Change  and  Safety:  Decision-­‐Making  from  Data  

Anna  Holloway,    Safety  Knowledge  and  Planning,  RSSB  London      William  Marsh,    School  of  Electronic  Engineering  and  Computer  Science,    Queen  Mary,  University  of  London  

•  Incident  data    risk  estimate    decision:  many  applications  •  Local  prediction  needed  •  Local  prediction  possible  with  data  and  knowledge  •  Bayesian  network:  example  application  

Boarding  and  Alighting  Accidents  

1.9

1.2 1.2 0.9

1.3

0.5 0.7

0.4 0.7 0.7

2.1 1.9 1.9

2.6 2.2

0.6 0.7 1.1

0.9 1.3

0.0

0.5

1.0

1.5

2.0

2.5

3.0

2006

/07

2007

/08

2008

/09

2009

/10

2010

/11

2006

/07

2007

/08

2008

/09

2009

/10

2010

/11

2006

/07

2007

/08

2008

/09

2009

/10

2010

/11

2006

/07

2007

/08

2008

/09

2009

/10

2010

/11

Fall between train and platform

Caught in train doors

Other alighting accident

Other boarding accident

FWI

Shock & trauma Minor injuries

  Accidents  to  passengers  geEng  on  and  off  trains  

  From  2011  Annual  Safety  Performance  Report    

  How  to  reduce  Boarding  /  AlighLng  harm    StaLon  staffing  (local)    StaLon  Design  (local)    Train  design  (more  global)      

  Incident  data  –  many  examples    Safety:  SPADs,  broken  rails,  bridge  strikes,  ….    Reliability:  signal  failure,  staff  absence,  ….  

Decision  Making  is  Local  

  Not  enough  data  for  local  decision  making  directly  

Just  Data  for  Decision  Making?  

0.E+00

2.E-07

4.E-07

6.E-07

8.E-07

1.E-06

1.E-06

VAL

WK

K

WA

F W

MW

W

WR

W

GV

W

RP

WFJ

W

EE

WLP

W

MS

WC

L W

HD

W

RU

W

SB

WES

W

YB

W

HM

W

HT

WB

D

WW

L W

DD

W

MI

WN

E W

IH

WSF

W

IV

WO

L W

DL

WSR

W

OF

WPL

W

RS

WM

D

FWI/

expo

sure

0

200

400

600

800

1000

1200

0 1 2 3 4 5 6 7 8 9 10

11-1

2 13

-14

15-1

6 17

-18

19-2

0 21

-25

26-3

0 31

-35

>35

Num

ber

of S

tatio

ns

Number of Incidents

Normalised  observed  harm  (FWI)  –  some  staLons:  spikey  

DistribuLon  of  incident  count:  many  zeros  

Model  Concept:  Data  and  Knowledge  

  Knowledge:  causal  analysis  of  incidents  

  EsLmated  effect  of  causes  on  incidents  

•  ORR Station Usage •  TSDB •  DfT – Significant

Steps Research •  DFT National Travel

Survey •  SRM Normalisers •  MET Office •  APRS •  T763 dispatch data

  How  railway  is  used    Prevalent  of  

causes  

  Incident  data    Presence  of  

causes  (e.g.  ice,  crowding)  

If  ice  is  a  cause  of  falls  then  we  expect  iciness  to  occur  in  incident  reports  more  o\en  than  we  expect  from  the  

prevalence  of  icy  condiLons  

Causes  

Incidents  

How  the  railway  is  used  

  Profile: region, station, …

  Distinguish individual and aggregate risk

Model Queries

5.2E-­‐08  5.3E-­‐08  5.4E-­‐08  5.5E-­‐08  5.6E-­‐08  5.7E-­‐08  5.8E-­‐08  

Individual  FWI  

Aggregate  FWI  (propor4onal)  

!"#$%&

!'("#$%&

!')"#$%&

!'*"#$%&

!'+"#$%&

!'!"#$%&

!',"#$%&

!'-"#$%&

!'%"#$%&

!'."#$%&

,"#$%&

/01&

/23&

405

&40"

&401&

406

&403

&407

&485

&481&

489

&483

&458

&45:

&451&

459

&45;

&4<8

&4<"

&4<:

&4<=

&4<>

&4<7

&4"0

&4":

&4"=

&4"7&

4??&

4?2&

4?1&

4@0

&4@6

&4@7

&4@4

&4:5

&4:"

&4:2&

4:=

&4:9

&4:A

&4:B

&42<&

42C&

42=

&42/&

4D<

&4D@

&4DD

&415&

41"&

41@

&41=

&

  ValidaLon    Strength  of  different  effect    SensiLvity  to  approximaLons    Consistency  with  data    

  Other  applicaLons  

  Tools  to  support  model  building  

Future  Work  

Thank  You  Thanks to the Safety Knowledge and Planning and Safety Intelligence groups at RSSB for their generous collaboration

•  Incident  data    risk  estimate    decision  •  Local  prediction  •  Data  and  knowledge  •  Bayesian  network:  example  application