated: Porto Alegre: test effectiveness in comparison to existing network. Demands are artificial. Berlin: use MATSim as the objective function. Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 33 / 35
  • 40. Thank you! Conclusion New methods and algorithms for the UTRP can make public transport better! Suggestions or questions? Thank you! Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 34 / 35
  • 41. Thank you! References Lang Fan, Christine L. Mumford, and Dafydd Evans. A simple multi-objective optimization algorithm for the urban transit routing problem. In Proceedings of the Eleventh conference on Congress on Evolutionary Computation, CEC’09, pages 1–7, Piscataway, NJ, USA, 2009. IEEE Press. Bruno Coswig Fiss (TU Berlin) Algorithms for the UTRP June 13, 2012 35 / 35