implementation of an equivalent consumption minimization ... · equivalent consumption minimization...
TRANSCRIPT
Implementation of anEquivalent ConsumptionMinimization Strategyin a Hybrid truck
D.J. van Mullem
CST.2010.19
Master’s thesis
Supervisory committee:
prof.dr.ir. M. Steinbuch 1
dr.ir. A.G. de Jager 1
dr.ir. J.T.B.A. Kessels2
ir. T.A.C. van Keulen1
1 EINDHOVEN UNIVERSITY OF TECHNOLOGY
DEPARTMENT OF MECHANICAL ENGINEERING
CONTROLS SYSTEMS TECHNOLOGY GROUP
2 TNO BUSINESS UNIT AUTOMOTIVE
ADVANCED POWERTRAINS TECHNOLOGY
Eindhoven, March 2010
Abstract
In their attempts to reduce fuel consumption, vehicle manufacturers have developed Hybrid Electric Vehicles
(HEV). These vehicles have two energy converters: usually an Internal Combustion Engine (ICE) and an
Electric Machine (EM). The EM enables energy recovery during braking or driving down hill, this energy can
be used at a latter, more convenient, time to propel the vehicle.
The algorithm that deals with the power split between the ICE and the Electric EM of a hybrid vehicle, is
called the Energy Management Strategy (EMS). The main objective of such an algorithm is to minimize
fuel consumption. EMS methods can be divided into two classes: non-causal methods that require exact
knowledge of the upcoming velocity and load trajecotries, and causal methods, that try to minimize fuel
consumption without this a-priori knowledge.
Optimal control theory can be used to rewrite the non-causal optimization problem to an optimization in-
dependent of time. The optimal solution is characterized by one constant Lagrange parameter of which the
value is determined by the velocity and load conditions. Online estimation of this Lagrange parameter makes
the method suitable for real-time implementation. Estimation can be done by feedback on the battery State-
Of-Charge (SOC). Hence, a real-time implementable strategy is obtained, in literature often referred to as
Equivalent Consumption Minimization Strategy (ECMS).
Nevertheless, after the simplification of the optimization problem, to find an optimal powersplit in real time,
it is still required to solve a nonconvex optimization problem with limited on-board computational power.
To cope with this constraint, the optimization problem is solved off-line for different values of the Lagrange
parameter, rotational speed and torque requests. The resulting optimal look-up tables can be stored in a
onboard vehicle Electronic Computation Unit (ECU).
To develop the ECMS for a hybrid DAF CF, a simulation model is used that contains the hybrid supervisory
controller and the several drivetrain components (ICE, EM and batteries). This simulation model is validated
on the chassis dynamometer at the Eindhoven University of Technology for several distance velocity profiles.
The cumulative fuel consumption error between simulation and experiment is within 2%.
The online estimation of the Lagrange parameters, requires feedback on the SOC. For real-time implementa-
tion of the ECMS, the feedback gain and initial value of the Lagrange parameter require tuning. The validated
simulation model is used to find an optimal setting, and to determine their influence on the vehicle’s fuel
consumption. Simulations result show that the ECMS has a fuel consumption reduction in the range of 1%
to 4%, when compared to the existing heuristic controller.
Samenvatting
Om het brandstofverbruik van een voertuig te reduceren hebben automobiel fabrikanten Hybride Elektrische
Voertuigen (HEV) ontwikkeld. Deze voertuigen zijn naast de standaard verbrandingsmotor (Internal Com-
bustion Engine, ICE) uitgerust met een Elektrische Machine (EM). De EM maakt het mogelijk om energie
terug te winnen tijdens remacties of tijdens bergaf rijden. Deze energie kan later gebruikt worden, om het
voertuig aan te drijven.
Het algoritme dat de vermogensvraag regelt tussen de verbrandingsmotor en de elektrische machine in een
hybride voertuig wordt een Energie Management Strategie (EMS) genoemd. Het voornaamste doel voor dit
algoritme is het minimaliseren van het brandstofverbruik. EMS methodes kunnen in twee groepen gesplitst
worden: niet-causale methodes die exacte kennis vereisen over het komende snelheidstraject en causale meth-
odes die zonder deze kennis proberen brandstofverbruik te reduceren.
Optimale Control theorie kan gebruikt worden om het niet-causale optimalisatieprobleem te herschrijven
naar een optimalisatie, onafhankelijk van tijd. De optimale oplossing wordt dan gegeven door één snelheids-
en last afhankelijk Lagrange parameter. Terugkoppeling op de referentiewaarde van de batterij wordt gebruikt
om deze Lagrange parameter online te schatten. Deze oplossingsmethode wordt Equivalente Consumptie
Minimalisatie Strategie (ECMS) genoemd, en maakt real-time implemenatatie mogelijk.
Hoewel hiermee het optimalisatieprobleem versimpeld wordt, blijft de beperkte rekenkracht van de controller
op het voertuig een beperkende factor bij het oplossen van het resterende non-convexe optimalistatieprob-
leem. Daarom is het probleem off-line opgelost voor verschillende waardes van de Lagrange parameter, het
toerental van de aandrijflijn en het gewenst koppel. De resulterende optimale vermogenssplit, kan vervolgens
worden opgeslagen in een voertuig controller (Electronic Computation Unit, ECU).
Om de ontwikkeling van de ECMS mogelijk te maken, is een simulatiemodel waarin alle aandrijflijn com-
ponenten (ICE, EM en batterijpakket) zijn gemodelleerd gebruikt. Dit simulatiemodel bevat de besturing-
shiërarchie zoals deze ook op de ECU van de DAF CF aanwezig is. Het simulatiemodel is gevalideerd voor
verschillende afstand snelheids trajectories. Brandstofverbruik over een ritcyclus kan daarmee voorspeld wor-
den binnen 2% nauwkeurigheid.
Om de Lagrange parameters online te schatten, is een terugkoppeling van de batterij refentiewaarde nodig. De
waardes die gebruikt worden om deze terugkoppeling te realiseren, moeten vooraf bepaald worden voor het
realtime implementeren van ECMS op een voertuig. Het gevalideerde simulatie model is gebruikt om deze
terugkoppelings kalibratiewaardes te vinden en hun invloed op het brandstofverbruik te onderzoeken. Simu-
laties tonen aan dat ECMS een brandstofbesparing van 1 tot 4% realiseert, in vergelijking met de bestaande
heuristische regelaar.
Contents
Abstract i
Samenvatting ii
1 Introduction 2
2 Hybrid vehicle specification 3
2.1 Drivetrain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Internal combustion engine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.3 Electric machine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.4 Batteries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.5 Control hierarchy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.6 Vehicle model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3 Hybrid electric vehicle control strategies 8
3.1 Problem statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.2 Rule based controllers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.3 Optimal Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.4 Equivalent consumption minimization strategy (ECMS) . . . . . . . . . . . . . . . . . . . . . 11
3.5 Energy managment strategy using route information . . . . . . . . . . . . . . . . . . . . . . . 12
3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4 Implementation of Equivalent Consumption Minimization Strategy 14
4.1 Off-line optimization for positive torque request . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.2 Negative torque request . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5 Simulation model validation 18
5.1 Test setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
CONTENTS 1
5.2 Road load settings and drive train properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
5.3 Engine map validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.4 Transient fuel consumption validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
5.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
6 Simulation results 29
6.1 Drive cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
6.2 Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
6.3 Comparison with heuristic controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
7 Conclusions and recommendations 34
7.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
7.2 Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
A Appendix A: Dynanometer Control Scheme 37
Bibliography 38
CHAPTER 1
Introduction
Hybrid Electric Vehicles (HEV’s) have, at least, two energy converters: usually an Internal Combustion Engine
(ICE) as prime mover, and an Electric Machine (EM) as secondary converter. The EM enables energy recovery
during braking or driving down hill. This energy can be used at a latter, more convenient, time to propel
the vehicle. The supervisory control algorithm, dealing with the balanced generation and re-use of the stored
energy, is called Energy Management Strategy (EMS).
Several contributions have been made regarding the EMS design for HEVs, see, e.g., [12, 16] for an overview.
EMS methods can be divided in two classes. Firstly, non-causal methods [4] that require exact knowledge
of the power and velocity trajectories and secondly, causal or real-time implementable methods [5] that try
to minimize fuel consumption without knowledge of the upcoming trajectories. In general, the non-causal
strategies are used to benchmark and to design real-time implementable strategies.
One method to obtain such a causal solution is the method of Lagrange, which simplifies the fuel minimiza-
tion problem to an optimization independent of time. This is exploited by real-time strategies that estimate
the Lagrange parameter based upon real-time available vehicle information, e.g., the battery State-Of-Charge
(SOC) [10]. Nevertheless, assuming that the Lagrange parameter could be predicted or estimated in a suitable
way, the resulting optimization is non-convex. Furthermore, for a real-time EMS, additional demands on the
EMS are imposed by limited on-board computational power and battery SOC estimation accuracy.
In this report the real-time EMS that estimates the feedback parameter online, as presented in [8, 10], is im-
plemented in a DAF CF. Since on-board computational power is limited, real time implementation requires
extra effort. Therefore, a computationally efficient method to implement the remaining non-convex optimiza-
tion problem is presented in which the optimization problem is solved off-line for different values of the
Lagrange parameter, crankshaft rotational speeds, and torque requests. The resulting optimal torque split is
stored in look-up tables that can be implemented in the onboard vehicle Electronic Computation Unit (ECU).
