You are on page 1of 14

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 60, NO.

7, SEPTEMBER 2011

2949

Energy-Optimal Control of Plug-in Hybrid Electric Vehicles for Real-World Driving Cycles
Stephanie Stockar, Vincenzo Marano, Marcello Canova, Giorgio Rizzoni, Fellow, IEEE, and Lino Guzzella, Fellow, IEEE

AbstractPlug-in hybrid electric vehicles (PHEVs) are currently recognized as a promising solution for reducing fuel consumption and emissions due to the ability of storing energy through direct connection to the electric grid. Such benets can be achieved only with a supervisory energy management strategy that optimizes the energy utilization of the vehicle. This control problem is particularly challenging for PHEVs due to the possibility of depleting the battery during usage and the vehicle-to-grid interaction during recharge. This paper proposes a model-based control approach for PHEV energy management that is based on minimizing the overall CO2 emissions produceddirectly and indirectlyfrom vehicle utilization. A supervisory energy manager is formulated as a global optimal control problem and then cast into a local problem by applying the Pontryagins minimum principle. The proposed controller is implemented in an energy-based simulator of a prototype PHEV and validated on experimental data. A simulation study is conducted to calibrate the control parameters and to investigate the inuence of vehicle usage conditions, environmental factors, and geographic scenarios on the PHEV performance using a large database of regulatory and real-world driving proles. Index TermsEnergy management, optimal control, plug-in hybrid electric vehicles.

N OMENCLATURE Cnom E f H Ibatt J K L m CO2 m equiv m f QLHV Pbatt Battery nominal capacity. Energy. Torque split factor. Hamiltonian. Battery current. Cost functional. Final state penalty term. Lagrangian. CO2 mass ow rate. Equivalent mass ow rate. Fuel mass ow rate. Fuel lower heating value. Battery power.

R s S t T u Vbatt Voc x ch i l CS x AER BSA CS CD ECMS EM HEV ICE PHEV SDP SoC SoE

Battery internal resistance. Equivalency factor. Target set for the nal state. Time. Torque. Control law. Battery voltage. Open circuit voltage. State variable. Efciency. Battery charger efciency. Lagrange multiplier. Initial condition for the Lagrange multiplier. Specic CO2 content. Scalar Lagrange multiplier. Fraction of driving cycle in CS mode. Angular velocity. Feasible set for the state. All electric range. Belted starter alternator. Charge sustaining. Charge depleting. Equivalent consumption minimization strategy. Electric motor. Hybrid electric vehicle. Internal combustion engine. Plug-in HEV. Stochastic dynamic programming. State of charge. State of energy.

I. I NTRODUCTION PHEVs are today considered to be a solution to reduce fuel consumption and CO2 emissions in the transportation sector. Compared with conventional hybrid vehicles, the high-capacity energy storage system of PHEVs and the ability to recharge the battery through connection to the electric grid provide the opportunity to control the battery depletion during vehicle utilization, ultimately improving fuel economy. Various studies have shown that the performance of PHEVs depends on several factors, many of which have little or no inuence on CS hybrids and conventional vehicles [1][7]. To name a few, the length of the driving path, the contribution of the electricity on the overall energy consumption of the vehicle, the cost of the electric energy, and its specic CO2 content have

Manuscript received July 4, 2010; revised January 6, 2011 and March 31, 2011; accepted May 8, 2011. Date of publication June 2, 2011; date of current version September 19, 2011. The review of this paper was coordinated by Dr. C. C. Mi. S. Stockar, V. Marano, M. Canova, and G. Rizzoni are with the Center for Automotive Research, The Ohio State University, Columbus, OH 43212 USA (e-mail: stockar.1@osu.edu). L. Guzzella is with the Department of Mechanical and Process Engineering, Eidgenoessische Technische Hochschule Zurich, 8092 Zurich, Switzerland. Color versions of one or more of the gures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identier 10.1109/TVT.2011.2158565

