Interdependencies through Bayesian Networks Preliminary results - - PowerPoint PPT Presentation

▶
interdependencies
SMART_READER_LITE
LIVE PREVIEW

Interdependencies through Bayesian Networks Preliminary results - - PowerPoint PPT Presentation

Sep 06, 2022 291 likes •656 views

Assessing ATM Performance Interdependencies through Bayesian Networks Preliminary results Andrea Ranieri, Andrada Bujor Advanced Logistics Group, Indra ICRAT 2014, Istanbul 29 th May 2014 ALG is part of Indra Business Consulting, the

slide-1
SLIDE 1

Assessing ATM Performance Interdependencies through Bayesian Networks

Preliminary results

Andrea Ranieri, Andrada Bujor – Advanced Logistics Group, Indra

ICRAT 2014, Istanbul – 29th May 2014

slide-2
SLIDE 2

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

ALG is part of Indra Business Consulting, the strategy consultancy division of Indra, a global leader in business and technology consulting

2

IT Systems and Services Transport and traffic systems Defence and Simulation Systems

(1) 2012 Data

  • Founded in 1994
  • Among the top management

consulting firms in Europe and Latin America

  • Projects in more than 40

countries

  • Ongoing expansion at

international level, with 14 Indra Business Consulting

  • ffices in Europe, Latin

America, Africa, the Middle East and Asia

  • Turnover of €66 m and a

team of more than 600 professionals

Strategic and Management Consultancy

  • Integration and

development of systems,

  • utsourcing
  • Innovative and high

technology solutions

  • Ticketing systems
  • ASP development and

e-business solutions

  • Centers of technological

competence:

  • Customer Relationship

Management

  • Supply Chain

Management

  • Public sector projects

(e.g. Voting machines and election management systems)

  • Projects for leading

clients in the transport sector (air traffic, terrestrial traffic and public transportation)

  • Others: e-ticketing,

ATMs, traffic control of high speed trains)

  • Leading Spanish

company in IT and Defence systems

  • Key programmes for

European and international defence

  • European leader in

surveillance and electronic intelligence

  • On the cutting edge of

simulation technology and excellence in maintenance systems

Top-tier European consultancy and IT firm Sales volume > €3,000 million€ (1) More than 40,000 employees (1) Presence in more than 115 countries Growth rate and EBITDA above sector average

slide-3
SLIDE 3

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

ALG is the Transportation and Infrastructure consulting practice

3

Transportation, Infrastructure & Logistics Consultancy

  • ALG is the Transportation, Infrastructure and Logistics practice within Indra

Business Consulting with:

  • Annual revenue of over 20 million $
  • More than 2,000 projects in more than 50 countries
  • Multidisciplinary team with 140 consultants (40% LatAm, 30% Europe, 15%

Middle East & Asia, 15% Africa)

  • Wide knowledge and experience in the complete life cycle of transportation and

infrastructure businesses

  • Our professional team combines a multi-disciplinary focus in the key strategic areas

(business management, engineering, operations, economics, information systems)

Transport Logistics

Sector expertise

Urban mobility and public transportation Aviation Ground transportation Traffic and tolls Supply chain Maritime transportation Railroads Logistic platforms

40% 18%

10% 12% 5% 5% 5% 5%

slide-4
SLIDE 4

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

ALG has acquired a strong international experience

4

We have delivered more than 2.000 projects carried out over the last 25 years

Offices Projects

35%

Latina America

30%

Europe

15%

Africa Middle East

10%

Asia

10%

Barcelona Beijing Bilbao Buenos Aires Caracas Casablanca Dubai Lima Lisboa London Madrid México Milan Miami Paris São Paulo

slide-5
SLIDE 5

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Our work model focuses on the achievement of tangible results, providing comprehensive service

  • ffering within the Indra portfolio

5

  • Technology consulting
  • Business Process Outsourcing (BPO)
  • Maintenance and Management of Applications
  • Outsourcing of distributed Systems and work positions
  • Outsourcing of the management of infrastructures and operation