To facilitate the development of an EMS on the DAF CF, a simulation model of the hybrid vehicle is used in
which the different controllers of the gearbox, battery, engine and electric machine are modeled. This model
is validated on several distance velocity trajectories, relevant for trucks on a chassis dynamometer. The goal
of the validation, is to predict fuel consumption within an accuracy of 5 %. Besides, the simulation model is
used to obtain tuning rules for the SOC feedback parameters.
This report is organized as follows: in Chapter 2 the vehicle model and the different components are pre-
sented, Chapter 3 gives an overview of different EMS methods. Chapter 4 describes the implementation of a
real-time EMS for the given vehicle configuration. The method that is used to validate the simulation model is
presented in Chapter 5. The simulation results to derive tuning rules for the feedback gain and a comparison
with the existing rule-based strategy are given in Chapter 6. Finally, conclusions and recommendations are
given in Chapter 7.
CHAPTER 2
Hybrid vehicle specification
The hybrid vehicle, used for testing and which is simulated, is a DAF CF 65. First the drivetrain components
as well as their models used for simulation Since the components of the drivetrain which are described in
this chapter are used both in the simulation model, as well as for the development of the Energy Management
Strategy (EMS) they are introduced first. After that the control hierarchy and the vehicle model is presented.
2.1 Drivetrain
The truck is an two axle configuration DAF CF, with a gross vehicle weight of 8230 kg. It is equipped with
a hybrid driveline, consisting of an Electric Machine (EM) and an Internal Combustion Engine (ICE). The
components have a parallel hybrid configuration, so the two sources may operate in parallel to accelerate the
vehicle. Physically, the EM is placed between the clutch and gearbox. A schematic overview of the driveline is
given in Figure 2.1.
Figure 2.1 / Schematic overview of the hybrid driveline
2.2 Internal combustion engine
The internal combustion engine is the primary power source of the vehicle. The main specifications of this
engine are given in Table 2.1. The engine is modeled by a non-linear static map, relating the engine torque
Tice and rotational speed ω towards mass fuel rate mf . Such a map does not take thermal and transient effects
into account, but has great advantages in terms of calculation speed. To obtain accurate values for this engine
map, a validation of the fuel map is performed and the resulting fuel map is given in Figure 2.2. More details
on the validation procedure are given in Chapter 5
2 Hybrid vehicle specification 4
Table 2.1 / Engine specificationsEngine displacement 5883 [cm3]
type 6 cylinder inline, 24 valves
max power 185 kW
max torque 910 Nm @ 1600 RPM
230
230
230
240
240
240
240
250
250
250
250
250
270
270
270
270
290
290
290
290
310310
310
330330
330
370 370
370
Brake Specific Fuel Consumption [g/kWh]
rotational velocity [RPM]
Tor
que
[Nm
]
800 1000 1200 1400 1600 1800 2000 2200 24000
100
200
300
400
500
600
700
800
900
Figure 2.2 / Fuel consumption in [g/kWh].
The fuel consumption is obtained by integrating the mass fuel flow at the instantaneous engine speed and
torque. In equation form it holds that:
Pf = mf (Tice, ω) HLV (2.1)
Here, Pf is the amount of power delivered by the fuel, and HLV the lower heating value of the fuel, for diesel
42.5 MJ/kg.
2.3 Electric machine
The secondary power source is an electric motor/generator coupled to a battery through an inverter. The
electric motor is an internal permanent magnet type synchronous motor. The combined efficiency of the
electric motor and inverter are depicted in Figure 2.3. Both the inverter and electric motor are liquid-cooled.
The required electric power Pb as function of the EM efficiency ηem(ω, Tem) is given by:
Pb(ω, Tem) = f(Tem, ω) (2.2)
In which Tem is the torque output from the EM. The model has the drawback that it does not take spinning
losses into account when Tem = 0. This does not correspond to the physical behavior of the EM. However,
2 Hybrid vehicle specification 5
20
20
20
20
4040
40
4050
50
50
5060
60
60
6070
70
70
70
80
80
80
80
85
85
88
88
90
90
Rotational speed ω [rpm]
Tor
que
[Nm
]
EM efficiency [%]
0 500 1000 1500 2000 2500
−400
−300
−200
−100
0
100
200
300
400
Figure 2.3 / Electric motor and inverter efficiency [%].
since characterization of the EM without the other drivetrain components is difficult, this model is used for
EMS development.
2.4 Batteries
The power for the electric motor is supplied by Magnesium type lithium-ion batteries with a total capacity
of 2.0 kWh. The total package consists of two parallel modules with 48 cells placed in series. The nominal
voltage of the package is 340 Volt. The State of Charge (SOC) of the battery depends on the battery current Isaccording to:
SOC(t) =Q(t)
Qmax=
∫ t
0
Is(t)
Qmaxdt+ SOC(0) (2.3)
in which Q(t) is the battery charge as function of time and Qmax is the maximum charge capacity of the
battery. For the losses in the battery, an internal resistance model is used:
Ps = Is V = Pb +
(PbV
)2
R(SOC) (2.4)
Here, V is the battery voltage, and R a SOC dependent resistance, the relation is shown in Figure 2.4.
2 Hybrid vehicle specification 6
0 0.2 0.4 0.6 0.8 12
2.5
3
3.5
4
SOC
Res
ista
nce
[mΩ
]
SOC depending resistance for battery cells
Figure 2.4 / SOC depending resistance of battery cells
2.5 Control hierarchy
The inverter, the batteries, the engine, the EM and the transmission all have their own control unit, supervised
by the hybrid controller. This is shown schematically in Figure 2.5. In this thesis, only decisions on supervisory
level are considered. Gear shifts, torque allocation and so on are performed by the underlaying controllers and
are therefore not taken into account.
Hybrid Controller
Transmissioncontroller
EMcontroller
ECUBattery
Controller
Gearbox Inverter ICE BatteryEM
Figure 2.5 / Structure of the control hierarchy on the truck
The main control variable for the controller is the split ratio u, which represents a power-split ratio between
the primary and secondary power source, here defined as:
Tice = u Treq (2.5)
Tem = (1− u) Treq
In which Treq represents requested torque by the driver. An overview of the drivetrain topology is given in
Fig. 2.6.
2 Hybrid vehicle specification 7
+
Figure 2.6 / Drive train topology.
2.6 Vehicle model
For simulation purposes of the hybrid truck and to compare fuel consumption for different velocity trajecto-
ries and EMS controllers, the vehicle dynamics are modeled by means of a point mass for the vehicle inertia.
As input for this model the current speed is compared with a desired setpoint for the speed, where a Propor-
tional Integrating (PI) controller actuates the accelerator and brake pedal.
y(t) = kp(v(t)− vdes) + ki
∫ t
0
(v(τ)− vdes)dτ (2.6)
Here y is a normalized pedal position, v the actual velocity and vdes the actual velocity. Note that the controller
will have to translate a positive value of y in a accelerator pedal actuation, and a negative value into brake pedal
actuation.
Road load and drive train losses:
the road load force is modeled with a second order function, depending on the vehicle speed. The drive train
and road load loss model in the simulation is compared with the losses encountered at the dynamometer.
More details about these models are in the validation procedure, presented in Chapter 5.
CHAPTER 3
Hybrid electric vehicle control strategies
HEVS can achieve better fuel economy by using two different energy sources. Their main advantage is that the
kinetic energy can be (partially) recovered during braking or driving down hill. To improve the performance
of such vehicles compared to regular internal combustion engine driven ones, they depend on their EMS, see
e.g. [3]. In this Chapter an overview of different control strategies for parallel hybrid-electric vehicles will be
presented. Addressed first are heuristic control and the current controller implemented on the DAF. Secondly,
various optimization techniques are covered and a derivation of the Equivalent Consumption Minimization
Strategy (ECMS) is presented.
3.1 Problem statement
The goal of the EMS is to minimize fuel consumption. It is not necessary to minimize the fuel mass-flow rate
at each time instant, but rather total fuel consumption. The aim is to find the value of the control input u(t)
which minimizes the cost function Jf (t, u) defined as:
Jf =
∫ tf
0
mf (t, u, ωeng) HLV dt (3.1)
For a hybrid vehicle, the drive mode that minimizes the problem stated in (3.1), corresponds with a purely
electric strategy where all the traction power is provided by the battery. However, the energy recovered by
regenerative braking is not sufficient to sustain battery SOC. In order to have a charge sustaining EMS, it is
required that the initial SOC is equal to the end SOC level. This is often referred to as the end-point constraint:
SOC(tf ) = SOC(t0)→∫ tf
0
Ps(u, t)dt = 0 (3.2)
Besides this constraint on the battery, the different components should deliver the requested amount of power
from the driver at all times:
Treq =(u Tice + (1− u) Tem
)(3.3)
There are several techniques to solve this problem for a given hybrid vehicle configuration. Fuzzy logic or
rule-based strategies can easy be implemented, but they don’t guarantee an optimal result for all situations.
Other solution techniques such as dynamic programming are able to calculate an optimal solution. However,
to find the optimal solution, knowledge of the complete driving cycle is required and computational efforts
are high.