0018-9545/$26.00 2011 IEEE

2950

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 60, NO. 7, SEPTEMBER 2011

been recognized as predominant factors in the assessment of fuel economy and emissions for PHEVs. A subject of strong interest on the part of the automotive industry is to understand the implications of different energy management strategies on fuel consumption, CO2 emissions, battery life, range, and performance. To this extent, one of the critical challenges for control design is to properly account for the grid energy in the vehicle energy optimization problem. Some methods have been so far proposed to design supervisory controllers for PHEVs, including the minimization of an equivalent fuel consumption, or the vehicle operating costs or the cumulative CO2 emissions [2][4], [8], [9]. Due to the complexity of the control problem, heuristic methods have often been considered [8], [10][12]. Although rule-based methods are often successfully employed in the industry, it is generally found that the controller design process is cumbersome and time consuming, and its results are limited to specic vehicle design and usage conditions. For this reason, model-based approaches can improve on all of these drawbacks and yield more cost-effective solutions. SDP appears more indicated, particularly when a small number of reference driving proles can be found as statistically representative of the vehicle utilization [13]. Recent results show improvements in fuel economy, operation costs, and emissions [8], [14]. The SDP approach, however, requires signicant amount of data to provide a statistically relevant validation framework. Further, the policy evaluation typically requires large computation time, partially overcome by estimating the control policy ofine and tabulate the results in the actual implementation. The ECMS is a well-known approach for the online energy management of HEVs that has recently been adapted to the supervisory control of PHEVs [15][20]. The proposed approach is based on assuming that the energy expended by the vehicle can be converted into an equivalent consumption of fuel. The results presented lead to the conclusion that near-optimal fuel economy can be achieved if the control algorithm depletes the battery proportionally to the driving distance. However, this implies that the vehicle velocity prole must be known a priori. Such condition prevents the ECMS to be generalized, requiring calibration of the equivalency factor for each driving prole. Furthermore, the assumption of converting the battery energy into an equivalent fuel mass ow rate is not formally applicable to PHEVs since the electric energy stored from the grid depends on the energy generation mix. This paper presents a novel supervisory energy management strategy for charge-depleting hybrid vehicles that accounts for the vehicle primary energy consumption, including the fuel energy and the electric energy from the grid. The structure of the proposed algorithm is general and adaptable to different vehicle architectures (series, parallel, and seriesparallel) and to any number of power splits. The proposed approach is based on the formulation of a global optimal control problem that minimizes the global CO2 emissions produced (directly and indirectly) by vehicle use. Pontryagins minimum principle is then applied to obtain a local minimization problem. The control strategy is applied to a forward-oriented simulator of a seriesparallel PHEV and used

Fig. 1.

Diagram of the prototype PHEV drivetrain. TABLE I DESCRIPTION OF THE VEHICLE DRIVETRAIN COMPONENTS

to conduct vehicle performance analysis, evaluating the impact of the control parameters for a variety of vehicle utilization and environmental scenarios. This paper is organized as follows: After an overview of the hybrid vehicle conguration and the model adopted for the control study, a description of the energy management strategy and its implementation into a control algorithm are given. Simulation results are presented to evaluate the sensitivity of the proposed control strategy to vehicle usage conditions and environmental and geographic scenarios, also providing an assessment of vehicle performance, fuel consumption, and CO2 emissions. II. D ESCRIPTION OF THE V EHICLE AND OF THE S IMULATOR The vehicle considered in this study is a seriesparallel prototype PHEV built on a midsize SUV platform [21], [22]. As shown in Fig. 1, the vehicle drivetrain includes a downsized diesel engine coupled with a BSA and a six-speed automatic transmission on the front axle and an EM on the rear axle. Table I describes the main vehicle components. The conguration chosen for this vehicle allows for a variety of operating modes, such as pure electric drive, electric launch, engine load shifting, motor torque assist, and regenerative braking [22]. A forward-oriented energy-based simulator was developed and validated using a combination of driving tests and laboratory test data [21][24]. Fig. 2 describes the information

STOCKAR et al.: ENERGY-OPTIMAL CONTROL OF PHEVs FOR REAL-WORLD DRIVING CYCLES

2951

steady-state and low-frequency behavior of the system. More complex and accurate battery models, for example, including the high-frequency dynamics and the effects of temperature, can be included in the same vehicle simulator, for example, to study the effects of driving conditions and energy management control on battery aging [29]. The battery model is here utilized to formulate the energy management control problem. In this case, the battery dynamics is described by the state equation
Fig. 2. Information ows in the forward-oriented vehicle simulator.

Ibatt (t) d SoC(t) = dt Cnom = batt , 1 batt , if Ibatt (t) 0 if Ibatt (t) > 0

(3) (4)

ows within the vehicle simulator [25][27]. The accelerator and brake pedal position commands from the driver are input to the controller, which determines the torque commands to the engine, EM, and BSA. The powertrain block computes the tractive force, which is the input to a vehicle longitudinal dynamics model that predicts the vehicle velocity. The drivetrain components included in the vehicle powertrain are detailed in Fig. 3. The energy-based (quasi-static) modeling approach is adopted to predict the overall vehicle fuel consumption over a driving cycle, neglecting the high-frequency dynamic effects [25]. The engine model is based on its steadystate fuel consumption map implemented in the simulator as a function of engine speed and input torque. Similarly, the electric machines are modeled as static elements, wherein the efciency is mapped as a function of their speed and input torque. The combined engine and BSA torque is transmitted through a torque converter and a six-speed automatic transaxle, whereas the EM is coupled to the rear axle through a xed gearbox. Losses in the transmission components are accounted for through the denition of efciency terms. The gear shifting strategy is determined by a simple scheduling controller based on the engine speed and the accelerator command. A. Energy-Based Model of the Battery As shown in Table I, the vehicle includes a 10-kWh Li-ion battery pack, which enables for an AER of 25 km [1]. A simplied model of the battery was built according to the equivalent circuit analogy [24], [26], [28]. In particular, a zeroorder model is considered here to describe the battery voltage output Vbatt (t) = Voc (SoC(t)) R (SoC(t)) Ibatt (t). (1)

where the battery efciency is dened as batt (SoC(t), t) = Vbatt (t) . Voc (SoC(t)) (5)