Understanding

  • f the value

creation

  • pportunities

Design of strategies Programme management

Value creation Top management consultancy

Analysis Conceptualization Implementation

  • Market and client

analysis

  • Diagnosis of

capabilities and competitive position

  • Description of

sector trends

  • Identification of

business

  • pportunities
  • Generation of

alternative solutions

  • Value quantification
  • f alternatives
  • Facilitation of

decision-making

  • Implementation

planning

  • Team management
  • Internal and external

negotiations

  • Execution of specific

tasks

  • Increasing

revenue

  • Raising margins
  • Developing

competitive advantage

  • Launching new

businesses

Generation of concrete results

Strategic Consultancy System Integration and Outsourcing

  • Strategic Planning of Systems and New Technologies
  • Diagnosis of the Systems Function
  • Selection of technologies and specific solutions
  • Implementation of packages commercial solutions of market
  • Development of tailored solutions
  • Cost modeling

Business consulting

slide-6
SLIDE 6

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Motivation State of the art The model Use of the model

6

slide-7
SLIDE 7

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Single European Sky & Performance Scheme

Motivation

7

European Commission’s ambitious initiative aiming to meet future ATM capacity and safety needs

  • Proposes a legislative approach Performance Scheme (Regulation)
  • European Commission establishes targets in key areas to be accomplished by ATM actors:

Traffic x3 Safety x 10 Co2 emissions per flight -10% ATM costs - 50%

*compared to 2005 levels

Human Factor pillar Airport pillar

Airports as integral part of the ATM network

Legislative pillar Technology pillar Safety pillar

SESAR

Performance Scheme

FABs Network Manager Extension of EASA competences

Safety, Capacity, Cost efficiency and Environmental Flight Efficiency

slide-8
SLIDE 8

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Performance regulations in European ATM

Motivation

8

2004

(EC) No 549/2004 SES Framework (EC) No 550/2004 Service provision

2009

(EC) No 1070/2009 SES II

  • Targets for Capacity,

Environment and Cost Efficiency

  • Determined Unit Rate
  • Risk-sharing mechanisms
  • Incentive schemes

(EC) No 691/2010 Performance scheme RP1 (EC) No 1191/2010 Charging Scheme Determined cost model

2013

  • Targets for Safety, Capacity,

Environment and Cost Efficiency

  • Mandatory financial incentive to

reach the capacity targets

  • Incentives on other targets

(EC) No 391/2013 Charging scheme Incentives (EC) No 390/2013 Performance scheme RP2 (EC) No 219/2007 SESAR JU (EC) No 1794/2006 Charging Scheme Full Cost recovery

  • Charges modulation to reflect

performance

  • Full Cost Recovery principle
  • Based on Service provision

regulation

  • Response to the worsen

saturation of European airspace and the increase of delays

  • Legal basis for the common

charging and performance schemes

  • Extension of the Regulation’s

scope to terminal ANS costs

  • Move to a system of binding

performance targets with incentives

(EC) No 551/2004 Airspace (EC) No 551/2004 interoperability

slide-9
SLIDE 9

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

KPAs and KPIs defined by SESII

Motivation

9

Capacity Cost effectiveness Safety Environment

Target assigned also for RP1

Effectiveness of safety management (“maturity”) Application of severity classification scheme Application of Just Culture Runway incursions ATM special technical events Separation infringements Level of occurrence reporting Application of automatic data recording for separation minima infringement monitoring Application of automatic data recording for runway incursion monitoring Target for RP2 Monitored

Safety

Arrival ATFM delays ATFM Slot adherence ATC pre-departure delay Additional time in taxi-out phase Additional time in arrival sequencing and metering area (ASMA) En-route ATFM delay

Capacity

Target for RP2 Monitored Horizontal flight efficiency of last filed flight plan (KEP) Horizontal flight efficiency of actual trajectory (KEA) Effectiveness of booking procedures for FUA Rate of planning of CDRs Effective use of CDRs Additional time in taxi-out phase Additional time in arrival terminal airspace (ASMA)