3 Hybrid electric vehicle control strategies 9
3.2 Rule based controllers
A straight forward EMS is the heuristic control strategy as stated in [12]. This strategy makes use of event-
triggered rules and uses the EM to assist the engine in torque production. Typically, heuristic rules are based
upon vehicle velocity, battery SOC and torque request. An example, given in [12] is as follows:
Heuristic controller
• if v < vmin the electric motor is used alone (u = 0)
• if v ≥ vmin and Treq < Tmax(ωeng), the engine is used alone (u = 1)
• If SOC is below threshold, the engine is forced to deliver excess torque to recharge the battery.
• If SOC is too high, the motor is used alone.
• Above Tmax(ωeng) the engine maximum torque is required and the EM is used to assist the engine.
In which v is the vehicle speed, vmin is a setpoint for minimum vehicle speed, Tmax(ω) is the maximum
engine torque at the current engine speed (ω) and SOC is the current SOC of the battery. The battery can
only be used between a certain minimum and maximum SOC level SOCmin and SOCmax. When the SOC
is outside this region, the lifespan of the battery will detoriate.
Existing controller
Heuristic controllers can be extended with more rules states and thresholds on e.g. brake and accelerator
pedal as stated in [12]. The hybrid DAF truck which is considered in this thesis also has a heuristic controller
implemented. This controller distinguishes two different modes:
Electrical vehicle mode:
in this mode the truck drives with the ICE disengaged, running solely on the Electrical motor. Split ratio u
is therefore set to zero, and the clutch is opened. This mode is enabled when operator demand torque is
between the limits of the electric motor and vehicle speed is lower than 50 km/h. This mode doesn’t improve
fuel consumption but is rather an extra functionality offered by a hybrid vehicle.
Drive mode:
In this mode the split value, u, is depending on operator torque, Treq and SOC of the battery. To keep the
SOC level between limits, first of all the amount of EM-torque is limited by the following rules:
• for SOC < SOCmin the ICE is run with slightly more torque in order to charge the battery. This
regeneration factor is depending on SOC and engine speed.
• for SOC > SOCmax no charging can occur, to maintain battery lifetime.
3 Hybrid electric vehicle control strategies 10
The resulting torque command is divided by the torque request to get the split ratio. The maximum regen-
eration torque, when the vehicle is slowing down, is determined by the EM limits. In order to use the stored
electrical energy, a rate limiter is used on the actual engine torque demand. This limit is reset on every clutch
opening and increases over time. By slowing down the response of the engine, the ‘gap’ in drive torque needs
to be filled with the electric machine. In this way, the EM runs most of the time with low vehicle speeds, so
inefficient use of the ICE is avoided.
Since thresholds for heuristic controllers have to be chosen in advance, these kind of strategies can not guar-
anteed to be optimal, especially when considering different driving cycles. Due to the nonlinear behavior of
the propulsion components, every cycle has another optimal solution and therefore rule-based strategies can
not be proven to be optimal in terms of fuel consumption. The main advantage is that heuristic controllers are
intuitive to conceive and easy to implement. If properly tuned, they can provide good results in terms of fuel
consumption reduction and charge sustainability. However, the behavior strongly depends on the choice of
thresholds that are involved. This may result in limited robustness of the heuristic controller, under changing
vehicle operation condition.
3.3 Optimal Control
Dynamic programming
The mathematical tool that is used for the optimization of (3.1) subject to (3.2) is the method of Dynamic
Programming (DP), as e.g. used by [4]. It is commonly used for optimization over a given time period. It gives
an optimal solution when there is a-priori knowledge about the driving cycle. It can be used to minimize the
total amount of fuel as defined in (3.1). Since dynamic programming requires gridding of the state and time
variables, discretization of these variables is required.
The problem requires the introduction of a cost-to-go matrix, defined as the cost along the optimal trajectory
passing through the ’point’ in the time-space to the target point with the desired SOC(tf ). Due to the dis-
cretization of the state space, the values of Ps are either interpolated or approximated by the nearest available
values on the grid. The energy that is introduced or deleted by this method determines whether the adopted
state-space discretization is acceptable or the number of points must be increased.
Analytical optimization method
Direct numerical optimization is inherently non-causal and requires substantial amounts of computational
time. A more practical approach is based on the minimum principle of Pontryagin. This method introduces
a Hamiltonian function to be minimized at each time. First, in order to minimize the cost function (3.1), with
the constraint over the final state of charge of the battery with (3.2). The dynamics from the system are defined
by:˙SOC = −Ps (3.4)
The Hamiltonian for this system can than be defined as:
H(u, t) = mf (u, t) HLV + λ(t)(−Ps) (3.5)
The parameter λ, which corresponds to the adjoint state in classical optimal control theory, is described by
3 Hybrid electric vehicle control strategies 11
the Euler-Lagrange equation:d
dtλ(t) = −∂H(u, t)
∂SOC= 0 (3.6)
When the influence of the SOC on the internal battery parameters is neglected, λ = 0 and therefore λ will be
constant and> 0. The necessary condition of optimality using the minimum principle are therefore met. With
this result the optimization can be reduced to searching for one constant parameter λ0 that gives minimal
fuel consumption for a given cycle, so with (2.1), (3.5) can be written as:
minu,λ
J(u, t) = minu,λ
(Pf (u, t) + λ Ps(u, t)) (3.7)
λ can here be physically interpretated as the relative incremental cost of the primary and secondary power
converter. For a known velocity and power trajectory, numerical methods can be used to find an optimal
value of the Lagrange parameter, λ∗, and therefore an optimal powersplit trajectory u∗(t), 0 ≤ t ≤ tf which
minimizes fuel can be defined as:
u∗, λ∗ = arg minu∈U,λ∈Λ
Jf (3.8)
Depending on the modeling of the components (3.7) can be solved with Dynamic Programming for nonlinear
and non convex functions or Quadratic Programming for quadratic functions. The non-convex optimization
problem is now reduced to finding the value of λ∗.
3.4 Equivalent consumption minimization strategy (ECMS)
The value of λ∗, which can be obtained with for instance DP or bisection, resembles to one drive cycle.
The calculated optimal value of λ∗ for a given trajectory can be used to find the optimal split ratio u for
any point in time in order to minimize the fuel consumption. However, over- or underestimation yields
towards under- or overcharging of the battery. Since exact knowledge of the upcoming driving conditions is
in real-time applications not available, a fixed value for λ∗ is not adequate to prevent the battery from over- or
undercharging and therefore a charge sustaining EMS.
However, once λ∗ has been estimated, it can also be used in a strategy without a-priori knowledge about the
cycle. There are various techniques to adapt the value of λ to the driving conditions at the current time. In
this way fuel consumption can be lowered substantially without the need of knowing the full driving cycle in
advance.
In literature [5], λ is referred to as the fuel equivalence factor, where the name Equivalant Consumption
Minimization Strategy is derived from. To keep the SOC between the desired limits, in Koot et al. [10] feedback
on the SOC is used. In this case it is a PI-controller with a limited integrating action towards a desired value
for the SOC.
λ(t) = λ0 +Kp(SOCdes − SOC(t)) +KI
∫ τ
0
(SOCdes − SOC(t))dτ (3.9)
As stated before, this algorithm can be used for online calculation of the equivalence factor and than compared
with the optimal solution obtained by the result of 3.1.
The estimation of λ(t) can be done in several more different ways, one way is pattern recognition on the
drivers profile. Another extension to the strategy of (3.9) is the adaptive strategy proposed by van Keulen et
al.[8]. This uses an estimate of the future recoverable electric energy using online available parameters velocity
3 Hybrid electric vehicle control strategies 12
and elevation. The estimated amount of brake power to stop a vehicle with a predefined velocity profile is equal
to:
Pbr = (m a+m g sin (α)) v(t) (3.10)
Where a and α represent the assumed deceleration rate and the expected road angle respectively,m the vehicle
mass and v is the expected velocity path. Since it involves a heavy duty, parallel hybrid truck, the approach
uses the assumption that there is a constant average deceleration rate. The recoverable brake power in this
model in time is dependent on the road load Prl (which is described by a third order polynomial of the
expected velocity path), the brake power Pbr and engine drag power Pdrag. Brake power is delivered until the
maximum generator power Pgenmax is reached:
Pr = max(
0, min(Pgenmax, Pbr − Prl − Pdrag))
(3.11)
The expected future recoverable electric energy, given the current vehicle velocity and elevation, can estimated
by integrating the recoverable brake energy over the estimated stop time tstop, while not exceeding maximum
battery capacity ∆SOCcap:
∆SOCrec(v, α, t) = min(∫ tstop
0
Pr V
Qmaxdτ , ∆SOCcap
)(3.12)
This finally leads to a set point for the equivalence factor which now becomes:
λ(SOC, v, h, t) = λ0 +K(SOC(0)−∆SOCrec − SOC(t)) (3.13)
Instead of the vehicle recoverable brake power, one could think of other vehicle parameters that can be used
to estimate the equivalent fuel cost.
3.5 Energy managment strategy using route information
The equivalent consumption minimization strategy can be extended even further if there is route informa-
tion available from on-board Geographical Information Systems (GIS). With the knowledge of the upcoming
route information, a future velocity profile can be estimated. This is done by van Keulen et al. [9] Especially
optimizing the vehicle braking trajectories can be used to maximize the energy recuperation.
Another advantage of the ECMS is that the desired SOC level, SOCdes, is time dependent, and therefore
making the reference SOC dependent for e.g. the elevation of the future driving profile. This could also help
to efficiently charge the battery before entering an enviromental zone, e.g. a city center.
