Big Data In Airport Operations

Size: px

Start display at page:

Download "Big Data In Airport Operations"

Eustace Cole
6 years ago
Views:

1 Big Data In Airport Operations ART Workshop Airport Capacity 21 st September 2016 Tom Garside Heathrow Airport Bert De Reyck UCL Xiaojia Guo - UCL

2 Improving Performance through Operating to Plan Forecast & Plan Objective Happy Passengers, on time, travelling with their bags ATMs Dynamic Modelling of Pax Flow Passengers DMAC (Dynamic Monitoring of Arrivals and Connections) identifying variances to plan of passenger flow in connections process Staff Integrated Plans, using Day Types, driving resourcing and operational preparedness Performance Review Benefits: Service Efficiency Capacity Prepared Informed Collaborative Proactive Live Performance Dashboard 1

Unlocking the Opportunity of Operating to Plan Data

Analysis PRM Security Liberated Data Pax Flow Ticket

identify passenger at risk of misconnecting Business

3 Unlocking the Opportunity of Operating to Plan Data Liberation Flight Buses Car park PTM Modelling & Analysis PRM Security Liberated Data Pax Flow Ticket Presentation Feedback Connections modelling to identify passenger at risk of misconnecting Business Change Airline Collaboration Prepared Informed Collaborative Proactive 2

Focus: Connecting Passengers Percent transfer PAX Percent covered in this study

Model Arrive at T5 Disembark Immigration Arrive at T2,3 or 4 Disembark Take

Driven Predictions Machine Learning Techniques Security Lane Resourcing TOBT

4 Focus: Connecting Passengers Percent transfer PAX Percent covered in this study International arriving PAX T5 outbound Flight Predictive model for Prescriptive Model Arrive at T5 Disembark Immigration Arrive at T2,3 or 4 Disembark Take connecting bus to T5 Ready to Fly Security Screening Boarding Departure Data Driven Predictions Machine Learning Techniques Security Lane Resourcing TOBT Adjustment BOSS BDD Confor mance data 12% IDAHO PTM 88% Airline 3 rd Party Border Force 3

5 Predictive Model 3.7M Passenger records over 2015 from the BOSS, BDD, and Conformance data sets. Significant predictors out of 33 tested. Passenger categories. Five Most Important Predictors 1. Whether or not the passenger arrives at T5 2. Inbound flight body type 3. Perceived connection time 4. Inbound flight travel class 5. Inbound flight stand type The Regression Tree Model Arrive at T5? Y Business/First class passenger? Y Connect to a domestic flight? Y Perceived connection time is less than 90 min? Y Categ. 1 Samples: 1% Median =

6 The full regression tree Predictive Model Distributions of each leaf Model accuracy 5

Live Trial 8H 5MIN 200S An eight hour live trial

7 Live Trial 8H 5MIN 200S An eight hour live trial took place on 19 July Predictions are made every five minutes. The script takes 200s to produce the upcoming two hours forecasts. Update decisions and wait for the next iteration Generate input data file from IDAHO Predict from the model and save the outputs 6

Prescriptive Model - TOBT 2PAX 3PAX 6PAX 1PAX 10 PAX 10 PAX 6PAX 4PAX BA054 BA092

Destination: ARN (Stockholm) STD: 09:15 Total PAX: 129 Transfer PAX: 79 Int.

3 2 1 1 Clear Risk of impacting TOBT N Predictions: Individual connection times

8 Prescriptive Model - TOBT 2PAX 3PAX 6PAX 1PAX 10 PAX 10 PAX 6PAX 4PAX BA054 BA092 BA216 BA058 BA246 BA294 BA116 AA730 1PAX BA901 6:00 7:00 8:00 9:00 Flight: BA 774 Destination: ARN (Stockholm) STD: 09:15 Total PAX: 129 Transfer PAX: 79 Int. transfer PAX: 43 Aircraft: Airbus 319 Predictions: Outbound flight late passengers 12% 88% Need to change TOBT? Y Make accurate adjustments Number of late PAX TOBT +5min +10min +15min +20min Clear Risk of impacting TOBT N Predictions: Individual connection times Ib flight Number of PAX median p75 p90 P(late) BA :45 8:51 8: AA :46 8:55 9: Identify and expedite late passengers 7

Prescriptive Model - Resourcing Predictions: Connecting passenger flows 15 min.

Number of passengers 300 250 200 150 100 50 90% chance < 350 PAX 50% chance < 330 PAX 10%

9 Prescriptive Model - Resourcing Predictions: Connecting passenger flows 15 min. intervals 350 9:45 to 10:00 a.m. Number of passengers % chance < 350 PAX 50% chance < 330 PAX 10% chance < 310 PAX Dynamic resourcing plans 0 8:00 8:20 8:40 9:00 Time 9:20 9:40 10:00 5 min. intervals min. intervals Number of passengers Number of passengers :00 8:20 8:40 9:00 Time 9:20 9:40 10:00 0 8:00 8:20 8:40 9:00 Time 9:20 9:40 10:00 Detailed passenger flow profiles Busyness level overview 8

Better passenger experience Potential reduction in flight delays How do we

10 Conclusions Big data Machine learning Data-driven decisions Robust and stable TOBT Better operational performance Efficient resourcing allocation Better passenger experience Potential reduction in flight delays How do we smooth the aircraft, passenger, and bag flows? How do we improve data collaborations? 9

Heathrow Current and Future Challenges Service & Efficiency Opportunities End

collaboration to enable predictable journeys Capacity Enabling passengers to

airport to unlock capacity Resilience Enhancing integrated situational

11 Heathrow Current and Future Challenges Service & Efficiency Opportunities End to end passenger delay reduction landside and airside Information collaboration to enable predictable journeys Capacity Enabling passengers to turn up at the airport at the right time Optimising passenger dwell at the airport to unlock capacity Resilience Enhancing integrated situational awareness during disruption Standardising approach to airport and airline information collaboration 10

Aviation ICT Forum OCT 2014

Airport Intelligence Enabler for Process Improvement Aviation ICT Forum OCT 2014 Jaipur S.Francis Rajan VP & Head ICT (Information and Communications Technology ) Aviation ICT Forum Jaipur 2014 Greetings