Environment

Target for RP2 Monitored Determined Unit Cost (DUC) for En route ANS (Determined Unit Rate DUR in RP1) Determined Unit Cost (DUC) for terminal ANS Terminal Unit Rate Terminal costs Costs of EUROCONTROL

Cost effectiveness

Target for RP2 Monitored

slide-10
SLIDE 10

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Motivation State of the art The model Use of the model

10

slide-11
SLIDE 11

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Relevant literature of performance interdependencies in ATM

State of the art

11

Capacity vs Cost efficiency

Assessing the true cost of delay to Airlines*

  • Study of the University of

Westminster in collaboration with PRU*

  • Cost of strategic and

tactical delay

  • Criteria for re-routing

decision

Econometric Cost Efficiency study

Econometric Cost-efficiency benchmarking of ANSP*

  • Estimates a cost function

for the provision of ANS

  • Demonstrates

the presence of economies of scale and density in ANS

  • Assesses

the level

  • f

ANSP inefficiency

Capacity vs Cost efficiency

The Optimum Capacity/Delay trade-off*

  • Future ATM Profile tool for

ATFM simulation

  • Cost model analysis

Traffic levels vs Cost effectiveness

A Network Pricing Formulation model*

  • Castelli et. al study the

dependencies of ANSP’s revenue as a function of the Unit rate

  • Results based on

simulations

  • Criteria for Unit Rate
  • ptimization

*Cook, G. Tanner, and Anderson, University of Westminster:

  • Evaluating

the true cost to airlines

  • f
  • ne

minute

  • f

airborne

  • r

ground delay”, study for Performance Review Commission (EUROCONTROL), 2004.

  • "European

airline delay cost reference values", study for EUROCONTROL Performance Unit, 2011. *Performance Review Unit Econometric cost-efficiency benchmarking

  • f

Air Navigation Service Providers”, May 2011 *Castelli, Lorenzo, Martine Labbé, and A. Violin. "A Network Pricing Formulation for the revenue maximization

  • f

European Air Navigation Service Providers", Transportation Research Part C: Emerging Technologies, 2012. *EUROCONTROL Performance Review Commission, “Performance Review Report 2010”, May 2011.

slide-12
SLIDE 12

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Capacity vs Cost efficiency

State of the art

12

Assessing the true cost of delay to Airlines

  • Study performed by the University of Westminster in collaboration with the PRU
  • Provides a number of figures useful to quantify the link between the capacity and the cost

efficiency

Airline delay cost

Tactical cost

  • Primary delay:
  • Fuel
  • Crew
  • Maintenance
  • Passenger
  • Secondary or reactionary
  • Crew
  • Maintenance
  • Passenger

Strategic cost

  • Fuel cost
  • Maintenance
  • Fleet
  • Crew
  • block-hour costs
  • Passengers (not considered in

the study)

  • Re-routing case  calculate the cost of the flight delay and the alternative route

Calculated Take-Off Time delay > 15 min

  • airborne extension n possibly accepted

if CTOT reduction is:

𝟐. 𝟐𝒐 ÷ 𝟐. 𝟒𝒐 + 𝟐𝟏 𝒏𝒋𝒐

Calculated Take-Off Time delay < 15 min Not worth a re-routing

slide-13
SLIDE 13

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Econometric regression model

State of the art

13

Variables (𝑫𝑗𝑢 ,𝑦𝑗𝑢 ) Definition Pit & Lee Random effect model True Random Effects model

  • Coeff. (𝜸)
  • Std. errors (𝒗𝒋𝒖)
  • Coeff. (𝜸)
  • Std. errors (𝒗𝒋𝒖)

C Total ATM/CNS provision costs Y Output measure i.e. number of composite flight-hours 0.57 0.06 0.73 0.01 W1 Average employment costs per hour for ATCOs in OPS 0.28 0.02 0.30 0.02 W2 Average employment costs for support staff 0.28 0.02 0.29 0.02 W3 Price of non-staff operating inputs (index for producer goods) 0.37

