Outline Vehicle Propulsion Systems Lecture 7

Supervisory Control Algorithms

Repetition Supervisory Control Algorithms Heuristic Control Approaches Optimal Control Strategies

Lars Eriksson Associate Professor (Docent)

Vehicular Systems Linkoping University

November 21, 2012

Analytical solutions to Optimal Control Problems ECMS Equivalent Consumption Minimization Strategy

1 / 33

3 / 33

Deterministic Dynamic Programming Basic algorithm

N1

Deterministic Dynamic Programming Basic Algorithm

Graphical illustration of the solution procedure

J(x0 ) = gN (xN ) +
k=0

gk (xk , uk )

xk+1 = fk (xk , uk ) Algorithm idea: Start at the end and proceed backward in time to evaluate the optimal cost-to-go and the corresponding control signal.
x
1 2 3
k= 0 ta

x
2 0 2 1 4 0 2

JN (xN ) 2

1 0 1 2 3 N
tb

N 1

k= 0 ta

N 1

N tb

4 / 33

5 / 33

Examples of Short Term Storage Systems

Pneumatic Hybrid Engine System

6 / 33

7 / 33

Outline

Parallel Hybrid Modes and Power Flows

Repetition Supervisory Control Algorithms Heuristic Control Approaches Optimal Control Strategies Analytical solutions to Optimal Control Problems ECMS Equivalent Consumption Minimization Strategy

The different modes for a parallel hybrid u Pbatt /Pvehicle Battery drive mode (ZEV)

Parallel Hybrid ZEV mode, u=1 E

8 / 33

9 / 33

Control algorithms

Strategies for the Parallel Hybrid

Power split u, Clutch Bc , Engine Be Mode ICE ZEV ZEV Power assist Recharge Regenerative braking Regenerative braking u 0 1 1 [0,1] <0 1 1 Be 1 0 0 1 1 0 0 Bc 1 0 1 1 1 0 1

Determining the power split ratio u uj (t) = Pj (t) Pm+1 (t) + Pl (t) (4.110)

1 2a 2b 3 4 5a 5a

Clutch engagement disengagement Bc {0, 1} Engine engagement disengagement Be {0, 1}

All practical control strategies have engine shut off when the torque at the wheels are negative or zero; standstill, coasting and braking.

10 / 33

11 / 33

Classication I Supervisory Control Algorithms

Classication II Vehicle Controllers

Non-causal controllers
Detailed knowledge about future driving conditions. Position, speed, altitude, trafc situation. Uses: Regulatory drive cycles, public transportation, long haul operation, GPS based route planning.

Heuristic controllers Causal State of the art in most prototypes and mass-production Optimal controllers Often non-causal Solutions exist for simplications Sub-optimal controllers Often causal On-going work to include optimal controllers in prototypes

Causal controllers
No knowledge about the future... Use information about the current state. Uses: The normal controller, on-line, in vehicles without planning

12 / 33

13 / 33

Some Comments About the Problem

Outline

Repetition Difcult problem Unsolved problem for causal controllers Rich body of engineering reports and research papers on the subject This can clearly be seen when reading chapter 7! Heuristic Control Approaches Optimal Control Strategies Analytical solutions to Optimal Control Problems ECMS Equivalent Consumption Minimization Strategy Supervisory Control Algorithms

14 / 33

15 / 33

Heuristic Control Approaches

Heuristic Control Approaches

Parallel hybrid vehicle (electric assist)

Operation usually depends on a few vehicle operation Rule based: Nested if-then-else clauses if v < vlow then use electric motor (u=1). else. . . Fuzzy logic based Classication of the operating condition into fuzzy sets. Rules for control output in each mode. Defuzzycation gives the control output.