For the zero-order model described by (1), the battery efciency can explicitly be calculated as batt (SoC(t), t) = 1 R (SoC(t)) Ibatt (t) . Voc (SoC(t)) (6)

III. OVERVIEW OF THE E NERGY M ANAGEMENT P ROBLEM FOR C HARGE S USTAINING H YBRID V EHICLES For a CS HEV, the supervisory controller formulates a control law u(t) that minimizes a cost function over a period of time [ta , tb ]. Commonly, the cost function is the vehicle fuel consumption
tb

JHEV (u) = mf =
ta

m f (t) dt

(7)

calculated over the entire mission of the vehicle. A constraint on the battery SoC at the end of the mission is also included to ensure nominally CS operations, i.e., SoC(tb ) = SoC(ta ). (8)

The open circuit voltage Voc and internal resistance R are polynomial functions of the battery SoC [28]. The SoC of the battery is dened as
t

SoC(t) = SoC0

1 Cnom
0

Ibatt (t) dt

(2)

The control sequence u(t) that satises the preceding state constraint is the solution of the optimal control problem [25], [30][34]. By applying Pontryagins minimum principle [35], [36], the constrained global optimization problem previously presented is cast into a local minimization problem given by a Hamiltonian function dened by the vehicle equivalent fuel consumption [25], [37] f (t) + s(t) m equiv (t) = m Pbatt (t) QLHV (9)

where Cnom represents the nominal battery capacity, as indicated in Table I. Although the foregoing model represents a strong approximation of the real battery behavior, it is consistent with the energy-based formulation, which limits the analysis to the

where s(t) is a fuel energy equivalency factor (nondimensional), Pbatt is the net power drawn from the battery, and QLHV is the fuel lower heating value. The cost functional dened by (9) is minimized at each time step. This allows one

2952

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 60, NO. 7, SEPTEMBER 2011

Fig. 3. Block diagram of the drivetrain power ows.

to nd a solution to the optimal control problem that can be implemented on a vehicle. Note that the approximation of converting an electrical energy utilization into a fuel mass ow rate introduces the equivalency factor s(t). This calibration parameter has a considerable impact on the battery SoC during a driving path. For this reason, the equivalency factor must be optimized or adapted based on the specic vehicle driving prole considered to achieve optimal fuel economy and CS operations [27], [37]. Experimental results, however, show that the ECMS performs close to the global optimum with modest calibration effort, with the advantage of being implementable online [21], [31]. IV. O PTIMAL C ONTROL P ROBLEM F ORMULATION FOR P LUG -I N H YBRID E LECTRIC V EHICLE E NERGY M ANAGEMENT To dene a supervisory energy management strategy for PHEVs, an optimal control problem for charge-depleting systems is formulated here. Compared with the energy management problem formulation previously presented, the constraint on the nal SoC dened in (8) must be eliminated to enable charge-depleting operations. Further, the equivalence between the battery energy usage and the fuel mass ow rate shown in (9) is formally incorrect for PHEVs, where the battery energy is mostly provided by the grid, hence decoupled from the fuel energy. This implies that the cost function must be redened for PHEVs. In this paper, the cost function is dened to account for the primary energy consumed by the vehicle during a driving path. The most representative indicator of the well-to-wheel energy utilization of a PHEV is given by the cumulative CO2 emissions produced by vehicle utilization, i.e.,
tb

To apply the optimal control theory to the PHEV energy management, the variables mCO2 ,f and mCO2 ,e must be related to vehicle system variables as follows: m CO2 ,f (t) = 1 Pfuel (t) m CO2 ,e (t) = 2 Pbatt (t) ch (11)

where, according to Fig. 3, Pfuel is the power associated with the fuel utilization and is determined as follows, assuming a lower heating value (QLHV = 43 MJ/kg): Pfuel (t) = m f (t) QLHV . (12)

The term ch = 0.86 in (11) represents the battery charger efciency (when the vehicle is connected to the grid) [38], and 1 and 2 are dened as the specic CO2 content in fuel and electricity per kilowatthour. Note that 1 corresponds to the engine brake specic CO2 (BSCO2 ), which can readily be calculated from fuel consumption data. The term 2 can reasonably be estimated based on the average CO2 content of the electricity generation mix for a specic geographic region [39]. To account for the energy stored in the battery, the SoE is introduced as SoE(t) = Ebatt (t) Enom (13)

where Enom = Cnom Voc is the nominal battery energy (in kilowatthours). Considering SoE as the new state variable instead of the SoC, the state equation of the system becomes Pbatt (t) d SoE(t) = (SoC(t)) dt Enom (14)

JPHEV (u) =
ta

m CO2 ,f (t) + m CO2 ,e (t) dt

(10)

where mCO2 ,f represents the mass CO2 produced by the consumption of fuel (when the engine is utilized), and mCO2 ,e results from the consumption of the electric energy stored onboard.

where is dened according to (5), and Pbatt is the battery power, which is dened as positive if discharging. Note that, if Vbatt (t) = Voc , then SoE = SoC.