3 Hybrid electric vehicle control strategies 13
3.6 Conclusion
In this chapter different methods to control HEVs are compared. In general, heuristic controllers provide sub-
optimal results. Besides that tuning these controllers takes a lot of effort, they cannot be proven to be optimal.
An ECMS controller with online estimation of the vehicle parameters can obtain results close to optimal,
while tuning efforts are minimal. However, information about the efficiency of the components is required.
Nevertheless, the focus in the remainder of this thesis lays on implementing and evaluation of an ECMS
controller for a hybrid vehicle. For ease of implementation there is chosen to not considering any integrating
action into the Lagrange parameter estimation, which simplifies (3.9) to:
λ(t) = λ0 +K(SOCdes − SOC(t)) (3.14)
CHAPTER 4
Implementation of Equivalent Consumption Minimization
Strategy
In the previous chapter it was shown that the method of Lagrange multipliers can be applied to arrive at an
optimization independent of time. Besides, it was noticed that the optimal Lagrange multiplier (λ) is constant
and depends on the actual vehicle and drive cycle characteristics. These important observations are used to
design a real-time EMS. Since processor load plays an important role in real-time implementation, efforts have
to be made in order to reduce computational burden: the EMS is implemented on a standard ECU. Therefore,
an accurate calculation of all possible split ratios, u, taking into account the drive trains components, takes
too much effort to perform online. To obtain acceptable computation times, the introduction of look-up tables
is proposed. For a complete set of Treq, λ0 and ω, an accurate calculation of the optimal split ratio based on
(3.7) is used (section 4.1). Since fuel consumption data for Treq < 0, is not available and hard to determine for
the given vehicle setup, implementation for negative torque requests is done on a heuristic base. To find an
instantaneous optimal split ratio, (3.8), the vehicle model of chapter 2 is used. The aim is to find a solution for
u∗, while still fulfilling the driver torque demand Treq which follows from the desired trajectory. The upper
and lower limits on Treq are given by:
(Tem,min(ω)− Tice,min(ω)) ≤ Treq ≤ Tice,max(ω) (4.1)
the subscripts ‘min’ and ‘max’, indicate the lower and upper limits of the EM and ICE, which are defined
as function of rotational speed, ω. Notice that the maximum drive train torque is limited by Tice,max and
therefore only depends on the ICE torque. Therefore, the hybrid system is not used to improve the acceleration
and gradability performance.
4.1 Off-line optimization for positive torque request
To obtain a real-time optimal split ratio, which is able to run on the ECU, an off-line numerical optimization is
performed. The resulting optimal solution for all ‘operating points’ is stored in lookup tables. The schematic
overview is presented in Figure 4.1. The discrete grid of operating points where the split ratio needs to be
defined is indexed by uni,nj . The limits on the EM are given by Tem,max and Trgn. For every point with
Treq > 0 the following calculation is performed:
• Step 1: define a vector of operating points for Treq, ω and λ. Vectors are indicated in boldface:
Treq = 0 +mi ∆T, mi ∈ [0 . . . ni] with ∆T =Tice,maxni
(4.2)
4 Implementation of Equivalent Consumption Minimization Strategy 15
Figure 4.1 / Schematic overview of implementation.
ω = ωmin +mj ∆ω, mj ∈ [0 . . . nj ] with ∆ω =ωmax − ωmin
nj(4.3)
λ = λmin +mk ∆λ, mk ∈ [0 . . . nk] with ∆λ =λmax − λmin
nk(4.4)
Here ni, nj and nk are a discrete number points for which the optimal split ratio will be calculated. Due
to memory limitations of the ECU, the number of points is limited. For ni, nj and nk the values are
48, 21 and 8 respectively. For both the speed ω and torque T , these numbers are taken from the already
available torque and speed vectors.
• Step 2: a feasible EM torque region for every (ni × nj) point given by (4.2) and (4.3) is defined by:
Tem = Tem,min +ml ∆Tem, ml ∈ [0 . . . nl] with ∆Tem =Tem,max(ω)− Trgn
nl(4.5)
here the maximum amount of regeneration torque Trgn is defined by:
Trgn = max(
(Treq − Ticemax(ω)), Temmin
(ω))
(4.6)
Immediately from (2.6), Tice follows:
Tice = Treq −Tem (4.7)
Since this calculation is performed off-line, and is therefore not limited by the ECU memory, ∆Tem can
be kept small, to ensure an accurate calculation of the optimal split ratio.
• Step 3: the equivalent costs can be calculated for the whole set of (ni × nj × nk) operating points by
substituting (4.7) into (2.1), (2.2) and taking into account battery losses by (2.4). The optimal split ratio
for positive torque requests can be calculated by:
u+i,j,k =
arg min(Pf + λ(k) Ps)
Treq(i) ω(j)(4.8)
4 Implementation of Equivalent Consumption Minimization Strategy 16
Rotational velocity ω [rpm]
Tor
que
requ
est [
Nm
]
λ = 2.1
1000 1500 2000 25000
100
200
300
400
500
600
700
800
900
Rotational velocity ω [rpm]
Tor
que
requ
est [
Nm
]
λ = 2.4
1000 1500 2000 25000
100
200
300
400
500
600
700
800
900
Rotational velocity ω [rpm]
Tor
que
requ
est [
Nm
]
λ = 2.7
1000 1500 2000 25000
100
200
300
400
500
600
700
800
900
Rotational velocity ω [rpm]
Tor
que
requ
est [
Nm
]
λ = 3
1000 1500 2000 25000
100
200
300
400
500
600
700
800
900
Full electric
Assist
no EM use
Charge
Figure 4.2 / Iso-split contours for various values of λ. The split-ratio is given as function of Treq and ω. For
λ = 2.2 the battery is nearly charged, and SOC decreases for increasing values of λ. A low speed
and torque region is present, where full electric drive is preferred. The recharging area ‘grows’
for a depleting battery, and vice versa for the assist area.
The result of this calculation for different values of λ is given in Figure 4.2. The split-ratio u+ is given as
function of Treq and ω. For a constantK, and SOCdes, λ is proportional to the SOC. For λ = 2.2 the battery is
charged, and SOC decreases for increasing values of λ. A low speed and torque region is present, where full
electric drive is preferred. The recharging area ‘grows’ for a depleting battery, and vice versa for the assist area.
In the figure, the area where the ICE has its maximum efficiency can be recognized: when engine speeds is
around 1200 RPM and there is a low value Treq, the strategy tries to charge the battery in order to get the
ICE at its maximum efficiency. Besides the operating area from the EM can be seen: when the battery is
completely charged (λ = 2.2), the split-ratio is set to zero only when the EM is able to deliver the required
amount of torque. Between operating points, u is determined by linear interpolation.
4.2 Negative torque request
For Treq < 0, two different situations can occur. Firstly, the area which is indicated gray in Figure 4.1, the drive
train is commanded ‘engine only’ mode. Secondly, when (Treq < Tdrag) the EM recuperates the remaining
4 Implementation of Equivalent Consumption Minimization Strategy 17
kinetic energy. Since fuel consumption data for negative torque output is not available, these cases are handled
in a heuristic way. It is assumed that engine braking is more efficient than idling the engine. This situation
changes when stop-start is enabled, than the engine stops when Treq < 0 and engine braking is not used.
u∗ (ω, Treq) =
u+ (Eq. 4.8) for Treq > 0
1 for Tdrag ≤ Treq ≤ 0
Tdrag/Treq for Treq < Tdrag
(4.9)
4.3 Conclusion
The instantaneous split ratio u, is controlled on supervisory level. Therefore, under some conditions the
desired values of this split ratio will be overruled by lower level controllers. This occurs for instance when
performing a gearshift. Another limiting factor is the driveability from the drivetrain: when a clutch closure
is performed, engine torque can not be changed instantaneously. This may lead to sub-optimal results. The
ECMS is implemented on the DAF CF, but for first experiments, a simulation model is preferred. In the next
chapter the validation of the simulation model is described. In Chapter 6, this simulation model is used to
find the values for the SOC-feedback algorithm.
CHAPTER 5
Simulation model validation
The simulation model as presented in chapter 2 consists of the hybrid controller, the drive train component
models, the vehicle model, and the road load characteristics. It is used to validate the existing heuristic con-
troller. To verify the simulation results, several speed-distance trajectories are run, both in the simulation
model and with the test vehicle on the chassis dynamometer, which is available at the TU/e.
Development of the EMS is more practical in a simulation environment than it is on a chassis dynamometer.
Furthermore, to reliably estimate the influence of EMS parameters on fuel consumption, a validation of the
simulation model is essential. Unfortunately, the EMS version implemented in the vehicle is unknown and
as a result, several threshold for the EMS controller need to be found. The validation is performed to match
threshold from the existing controller, have rotational speeds (and therefore drive train characteristics) within
acceptable limits, and check the cycle tracking behavior. The final goal is to obtain a fuel accuracy between
simulation model and chassis dynamometer within 5%.
Firstly, road load and drive train properties, of the vehicle and in the simulation model, are compared with
each other. Secondly, the engine fuel consumption is measured at different engine operating points. Both the
ECMS, and the simulation fuel consumption results, use a the static engine map. Finally, using the results
obtained for drive train efficiency and vehicle characteristics, two velocity profiles are repeatedly driven to
validate the simulation model and the reproducibility of the dyno measurements.