  • 0.34
  • W4

Capital input price 0.07 0.02 0.07 0.01 VAR Traffic variability 1.27 0.20 0.90 0.08 NET Network Concentration

  • 0.34

0.09

  • 0.41

0.02 Size Size of airspace controlled 0.28 0.11

  • 0.01

0.01 COMP Structural traffic complexity

  • 0.04

0.09

  • 0.06

0.02 BUS Business environment quality

  • 0.22

0.04

  • 0.19

0.02 T Time trend

  • 0.02

0.00

  • 0.03

0.00 Constant (𝜷) 4.27 1.45 Inefficiency ANSP level of inefficiency 60% 13%

Cobb-Douglas cost function

for ANSP “i” and year “t”

slide-14
SLIDE 14

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Delay, capacity and costs

State of the art

14

The Optimum Operating Point represents the best trade-off between the cost

  • f providing capacity and the cost of delay for a particular ACC

Delay costs

Costs M€

Total Cost “Static optimum capacity level”

Costs M€

Capacity / Traffic Volume Capacity / Traffic Volume

M€

Cost of ANS capacity (en route) Minimise total costs to airspace users Traffic demand Cost of ATFM delay (en route)

Source: EUROCONTROL Capacity Enhancement Functional (CEF) ― Capacity Assessment & Planning Guidance.‖, an overview of the European Network Capacity Planning, Edition September 2007

slide-15
SLIDE 15

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Traffic levels vs Cost effectiveness

State of the art

15

A Network Pricing Formulation model

  • European ANSP sets its Unit Rate annually, according to EC Regulation (1191/2010)
  • Castelli et. al study the dependencies of ANSP’s revenue as a function of the Unit Rate

Route Charge

𝑺𝑫 = 𝒎 ∗ 𝒙 ∗ 𝑼

  • l (distance factor):
  • w(weight factor):
  • T(Unit Rate): charge imposed on a flight per 100 km flown in a

given charging zone and per 50 metric tonnes of aircraft weight 𝑚 = 𝑒 (𝐻𝑠𝑓𝑏𝑢 𝐷𝑗𝑠𝑑𝑚𝑓 𝐸𝑗𝑡𝑢𝑏𝑜𝑑𝑓) 100 𝑥 = 𝑁𝑈𝑃𝑋 50 Determine the optimal T for both ANSP and Airspace Operator through a Bilevel Programming framework ANSP Revenue for one simulation Cost of one flight

slide-16
SLIDE 16

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Methodology approach – Bayesian Networks

State of the art

16

  • Graphical model that encodes

probabilistic relationships

  • Considers dependencies among

all variables Bayesian Network - an influence diagram where each node represents a discrete variable, an arc represents a causal influence between the linked variables and the strength of this influence can be quantified using probabilities

A B

A P(A) High 0.7 Moderate 0.2 Low 0.1 P(B|A) A

High Moderate Low

B TRUE 0.8 0.4 0.1 FALSE 0.2 0.6 0.9

Bayesian Network Model

B Computation P(B) TRUE

0.8*0.7+0.4*0.2+0.1*0.1

0.65 FALSE

0.2*0.7+0.6*0.2+0.9*0.1

0.35 Conditional probability of B Marginal probability of A Marginal probability of B

  • Calculate posterior probabilities

P(X|Ɛ) with Ɛ collected evidence Bayesian Inference (belief update)

[High, Mod, Low] [True, False]

slide-17
SLIDE 17

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Strengths and Weaknesses of Bayesian Networks

