[PPT] - MORP: Data-Driven Multi-Objective Route Planning and Optimization PowerPoint Presentation

SLIDE 1

MORP: Data-Driven Multi-Objective Route Planning and Optimization for Electric Vehicles

Ankur Sarker, Haiying Shen, and John A. Stankovic

Department of Computer Science, University of Virginia Charlottesville, Virginia, USA

SLIDE 2

Introduction
System Design
Performance Evaluation
Conclusion

Outline

2

SLIDE 3

3

Introduction

Wireless power transfer system

The Online Electric Vehicle (OLEV) is

an electric vehicle that charges wirelessly while moving using electromagnetic induction.

SLIDE 4

4

Introduction

Wireless power transfer system

The Online Electric Vehicle (OLEV) is

an electric vehicle that charges wirelessly while moving using electromagnetic induction.

The Korean Advanced Institute of

Science and Technology (KAIST) developed first recharging road on March 9, 2010.

SLIDE 5

5

Introduction

Wireless power transfer system

Long Queue

SLIDE 6

6

Introduction

Wireless power transfer system

Long Queue Time-Consuming

SLIDE 7

7

Introduction

Wireless power transfer system

Long Queue Time-Consuming Range Anxiety

SLIDE 8

8

Introduction

Wireless power transfer system

Long Queue Time-Consuming Range Anxiety Maintain SoC

SLIDE 9

9

Introduction

Wireless power transfer system

A B RSU EVi EVk RSU: Road Side Unit EV: Electrical Vehicle

SLIDE 10

10

Introduction

Wireless power transfer system

A B Charging sections RSU EVi EVk RSU: Road Side Unit EV: Electrical Vehicle

SLIDE 11

11

Introduction

Wireless power transfer system

A B Charging sections RSU EVi EVk RSU: Road Side Unit EV: Electrical Vehicle

For a given EV driving from a source to a destination, how to choose a route so that:

SLIDE 12

12

Introduction

Wireless power transfer system

A B Charging sections RSU EVi EVk RSU: Road Side Unit EV: Electrical Vehicle

For a given EV driving from a source to a destination, how to choose a route so that: i) Arrive at its destination with sufficient power supply on the way

SLIDE 13

13

Introduction

Wireless power transfer system

A B Charging sections RSU EVi EVk RSU: Road Side Unit EV: Electrical Vehicle

For a given EV driving from a source to a destination, how to choose a route so that: i) Arrive at its destination with sufficient power supply on the way ii) Consider current traffic flow and minimize: the driver’s range anxiety

SLIDE 14

14

Introduction

Wireless power transfer system

A B Charging sections RSU EVi EVk RSU: Road Side Unit EV: Electrical Vehicle

For a given EV driving from a source to a destination, how to choose a route so that: i) Arrive at its destination with sufficient power supply on the way ii) Consider current traffic flow and minimize: the driver’s range anxiety, charging monetary cost

SLIDE 15

15

Introduction

Wireless power transfer system

A B Charging sections RSU EVi EVk RSU: Road Side Unit EV: Electrical Vehicle

For a given EV driving from a source to a destination, how to choose a route so that: i) Arrive at its destination with sufficient power supply on the way ii) Consider current traffic flow and minimize: the driver’s range anxiety, charging monetary cost, travel time

SLIDE 16

16

Introduction

Wireless power transfer system

A B Charging sections RSU EVi EVk RSU: Road Side Unit EV: Electrical Vehicle

For a given EV driving from a source to a destination, how to choose a route so that: i) Arrive at its destination with sufficient power supply on the way ii) Consider current traffic flow and minimize: the driver’s range anxiety, charging monetary cost, travel time, and energy consumption

SLIDE 17

17

Introduction

State-of-the-Art

IEEE TSG’12 IEEE TPS’14 IEVC’14 IEEE TSG’14 IEEE TPD’13 IEEE TPS’12 IEEE TPS’14 Plug-in charging station

SLIDE 18

18

Introduction

State-of-the-Art

IEEE TSG’12 IEEE TPS’14 IEVC’14 IEEE TSG’14 IEEE TPD’13 IEEE TPS’12 IEEE TPS’14 Plug-in charging station Wireless power transfer IEEE Systems Journal’16 Annals of Physics’08 IEEE ICDCS’17 ICPP’16

SLIDE 19

19

Introduction

State-of-the-Art

Wireless power transfer IEEE Systems Journal’16 Annals of Physics’08 IEEE ICDCS’17 ICPP’16