STOCKAR et al.: ENERGY-OPTIMAL CONTROL OF PHEVs FOR REAL-WORLD DRIVING CYCLES

2953

Based on the preceding state equation, the control variable u(t) for the energy management problem can be dened as the vector PEM,el (t) u(t) = Pbatt (t); Pbatt (t) (15)

which has to be minimized at each time t to provide the optimal control policy. If uo (t) is the optimal control policy, then the following necessary conditions must be satised: dxo (t) = H |o = f (xo (t), uo (t), t) dt do (t) = x H |o ii) dt o iii) x (ta ) = xa iv) xo (tb ) S Rn v) H (xo (t), uo (t), o (t), t) H (xo (t), u(t), o (t), t) i) If the state x(t) is bounded, namely vi) xo (t) x (t)t [ta , tb ] x (t) = {x Rn |G(x, t) 0; G : Rn x[ta , tb ] R} where G(x, t) denes the inequality constraints, an additional term is introduced in the Hamiltonian function to account for this limitation. The corresponding Lagrange multiplier is a scalar denoted by l and subject to the KuhnTucker condition vii) o l 0. For the PHEV control problem, an extended Hamiltonian function is dened based on the cost functional in (10) and the state constraints on the battery SoE H (x(t), u(t), (t), (t), t) = 1 Pfuel (t)+ + Pbatt (t) 2 (t) (SoC(t)) (t) (SoC(t)) ch Enom Enom (21)

where the second element represents the power split between the rear EM and the BSA electric power outputs. Since the mechanical power demand to the drivetrain is known, the electric EM and BSA power can be obtained from the efciency maps of the two components and simple energy balances, according to the power ow diagram in Fig. 3. To respect the physical limitations imposed by the drivetrain components, the control and state variables are subject to constraints. In particular, the battery SoE and power must be limited to prevent abuse and aging-related issues [40], i.e., SoEmin SoE(t) SoEmax Pbatt,min Pbatt (t) Pbatt,max (16)

where, usually, SoEmin = 0.25, and SoEmax = 0.95. Further constraints stem from the power limits of the drivetrain components PEM,min PEM (t) PEM,max PBSA,min PBSA (t) PBSA,max (17)

where the power limits are functions of the EM and BSA speed.

V. S OLUTION OF THE P LUG -I N H YBRID E LECTRIC V EHICLE E NERGY M ANAGEMENT P ROBLEM The optimization problem previously dened is tackled using Pontryagins minimum principle [35], [36], which, in principle, allows one to obtain closed-form expressions for locally optimal control signals. In the general case, an explicit control signal can only be found by solving a two-point boundary value problem. For the specic problem at hand, an optimal solution can be found by adopting the reasonable simplications shown below. The starting point is a description of the system dynamics dx(t) = qf (x(t), u(t), t) dt with the cost functional
tb

with

if SoE(t) SoEmax l , (t) = l , if SoE(t) SoEmin 0, else

(22)

(18)

J (u) =
ta

L (x(t), u(t), t) dt + K (xb , tb ).

(19)

where Pfuel can be calculated from the engine fuel consumption maps, as in Fig. 3, and l is the scalar Lagrange multiplier for the inequality constraints on the SoE. The extended Hamiltonian function allows one to include the state constraints within the same optimal control problem. Note that (21) provides necessary conditions for optimality according to the previously mentioned conditions. Such formulation, however, can lead to suboptimal results if the state constraints are active. When this occurs, the optimal value for the parameter l is unknown and should be determined by applying conditions i)vii). Since the time intervals during which the state is sliding along the upper or lower boundary are limited in occurrence, the value for 0 l is determined here by a trial-and-error procedure [35]. The necessary condition for the costate o (t) is do (t) =x H |o= (1 Pfuel (t)) dt x x + x 2 Pbatt (t) ch + (23)

The theorem introduces the Hamiltonian function H (x(t), u(t), (t), t) = L (x(t), u(t), t)+ (t) f (x(t), u(t), t) (20)