State of the art

17

  • Intuitive representation of the cause-effect relationships

among involved variables

  • Uncertainty about problem domain can be encoded

in a probabilistic interaction model

  • White box model permitting different possible analysis:

independence analysis, sensitivity analysis, value of information

  • Allows building knowledge from historical data in a quantitative way
  • Many efficient algorithms to learn Bayesian Network from data and fuse

with expert knowledge

Strengths Weaknesses

  • Requires an extensive set of data to derive both structure and

parameters

  • Uses a large number of probabilities in numerical form
  • Computational complexity grows exponentially with the

number of nodes

slide-18
SLIDE 18

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Motivation State of the art The model Use of the model

18

slide-19
SLIDE 19

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

High level ANS system – the economic view

The Model

19

ANS Delivery

(Quantitative & Qualitative Production) Capacity of ANS production Cost of ANS production Demand for ANS ANS Revenue ANS-related delays

Demand-capacity balancing Financial sustainability Inputs/Outputs relationships in the provision of ANS

  • Economies of scale:
  • Larger ANSPs tend to have

lower unit costs

  • Economies of density:
  • Additional traffic can be

absorbed using same resources

slide-20
SLIDE 20

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014 20

Building a Bayesian Network – Preliminary BN

Dataset Learn New Structure Preliminary Bayesian Network Expert judgment Knowledge from literature Structure Rules Final Bayesian Network Data driven process Model refinement

Data driven process

The Model ANSP ANSP Size ATFM Delay/fl-hr Flight Hours ATCO productivity Adjusted ATCO Cost/fl-hr Traffic Complexity

Local factors Controlled IFR traffic Traffic density*structural index Flight Hours / ATCO hours on-duty Area of airspace in km2 Area of airspace in km2 PPP adjusted hourly ATCO cost / Flight Hours

Power Constructor algorithm [Cheng,1998]

based on independence tests

slide-21
SLIDE 21

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Building a Bayesian Network – Final BN

The Model

21

Model refinement

  • Expert judgement – added value from team expertise
  • Knowledge from literature - gained from analysing the background of the project
  • Structure rules – absence of cycles and minimal number of variable’s parents to limit

computational complexity

ANSP ANSP Size ATFM Delay/fl-hr Flight Hours ATCO productivity Adjusted ATCO Cost/fl-hr Traffic Complexity

slide-22
SLIDE 22

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Scope of the model

The Model

22

Variable (Values per year and per ANSP) Data source ATFM Delay Performance Review Report Traffic Complexity Flight hours ATM Cost Efficiency report En Route ATCO hours on-duty En Route ATCO cost per hour adjusted by PPP

Air Navigation Service Providers Aena IAA ANS CR LFV Austro Control LPS Avinor LVNL Belgocontrol MATS BULATSA M-NAV Croatia Control MUAC DCAC Cyprus NATS DFS NAV Portugal

DSNA NAVIAIR EANS ROMATSA ENAV Skyguide HCAA Slovenia Control HungaroControl

Geographical scope: 27 ANSPs – En-route airspace only Temporal scope: 9 years

Annual values for 2003 - 2011

Performance data availability

slide-23
SLIDE 23

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

  • Discretization  Convert continuous variables into discrete ones

Probabilistic model - Discretization

The Model

23

Equivalent number of

  • bservations

3 States States’ limits

Example – Flight Hours node Definition

Observation frequency States distribution Greater sample dispersion for High State 5 biggest ANSPs in Europe

  • AENA
  • DFS
  • ENAV
  • NATS
  • DSNA
slide-24
SLIDE 24

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Motivation State of the art The model Use of the model

24

slide-25
SLIDE 25

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Use of the model – analyze and predict

Use of the model

25

Bayesian inference – computing the impact of observing values of a subset of the model variables on the probability distribution over the remaining variables.

