Reinventing Crew Scheduling At Netherlands Railways Erwin Abbink, NS Reizigers bv, The Netherlands (NL) Matteo Fischetti, University of Padua, Italy Double.

Reinventing Crew SchedulingAt Netherlands Railways

Erwin Abbink, NS Reizigers bv, The Netherlands (NL)

Matteo Fischetti, University of Padua, Italy

Double Click sas, Italy

Leo Kroon, Erasmus University Rotterdam, NL

NS Reizigers bv, NL

Gerrit Timmer, Free University of Amsterdam, NL

ORTEC International bv, NL

Michiel Vromans, Erasmus University Rotterdam, NL

Contents

1. Introduction

2. History

3. Development of an alternative model

4. Sharing Sweet&Sour

5. TURNI and solution methods

6. Efficiency improvements

7. Conclusions

NS Reizigers (Netherlands Railways)

• Main Dutch operator of passenger trains

• 5,000 timetabled trains per weekday

• 1 million passenger trips per weekday

• 112 million train kilometers per year

• 3,000+ drivers and 3,500+ conductors

• 29 crew depots

• 2,600+ carriages

0 50km

• 29 Depots • Duties are created in Utrecht (Ut)

• Rosters are created

locally in the depots

• Focus: Duties

NS Reizigers (Netherlands Railways)

Ut

Crew management

Efficiency

Acceptance

RobustnessPunctuality

Find: - Balance

- Trade-off

Minimum transfer

time

Pre- and post times

Meal break rule

Maximum duty length

- Route knowledge

- Rolling stock knowledge

Duty examples

Rostering Rules Maximum percentage night duties

Maximum percentage long duties (>9 hrs)

Maximum average duty

length

History

• June 10th 2001: introduction of the “Church Circles”

• Aim of the management (TOP DOWN):

• Improve robustness / punctuality and service

• How:

• Less different trains / routes per duty

• Train change only during meal break

• Better knowledge of local situation

History

• Drivers and conductors were quite unhappy

• Decrease in quality / variation of their work

• Unfair division of aggression work over depots

• “Secret agenda of management”

• Punctuality down

• Motivation down / Sickness up

Involved parties

Union A

Union B Depot B

Depot A

WorksCouncil

ManagementSTRIKES

STRIKES

Union A

Union B Depot B

Depot A

WorksCouncil

Management

Involved parties

Development of alternative model

Union A

Union B Depot B

Depot A

WorksCouncil

Management

Participative approach (BOTTOM UP)

Alternative 1

Alternative n

‘Sharing Sweet & Sour’....

Selection by Works Council

Acceptance by Management

2-day conferences

Hundreds of optimization runs with TURNI

Sharing Sweet&Sour

Additional variation rules:

• Max Repetitions In Duties (RID)

