Transportation TechnologyR&D Center

Route-Based Energy Management for

PHEVs: A Simulation Framework for

Large-Scale Evaluation

Dominik Karbowski, Namwook Kim, Aymeric Rousseau

Argonne National Laboratory, USA

Optimal Energy Management of xEVs Needs Trip

Prediction

2

GPS Sensors

Cloud Computing

On-board Computing

Geographical Information

Live Traffic Situation

0 10 20 300

20

40

60

Milesm

ph Trip prediction

Increased connectivity and increased availability of data opens the door to trip prediction

Vehicle energy use can be reduced through application of control theory or fine tuning:• Dynamic Programming (DP): finds the global optimum for the command law• Instantaneous optimization:

o ECMS: Equivalent Minimization Consumption Strategyo PMP: Pontryagin Minimization Principle

• All techniques require knowledge of the trip!

Our Vision for Route-Based Control

3

0 10 20 300

20

40

60

Miles

mph

Itinerary Computation

Driver’s Input

Pattern Recognition

OR

GPS

Live Traffic

Route Prediction

Optimal Energy MgmtDestination

Current Position

Average traffic speed

Detailed Segment-by-Segment Information

Speed & Grade

Route-based Optimization

Optimal Control

Scope of Argonne’s Research

• Original research on speed prediction, an often overlooked problem

• Research on implementable solutions for route-based control

• Evaluation of real-world benefits of route-based control

Speed Prediction

4

Future Speed: Stochastic or Deterministic? Both!

5

Impossible to know the exact future speed profile: driving is not deterministic!

0 20 40 60 80 100-2.5

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

2.5

Speed [km/h]

Accel. [

m/s

2]

DETERMINISTIC

the driver follows an itinerary selected at the beginning of the trip

OR the driver selects an itinerary in the navigation unit, and follows directions

STOCHASTIC

Free flow: natural variations (accelerations vary, not always same speed)

Interactions with other cars & environment

But not completely random either…

Maps / GIS Can Provide Information About a Given

Itinerary

6

– Traffic pattern speed: average traffic speed for a given time/day

– Road slope: modeled with splines, not simply from GPS altitude data

– Speed limitations

– Position of traffic lights, stop signs, intersections, and other signs

– Category of road

– Number of lanes

– Etc.

Distance

Vehicle Speed

ADAS-RP

ADAS = Advanced Driver Assistance SystemsRP = Research Platform

But not enough to predict fuel consumption!

0 500 1000 1500 2000 25000

10

20

30

40

Chicago Real-World Data Provides Stochastic Aspect

7

0 100 200 300 400 5000

5

10

15

20

25

0 100 200 300 400 5000

5

10

15

20

25

2007 travel survey for the Chicago Metropolitan Agency for Planning (CMAP)

GPS loggers

267 households surveyed

10k vehicle trips

6M data points

Processing0 20 40 60 80 100 120 140

-4

-3

-2

-1

0

1

2

3

Speed [km/h]

Accele

ration [

m/s

2]

% of total

0.0001

0.001

0.01

0.1

1

time density Top envelope Bottom envelope

59% of points deemed valid (=1000h)

From Database to Actual Speed Profiles:

Constrained Markov Chains

8

Markov Chains

…

0.02

0.050.01

0.02

0.050.01

0.02

0.050.01

……

… … …

…0.03

0.07

0.15

Valid Real-World Micro-Trips

Transition Probability

Matrix

14.5

15.5

16.0

16.5

13.5

14.0

17.0

t+1tt-1t-2

Spee

d (

m/s

)

Time (s)

P=0.05

P=0.3

P=0.2

P=0.15

P=0.15

P=0.1

P=0.05

Initialization (t=0, a=0, v=0)

TPM

Random number generation

Compute next state

v=vend?

d>Dtarget?

Metadata matches target?

Speed Profile

Yes

Yes

Yes

No

No

No

0 200 4000

50

100

0 100 200 300 4000

50

100

0 100 200 300 4000

50

100

Constrained Markov Chain

Examples of Synthesized Speed Profiles

9

Target Speed32 km/h

0 100 200 300 400 500 6000

10

20

30

40

50

60

70

80

90

Time (s)

Sp

ee

d (

km

/h)

/ T

ime

(s)

V Vmax V

avg

tgtV

avg

act tstop

One synthetic speed profile for one entire itineraryMultiple stochastic speed profiles for the same target micro-trip

Speed Limit50 km/h

Optimal Energy Management

10

Pontryagin’s Minimization Principle

11

𝑃𝑏∗ = argmin

𝑃𝑏

( 𝑃𝑓 𝑃𝑏 + 𝒓(𝒕)𝜃 𝑃𝑏 𝑃𝑏)

Hamiltonian

Fuel PowerFunction of 𝑃𝑏 through optimal operation maps

Equivalence Factor (EQF)

Term close to 1

Battery Power Command

In our study we make the assumption that EQF 𝑟 𝑡 = 𝑟0 Optimal EQF: one that results in SOC=30% for the first time at the end of the trip

System: one-mode power-split PHEV (similar to Toyota Prius PHEV)

Not influenced by control: Vehicle speed, torque demand decided by driver

State of the system: battery state-of-charge (SOC)

At any given time, for any given vehicle speed / wheel torque / battery power, one optimal operating point minimizing fuel consumption exists

Battery power 𝑃𝑏 = command variable

Constraints: final SOC is 30%

PMP:

Optimal Command

PMP Implementation in the Vehicle Controller

12

Battery power “Candidates”

Optimal Operating Points

Fuel and battery power computation

Minimization of cost function Filtering of the

optimal power demand

Computation of corresponding torque/speed targets

PMP w/ ICE

EV Mode

Optimal Speed/Torque Targets (HEV)