Pbatt (t) (SoC(t)) (o (t) + (t)) Enom

2954

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 60, NO. 7, SEPTEMBER 2011

with the initial condition o (t = ta ) = i . Since no explicit condition is given for i , this parameter needs to be calibrated. The ordinary differential equation (ODE) for the costate o (t) can further be simplied since Pfuel (t) and Pbatt (t) do not depend on the battery SoE (or SoC). However, the preceding assumption is not valid for the battery efciency. In fact, the battery power is Pbatt (t) = Ibatt (t) Vbatt (t), whereas the maximum battery power during discharging is Pmax (t) = Ibatt (t) Voc (SoC(t), t). This will further penalize any operation at low SoE when the battery efciency is lower. Inserting this expression in (23), the costate equation can be rewritten as follows:
batt if Pbatt (t) < 0 d (t) x , o = Pbatt (t)( (t)+(t)) batt dt 2 x , if Pbatt (t) 0. Enom batt (SoC(t),t) (24) o

o Pbatt (t)( (t)+(t)) Enom

According to the minimum principle, the control policy denoted by uo (t) is optimal if H (xo (t), u(t), o (t), t) presents a global minimum with respect to uo (t). As a nal remark, a proof of equivalence between the ECMS and the solution of the optimal control problem through Pontryagins minimum principle was obtained for the CS HEV case in [27] and [31]. This proof is extended here to the chargedepleting PHEV case. In fact, the ECMS formulation presented in (9) can be made equivalent to the Hamiltonian function dened by (21) if the equivalency factor s(t) is dened as (SoC(t), t) 2 ((t) + (t)) . s(t) = 1 ch Enom 1
Fig. 4. Flowchart describing the implementation of the energy management algorithm.

(25)

at the driveshaft Treq is evaluated using the driver accelerator and brake commands and as follows:
+ Treq (t) = (t) Tmax + (t) Tmax

(27)

VI. I MPLEMENTATION OF THE E NERGY M ANAGEMENT S TRATEGY The solution of the optimal control problem dened by (21) can be applied to forward-oriented models or to a vehicle control system. Fig. 4 illustrates a procedure for the implementation of the solution into a control algorithm. Note that, although the vehicle drivetrain includes three propulsion systems (namely, an engine and two EMs), the proposed implementation allows for the optimal torque split between an arbitrary number of power generation elements. According to Fig. 4, the variables fICE and fBSA dene the fraction of the torque demand to the drivetrain Treq that is commanded to the engine and to the BSA, respectively. By conducting an energy balance on the system in Fig. 3, three matrices containing all the possible torque combinations that satisfy the drivetrain demand are generated, i.e., TICE (t) = fICE Treq (t) R
nxm

+ where Tmax is the maximum positive torque that the powertrain can generate combining ICE, BSA, and EM, whereas Tmax is the maximum negative torque that can be absorbed by the electric machines (BSA and EM), accounting for battery power limitations. The torque delivered by each component is then limited according to (17). Note that the torque variables dened are considered as mechanical and, hence, calculated at the shaft of each component. The electrical power provided by the battery and the power associated with the engine fuel utilization are then computed to evaluate the Hamiltonian function in (21). Specically, Pfuel is determined from the engine fuel consumption, according to (12), whereas the power of the electric machines is computed from the efciency maps for the BSA and EM, i.e.,

PEM,el (t) = TEM (t) EM (t) EM,el PBSA,el (t) = TBSA (t) ICE (t) BSA,el Rnxm Rnxm (26) (28) TBSA (t) = fBSA (1 fICE ) Treq (t)

TEM (t) = (1 fBSA ) (1 fICE ) Treq (t)

where the dimensions m and n are related to the chosen resolution for the factors fICE and fBSA . The torque request

where, for the rear EM, EM,el = 1/EM if the machine is working as a motor, and EM,el = EM if it is working as a generator. For each torque split combination that satises the preceding constraints, the Hamiltonian function is dened based on (21).

STOCKAR et al.: ENERGY-OPTIMAL CONTROL OF PHEVs FOR REAL-WORLD DRIVING CYCLES

2955

In doing so, the expression (batt / SoE) in (24) is explicitly calculated according to (5), i.e., dbatt (SoC(t)) Ibatt (t) = dSoE Voc (SoC(t)) R (SoC(t)) Ibatt (t) R (SoC(t)) R (SoC(t)) Voc (SoC(t)) SoC(t) Voc (SoC(t)) SoC(t) . (29)
Fig. 5. Example of velocity prole for the controller validation (indicated as Path 3 in Table V).

Note that, since the parameters Voc and R are continuous piecewise polynomial functions [28], they can be differentiated in the entire SoC range. o o and fBSA that At any time step, the combination fICE minimizes the Hamiltonian function matrix is chosen as the solution of the optimization problem. It is worth observing that the proposed algorithm, although suitable for implementation into forward-oriented simulators or hardware-in-the-loop systems for control development and testing, cannot directly be applied to real-time control due to the required computation and numerical optimization of the Hamiltonian function at each time step. However, the computation effort can signicantly be decreased by precomputing the Hamiltonian function and importing the results as maps in the vehicle control system. A similar approach was adopted for the implementation of ECMS to a CS HEV [21], [33], [37].