5.1 Test setup
The Heavy Duty Chassis Dynamometer‘, at the Eindhoven University of Technology, is build in order to
fulfil a broad spectrum of automotive test applications. Depending on the application, testing can be done at
circumferential speeds up to 225 km/h. Figure 5.1 shows the test vehicle on the dynamometer. For the purpose
of this research, several safety precautions are added to the system and the velocity tracking controller is tuned.
The test vehicle is equipped with a fuel flow meter to accurate measure fuel consumption. A schematic
overview of the complete setup of the chassis dynamometer and the safety systems is given in Appendix A.
To enhance reproducibility of the test results, the test vehicle is equipped with a drive robot, which is able to
actuate the accelerator and brake pedal. The velocity controllers, that are used in the simulation model and
on the dyno, have a PI-velocity controller and track a reference velocity as function of the driven distance.
Due to the stop-times (idling), it is more difficult to implement a distance based controller, than a time based
controller. However, this way of drive cycle testing has an advantage over the usually used time-speed mode:
different vehicle configurations always drive the same distance, but when vehicle weight is higher, it takes
a longer time to reach the target destination. This is more comparable to real life situations, where drivers
accept longer driving times when a truck is loaded.
5 Simulation model validation 19
Figure 5.1 / DAF CF on the chassis dynamometer at the Eindhoven University of Technology.
5.2 Road load settings and drive train properties
testing procedure
First of all, the losses in the drive train need to be characterized. By means of a coast down measurement,
the drive train steady state losses (independent of the vehicle velocity) are determined. The hybrid system is
unplugged and the gearbox is set to neutral. The chassis dynamometer speeds up the wheels to 95 km/h. The
deceleration time can be used to find the steady state drive train losses. These proved to be:
Tloss,wheels = cr mdyno g = 378 [Nm] (5.1)
here, cr is the rolling resistance coefficient, mdyno is the vehicle mass on the dyno, and g is the gravitational
constant. Since the weight distribution of the vehicle is known, cr can easily be deduced from these tests. In
the second part of the validation, the engine map is validated on the chassis dynamometer and, therefore, the
torque and speed dependent losses require modeling as well. The drivetrain losses are model by a torque loss
model for the differential:
Tloss,diff = Tss + ωinput kω + Tinput kT(5.2)
in which, ωinput and Tinput represent the final drive input speed and torque, respectively. Tss is the steady
state loss torque, kω and kT
are an input speed dependent and input torque dependent loss factor. To find
the appropriate values for these loss factors, the cycle in Figure 5.2 is driven. The torque at the wheels, on the
dyno, is compared with the torque in the simulation model. The simulation model is be tuned to match the
drivetrain losses. The results are given in the next section. For the torque at the wheels it holds that:
Twheels = Tice1
Rwrgb fd + Tloss,diff + Tloss,wheels (5.3)
5 Simulation model validation 20
Where Tdyno is the torque measured at the dynamometer, Rw is the wheel radius, rgb is the gearbox ratio,
and fd is the final drive ratio of the truck.
0 1000 2000 3000 4000 5000 60000
10
20
30
40
50
60
70
80
90
100Distance − Velocity profile
distance [m]
Vel
ocity
[km
/h]
SimulationDyno
Figure 5.2 / Validation profile
Besides the loss models (5.2) and (5.3), the road load settings for the simulation model and on the dyno have
to be in accordance with each other. The total driving resistance has to be specified for the simulation model
as well as for the dyno. Although they should deliver the same resistance force there are differences in the
calibration values for both systems. The resistance experienced by the vehicle is given by:
Fres = Froll + Faero (5.4)
= m g cr1 + cr2 v + 1/2 ρair Cw A v2
The vehicle velocity v, for all the other variables, see Table 5.1. Between the simulation and dyno tests there
are some differences, the most important ones are:
• in the simulation drivetrain losses have to be modeled, while during dyno tests, these are experienced.
• on the dyno the rolling resistance for the front tires must be added to the steady state losses to get
comparable results.
• the vehicle mass on the dyno, mdyno should be slightly less than msim, since it has to compensate for
the inertia of the rear wheels.
The various parameters and settings are given in Table 5.1.
5 Simulation model validation 21
0 1000 2000 3000 4000 5000 6000
−4000
−2000
0
2000
4000
6000
Torque at wheels
distance [m]
Tor
que
[Nm
]
dynosimulation
Figure 5.3 / Torque at the wheels for simulation and on the dyno
Table 5.1 / Road Load settingsSymbol description value
msim simulation vehicle mass 8445 [kg]
mdyno dyno mass setting 8320 [kg]
g gravitational constant 9.81 m/s2
A frontal vehicle area 7.68 m2
Cw Drag coefficient 0.673 [-]
ρair Air density 1.29 [kg/m3]
Cr1 1st order rolling resistance factor 0.0075 [N/N]
Cr2 2nd order rolling resistance factor 3.24 [Ns/m]
results
Now that the road load settings coincide with each other, the differential loss model from (5.2) can be tuned,
by looking at the torque at the wheels. The comparison for the amount of wheel torque is given in Figure 5.3.
Notice that for the simulation model, the estimated front wheel resistance (300 Nm) is subtracted to compare
the results.
The final calibration values for the loss model are presented in Table 5.2,
5 Simulation model validation 22
Table 5.2 / Vehicle drive train.Symbol description value
Rw wheel radius 0.519 [m]
rgb gearbox ratios 7.1, 4.1, 2.5, 1.4, 1, 0.8 [-]
fd final drive ratio 5.13 [-]
Tss steady state differential losses 10 [N]
kω speed dependent loss factor 0.12 [Ns/rad]
kT
torque dependent loss factor 250 [N/N]
5.3 Engine map validation
testing procedure
The losses in the drivetrain are properly modeled, so the chassis dynamometer can be used to test the engine
fuel economy at different operation points. The dyno is set to a desired velocity to keep engine speed at a
constant level. The throttle position from the ICE is held stationary for about 10 seconds and than increased
with 5%. This is repeated for several engine speeds with intermittent values of 150 RPM, the evaluated points
are indicated in Figure 5.4(a). As an example, the result for one engine speed is given in Figure 5.4(b). The data
is filtered with a lowpass filter in order to remove signal noise. The fuel flow is measured with a fuelflowmeter
and compared with the CAN data as an extra check to avoid errors.
0
100
200
300
400
500
600
700
800
900
1000
800 1000 1200 1400 1600 1800 2000 2200 2400
Tor
que
[Nm
]
Rotational velocity [rpm]
Maximum TorqueMeasurement points
(a) Measurement points
0 50 100 150 200 250 3000
5
10
15
20
25
30
35
40
time [s]
fuel
use
[L/h
]
Fuelflow for 1600 RPM
FuelFlowmeterCAN data
(b) Fuel consumption at 1600 RPM
Figure 5.4 / Engine map validation procedure.
The measured torque at the dyno can be calculated back towards Tice with (5.3). A Matlab function is now used
to split the measurements, to use only the steady state parts, where fuel consumption is stabilized, at different
speed and torque points. The measured fuel flow mf is related to the Brake Specific Fuel Consumption (BSFC)
with the following formula:
BSFC(Tice, ωice) =mf (Tice, ωice)
ρdiesel Tice ωice36 · 106 (5.5)
5 Simulation model validation 23
for diesel a density, ρdiesel of 835 kg/m3 is assumed.
results
The resulting BSFC map is already presented in Figure 2.2 and this map is used in the simulation model.
The static engine modeling however, has some limitations which can be seen when driving the velocity profile
from 5.2. When the truck is driving in ‘engine only’ mode, so when power is solely provided by the ICE, driving
torque is not a direct function of pedal position. The engine torque is limited in rise and fall rate, especially at
low speeds ( < 1300 RPM). Due to emission legislation (soot) and turbo lag, a fair amount of time is required
to get towards maximum torque. Details can be seen in Figure 5.5(a). Here, the torque demand (accelerator
pedal position) and the actual engine percentage torque are given. The delay in engine torque is not modeled
in simulation, which causes a small error in trajectory tracking and fuel consumption. Furthermore, the pedal
position is limited at 89 % in order to prevent the gearbox from doing a kickdown: when driving a distance-
velocity profile, shifting down before accelerating, makes the response of the vehicle more non-linear, and is
therefore not desired.
2200 2400 2600 2800 3000 32000
10
20
30
40
50
60
70
80
90
100
perc
enta
ge %
distance
Pedal response from engine
Pedal positionTorque percentage
(a) Turbo lag
0 1000 2000 3000 4000 5000 6000
800
1000
1200
1400
1600
1800
2000
Engine Speed during constant speed cycle
distance [m]
Eng
ine
spee
d [R
PM
]
SimulationDyno
(b) Engine speeds
Figure 5.5 / Engine response during validation cycle.
An extra check can be made to ensure that gearbox ratios and wheel radius are correct, by comparing the
measured velocity from the dynamometer to the calculated velocity from the engine speed using the drivetrain
ratios from Table 5.2. From Figure 5.5(b) we can conclude that gearbox ratios and the wheel radius agree with
each other. One remark is made concerning the shift actions, which can be seen in Figure 5.5(b). Since the
truck on the dyno uses another software version than the simulation model, gear shifts are carried out in a
different way. The simulation model shifts without loss in velocity. The truck on the dyno, however, falls back
in vehicle velocity after each shift action. As a result, transient response during acceleration will differ between
dyno measurements and simulation. Using the measured drivetrain loss and engine map, the transient fuel
consumption and control decisions of the hybrid system can now be validated.