Determine control output as function of some selected state variables: vehicle speed, engine speed, state of charge, power demand, motor speed, temperature, vehicle acceleration, torque demand.
16 / 33 17 / 33

Heuristic Control Approaches Concluding Remarks

Outline

Easy to conceive Relatively easy to implement Result depends on the thresholds Proper tuning can give good fuel consumption reduction and charge sustainability Performance varies with cycle and driving condition Not robust Time consuming to develop an tune for advanced hybrid congurations

Repetition Supervisory Control Algorithms Heuristic Control Approaches Optimal Control Strategies Analytical solutions to Optimal Control Problems ECMS Equivalent Consumption Minimization Strategy

18 / 33

19 / 33

Consider a driving mission

Variables. Control signal u(t), System state x(t), State of charge q(t) (is a state).

Formulating the Optimal Control Problem

What is the optimal behaviour? Denes Performance index J. Minimize the fuel consumption
tf

J=
0

mf (t, u(t))dt

Balance between fuel consumption and emissions

tf

J=
0

mf (t, u(t)) + CO mCO (x(t), u(t))+ NO mNO (x(t), u(t)) + HC mHC (x(t), u(t)) dt

Include driveability criterion

tf

J=
0
20 / 33

mf (t, u(t)) +

d a(t) dt

dt
21 / 33

First Solution to the Problem

Including constraints
Hard or soft constraints
tb

min J(u) =
ta

L(t, u(t))dt

s.t. q(0) = q(tf ) Minimize the fuel consumption

tb tf

J=
0

mf (t, u(t))dt

min J(u) = (q(tf )) +

ta

L(t, u(t))dt

How to select (q(tf ))? (q(tf )) = (q(tf ) q(0))2 penalizes high deviations more than small, independent of sign (q(tf )) = w (q(0) q(tf )) penalizes battery usage, favoring energy storage for future use One more feature from the last one
22 / 33 23 / 33

Including constraints

Constraints That are Also Included

Including battery penalty according to

tf

State equation x = f (x) is also included From Lecture 5 Consider hybrid with only one state, SoC q(t)dt
0 tb

(q(tf )) = w (q(0) q(tf )) = w enables us to rewrite

tb

min J(u) = (q(tb ), tb ) +

ta

L(t, u(t))dt

s.t. L(t, u(t)) + w q(t)dt

min J(u) =
ta

d q = f (t, q(t), u(t)) dt u(t) U(t) q(t) Q(t)

24 / 33

25 / 33

Outline

Analytical Solutions to Optimal Control Problems

Repetition Core of the problem Supervisory Control Algorithms min J(u) = (q(tb ), tb ) + Heuristic Control Approaches Optimal Control Strategies Analytical solutions to Optimal Control Problems ECMS Equivalent Consumption Minimization Strategy
ta tb

L(t, u(t))dt

s.t. q(t) = f (t, q(t), u(t)) Hamiltonian from optimal control theory H(t, q(t), u(t), (T )) = L(t, u(t)) + (t) f (t, q(t), u(t))

26 / 33

27 / 33

Analytical Solutions to Optimal Control Problems

Hamiltonian H(t, q(t), u(t), (T )) = L(t, u(t)) + (t) f (t, q(t), u(t)) Solution (theory from chapter 9) u(t) = arg min H(t, q(t), u(t), (T ))
u

Analytical Solutions to Optimal Control Problems

0 depends on the (soft) constraint 0 = (q(tf )) = /special case/ = w q(tf )

Different efciencies 0 = (q(tf )) = q(tf ) wdis , q(tf ) > q(0) wchg , q(tf ) < q(0)

with f (t, q(t), u(t)) q q(t) =f (t, q(t), u(t)) (t) =

If q f (t, q(t), u(t)) = 0 the problem becomes simpler becomes a constant 0 , search for it when solving

Introduce equivalence factor (scaling) by studying battery and fuel power s(t) = (t) HLHV Vb Qmax

ECMS Equivalent Consumption Minimization Strategy

28 / 33 29 / 33

Determining Equivalence Factors I

Determining Equivalence Factors II

Collecting battery and fuel energy data from test runs with constant u gives a graph

Constant engine and battery efciencies sdis = 1 e f e f

schg =

Slopes determine sdis and schg .

30 / 33 31 / 33

ECMS On-line Implementation

Flowchart

There is also a T-ECMS (telemetry-ECMS)

32 / 33