1

Not applicable for dynamic wireless charging

SLIDE 20

20

Introduction

State-of-the-Art

1

Not applicable for dynamic wireless charging Cannot maintain the SoC of vehicles in a metropolitan road network

2

SLIDE 21

Introduction
System Design
Performance Evaluation
Conclusion

Outline

21

SLIDE 22

22

System Design

verview

Data Cleaning Traffic Data Gridded Roadmap

1

SLIDE 23

23

System Design

verview

Temporal Analysis Data Cleaning Traffic Data Gridded Roadmap

1

SLIDE 24

24

System Design

verview

Temporal Analysis Spatial Analysis Data Cleaning Traffic Data Gridded Roadmap

1

SLIDE 25

25

System Design

verview

Temporal Analysis Spatial Analysis Spatio- Temporal Correlation Data Cleaning Traffic Data Gridded Roadmap

1

SLIDE 26

26

System Design

verview

Temporal Analysis Spatial Analysis Spatio- Temporal Correlation Data Cleaning Traffic Data Gridded Roadmap

1

Traffic Prediction

2

SLIDE 27

27

System Design

verview

Temporal Analysis Spatial Analysis Spatio- Temporal Correlation Data Cleaning Traffic Data Gridded Roadmap

1

Traffic Prediction Velocity Prediction

2

SLIDE 28

28

System Design

verview

Temporal Analysis Spatial Analysis Spatio- Temporal Correlation Data Cleaning Traffic Data Gridded Roadmap

1

Traffic Prediction Velocity Prediction Power Consumption

2

SLIDE 29

29

System Design

verview

Temporal Analysis Spatial Analysis Spatio- Temporal Correlation Data Cleaning Traffic Data Gridded Roadmap

1

Traffic Prediction Velocity Prediction Power Consumption

2

Objective Functions

3

SLIDE 30

30

System Design

verview

Temporal Analysis Spatial Analysis Spatio- Temporal Correlation Data Cleaning Traffic Data Gridded Roadmap

1

Traffic Prediction Velocity Prediction Power Consumption

2

Objective Functions Multi- Objective Route Planning

3

SLIDE 31

31

System Design

Traffic data analysis

We collected 212 consecutive day- long historical hourly traffic flow data:

SLIDE 32

32

System Design

Traffic data analysis

We collected 212 consecutive day- long historical hourly traffic flow data:

1. From December 1, 2016 to June

30, 2017

SLIDE 33

33

System Design

Traffic data analysis

We collected 212 consecutive day- long historical hourly traffic flow data:

1. From December 1, 2016 to June

30, 2017

2. 20 locations in 3 interstate routes,

9 US routes, and 6 state routes

SLIDE 34

34

System Design

Traffic counts prediction

Traffic locations with historical traffic data
Traffic locations without historical traffic data

SLIDE 35

35

System Design

Traffic counts prediction

Traffic locations with historical traffic data

SLIDE 36

36

System Design

Traffic counts prediction

Traffic locations with historical traffic data

Hourly traffic counts

SLIDE 37

37

System Design

Traffic counts prediction

Traffic locations with historical traffic data

Correlation of two locations

SLIDE 38

38

System Design

Traffic counts prediction

Traffic locations with historical traffic data

Weekly traffic counts

SLIDE 39

39

System Design

Traffic counts prediction

Traffic locations with historical traffic data

Change rate of traffic counts

SLIDE 40

40

System Design

Traffic counts prediction

Traffic locations with historical traffic data

Horizontal spatio-temporal autoregressive integrated moving average model

SLIDE 41

41

System Design

Traffic counts prediction

Traffic locations with historical traffic data

Horizontal spatio-temporal autoregressive integrated moving average model

SLIDE 42

42

System Design

Traffic counts prediction

Traffic locations with historical traffic data

Horizontal spatio-temporal autoregressive integrated moving average model

Autoregressive term w.r.t. day

SLIDE 43

43

System Design

Traffic counts prediction

Traffic locations with historical traffic data

Horizontal spatio-temporal autoregressive integrated moving average model

Autoregressive term w.r.t. day Moving average term w.r.t. day

SLIDE 44

44

System Design

Traffic counts prediction

Traffic locations with historical traffic data

Horizontal spatio-temporal autoregressive integrated moving average model

Autoregressive term w.r.t. day Moving average term w.r.t. day Autoregressive term w.r.t. location

SLIDE 45

45

System Design

Traffic counts prediction

Traffic locations with historical traffic data

Horizontal spatio-temporal autoregressive integrated moving average model

Autoregressive term w.r.t. day Moving average term w.r.t. day Autoregressive term w.r.t. location Moving average term w.r.t. location