5 Simulation model validation 24
5.4 Transient fuel consumption validation
Road load, drivetrain losses and the engine map are matched between the simulation model and dyno. Hence,
it is possible to drive distance velocity trajectories to validate decisions of the existing rule based controller
and fuel consumption over a cycle. To see whether the velocity controller coincide with each other, the velocity
error for the profile driven in Figure 5.2 is depicted in Figure 5.6.
0 1000 2000 3000 4000 5000 6000 7000−1.5
−1
−0.5
0
0.5
1
1.5
Vel
ocity
err
or [m
/s]
Error signal
distance [m]
SimulationDyno measurements
Figure 5.6 / Velocity error for validation cycle.
From this figure we can conclude that the steady state error remains within 0.2 m/s on the dyno. Transient
responses are in good agreement with each other. However, the experimental results have a more transient
behavior. Overall, the difference in dynamic response can be explained by:
• the dyno velocity signal needs to be filtered to be useful as control input,
• in general, drivetrain and dynamometer dynamics are not modeled.
testing procedure
Since a hybrid vehicle is considered, a compensation for fuel consumption depending on the SOC difference
has to be made. When this compensation would not be performed, battery depletion is the most fuel effi-
cient. The compensation is done by taking into account the difference in SOC as a correction for the fuel
consumption:
mf = mf,original + ∆SOCE%SOC
EDieselλ (5.6)
here, ∆SOC is defined as SOC(t) − SOC(t0), E%SOC is the energy content for 1 % SOC, and EDiesel
the energy content of diesel in kWh/L. λ is the Lagrange parameter corresponding to the cycle dependent
5 Simulation model validation 25
incremental cost, see section 3.3. Hence, the fuel consumption and SOC trajectory for the cycle presented in
Figure 5.2 can be validated. The SOC and fuel consumption results are given in Figure 5.7.
0 1000 2000 3000 4000 5000 60000.3
0.35
0.4
0.45
0.5
0.55
0.6
0.65
Battery SOC
distance [m]
SO
C [−
]
SimulationDyno
(a) SOC trajectory
0 1000 2000 3000 4000 5000 60000
0.5
1
1.5Corrected cumulative fueluse
distance [m]
Fue
luse
[L]
SimulationDyno
(b) Cumulative fuel consumption
Figure 5.7 / SOC and fuel consumption for validation cycle.
Overall, good agreement exists between EMS control actions for the two systems. The cumulative fuel con-
sumption is within 2% difference between simulation and dyno tests. Since this profile is driven only once
and merely driven to, synchronize controller settings, adapt drivetrain properties, and to check fuel consump-
tion. With the settings adapted for this cycle, two more cycles are driven repeatedly. Their velocity profiles are
depicted in Figure 5.8. The limited dynamics cycle on the dynamometer is driven seven consecutive times,
while UDDS is repeated five times.
−500 0 500 1000 1500 2000 2500 3000 3500 40000
10
20
30
40
50
60
70
80
90Distance − Velocity profile
distance [m]
Vel
ocity
[km
/h]
SimulationDyno measurements
(a) Limited dynamics cycle
0 1000 2000 3000 4000 5000 6000 7000 8000 9000−20
0
20
40
60
80
100Distance − Velocity profile
distance [m]
Vel
ocity
[km
/h]
SimulationDyno
(b) UDDS cycle
Figure 5.8 / Fuel consumption validation cycles.
5 Simulation model validation 26
Results
For both the limited dynamics cycle and UDDS cycle, the corresponding SOC trajectories, for all the consec-
utive tests, are given in Figure 5.9.
0 500 1000 1500 2000 2500 3000 35000.3
0.35
0.4
0.45
0.5
0.55
0.6
0.65
Battery SOC
distance [m]
SO
C [−
]
SimulationDyno
(a) SOC trajectory for limited dynamics cycle
0 1000 2000 3000 4000 5000 6000 7000 80000.25
0.3
0.35
0.4
0.45
0.5
0.55
0.6
0.65
0.7Battery SOC
distance [m]
SO
C [−
]
SimulationDyno
(b) SOC trajectory for UDDS cycle
Figure 5.9 / SOC trajectories for fuel consumption validation cycles.
The SOC trajectories show that the decisions from the heuristic controller on the dyno and in the simulation
agree well with each other. For the limited dynamics cycle the battery gets charged during the cycle. To make
a fair comparison, the truck is driven in electric mode before starting a new test. This ensures that the battery
boundaries are not hit during the cycle. When the measurements are started with a low SOC, we can see that
the EMS tries to charge the battery towards 0.4. Looking at the regeneration actions, we can conclude that
the EM and battery model that is present in the simulation model, agrees very well with the real components:
nearly the same amount of energy is recovered when the vehicle decelerates in the simulation model and
on the dyno. However, the measurements on the dyno have the drawback that their SOC-level is discretized,
and therefore, it is hard to exactly compare the amount of regeneration energy. From the SOC trajectories on
the UDDS cycle, the same result can be concluded: regeneration and the decision to charge the batteries are
based on the same thresholds of the heuristic controller. Notice that it is hard to compare the strategies with
each other, since only SOC trajectories can be evaluated and no intermediate control decisions are available.
Finally, fuel consumption can be validated over the series of tests. With the aid of (5.6) the corrected cumu-
lative fuel consumption over a cycle can be presented. For the two validation profiles, the result is given in
Figure 5.10.
5 Simulation model validation 27
0 500 1000 1500 2000 2500 3000 3500 4000−0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9Corrected cumulative fueluse
distance [m]
Fue
luse
[L]
SimulationDyno measurements
(a) Cumulative fuel for limited dynamics cycle
0 1000 2000 3000 4000 5000 6000 7000 80000
0.5
1
1.5
2
2.5
Corrected cumulative fueluse
distance [m]
Fue
luse
[L]
SimulationDyno
(b) Cumulative fuel for UDDS cycle
Figure 5.10 / Cumulative fuel consumption for validation cycles.
For the tests that are driven multiple times on the chassis dynamometer, the average fuel consumption from
the consecutive measurements is taken and compared to the simulation. Results are given in Table 5.3.
Table 5.3 / Cumulative fuel consumption for dyno and simulation.Cycle Dyno mean [L] Simulation fuel[L] Difference [%]
Validation 1.246 1.264 + 1.5
Limited dynamics 0.817 0.799 - 1.6
UDDS 2.503 2.529 + 1.1
Note that the validation profile is driven only once, while the limited dynamics and UDDS are driven 7 and
5 times respectively. Fuel usage for the validation cycle is not averaged. Overall simulated fuel consumption
prediction is within 2 % of the experimental results, well within the desired 5 %. Besides, the reproducibility
of the experiments on the chassis dynamometer is between 3%.
5 Simulation model validation 28
5.5 Conclusion
Validation of the simulation model is done in a several steps, to reach the final goal, a fuel accuracy within
5%. First, coast down-measurements and a validation cycle are used to match the road load and drivetrain
properties between the simulation model and the actual vehicle. The torque at the wheels and engine speed
are measured to get these properties into good accordance with each other. Secondly, measurements are
performed to obtain a static engine map. This map suffices when predicting fuel consumption, however the
dynamical aspects of the engine, are not taken into account. Moreover, drivetrain and dynamometer dynamics
are not modeled and in combination with the different shifting procedure, this causes a discrepancy in
velocity tracking between simulation model and dyno.
Overall, from the SOC trajectories it can be concluded from the various validation profiles that the decisions
from the strategy are in good agreement between the simulation model and on the dyno. The drivetrain char-
acteristics and road load settings are comparable with each other and the fuel accuracy from the simulation
model agrees with the measurements well between the desired 5 %.
CHAPTER 6
Simulation results
The adaptive EMS, as presented in Chapter 4, is implemented in the simulation model that is validated in
the previous chapter. The model is used to (i) obtain tuning rules for the feedback parameters (3.14), and (ii)
evaluate the ECMS with the existing rule-based controller.
6.1 Drive cycles
The simulation model is validated, however, for real-time implementation of the ECMS, the parameters λ0
and K of (3.14) have to be chosen in advance. The aim is to find appropriate values for λ0 and K for the
real-time controller. To complete the overview of the ECMS performance, on a wide range of conditions, four
different distance velocities trajectories are evaluated. This results in tuning rules for the feedback parameters
and a comparison of the ECMS and the existing controller. A short description of the cycles and their distance
velocity profile is given in the next Section. The four cycles are firstly, the UDDS cycle, secondly is the Man-
hattan Bus Cycle, thirdly is the Federal Test Procedure (FTP - 75), and finally is the European Transient Cycle
(ETC).
Distance velocity trajectories
A short description of the cycles is given. Besides, for all the cycles their distance velocity trajectories are
depicted in Figure 6.1, the main parameters are given in Table 6.1.
• The Urban Dynamometer Driving Schedule (UDDS) for heavy duty vehicles UDDS, already used for
the validation of the simulation model, is a test with relative high speeds and a total length of 8.9 km.
• The Manhattan bus cycle, is a cycle which has a more transient behavior and is representing typical bus
start stop actions. From the four cycles that are considered, Manhattan has the highest acceleration and
deceleration.