  • Inputs are set as evidence for the desired nodes
  • The remaining variables, with no evidence set, are Outputs

Behavioral analysis of past performance

  • Getting insights on the complex relationships among factors affecting

performance, by applying Bayesian inference

1

  • Flight hours
  • Traffic Complexity
  • Adjusted ATCO

cost per flight hour Influence

  • f ANSP

size

  • ATFM Delay per

flight hour Influence

  • f Traffic

Complexity

  • ATFM Delay per

flight hour

  • Adjusted ATCO cost

per flight hour Influence

  • f ATCO

productivity

OUTPUT INPUT

Predictive model

  • Assess the probabilities of compliance with the performance targets imposed by

the Performance Scheme Regulation at:

  • European level
  • National level

2

slide-26
SLIDE 26

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

37% 11% 4% 13% 19% 35% 14% 20% 29% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% < 152465 fl-hr 152465 - 315942 fl- hr > 315942 fl-hr < 152465 fl-hr 152465 - 315942 fl- hr > 315942 fl-hr < 152465 fl-hr 152465 - 315942 fl- hr > 315942 fl-hr Flight hours in ANSP Size < 92800 km2 Flight hours in ANSP Size 92800 - 389895 km2 Flight hours in ANSP Size > 389895 km2

Adjusted ATCO cost/fl-hr probability of occurrence

> 82.24 €/fl-hr 62.45 - 82.24 €/fl-hr < 62.45 €/fl-hr Use of the model

26

Presence of economies of density and scale

P(High costs) ↓ P(High costs) ↑ P(High costs) ↑

Small ANSPs Medium ANSPs Large ANSPs

Traffic ↑ Traffic ↑ Traffic ↑

slide-27
SLIDE 27

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

0% 4% 43% 30% 89% 0% 70% 0% 0% 0% 0% 18% 0% 0% 100% 29% 100% 100% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% < 2.8 2.8 - 5.8 > 5.8 < 2.8 2.8 - 5.8 > 5.8 < 2.8 2.8 - 5.8 > 5.8 Traffic Complexity in ANSP Size < 92800 km2 Traffic Complexity in ANSP Size 92800 - 389895 km2 Traffic Complexity in ANSP Size > 389895 km2

Flight hours probability of occurrence

> 315942 fl-hr 152465 - 315942 fl-hr < 152465 fl-hr Use of the model

27

Traffic complexity vs number of Flight Hours

Traffic Complexity ↑

Small ANSPs Medium ANSPs Large ANSPs

Traffic Complexity ↑ Traffic Complexity ↑ P(High Traffic) ↑ P(High Traffic) ↑ P(High Traffic)↑↑

slide-28
SLIDE 28

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

17% 33% 50% 27% 34% 39% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% < 2.8 2.8 - 5.8 > 5.8 < 152465 152465 - 315942 > 315942 Traffic Complexity Flight Hours ATFM Delay/fl-hr probability of occurrnece < 0.07min/fl-hr 0.07 - 0.72 min/fl-hr > 0.72 min/fl-hr Use of the model

28

Traffic complexity vs ATFM delay

P(High delays) ↑

↑ ↑

P(High delays) ↑

slide-29
SLIDE 29

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Use of the model

29

Traffic complexity vs ATFM delay and Adjusted ATCO costs

Scenario 1 High flight hours and High ATCO productivity Small variation in costs Big increase in Delays Scenario 2 High flight hours and Low ATCO productivity Big increase in costs moderate increase in Delays

20% 11% 52% 8%

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% ATFM Delay/fl-hr Adjusted ATCO cost/fl- hr ATFM Delay/fl-hr Adjusted ATCO cost/fl- hr Traffic Complexity < 2.8 Traffic Complexity > 5.8

Probability of occurrence

High State Medium State Low State

10% 20% 24% 45%

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% ATFM Delay/fl-hr Adjusted ATCO cost/fl- hr ATFM Delay/fl-hr Adjusted ATCO cost/fl- hr Traffic Complexity < 2.8 Traffic Complexity > 5.8

Probability of occurrence

High State Medium State Low State

slide-30
SLIDE 30

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Use of the model

30

Traffic complexity vs ATFM delay and Adjusted ATCO costs per ANSP size

Presence of economies of scale:

1. Similar probabilities for delays BUT Higher probabilities of High costs for low complexity 2. Probabilities of high costs reduce as complexity increases

