By Roberto Venturini - - - PowerPoint PPT Presentation

▶
by roberto venturini https www flickr com photos robven
SMART_READER_LITE
LIVE PREVIEW

By Roberto Venturini - - - PowerPoint PPT Presentation

Jul 29, 2023 144 likes •540 views

1 By Roberto Venturini - https://www.flickr.com/photos/robven/1953413479, CC BY 2.0, https://commons.wikimedia.org/w/index.php?curid=57831577 Gerrit-Jan de Bruin Supervision by Jasper van Vliet M.Sc. en dr. Johan Westerhuis Efficient compliance

slide-1
SLIDE 1

By Roberto Venturini - https://www.flickr.com/photos/robven/1953413479, CC BY 2.0, https://commons.wikimedia.org/w/index.php?curid=57831577 1

slide-2
SLIDE 2

Efficient compliance monitoring:

Comparison of both airborne and landside sniffing and spectrometric methods to provide direct control on the sulfur emission of ships.

Gerrit-Jan de Bruin Supervision by Jasper van Vliet M.Sc. en dr. Johan Westerhuis

slide-3
SLIDE 3

Contents

 Introduction  Aim  Analytical techniques  Statistical techniques  Classification with linear boundary  Classification using Z-score  EM algorithm  Outlook

Efficient compliance monitoring 3

slide-4
SLIDE 4

The emission of SO𝟑 over time.

4

50 100 150 200 250 1990 1995 2000 2005 2010 2013 2014 2015 SO₂ (kton)

SO₂ emissions in the Netherlands

Total Transport Shipping

Efficient compliance monitoring Introduction

60 000 premature deaths, Corbett 2 year loss, CAFE

slide-5
SLIDE 5

5

1 2 3 4 5 01-2010 01-2012 01-2014 01-2016 01-2018 01-2020

FSC [% (m/m)]

Maximum allowed FSC

Within SECA Global

Left: Image courtesy of D.J. Oostwoud Wijdenes and National Geographic Society. Efficient compliance monitoring Introduction

$ 40 000 dayˉ¹

slide-6
SLIDE 6

Fuel Sulfur Content

 FSC = weight of sulphur

weight of fuel

 FSC = 16 64.066×𝑁 S × ׬ SO2 − SO2 bg 𝑒𝑢 12 44×

Τ 𝑁 C 0.87 ×׬ CO2 − CO2 bg 𝑒𝑢

 FSC = 0.232

׬ SO2 − SO2 bg 𝑒𝑢 ׬ CO2 − CO2 bg 𝑒𝑢

6 Efficient compliance monitoring Introduction Image courtesy of ILT.

slide-7
SLIDE 7

Aim

 Compare different techniques and operators for future use for

the inspectorate.

 Explore the measurements performed so far by all inspectorates

in Northern Europe.

 What are the compliance rates?  What are the type I and type II errors? I.e. how sure are we that

a ship is (non-)compliant?

7 Efficient compliance monitoring Introduction

slide-8
SLIDE 8

8 Image courtesy: ILT Efficient compliance monitoring Introduction

slide-9
SLIDE 9

9 Image courtesy: ILT Efficient compliance monitoring Introduction

slide-10
SLIDE 10

TNO/ ILT sniffer

10 Efficient compliance monitoring Analytical instrument Image courtesy: ILT

slide-11
SLIDE 11

BSH, 3564 Denmark, 354 DFDS-Maersk, 10 Explicit, 327 ILT, 743 MUMM, 1390 TNO, 1661

15

N = 8049

Efficient compliance monitoring Campaigns

slide-12
SLIDE 12

What fraction is non-compliant?

16

200 400 600 800 2000 4000 6000 8000 < -0.1 0.1 0.2 > 0.3

FSC (% m/m) Count Cumulative count

Efficient compliance monitoring Campaigns

slide-13
SLIDE 13

What fraction is non-compliant?

17

200 400 600 800 2000 4000 6000 8000 < -0.1 0.1 0.2 > 0.3

FSC (% m/m) Count Cumulative count

Efficient compliance monitoring Campaigns

slide-14
SLIDE 14

What fraction is non-compliant?

18

Classification True value

7 4 6 2

N = 19 Accuracy = 47%

Efficient compliance monitoring Campaigns

slide-15
SLIDE 15

Intermezzo – type I and type II errors

19

Classification True value

7 4 6 2

N = 19 Accuracy = 47% Type 1 Type 2

Efficient compliance monitoring Intermezzo

slide-16
SLIDE 16

Intermezzo – type I and type II errors

20

Classification True value

7 4 6 2

N = 19 Accuracy = 47% Type 1: wrongly accusing Type 2: overlooking non-compliance

Efficient compliance monitoring Intermezzo

slide-17
SLIDE 17

Intermezzo – type I and type II errors

 What do we want?

21

Low type-I error High type-II error Equal type-I and type-II errors High type-I error Low type-II error Court

Efficient compliance monitoring Intermezzo

slide-18
SLIDE 18

Intermezzo – type I and type II errors

 What do we want?

22

Court Preselection Low type-I error High type-II error Equal type-I and type-II errors High type-I error Low type-II error

Efficient compliance monitoring Intermezzo

slide-19
SLIDE 19

Intermezzo – type I and type II errors

 What do we want?

23

Court Climate modeling Preselection Low type-I error High type-II error Equal type-I and type-II errors High type-I error Low type-II error

Efficient compliance monitoring Intermezzo

slide-20
SLIDE 20

Intermezzo – type I and type II errors

 What do we want?

24 Efficient compliance monitoring Intermezzo

Type I error Type II error

slide-21
SLIDE 21

What fraction is non-compliant?