• The Federal Test Procedure (FTP) 75 is a cycle used for emission certification of light duty vehicles in
the United States.
• The ETC cycle has been developed by TUV Automotive and is based on road cycle measurements
of heavy duty vehicles. Different driving conditions are represented by three parts of the ETC cycle,
including urban, rural and motorway driving.
6 Simulation results 30
0 1 2 3 4 5 6 7 8 90
20
40
60
80
100
Distance [km]
Vel
ocity
[km
/h]
(a) UDDS
0 0.5 1 1.5 2 2.5 3 3.50
10
20
30
40
50
Distance [km]
Vel
ocity
[km
/h]
(b) Manhattan
0 2 4 6 8 10 12 14 16 180
20
40
60
80
100
Distance [km]
Vel
ocity
[km
/h]
(c) FTP-75
0 5 10 15 20 25 300
20
40
60
80
100
Distance [km]
Vel
ocity
[km
/h]
(d) ETC
Figure 6.1 / Different distance velocity trajectories used for simulation.
Table 6.1 / Parameters for different cycles.Parameter UDDS Manhattan FTP - 75 ETC
total length [km] 8.93 3.32 17.8 29.4
average speed [km/h] 30.3 6.8 34.1 59.0
maximum speed [km/h] 93.0 40.9 91.2 91.1
average acceleration [m/s2] 0.41 0.63 0.56 0.25
average deceleration [m/s2] -0.51 -0.80 -0.65 -0.27
Simulation procedure
From(3.6) it follows that there is a constant λ for which the result is optimal for the given vehicle and driving
cycle combination. For all the cycles, bisection is used to find the value of λ0, with feedback set to zero, for
which the following condition holds:
SOC(t0) = SOC(tf ) (6.1)
The bisection method starts with SOC(t0) = 0.5 and a guess of the Lagrange parameter. The distance velocity
trajectory is than driven, with these settings. SOC(tf ) is used to correct the initial guess of λ0 until (6.1) holds.
According to the theory of Lagrange, this value of λ is considered to be optimal. However, a few remarks can
be made about this assumption:
• implementation of negative Treq is done in a heuristic way,
• the split ratio u is only controlled on supervisory level and is overruled, when for example a gearshift
is performed. Furthermore, the optimal control solution assumes that there are no state constraints on
the battery.
• the calculation is performed with discrete look-up tables. This can cause the solution to differ from the
optimal solution,
6 Simulation results 31
• the SOC dependent characteristics of the battery are neglected in the derivation of (3.7), while the sim-
ulation model takes a SOC depending resistance charge into account.
• To maintain driveability, the ICE is limited in its power during clutch closings.
the theory of Lagrange
Despite these differences, the value of λ is used to correct cumulative fuel usage with the aid of (5.6).
6.2 Simulation results
Cumulative fuel consumption for the optimal solution of the four cycles is given in Table 6.2. The cumulative
fuel consumption for the ECT cycle is depicted in Figure 6.2. The value of the feedback parameters for which
the cumulative fuel consumption is the lowest is used to clarify and compare the results for the other cycles.
The relative increase of fuel consumption compared to this ’lowest’ solution is depicted in Figure 6.3.
From Figure 6.3, it can be concluded that the general pattern for the four cycles is the same: with K = 0, the
choice of λ0 has large influence on the fuel consumption, while for larger numbers of K, the choice of λ0
has smaller influence on total fuel consumption. For even higher numbers of K, the fuel consumption rises
slightly and deviates from the optimal solution. Since for real-time implementation an open-loop controller,
(K = 0), is not preferable, some feedback should be implemented.
Although for higher values of K it was expected to see a higher fuel consumption and to follow the desired
SOC more strictly, still energy recuperation is done in a heuristic way. Besides, the look-up tables containing
the optimal split value in the controller are defined preceding the simulation. The creation of this split-maps
is done with predefined settings of the feedback values, which puts a limit on the feedback parameters. For
high values, K > 3 the ‘lookup solution’ will provide the same answer than a smaller value of K gives.
Therefore the influence of very high values of K is limited. However, choosing K very high will result in
very strict tracking the desired SOC and will therefore be less advantageous for battery wear. Besides, these
feedback settings do not provide possibilities for the controller to transfer energy over a period of time: all
the energy that is recuperated will be used immediately. From the fuel consumption figures we can conclude
that tuning parameters for this controller and these cycles should be with λ0 ' 2.55 and K ≥ 1.5
Table 6.2 / Parameters for different cycles.Parameter UDDS Manhattan FTP - 75 ETC
λopt 2.59 2.43 2.47 2.48
Total fuel consumption [L] 2.38 1.09 4.40 2.81
Fuel consumption [L/km] 0.258 0.332 0.247 0.259
6 Simulation results 32
01
23
4
2
2.5
32.8
2.9
3
3.1
3.2
K [−]
Fuel consumption for ETC cycle
λ0 [−]
Cum
ulat
ive
fuel
usa
ge [L
]
Figure 6.2 / Fuel consumption for various ECMS feedback parameters on the ETC cycle.
0.5
0.50.5
0.5
0.5
0.5
1
1
1
1
1
1
1.5
1.5
1.5
1.5
1.5
2
2
2 2
2
3
3
34
K
λ0
2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 30
0.5
1
1.5
2
2.5
3
(a) UDDS
0.5
0.5
0.5
11
1
1
1
1.5
1.5
1.5
1.5
1.5
2
2
2
2
2
3
3
3
3
4
4
4
4
55
5
7 7
7
10
K
λ0
2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 30
1
2
3
4
5
6
7
8
9
(b) Manhattan
0.5
0.5
0.5
0.5
0.5
0.5
1
1
1
1
1
1
1.5
1.5
1.5
1.5
1.5
2
2
2
2
2
3
3
3
3
4
4
4
5
5
7
7
K
λ0
2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 30
0.5
1
1.5
2
2.5
3
3.5
4
(c) FTP-75
0.5
0.5
0.5
0.5
0.5
0.5
1
1
1
1
1
1
1.5
1.5
1.5
1.5
1.5
1.5
2
2
2
2
2
3
3
3
3
3
4
4
4
4
5
5
5
7
710
K
λ0
2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 30
0.5
1
1.5
2
2.5
3
3.5
4
(d) ETC
Figure 6.3 / Percentage deviation in fuel consumption for different distance velocity trajectories compared
with lowest fuel consumption solution.
6 Simulation results 33
6.3 Comparison with heuristic controller
In the previous section, the settings for the ECMS feedback parameters for real-time implementation are
presented. Hence, it is possible to compare the original heuristic controller with the ECMS to see if the
ECMS controller provides any advantage in terms of fuel consumption. All four cycles are simulated with the
heuristic controller and with the ECMS strategy. For the ECMS, feedback settings of λ0 = 2.55 and K = 1.5
are chosen. The heuristic controller is driven with an initial SOC of 0.35. The results are presented in Table
6.3.
Table 6.3 / Fuel consumption on different drive cyclesCycle Heuristic [L] ECMS [L] difference [%]
UDDS 2.94 2.82 -3.9%
Manhattan 4.45 4.39 -1.4%
FTP - 75 2.35 2.32 -1.0%
ETC 1.14 1.09 -3.9%
From this data we can conclude that the ECMS strategy provides a benefit in fuel consumption of at least 1%.
The reduction in fuel consumptions is highly dependent from the cycle that is driven. ECMS seems better on
highway trajectories. A possible explanation for this can be given by the fact that the ECMS controller is not
always free to change the split ratio instantaneous because of driveability limitations . Due to the configuration
of the components, the clutch has to close when going from split ratio u = 0 to higher values. For distance
velocity profiles with a high dynamic behavior, this situation occurs more often than for low dynamic profiles.
However, the ECMS provides better fuel consumption for all cycles. Another advantage over the heuristic
strategy, is that ECMS can cope with a varying SOCdes over time. This makes implementation especially
beneficial when upcoming route profiles (with, e.g., slopes) can be taken into account.
CHAPTER 7
Conclusions and recommendations
7.1 Conclusions
This thesis presents a practical implementation of an Energy Management Strategy (EMS) for a hybrid DAF
CF. The goal of the EMS is to reduce cumulative fuel consumption, while the on board computational power,
limits the complexity of the controller. The method of Lagrange is used to simplify the fuel minimization
problem towards the Equivalent Consumption Minimization Strategy (ECMS).
The simulation model that is used for the EMS development is validated with a test vehicle on the chassis dy-
namometer. As a first step, the chassis dynamometer at Eindhoven University of Technology is made suitable
for validation. Therefore, several adaptations to the control system of the chassis dynamometer have been
made. Extra safety precautions are included and the velocity tracking controller is tuned. Furthermore, for
future development, the dyno is adjusted to simulate elevation in the distance velocity profiles.
The validation is done in three steps: first, the model of the drivetrain losses is validated. Secondly, the fuel
consumption, as function of engine speed and torque is measured and used to predict fuel consumption.
Finally, transient velocity distance profiles are used to match thresholds for the existing heuristic controller
and in this way cumulative fuel consumption for several distance velocity trajectories is predicted within 2%.
Extensive simulations have been performed to find good initial settings for the feedback parameters for the
online estimation of the Lagrange parameter. When providing some feedback towards the desired battery
state-of-charge, the influence of the initial value of the Lagrange parameter, on fuel consumption becomes
limited. It is shown that the ECMS is able to perform well, for a given set of feedback parameters on different
distance velocity trajectories. Fuel savings are at least 1% compared to a properly tuned heuristic controller.