Speed/Torque Targets (EV)

ICE ON/OFF

Logic

Cost HEV

Cost EV

ICE ON/OFF

Speed &Torque Targets

Simulation Framework

13

Driver & Powertrain

14

Power-Split Hybrid-Electric (Toyota Prius Hybrid System)

Driver presses on pedals

Vehicle energy management computes torque demands

Powertrain = all components

200 300 400

-50

0

50

100

150

0.05 0.05 0.050.1 0.1 0.10.15 0.15 0.150.2 0.2 0.2

0.25 0.25 0.250.3 0.3 0.3

0.35

0.35 0.350.35

Speed (rad/s)

Torq

ue (

N.m

)

Components: dynamics + look-up tables from test data

A forward-looking model of the Prius PHEV in Autonomie

Route Selection

15

ADAS-RP

User can select route in HERE’s ADAS-RP

Route export plug-in

Speed Prediction

16

Input = Route from ADAS-RP Output = n speed profiles + grade

Route-Dependent Optimization

17

Start

Run EV+CS

∃ tSOC=30%? No PMP

Run PMP

tSOC=30%<tend - δt ?

∃ tSOC=30%? SOCend>SOCtgt+δSOC?

tSOC=30%<tend - δt ? EQF Found

Increase EQF Decrease EQF

No

No No

No

Yes

Yes

No

Yes

Yes

Yes

SOCtgt

tend

SOCtgt

tend

SOCtgt

tend

SOC drops too fast

Battery not used enough

Optimal SOC drop

Actual Driving

18

0

50

100Vehicle Speed (km/h)

0

50

100

0

50

100

0 5 10 15 200

50

100

Distance (km)

0

50

100Vehicle Speed (km/h)

0

50

100

0

50

100

0 5 10 15 200

50

100

Distance (km)

P Speed profiles for EQF optimization (before driving)

Q Speed profiles for actual driving≠

In the real-world, actual speed prediction will be different from prediction…

Large-Scale Evaluation

19

SOCtgt

tend

0

50

100

Vehicle Speed (km/h)

0

50

100

0

50

100

0

50

100

0

50

100

0

50

100

0

50

100

0 5 10 15 200

50

100

Distance (km)

Start

Run EV+CS

∃ tSOC=30%? No PMP

Run PMP

tSOC=30%<tend - δt ?

∃ tSOC=30%? SOCend>SOCtgt+δSOC?

tSOC=30%<tend - δt ? EQF Found

Increase EQF Decrease EQF

No

No No

No

Yes

Yes

No

Yes

Yes

Yes

30 itineraries 8 Generations 3 SOCinit 9 EQFs

× × ×

1 EQF Optimization

+ 8 suboptimal valuesaround

optimal EQF

Example of Result (1 Itinerary, 1 generation)

20

0 500 1000 1500 2000 2500 3000-80

-60

-40

-20

0

20

40

60

80

100

120

Cycle: Urban_0020_3 Dist: 35.1505 km Avg Spd: 43.7861 km/h

Time (s)

0 500 1000 1500 2000 2500 300020

30

40

50

60

70

80

90

Time (s)

SO

C (

%)

0 500 1000 1500 2000 2500 30000

50

100

150

200

250

300

350

400

Time (s)

Fu

el(

g)

2.52 2.54 2.56 2.58 2.6 2.62 2.64-15

-10

-5

0

5

10

15

EqF

Fu

el

Savin

g (

%)

8 8.5 9 9.5 1012.5

13

13.5

14

14.5

15

15.5

Battery Energy (MJ)

Fu

el

En

erg

y (

MJ)

Speed (km/h)

grade (%)(x10)

elev. (m)

Ref

Opt

2.53

2.54

2.55

2.56

2.58

2.59

2.6

2.61

Ref

Opt

2.53

2.54

2.55

2.56

2.58

2.59

2.6

2.61

unadj.

adj

EqF

2.53

2.54

2.55

2.56

2.57

2.58

2.59

2.6

Ref.

Opt.

Tgt SOC

Fuel savings need to be SOC adjusted: final SOC in optimal case is always 30%, but it varies for reference case (stays in the [28,32] range)

Preliminary Results Show Strong Benefits (Best Case

Scenario)

21

16 18 20 22 24 26 28 30 32 34 36-10

-5

0

5

10

15

20

25

30

Trip Distance (km)

Adj. F

uel S

avin

gs (

%)

SOC0=50%

SOC0=70%

SOC0=90%

Conclusion

Optimal energy management for xHEVs theoretically requires full knowledge of duty cycle, which is not possible in the real-world

A realistic and stochastic prediction can be achieved, through a combination of Markov chains and data from digital maps

PMP is a convenient way of achieving optimal control in a real-world controller

Efficacy depends on one tuning parameter, the equivalence factor (EQF)

EQF depends on the future route

A framework was designed to evaluate route-based control along with its uncertainties:

– Detailed powertrain model in Autonomie

– Vehicle speed prediction for a given itinerary

– Optimal route-based tuning

– Large-scale simulation to evaluate benefits in a broad range of situations

Future work:

– Further statistical analysis: can the optimal EQF be inferred from simple parameters, or full simulation is needed?

– Make the controller adaptive, i.e. update EQF periodically (vs. simply at start of the trip)

22

Acknowledgement

23

Contact

Dominik Karbowski (Principal Investigator): dkarbowski@anl.gov / 1-630-252-5362

Aymeric Rousseau (Systems Modeling and Control Manager): arousseau@anl.gov

Funded by the Vehicle Technology Office

Program Manager: David Anderson

www.transportation.anl.govwww.autonomie.net

HERE (a Nokia company) provided free license for ADAS-RP