16% 34% 49% 18%

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

ATFM Delay/fl-hr Adjusted ATCO cost/fl- hr ATFM Delay/fl-hr Adjusted ATCO cost/fl- hr Traffic Complexity < 2.8 Traffic Complexity > 5.8

Probability of occurrence

High State Medium State Low State

16% 18% 52% 35%

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

ATFM Delay/fl-hr Adjusted ATCO cost/fl- hr ATFM Delay/fl-hr Adjusted ATCO cost/fl- hr Traffic Complexity < 2.8 Traffic Complexity > 5.8

Probability of occurrence

High State Medium State Low State

18% 18% 52% 27%

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

ATFM Delay/fl-hr Adjusted ATCO cost/fl- hr ATFM Delay/fl-hr Adjusted ATCO cost/fl- hr Traffic Complexity < 2.8 Traffic Complexity > 5.8

Probability of occurrence

High State Medium State Low State

Small ANSPs Medium ANSPs Large ANSPs

slide-31
SLIDE 31

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Predictive use of the model – new structure

Use of the model

31

Aim of this second application of the model assess the probabilities of compliance with the Performance Scheme targets Align the model with the performance indicators: ATFM Delay/flight and Unit Rate

Original structure New structure New variable ANSP ANSP Size ATFM Delay/flight Flights ATCO productivity Unit Rate Traffic Complexity Adjusted ATCO Cost/fl-hr

Flight Hours Flights ATFM Delay/fl-hr ATFM Delay/flight Unit Rate

slide-32
SLIDE 32

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Performance Scheme targets for the second Reference Period (2015-2019)

Use of the model

Performance Scheme

Establishes the targets to be accomplished by the European Members in the areas of:

  • Safety
  • Capacity
  • Cost efficiency
  • Environmental flight efficiency

ATFM Delay/flight Unit Rate 3 States High Medium Low 2 States High Low ATFM Delay/flight Low ≤ 0,5 min/flight < High Unit Rate Low ≤ 49,10 €/SU < High

KPA KPI Union wide targets for RP2 (2015-2019) Capacity Average En-route ATFM delay 0,5 minutes per flight to be reached for each calendar year Cost- efficiency DUC (Determined Unit Cost) 2015 2016 2017 2018 2019 56,64 54,95 52,98 51,00 49,10

New discretization

Change the discretization according to the target values in order to determine their probability of compliance

32

slide-33
SLIDE 33

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Probabilities of compliance with both targets

Use of the model

33

General European level Targets seem to be ambitious: 38% probabilities of compliance with both targets National level Notable differences among ANSPs

38% 36% 62% 64% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 0.5 min/fl < > 0.5 min/fl

Unit Rate probability of occurrence

ATFM Delay/flight

> 49.10 €/SU < 49.10 €/SU 70% 20% 33% 7% 65% 31% 43% 78% 32%31% 55% 29% 82% 85% 79% 45%49% 36% 21% 77% 71% 45%45% 32% 66% 35% 27%

0% 10% 20% 30% 40% 50% 60% 70% 80% 90%

slide-34
SLIDE 34

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014

Conclusions and next steps

Use of the model

34

  • Building knowledge from historical data in a quantitative way
  • An intuitive representation of the cause-effect relationships among

involved variables

  • Dealing with the stochastic nature of the underlying system in a natural

and direct way

  • Supporting decision making under uncertainty, e.g. when configuring

resources for an ANSP

  • Predicting future behavior based on past observations

The use of Bayesian Networks in ATM performance analysis allows: Overcoming the current limitations:

  • Increasing variables’ discretization levels trading off model complexity
  • Adding more variables to comprehensively represent all influences
slide-35
SLIDE 35

Assessing ATM Performance Interdependencies through Bayesian Networks ICRAT 2014, Istanbul - May 2014 35

Andrea Ranieri aranieri@alg-global.com

Advanced Logistics Group, S.A. Tánger 98, 3 08018 Barcelona Spain T +34 93 463 23 12 www.indra.es