Besides EMS development, the simulation model enables modification of the vehicle Electronic Computa-
tional Unit. Therefore, the ECMS can be tested on the chassis dynanometer as well. Material breakdown
caused some delay in the testing and therefore testing of the ECMS on the dyno has not been finished yet.
However, the first test results with the ECMS on the chassis dynamometer are promising.
7.2 Recommendations
The static map that is used to model the EM efficiency, does not take not include torque losses at zero power
output into account, so when Tem = 0. This model does not correspond very well to the physical behavior of
the component, so for future research a more detailed model of the EM is preferred.
Since not all the appliances are electrified, in the current drivetrain, switching off the engine when it is not
7 Conclusions and recommendations 35
used, is not possible. Together with the limitation on engine torque when a clutch closure occurs, makes that
the hybrid system can be even more efficient with the given vehicle configuration. Switching off the engine,
however, should be penalized by the EMS to maintain driveability and prohibit excessive component wear.
Further research is required to find out what the estimated fuel savings of this measure are.
The test vehicle and dynanometer at Eindhoven University of Technology are fully operational, which enables
validation of the ECMS strategy on a chassis dynamometer in the near future. Since the controller is able to
receive a time dependent battery state of charge trajectory, development of algorithms to calculate optimal
SOC trajectories (e-horizon) along a velocity profile, could make an ECMS even more beneficial. This is
especially the case when route profiles with elevation are considered.
An extension of the ECMS strategy with a gear shifting algorithm could provide an even more fuel economic
EMS, however driveability issues arise when real time implementation is considered: shifting to a higher gear
number will be beneficial under most conditions. On the other hand: the amount of torque available from
the engine will decrease after this shift action. Therefore the driver will encounter a slower dynamic response
from the vehicle. More research is necessary to find an optimum for the trade off between fuel economy and
driveability.
Dankwoord
Het afgelopen jaar is heel erg vlug voorbijgegaan, en na zo’n lange tijd aan één project werken, kan het niet
kwaad om even terug te kijken. Mijn afstudeeropdracht heeft mij bijzonder geboeid en had alles in zich om
een bijzonder veelzijdige, interessante en uitdagende opdracht te zijn. Zo’n opdracht kun je niet alleen doen
en daarom wil ik vanaf deze plaats een paar mensen bedanken. Allereerst, de mensen bij TNO Automotive,
voor de vele mogelijkheden en de leuke tijd die ik in Helmond gehad heb. Hoewel de economische situatie
zo nu en dan zijn weerslag hadden op de afdeling, was het bijzonder leerzaam en interessant om daar een
opdracht te mogen verrichten. De hele afdeling heeft mij daar enorm geholpen, maar een speciaal woord van
dank gaat uit naar Finnis en John. Bij beide kon ik altijd aankloppen voor vragen of extra informatie, hoe vaak
dat ook mocht zijn en immer werden die vragen beantwoord met een niet aflatend enthousiasme.
De laatste maanden van het project vonden voornamelijk plaats op de TU/e. Het in de vingers krijgen van
zo’n gecompliceerde testopstelling heeft aardig wat geduld en hoofdbrekens gekost, maar eindigen met een
werkende testopstelling geeft veel voldoening. Ook de mensen van de Technische Universiteit die het project
ondersteunden, ben ik erg dankbaar. Mijn begeleider Thijs wil ik bijzonder bedanken voor de fijne samen-
werking. De deur die steeds open stond voor verdere uitleg en de vele uren aan de rollenbank heb ik bijzonder
gewaardeerd.
Natuurlijk doe je studeren niet alleen en daarom ook bedankt aan alle studiegenoten voor de broodnodige
bakjes koffie en gezelligheid tussendoor. Bedankt aan alle familie en vrienden die me gesteund hebben
tijdens mijn studie. En als laatste, maar zeker niet als minste: papa en mama voor alle mogelijkheden die ik
gekregen heb tijdens mijn studie, Sophie, die altijd voldoende kan relativeren en Jolanda voor het geduld wat
je met mij wel eens moet hebben.
Dominique
Appendix A
Schematic overview of the different control systems
on the dynanometer
C1 C2
Horiba
SP
iABB roll
inertiaTset
Tm
eT ω
Tm
Trol
chassis dyno
testvehicle
MACS
pedalpositions
iFF
vr(x)
referencetrajectory
-
-
ev
FeedForward
velocitycontrol
αr(x) Tvehvrol
-+
ar, vr, αrγdyno
γpedal
1s
ω+
SOCr(x)
x
Dyno Safety systemSafety control onmax. velocity
ω
ωhybrid
usafe
Figure A.1 / Schematic overview of the different control systems on the dynanometer
In Figure A.1 the different control schemes for the chassis dynanometer are depicted. The MACS-controller
takes care of the velocity trajectory tracking, and handles the dyno safety. For the dyno safety one of the analog
outputs (AO) is coupled towards the Dyno safety input. Giving a high voltage (>5 Volts) on this output results
in switching off the Horiba controller. This output is generated when ωmot or vdyno is above threshold level.
Dynamic heigth profiles
The Horiba controller is able to simulate a dynamica heigth profile (road slope α as function of driven dis-
tance). The MACS-controller AO generates a voltage corresponding to the desired road slope. This voltage is
depending on the driven distance.
Bibliography
[1] S. Delprat, J. Lauber, T.M. Guerra and J. Rimaux Control of a Parallel Hybrid Powertrain: Optimal Control.
IEEE Transactions on Vehicular Technology, Vol. 53, No. 3, May 2004.
[2] L. Guzzella, A. Sciarretta. Vehicle Propulsion Systems. ISBN 3-540-25195-2, Springer, 2005.
[3] T. Hofman, R. Van Druten, A. Serrarens, M. Steinbuch. Rule-based energy management strategies for
hybrid vehicles Int. Journal of Electric and Hybrid Vehicles, Vol. 1, No. 1, pp. 71-94, 2007.
[4] C. Lin, H. Peng, J. Grizzle, J. Kang. Power Management Strategy for a Parallel Hybrid Electric Truck
IEEE Trans. on Control Systems Tech.., Vol. 11, No.6, pp. 839-349, November 2003.
[5] C. Musardo, G. Rizzoni, B. Staccia A-ECMS: An Adaptive Algorithm for Hybrid Electric Vehicle Energy
Management. Proceedings of the 44th IEEE Conference on Decision and Control, and the European Control
Conference 2005 Seville, Spain, December 12-15, 2005.
[6] J. Kessels Energy Management for Automotive Power Nets ISBN 90-386-1963-4, Unversiteits-
drukkerij,Technische Universiteit Eindhoven, 2007.
[7] J. Kessels, P. van den Bosch. Electronic Horizon: Road Information used by Energy management Strate-
gies. Int. Journal of Intelligent Information and Database Systems, Vol.2, No. 2, pp. 187-203, 2008.
[8] T. van Keulen, B. de Jager, M. Steinbuch An Adaptive Sub-Optimal Energy Management Strategy for
Hybrid Drive-Trains. Proceedings of the 17th IFAC World Congress, Seoul, Korea, July 6-11, 2008.
[9] T. van Keulen, B. de Jager, A. Serrarens, M. Steinbuch Optimal Energy Management in Hybrid Electric
Trucks using Route Information. Int. Conf. on Advances in Hybrid Powertrains, Rueil Malmaison, 2008.
[10] M. Koot, J. Kessels, B. de Jager, W. Heemels, P. van den Bosch, M. Steinbuch Energy Management
Strategies for Vehicular Power Systems. Transaction on vehicular technology, Vol. 54, No. 3, pp. 1-11, May
2005.
[11] G. Paganelli, T. Guerra, S. Delprat, J. Santin, M. Delhom, E. Combes, Simulation and Assessment of
Power Control Strategies for a Parallel Hybrid Car. Proceedings of the Institution of Mechanical Engineers,
Part D: Journal of Automobile Engineering, Vol. 214, No. 7, pp. 705-717, 2000.
[12] P. Pisu, G. Rizzoni. A Comparative Study of Supervisory Control Strategies for Hybrid Vehicle Energy
Managment. IEEE Transactions on Control Systems Technology, Volume 15, No. 3, pp. 506-518, May 2007.
[13] P. Pisu, K. Koprubasi, G. Rizzoni. Energy Management and Drivability Control Problems for Hybrid Elec-
tric Vehicles. Proc. of the 44th IEEE Conf. on Decision and Control, Seville, Spain, pp. 1824-1830, December
2005.
BIBLIOGRAPHY 39
[14] SAE J-1506 Emission Test Driving Schedules. SAE Standard, 2002.
[15] SAE J-2711 Recommended Practice for Measuring Fuel Economy and Emissions of Hybrid-Electric and
Conventional Heavy-Duty Vehicles Issued 2002-09.
[16] A. Sciarretta, L. Guzella. Control of Hybrid Electric Vehicles. IEEE Transactions on Control Systems
Technology, volume 27, No. 2, pp. 60-70, April 2007.
[17] R.T.M Smokers, S. Ploumen, M. Conte, L. Buning, K. Meier-Engel, Test Methods for Evaluating Energy
Consumption and Emission of Vehicles with Electric, Hybrid and Fuel Cell Power Trains. Proc. of the EVS
17, Driving new visions Montreal, Canada, October 2000.