• Max percentage of aggression work per depot

• Max standard deviation on aggression

• Min percentage of preferred trains per depot

• Max standard deviation on preferred trains

• Min number of routes per depot

• Min average number of routes

• Max percentage of Rolling Stock cluster per depot

Sharing Sweet&Sour

Hdr

Lw

Gn

Emn

EkzHnAmr

LlsZl

Kpn

Hgl

Bh

EsZp

Dv

AmfAmfs

Asd

Hnk

Apd

HlmZvt

Ledn

Apn

WpNdb

Brn

Ut Ed Ah

Nm

RhnGdGdg

ZLGvcGv

Br

Gdm

Hlds

Rth

Rtd

Ddr Ht

Rsd

Vs

Ehv Vl

Mt

Hrl

Rm

Mtr

Krd

Stola

Luik

Hfdo

Hdr

Lw

Gn

Emn

EkzHnAmr

LlsZl

Kpn

Hgl

Bh

EsZp

Dv

AmfAmfs

Asd

Hnk

Apd

HlmZvt

Ledn

Apn

WpNdb

Brn

Ut Ed Ah

Nm

RhnGdGdg

ZLGvcGv

Br

Gdm

Hlds

Rth

Rtd

Ddr Ht

Rsd

Vs

Ehv Vl

Mt

Hrl

Rm

Mtr

Krd

Stola

Luik

Hfdo

Hdr

Lw

Gn

Emn

EkzHnAmr

LlsZl

Kpn

Hgl

Bh

EsZp

Dv

AmfAmfs

Asd

Hnk

Apd

HlmZvt

Ledn

Apn

WpNdb

Brn

Ut Ed Ah

Nm

RhnGdGdg

ZLGvcGv

Br

Gdm

Hlds

Rth

Rtd

Ddr Ht

Rsd

Vs

Ehv Vl

Mt

Hrl

Rm

Mtr

Krd

Stola

Luik

Hfdo

Sharing Sweet&Sour

----------------------------------------------------------Crew | | RID |% Pref.|%Aggres| RS | Nasty RSDepot |Routes | Avg | Train | Train |Clust| (1) (2)----------------------------------------------------------Ah | 19 | 2.4 | 63.9 | 9.6 | 7 | 0.8 15.2Amf | 22 | 2.4 | 46.0 | 29.5 | 6 | 29.7 25.3Amr | 22 | 2.5 | 67.9 | 42.6 | 6 | 26.6 3.9Asd | 40 | 2.5 | 58.0 | 29.9 | 6 | 36.4 8.9Ddr | 13 | 2.6 | 35.0 | 24.9 | 4 | 15.9 46.6Ehv | 17 | 2.6 | 38.5 | 4.1 | 6 | 0.3 58.7Ekz | 23 | 2.9 | 28.9 | 27.4 | 6 | 32.0 8.8Es | 16 | 2.7 | 47.9 | 6.7 | 6 | 0.0 6.1 . . . . . . . . . . . . . . . . . . . . . . . .Llso | 21 | 2.3 | 44.0 | 44.4 | 5 | 44.4 15.6Mt | 11 | 2.7 | 42.3 | 2.0 | 3 | 0.0 54.6Nm | 18 | 2.3 | 44.0 | 7.6 | 7 | 0.4 35.6Rtd | 32 | 2.3 | 32.2 | 20.5 | 6 | 14.6 32.2Ut | 44 | 2.5 | 27.1 | 14.6 | 6 | 8.6 51.8Vl | 14 | 2.8 | 40.3 | 2.6 | 6 | 0.0 38.4Vs | 17 | 2.5 | 64.1 | 15.3 | 4 | 8.3 26.4Zl | 26 | 2.3 | 54.4 | 3.0 | 7 | 13.0 5.8----------------------------------------------------------Global | 21.2 | 2.4 | 12.5 | 12.8 |

Sharing Sweet&Sour: variation statistics

TURNI: crew scheduling system by Double Click sas

• TURNI based on a set covering model

with many additional nasty “depot” constraints

• Typical instance of NS Reizigers (drivers):

- 14,000 timetabled trips

- 1,000 duties from 29 crew depots

• Extensive customizations for NS Reizigers

(Extended) Set Covering model

1

D

tDddx

}1,0{dx

otherwise.

selected, is duty potential if

0

1 dxd

Tt ,...,1

cd

D

ddc uxb

1, Cc ,...,1

Dd ,...,1

d

D

dd xk

1min subject to

Solution techniques

• Dynamic column generation

• Column generation based on dynamic programming

• Lagrangian optimization instead of LP

• Fast heuristics using Lagrangian dual information

• Intensification through variable fixing / local branching

• Solution refinement through matching model

Railways vs. Airlines

• Large instances in comparison with airlines:

- more activities / legs per instance (14,000)

- more activities / legs per duty / pairing (avg. 14)

• Crew qualifications cannot be used to partition instances

• More complex and detailed rules:

- rolling stock circulation

- complex variation rules

- depot constraints

• Generation of the duties subject to the new rules would have been impossible without TURNI

• Total amount of work: + 3.2%

• Total # duties: + 1.2%

• Initial savings: 2.0% $ 7 million

$ 7 mill.

$ 4.8

mill.

• Real savings: $ 4.8 million as variation was larger than agreed earlier in order to increase personnel acceptance

Efficiency improvement 2003 - 2004

Efficiency improvement 2005

• Reduction of the transfer time from 25 to 20 minutes

(only a minor negative effect on robustness / punctuality)

• Efficiency improvement of 2.5% or $8 million per year

• Total savings: $7 + $8 = $15 million per year

• Using TURNI, changing the duty structures is easy

(both for analysis and for production planning)

Conclusions

• Application of TURNI led to:

- a new production model (Sharing Sweet&Sour)

- more efficient duties (4.5%)

• Unpunctuality decreased by 25% (Punctuality 80% → 85%)

• Motivation of personnel up

• Sickness rate down by 50%

• BOTTOM UP > TOP DOWN

Conclusions

• Initially nobody believed in the existence of a solution for

the conflicts between all parties

• Combination of a participative approach together with

expertise in Operations Research led to the success

• Operations Research led to quantitative and objective

results accepted both by personnel and management

Union A

Union B Depot B

Depot A

WorksCouncil

Management

Conclusions

Union A

Union B Depot B

Depot A

WorksCouncil

Management

Conclusions

O O

Union A

Union B Depot B

Depot A

WorksCouncil

Management

Conclusions

O --

Union A

Union B Depot B

Depot A

WorksCouncil

Management

Conclusions

O R