VII. R ESULTS AND A NALYSIS The energy management algorithm was applied to the forward-oriented PHEV simulator to conduct an evaluation of the vehicle performance for a variety of usage conditions. The focus of the study conducted is on the effects of the control parameters on the vehicle fuel economy and CO2 emissions and the inuence of driving conditions and energy generation scenarios.

Fig. 6. Summary of electricity generation mix for four countries (sources: [43][47]).

A. Vehicle Driving Scenarios The characteristics of the driving prole have a strong impact on the calibration of the PHEV control algorithm [10], [11], [15], [41]. In this paper, a rich set of driving proles was adopted as a validation framework for the energy management control algorithm, analyzing scenarios consistent with the driving behavior of customers and improving the generality of the results. The simulations were conducted on a set of regulatory and real-world driving proles extracted from a database of eet study data to statistically represent typical usage conditions of a PHEV, including urban, extra-urban, and highway segments with variable driving length [42]. Table V in the Appendix lists the main characteristics (velocity and energy demand at the wheel) of all the driving cycles considered in this study. The cycles are all characterized by a driving distance greater than the vehicle AER. This allows for the possibility of depleting the battery, depending on the calibration of the energy management strategy.

Fig. 5 shows the velocity prole of one of the nonregulatory cycles considered. This pattern is representative of mixed-mode driving conditions, alternating urban driving and a highway segment.

B. Electricity Generation Scenarios The impact of the electricity generation mix on the PHEV utilization was evaluated by varying the specic CO2 emission coefcient 2 to encompass different energy generation scenarios, including electricity production from both fossil fuel and renewable sources. Four of the values considered for 2 are representative of the energy generation mix for the U.S., Switzerland, France, and Germany, as summarized in Fig. 6. For simplicity, it will be assumed that the grid energy consumed by the PHEV has the same specic CO2 content as the generation mix. Note that this must be considered an approximation for the European countries, where the open energy market may cause differences between the CO2 content of the electricity produced by each country and that consumed by the vehicle.

2956

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 60, NO. 7, SEPTEMBER 2011

C. Denition of Controller Parameters and Performance Metrics Based on the optimal control problem in (21), the parameters requiring calibration are the initial condition for the Lagrange multiplier i and the scalar Lagrange multiplier l (which varies the penalty on the battery SoE constraints). The impact of the foregoing parameters will be evaluated through three different metrics. First, the evolution of the battery SoE during the driving pattern and its nal value SoEnal (i , l ) will be considered. Then, the overall CO2 mass calculated with (10) and (11) will be evaluated as mCO2 = 1 mf QLHV + 2 1 Enom SoE ch (30)

where QLHV is the fuel lower heating value, and SoE is the difference between initial and nal SoEs. Another variable is introduced to indicate whether the vehicle is operating in CD or CS mode, hence identifying how fast the control strategy depletes the battery. The variable CS denes the fraction of the driving cycle where the vehicle operates in CS mode at its lower SoE bound, i.e., CS = tCS . tcyc (31)

Fig. 7. Final value of the battery SoE as function of i and l for the case study (cycle Path 3, U.S. scenario).

Specically, tCS is calculated by considering the time during which the vehicle operates within a 5% window around SoE = SoEmin . In the following results, the battery is assumed at SoE = SoEmax = 95% at the beginning of each cycle. Knowledge of the fraction of the driving cycle in CS mode is not only relevant for energy optimization but for reliability, safety, and aging issues as well [48]. D. Analysis of Simulation Results for One Driving Cycle and One Energy Scenario To illustrate the results, one case study will be analyzed in detail with reference to the driving cycle shown in Fig. 5 and the U.S. energy generation scenario. Simulations were conducted to evaluate the vehicle CO2 emissions, the fuel economy, and the utilization of the battery energy in relation to the control parameters. Fig. 7 reports the values of the nal battery SoE obtained by varying the parameters i and l . Note that an undesired complete depletion of the battery is possible for certain combinations of the control strategy parameters. It is evident that l affects the ability of the controller to respect the state constraints. In particular, the SoE exceeds its boundaries when l is below a threshold (for the considered scenario, l < 10). The parameter i determines whether the lower or upper SoE bound is violated. In the rst case, the vehicle uses more battery energy than the one allowed. In the second case, the controller requires the engine to produce more power to further charge the battery. If any of the above cases occurs, the corresponding solution is not part of the feasible domain and cannot be considered for the control problem. For this reason, the points violating the state constraints will be removed from the following results.

