Big Data In Airport OperationsART Workshop ‘Airport Capacity’ – 21st September 2016

Tom Garside – Heathrow AirportBert De Reyck – UCL

Xiaojia Guo - UCL

Integrated Plans, using Day Types, driving resourcing and operational preparedness

Dynamic Modelling of Pax Flow

Forecast & Plan

Performance Review

DMAC (Dynamic Monitoring of Arrivals and Connections) identifying variances to plan of passenger flow in connections process 

Prepared | Informed | Collaborative | ProactiveLive Performance Dashboard

ATMs

Passengers

Staff

Benefits: 

• Service• Efficiency• Capacity

Improving Performance through ‘Operating to Plan’

1

Happy Passengers, on time, travelling with their bags 

Objective

Business Change

Liberated Data

FlightPTM

PRM

Security

Pax Flow

Feedback Ticket

Presentation

Car parkBuses 

Data Liberation  Modelling & Analysis 

Airline Collaboration

Prepared | Informed | Collaborative | Proactive

Connections modelling to identify passenger at risk of misconnecting

Unlocking the Opportunity of ‘Operating to Plan’

2

Arrive atT2,3 or 4

Takeconnectingbus to T5

Ready toFly

BOSS

PTM

Confor‐mancedata

BDD

Data‐Driven Predictions

IDAHO

Arrive at T5

Disembark

DisembarkSecurityScreening Boarding Departure

Machine Learning Techniques Security LaneResourcing

TOBT Adjustment

Predictive model for ∆ Prescriptive Model

Percent transfer PAX

Percent covered in this study

Focus: Connecting Passengers

12%

88%

Immigration

3

T5 outbound Flight

Internationalarriving PAX

Border Force3rd PartyAirline

Predictive Model

3.7M 10 47Passenger records over 2015from the BOSS, BDD, and Conformance data sets. 

Significant predictors out of33 tested.

Passenger categories. 

Five Most Important Predictors

1. Whether or not the passenger arrives at T5 

2. Inbound flight body type

4. Inbound flight travel class

3. Perceived connection time

5. Inbound flight stand type

4

Y

Y

YArrive at T5 ?

Business/Firstclass

passenger?

Connect to adomesticflight ?

Perceived connection time is lessthan 90 min?

Categ. 1Samples: 1%Median = 34.0

The Regression Tree Model

Y

Predictive Model

∆

The full regression tree

5

Model accuracy

Distributions of each leaf

Live Trial

8H 5MIN 200SAn eight‐hour live trial took 

place on 19 JulyPredictions are made every 

five minutes.The script takes 200s to

produce the upcoming twohours’ forecasts.

Generate input datafile from IDAHO

Predict from the model andsave the outputs

Update decisions andwait for the next iteration

6

Ib flight Number of PAX median p75 p90 P(late)BA901 1 8:45 8:51 8:56 0.75AA730 4 8:46 8:55 9:04 0.70

Prescriptive Model - TOBT

AA730BA246 BA294

Flight: BA 774Destination:  ARN (Stockholm) STD: 09:15Total PAX:  129Transfer PAX: 79Int. transfer PAX: 43Aircraft: Airbus 319

6:00 7:00 8:00 9:00

BA054 BA092 BA216 BA058 BA1162 PAX 3 PAX 6 PAX 1 PAX 10 PAX 10 PAX 6 PAX 4 PAX

• Clear• Risk of impacting TOBT

Need to changeTOBT ?

Identify and expeditelate passengers

YMake accurateadjustments

N

Number of late PAXTOBT +5min +10min +15min +20min5 3 2 1 1

Predictions: Outbound flight late passengers

Predictions: Individual connection times

BA9011 PAX

12%

88%

7

15 min. intervals

8:00 8:408:20 9:00

Num

ber o

f pas

seng

ers

250

300

350

50

9:20 9:40Time

200

150

100

10:000

Predictions: Connecting passenger flows

Dynamicresourcing plans

Busyness level overviewDetailed passenger flow profiles

5 min. intervals

8:00 8:408:20 9:00

Num

ber o

f pas

seng

ers

80

100

120

09:20 9:40

Time

60

40

20

10:00

60 min. intervals

8:00 8:408:20 9:00

Num

ber o

f pas

seng

ers

600

800

1000

9:20 9:40Time

400

200

010:00

Prescriptive Model - Resourcing

8

10% chance< 310 PAX

90% chance< 350 PAX50% chance< 330 PAX

9:45 to 10:00 a.m.

Robust and stable TOBT

Better operational performanceEfficient resourcing allocation

Better passenger experiencePotential reduction in flight delays

Conclusions

Big data

How do we smooth the aircraft, passenger, and bagflows?

9

Machine learning

Data-drivendecisions

How do we improve data collaborations?

Service & Efficiency

Capacity

Resilience

10

Heathrow Current and Future Challenges

• End to end passenger delay reduction – landside and airside• Information collaboration to enable predictable journeys

• Enabling passengers to turn up at the airport at the ‘right time’• Optimising passenger dwell at the airport to unlock capacity

• Enhancing integrated situational awareness during disruption • Standardising approach to airport and airline information 

collaboration

Opportunities