SLIDE 46

46

System Design

Traffic counts prediction

Traffic locations with historical traffic data
Traffic locations without historical traffic data

SLIDE 47

47

System Design

Traffic counts prediction

Traffic locations without historical traffic data

Spatio-temporal ordinary-kriging

SLIDE 48

48

System Design

Traffic counts prediction

Traffic locations without historical traffic data

Spatio-temporal ordinary-kriging

SLIDE 49

49

System Design

Traffic counts prediction

Traffic locations without historical traffic data

Spatio-temporal ordinary-kriging

Weights chosen to minimize the prediction error variance

SLIDE 50

50

System Design

Traffic counts prediction

Traffic locations without historical traffic data

Spatio-temporal ordinary-kriging

Weights chosen to minimize the prediction error variance Vehicle counts xi at time ti

SLIDE 51

51

System Design

Traffic counts prediction

Traffic locations without historical traffic data

Spatio-temporal ordinary-kriging

Weights chosen to minimize the prediction error variance Vehicle counts xi at time ti

SLIDE 52

52

System Design

Traffic counts prediction

Traffic locations without historical traffic data

Spatio-temporal ordinary-kriging

Weights chosen to minimize the prediction error variance Vehicle counts xi at time ti Original mean Random quality with mean zero

SLIDE 53

53

System Design

Velocity prediction

Relationship among free-flow velocity, average velocity, and traffic counts

SLIDE 54

54

System Design

Velocity prediction

Relationship among free-flow velocity, average velocity, and traffic counts

Spatio-temporal ordinary-kriging Hourly average velocity

SLIDE 55

55

System Design

Velocity prediction

Relationship among free-flow velocity, average velocity, and traffic counts

Spatio-temporal ordinary-kriging Weekly average velocity

SLIDE 56

56

System Design

Velocity prediction

Relationship among free-flow velocity, average velocity, and traffic counts

Spatio-temporal ordinary-kriging Average velocity distribution

SLIDE 57

57

System Design

Velocity prediction

Relationship among free-flow velocity, average velocity, and traffic counts

Free flow velocity

SLIDE 58

58

System Design

Velocity prediction

Relationship among free-flow velocity, average velocity, and traffic counts

Free flow velocity Function of traffic counts

SLIDE 59

59

System Design

Velocity prediction

Relationship among free-flow velocity, average velocity, and traffic counts

Free flow velocity Function of traffic counts

SLIDE 60

60

System Design

Velocity prediction

Relationship among free-flow velocity, average velocity, and traffic counts

Free flow velocity Function of traffic counts Kernel regression

SLIDE 61

61

System Design

Power consumption prediction

Given EV 𝑙’s battery power 𝑄batt,𝑙(𝑢 − 1) at the time (𝑢−1), its battery power at time 𝑢 is calculated by:

SLIDE 62

62

System Design

Power consumption prediction

Stored battery power

Given EV 𝑙’s battery power 𝑄batt,𝑙(𝑢 − 1) at the time (𝑢−1), its battery power at time 𝑢 is calculated by:

SLIDE 63

63

System Design

Power consumption prediction

Stored battery power

Given EV 𝑙’s battery power 𝑄batt,𝑙(𝑢 − 1) at the time (𝑢−1), its battery power at time 𝑢 is calculated by:

Added battery power

SLIDE 64

64

System Design

Power consumption prediction

Stored battery power

Given EV 𝑙’s battery power 𝑄batt,𝑙(𝑢 − 1) at the time (𝑢−1), its battery power at time 𝑢 is calculated by:

Added battery power Power required to drive

SLIDE 65

65

System Design

Power consumption prediction

Stored battery power

Given EV 𝑙’s battery power 𝑄batt,𝑙(𝑢 − 1) at the time (𝑢−1), its battery power at time 𝑢 is calculated by:

Added battery power Power required to drive EV properties, velocity, and acceleration

SLIDE 66

66

System Design

Multi-objective optimization

We can formulate the following multi-objective optimization problem: min(f1, f2, f3, f4) Where: f1 – Power consumption of EVs f2 – Travel time of EVs f3 – Charging monetary cost of EVs f4 – Range anxiety of EVs (more details in the paper)

SLIDE 67

67

System Design

Multi-objective optimization

Adaptive epsilon constraint method:

SLIDE 68

68

System Design

Multi-objective optimization

Adaptive epsilon constraint method:

Alternates the multi-objective optimization

problem into a single hyper-grid searching problem

SLIDE 69

69

System Design

Multi-objective optimization

Adaptive epsilon constraint method:

Alternates the multi-objective optimization

problem into a single hyper-grid searching problem

Turns into a 3-dimensional hyper-grid

searching problem

SLIDE 70

70

System Design

Multi-objective optimization

Adaptive epsilon constraint method:

Alternates the multi-objective optimization

problem into a single hyper-grid searching problem

Turns into a 3-dimensional hyper-grid

searching problem

The grid cell coordinates are identified by

the solutions of these objective functions

SLIDE 71

71

System Design

Multi-objective optimization

Adaptive epsilon constraint method:

Alternates the multi-objective optimization

problem into a single hyper-grid searching problem

Turns into a 3-dimensional hyper-grid

searching problem

The grid cell coordinates are identified by

the solutions of these objective functions

(more details in the paper)

SLIDE 72

72

System Design

Multi-objective optimization

SLIDE 73

73

System Design

Multi-objective optimization

SLIDE 74

74

System Design

Multi-objective optimization

SLIDE 75

75

System Design

Multi-objective optimization

SLIDE 76

Introduction
System Design
Performance Evaluation
Conclusion

Outline

76

SLIDE 77

77

Evaluations

Experimental settings

1. Energy consumption rate: Chevrolet Spark EV

❑ Based on the parameters of Chevrolet Spark EV

SLIDE 78

78

Evaluations

Experimental settings

1. Energy consumption rate: Chevrolet Spark EV

❑ Based on the parameters of Chevrolet Spark EV

2. Multi-objective route planning

❑ Using OpenStreetmap and SUMO traffic simulator ❑ NetworkX ❑ 15 charging sections ❑ 10 EVs ❑ 2 other existing methods

SLIDE 79

79

Evaluations

Experimental settings

1. Energy consumption rate: Chevrolet Spark EV

❑ Based on the parameters of Chevrolet Spark EV

2. Multi-objective route planning

❑ Using OpenStreetmap and SUMO traffic simulator ❑ NetworkX ❑ 15 charging sections ❑ 10 EVs ❑ 2 other existing methods

3. Vehicle count prediction

❑ Based on the collected data from SC DOT ❑ 20 different traffic locations ❑ 212 days

SLIDE 80

80

Evaluations

Experimental settings

1. Energy consumption rate: Chevrolet Spark EV

❑ Based on the parameters of Chevrolet Spark EV

2. Multi-objective route planning

❑ Using OpenStreetmap and SUMO traffic simulator ❑ NetworkX ❑ 15 charging sections ❑ 10 EVs ❑ 2 other existing methods

3. Vehicle count prediction

❑ Based on the collected data from SC DOT ❑ 20 different traffic locations ❑ 212 days

4. Average velocity prediction

❑ Using Kernel regression

SLIDE 81

81

Evaluations

EV energy model

Metric: Energy consumption rate Observation: Energy consumption is lower during deacceleration Reason: Due to the deacceleration force

SLIDE 82

82

Evaluations

Traffic counts prediction

Metric: Traffic counts Observation: Predicted value is close to actual value Reason: It can considers the spatio-temporal relations horizontally

Actual and predicted traffic counts

SLIDE 83

83

Evaluations

Multi-objective route planning

Metric: Charging monetary cost Observation: Charging monetary cost is lower Reason: Proposed method satisfies charging monetary cost constraints Metric: Range anxiety Observation: Range anxiety is lower Reason: It can also satisfy charging monetary cost constraints

SLIDE 84

84

Evaluations

Multi-objective route planning

Metric: Energy consumption Observation: Energy consumption is lower in EcoDrive Reason: EcoDrive does not satisfy all four objective functions together Metric: Travel time Observation: Travel time is lower in the proposed method Reason: Same

SLIDE 85

Introduction
System Design
Performance Evaluation
Conclusion

Outline

85

SLIDE 86

86

Conclusion

1. We predicted traffic counts and average velocity
2. We proposed a multi-objective route planner for OLEVs
3. We evaluated the proposed system using real traffic data

SLIDE 87

87

Conclusion

1. We predicted traffic counts and average velocity
2. We proposed a multi-objective route planner for OLEVs
3. We evaluated the proposed system using real traffic data

Future work

1. Devise more sophisticated regression models in traffic

predictions

2. Build an online optimization framework based on more complex

traffic scenarios

SLIDE 88

88

Thank you! Questions & Comments?

Ankur Sarker as4mz@Virginia.edu Ph.D. Candidate Pervasive Communication Laboratory University of Virginia