Fig. 8(a) and (b) shows the contribution from the fuel energy and the electric energy to the total vehicle CO2 emissions. Compared with Fig. 7, a nal SoE close to its upper bound implies that most of the energy consumed by the vehicle was supplied by the ICE, leading to high fuel consumption and engine CO2 contribution. Conversely, a low nal SoE results in lower engine CO2 emissions. Combining the CO2 from fuel energy and battery energy, it is possible to obtain the overall CO2 emissions for the PHEV, as shown in Fig. 8(c). The response to the control parameters is almost at, indicating rather limited benets on the vehicle CO2 emissions. This behavior can be explained by the high specic CO2 content of the electricity in the scenario considered [43], [44]. In fact, the production of energy from coal (a carbon-rich fuel) and the low well-to-tank efciency of the electricity generation ultimately offset the higher tank-to-wheel efciency of the electric traction to the point where the CO2 produced from the battery energy use is comparable with the fuel energy utilization. Fig. 9 shows the PHEV fuel consumption over the sample cycle. Since the factor 1 is constant, the fuel consumption is directly related to the CO2 contribution from the fuel energy. This implies a simplication in the study since the brakespecic CO2 of the engine varies based on the engine operating condition. However, for a PHEV, such differences would be minimal as the engine operating range is limited compared with a conventional vehicle. Comparing Fig. 9 with Fig. 8(c), it is evident that the best engine fuel consumption is achieved whenever the battery is completely depleted at the end of the cycle. However, the optimal value of the parameter i is determined by minimizing the cumulative CO2 emissions, which does not necessarily correspond to the best fuel economy. On the other hand, if the specic CO2 content of the grid 2 tends to zero, then the minimum CO2 and the best fuel economy would be coincident. This corresponds to an ideal case where the electric energy is entirely produced from renewable sources.

STOCKAR et al.: ENERGY-OPTIMAL CONTROL OF PHEVs FOR REAL-WORLD DRIVING CYCLES

2957

Fig. 9. Fuel economy of the PHEV as a function of i and l for the case study (cycle Path 3, U.S. scenario).

Fig. 10. Percentage of cycle in CS mode at the low SoE bound as a function of i and l for the case study (cycle Path 3, U.S. scenario).

Fig. 8. Overall vehicle CO2 emissions as a function of i and l for the case study (cycle Path 3, U.S. scenario). (a) Contribution from fuel energy. (b) Contribution from electric energy. (c) Combined.

Fig. 10 shows the fraction of cycle duration where the vehicle operates in CS mode at its lower SoE bound as a function of the control parameters. For high values of i , the vehicle

is operated in CD-CS mode, and the SoE reaches the lower bound before the end of the driving path. For the driving cycle considered, CS is slightly below 40%, meaning that approximately 60% of the energy requested to the drivetrain can be satised with the battery. For i > 10, the control strategy forces the vehicle to deplete the battery and, when the lower SoE bound is reached, switches to CS mode. Conversely, as i decreases, CS decreases steeply to zero, and when i 0, the control strategy is no longer able to deplete the battery. At this condition, the nal SoE is near the same value as the initial value, hence, the control strategy tends to operate the system in CS mode at the higher SoE bound. This is conrmed in Fig. 11, where the evolution of the battery SoE during the driving cycle is represented for four different values of i , whereas l is set constant. Intermediate solutions are observed for values of i included within the two bounds. In particular, a value i = 6 allows the battery to be gradually depleted during the cycle, reaching the lower SoE bound only at

2958

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 60, NO. 7, SEPTEMBER 2011

TABLE II SENSITIVITY A NALYSIS OF THE C OST F UNCTIONAL J (u) TO THE PARAMETER i (U.S. S CENARIO )

Fig. 11. Battery SoE prole during the driving cycle for l = 18 varying i (cycle Path 3, U.S. scenario).

Fig. 12. Optimal value of the initial condition i as a function of vehicle energy demand for different driving cycles (U.S. scenario).

the end of the driving pattern and avoiding any CS operations. This operation, which is known as blended mode, allows for the achievement of the minimum vehicle fuel consumption along a prescribed driving cycle [17]. E. Effects of Driving Cycle Characteristics For all the vehicle driving proles listed in Table V and the U.S. energy generation scenario, a full factorial design of experiments was generated, varying the control parameters i and l of the supervisory controller. Then, the optimal combination (i , l )opt was determined by minimizing the CO2 emissions produced by the PHEV. Fig. 12 summarizes the results of the simulations, representing the optimal value of the Lagrange multiplier i against the vehicle energy demand at the wheel calculated for the driving cycles considered in the study. The parameter l was set to a constant value to ensure that the constraints on the battery SoE are always respected. As Fig. 12 shows, the results tend to cluster within a limited range of values for the parameter i and are almost independent on the energy demand at the wheel. A sensitivity study was conducted to evaluate the inuence on the cost functional J (u) of errors in the optimal value of the control parameter i . The analysis was conducted with reference to ve specic driving patterns, representing the limit scenarios for Fig. 12. Table II summarizes the sensitivity results to variations in i around the optimal value corresponding to each of the ve driving cycles considered. In all the cases, the sensitivity of the vehicle CO2 emissions is very limited. The results conrm