25

200 400 600 800 2000 4000 6000 8000 < -0.1 0.1 0.2 > 0.3

FSC (% m/m) Count Cumulative count

Low type II error Low type I error

Efficient compliance monitoring Campaigns

slide-22
SLIDE 22

Z-score

 𝐼0: The ship has a FSC of 0.1 wt. % or less.  𝐼1: The ship has a higher FSC than 0.1 wt. %.  𝑨 =

ҧ 𝑦−𝜈0 Τ 𝑡𝑦 𝑜

 Z-score can be calculated to p-value with a significance level

26

N = 5552 (69%)

Efficient compliance monitoring Campaigns

slide-23
SLIDE 23

Z-score with 𝜷 = 𝟏. 𝟏𝟔

27

9 % 91 %

Efficient compliance monitoring Campaigns

slide-24
SLIDE 24

Z-score with 𝜷 = 𝟏. 𝟏𝟔

28

Classification True value

11

4 2 2

N = 19 Accuracy = 68%

Efficient compliance monitoring Campaigns

slide-25
SLIDE 25

Another approach

30 Efficient compliance monitoring Another approach

slide-26
SLIDE 26

What fraction is non-compliant?

31

 How many port state controls should take place?  How reliable are climate modellings assuming 100% compliance?  What is the catch rate?

Efficient compliance monitoring Another approach

slide-27
SLIDE 27

EM-algorithm

32 Efficient compliance monitoring Another approach

slide-28
SLIDE 28

EM algorithm

 Guess initial parameters  Calculate responsibility  Maximize likelihood of all parameters

33 Efficient compliance monitoring Another approach

slide-29
SLIDE 29

34

𝛿𝑗,0 = 1 𝛿𝑗,1 = 0 𝛿𝑗,0 = 0.5 𝛿𝑗,1 = 0.5 𝛿𝑗,0 = 0 𝛿𝑗,1 = 1

𝛿𝑗,0 + 𝛿𝑗,1 = 1 For each datapoint i

Efficient compliance monitoring Another approach

slide-30
SLIDE 30

EM algorithm

 Guess initial parameters  Calculate responsibility  Maximize likelihood of all parameters

ෞ 𝜈𝑙 = 1 𝑜𝑙 ෍

𝑗∈𝑙 𝑜𝑙

𝑦𝑗 ෞ 𝜏𝑙 = 1 𝑜𝑙 ෍

𝑗∈𝑙 𝑜𝑙

𝑦𝑗 − 𝜈𝑙 2

35 Efficient compliance monitoring Another approach

slide-31
SLIDE 31

EM algorithm

 Guess initial parameters  Calculate responsibility  Maximize likelihood of all parameters

ෞ 𝜈𝑙 = 1 𝑜𝑙 ෍

𝑗∈𝑙 𝑜𝑙

𝑦𝑗 ෞ 𝜏𝑙 = 1 𝑜𝑙 ෍

𝑗∈𝑙 𝑜𝑙

𝑦𝑗 − 𝜈𝑙 2

36 Efficient compliance monitoring Another approach

slide-32
SLIDE 32

EM algorithm

 Guess initial parameters  Calculate responsibility  Maximize likelihood of all parameters

Iterate until convergence

37 Efficient compliance monitoring Another approach

slide-33
SLIDE 33

EM-algorithm

38

N = 5552 (69%) 𝜈1 = 0.06 wt−% 𝜏1 = 0.04 wt−% 𝜈2 = −1.1 wt−% 𝜏2 = 0.8 wt−%

96 % 4 %

Efficient compliance monitoring Another approach

slide-34
SLIDE 34

What fraction is non-compliant?

39

200 400 600 800 2000 4000 6000 8000 < -0.1 0.1 0.2 > 0.3

FSC (% m/m) Count Cumulative count

Efficient compliance monitoring Another approach

slide-35
SLIDE 35

EM algorithm

 Guess initial parameters  Calculate responsibility

ෞ 𝛿𝑗,𝑙 = ฏ 𝜌𝑙 prior 𝒪(𝑦𝑗| ෞ 𝜈𝑙, ෞ 𝜏𝑙

2)

likelihood 𝜌1𝒪(𝑦𝑗| ෞ 𝜈1, ෞ 𝜏1

2) + 𝜌2Lognormal(𝑦𝑗 − 0.1| ෞ

𝜈2, ෞ 𝜏2

2)

evidence

 Maximize likelihood

40 Efficient compliance monitoring Another approach

slide-36
SLIDE 36

ෞ 𝜈𝑑 = σ𝑗∈𝑑

𝑜𝑙 𝑦𝑗

𝑂𝑑 ෞ 𝜏𝑑 = σ𝑗∈𝑑

𝑜𝑙 𝑦𝑗 − 𝜈𝑑 2

𝑂𝑑

Type your footer here 41

ෞ 𝜈𝑜𝑑 = 1 𝑂𝑜𝑑 ෍

𝑗∈𝑜𝑑 𝑜𝑙

log 𝑦𝑗 − 0.1 ෞ 𝜏𝑜𝑑 = 1 𝑂𝑜𝑑 ෍

𝑗∈𝑜𝑑 𝑜𝑙

log 𝑦𝑗 − 0.1 − 𝜈𝑜𝑑 2

slide-37
SLIDE 37

42 Efficient compliance monitoring Another approach

slide-38
SLIDE 38

43

3 % 97 %

Efficient compliance monitoring Another approach

slide-39
SLIDE 39

Outlook

 Determine the relation between type I and type II errors more

precisely.

 Better instruments will result in better accuracy.  Better validation makes the introduction of supervised methods

possible.

44 Efficient compliance monitoring Outlook