Machine Learning & AI in Transport and Logistics

Transcription

1 Machine Learning & AI in Transport and Logistics Frank Salliau & Sven Verstrepen Logistics Meets Innovation Vlerick Brussels Nov. 15th 2017

2 Sci-fi in 2002

3

4 Reality in 2017

5

6 Sci-fi in 1984

7

8 Reality in 2017

9

10

11

12 Alexa, check my calendar Alexa, what is machine learning? Alexa, play Spotify

13

14 What is it all about?

15

16

17 What is Machine Learning? The science of getting computers to act without being explicitly programmed - Andrew Ng (Stanford/DeepLearning AI)

18 Challenge Recognize dogs in images

19 What a human sees DOG NOT A DOG

20 What a computer sees??????

21 What do we need? TRAINING DATA

22 Training phase DOG (1) NOT DOG (0) Labeled training set (dog/not dog) > 1000 images Untrained Neural Network

23 Prediction phase 87% DOG 13% NOT DOG Unlabeled image Trained Neural Network

24 Why is this booming now?

25

26 What are the drivers?

27 (BIG) DATA

28 Byte : one grain of rice Byte Credits: David Wellman,

29 Byte Kilobyte : one grain of rice : cup of rice Kilobyte

30 Byte : one grain of rice Kilobyte : cup of rice Megabyte : 8 bags of rice Megabyte

31 Byte : one grain of rice Kilobyte : cup of rice Megabyte : 8 bags of rice Gigabyte : 3 Semi trucks Gigabyte

32 Byte : one grain of rice Kilobyte : cup of rice Megabyte : 8 bags of rice Gigabyte : 3 Semi trucks Terabyte : 2 Container Ships Terabyte

33 Byte : one grain of rice Kilobyte : cup of rice Megabyte : 8 bags of rice Gigabyte : 3 Semi trucks Terabyte : 2 Container Ships Petabyte : Blankets Manhattan Petabyte

34 Byte : one grain of rice Kilobyte : cup of rice Megabyte : 8 bags of rice Gigabyte : 3 Semi trucks Terabyte : 2 Container Ships Petabyte : Blankets Manhattan Exabyte : Blankets west coast states One Exabyte Byte

35 Byte : one grain of rice Kilobyte : cup of rice Megabyte : 8 bags of rice Gigabyte : 3 Semi trucks Terabyte : 2 Container Ships Petabyte : Blankets Manhattan Exabyte : Blankets west coast states Zettabyte : Fills the Pacific Ocean Zettabyte

36 Byte : one grain of rice Kilobyte : cup of rice Megabyte : 8 bags of rice Gigabyte : 3 Semi trucks Terabyte : 2 Container Ships Petabyte : Blankets Manhattan Exabyte : Blankets west coast states Zettabyte : Fills the Pacific Ocean Yottabyte : A EARTH SIZE RICE BALL! Yottabyte

37 Byte : one grain of rice Kilobyte : cup of rice Megabyte : 8 bags of rice Gigabyte : 3 Semi trucks Terabyte : 2 Container Ships Petabyte : Blankets Manhattan Exabyte : Blankets west coast states Zettabyte : Fills the Pacific Ocean Yottabyte : A EARTH SIZE RICE BALL! Hobbyist Desktop Internet Big Data The Future

38 1 Yottabyte 1 Xenottabyte 1 Shilentnobyte 1 Domegemegrottebyte 1 Icosebyte 1 Monoicosebyte

39 Where does all this data come from?

40

41 The Power of the Crowd

42

43

44

45

46

47

48

49 Machine Learning Supervised Learning Unsupervised Learning Reinforcement Learning Deep Learning Deep Reinforcement Learning

50 Supervised Learning

51 Regression

52 Classification

53 Deep Learning

54

55

56 Google Deep Dream (sometimes nightmare)

57

58

59 Unsupervised Learning

60

61 Reinforcement Learning

62

63

64

65 Getting Started

66 CRISP Methodology Exploratory Iterative

67

68 Quality of data: Garbage In Garbage Out

69 Multi-disciplinary team

70

71

72 D3.js - Open Source

73 Examples in Transport & Logistics

74 IBM Watson

75 Ahlers Supply Network Innovation & Analytics (ASNIA) Using ML to identify horizontal collaboration synergies between multiple shipper networks

76 Transmetrics: replacing budgets with prediction Transmetrics big data cargo platform a rigorous approach deriving benefits from current and future data Data uptake Demand AI Execution cleansing forecast optimizatio Controlling and modelling n enrichment Traditional tactical management relies on budgets Transmetrics: predictive tactical management Budgeting process: Flying on paper maps, experience and gut feeling Transmetrics: High precision flying assisted by data and AI

77 Data quality improvement status of data after AUTOMATED / AI processing Preliminary data quality assessment after AUTOMATED / AI processing Challenging Good to very good Original situation Ambiguity in interpretation of orders and pieces records Missing size information at order and piece level Mix of measurement units (ldm, m 3, kilos, pallets) Missing or unreliable capacity information for linehauls Other issues (e.g. senders with multiple name spellings) Achieved situation Clear and correct identification of all shipments and pieces (including ignoring of non-piece lines) For each piece calculated full set of measurements (height, width, length, weight, volume) which enables 3D loading factors! Complete set of measurements: loading meters, surface/pallets (m 2 ), volume (m 3 ) Built AI algorithm to estimate missing capacity information Grouped senders by AI; among others found a customer with 330+ different accounts/names

78 Methodology / Behind the scenes 1. Preprocess data cleanse and disaggregate pieces/volume/weight Remove unnecessary rows Ensure one row=one piece 2. Estimate volume Predict missing volume data 3. Estimate dimensions Mine text fields, Apply industry standards and business rules Approaches used Combinatorial optimization Quadratic optimization Gradient boosting trees and other regressions Natural language processing Named identity disambiguation Expert input / business rules 4. Predict missing linehaul capacity use historical data to predict capacities 95% accuracy 3D Loading Factor estimation Loading meters (m) Loading area, pallets (m 2 ) Volume (m 3 ) Expected 90% accuracy after calibration

79 3D loading visualization

80 Loading factor optimization results Real situation (as observed on the warehouse floor) Loading factor calculated by using the data Loading meters 13.6 m 100% Floor 30.4 m 2 91% Volume 29.8 m 3 34%

81 Sven Verstrepen Head of Supply Network Innovation & Analytics

82 Frank Salliau Independent Data Scientist & Machine Learning Expert