that, for the energy generation scenario considered, the control strategy is relatively insensitive to the characteristics of the driving pattern [26]. The behavior can be justied by considering that the parameter i is the initial condition of the costate ODE of the optimal control problem. Therefore, its inuence on the optimal solution progressively decreases with the duration of the driving cycle as (t) converges. In summary, the simulation results show that the vehicle CO2 emissions are relatively insensitive to the Lagrange multiplier i for the considered energy generation scenario. Furthermore, the optimal value of the control parameter, which allows the vehicle to operate in blended mode with minimum energy consumption, is nearly independent from the driving cycle duration and vehicle energy demand. Conversely, the parameter l has no impact on the vehicle performance but ensures satisfaction of the constraints on the battery SoE bounds. Specically, a threshold value can be identied for l so that the state constraints are always respected, allowing one to reduce the controller calibration problem to the sole parameter i . This presents advantages for parameter tuning as nearoptimal results can be achieved with minimal calibration effort, in particular, without the need for information, such as the driving length. F. Effects of Energy Generation Scenarios To extend the validation framework, different scenarios were considered to evaluate the sensitivity of the control parameter i to different values of the energy generation mix. As an example, this analysis was initially limited to the sample driving cycle shown in Fig. 5. Fig. 13 represents the vehicle CO2 emissions and engine fuel consumption against the parameter i for the four different energy generation scenarios shown in Fig. 6. The U.S. and German energy production scenarios are relatively similar, with the high specic CO2 content of the electric generation mix causing a relatively at response of the overall vehicle emissions to the control parameter i . Conversely, the case of Switzerland and France is rather different, as the energy generation is predominantly composed by renewable or low CO2 primary sources. These two scenarios offer promising opportunities for a large PHEV penetration. Here, a higher sensitivity in the vehicle CO2 emissions can be observed with respect to the control strategy parameter. Fig. 14 illustrates the inuence of the specic CO2 content of the grid energy on the optimal value of the Lagrange multiplier

STOCKAR et al.: ENERGY-OPTIMAL CONTROL OF PHEVs FOR REAL-WORLD DRIVING CYCLES

2959

TABLE III SENSITIVITY A NALYSIS OF THE C OST F UNCTIONAL J (u) TO THE PARAMETER i (S WISS S CENARIO )

Fig. 13. Impact of the energy generation mix on CO2 and fuel consumption against the parameter i for different energy generation scenarios (cycle Path 3).

This behavior indicates that the optimality of the control strategy (and, consequently, the PHEV fuel consumption and CO2 emissions) is more affected by the driving cycle characteristics as the electric energy is predominantly generated from renewable sources. Similar to the U.S. scenario, a sensitivity study was conducted on the cost functional J (u) varying the parameter i for different driving patterns. It is possible to notice here the increased sensitivity of the vehicle CO2 emissions to errors in the optimal value of the control parameter. On the other hand, a considerably large error must be given to i to detect sufciently high variations in the cost functional J (u). This indicates the presence of a relatively large region around the sweet-spot, where the CO2 emissions and the vehicle performance vary only marginally. VIII. C ONCLUSION

Fig. 14. Inuence of the energy generation mix parameter 2 on the optimal value of the parameter i (cycle Path 3).

Fig. 15. Optimal value of the initial condition i as a function of vehicle energy demand for different driving cycles (Swiss scenario).

i with reference to the sample driving cycle, indicating a linear correlation between 2 and i . This suggests that the calibration of the PHEV supervisory controller could be done when the battery is connected to the grid based on the specic CO2 content of the electricity generation during the charging operation. Fig. 15 summarizes the optimal value of the Lagrange multiplier i against the vehicle energy demand at the wheel for all the driving cycles considered. The low specic CO2 content of the electric energy generation in Switzerland causes a different behavior compared with that observed in Fig. 12 for the U.S. scenario. Although the results are still clustered in a limited range of i , a slightly increase dependence of the optimal parameter value with the driving cycle energy demand can be observed (see Table III).

This paper has presented a novel approach to the supervisory energy management of PHEVs. This paper has addressed the fuel consumption and CO2 emissions associated with the PHEV use through a well-to-wheel energy balance that explicitly accounts for the fuel energy and grid energy utilization. An optimal control problem was formulated by dening a cost functional based on the cumulative CO2 produced directly and indirectlyby the vehicle. Pontryagins minimum principle was then applied to reduce a global optimization problem to a local minimization, allowing for the control problem to be solved and implemented in an algorithm. The control algorithm was then implemented on a validated energy-based PHEV simulator. Simulations were conducted to evaluate the sensitivity of the supervisory controller to different vehicle utilization and energy generation scenarios. A large database of driving proles, including regulatory cycles and real-world vehicle velocity proles extracted from eet studies data, was considered to provide a validation framework. Based on the analysis conducted, the proposed energy management strategy presents a relatively low sensitivity to the driving prole characteristics (i.e., the energy demand at the wheel or the driving distance). This result was achieved because of the denition of a cost functional that formally accounts for the different mix of primary energy forms utilized by the PHEV, representing an improvement over the conventional control approaches that approximate the energy utilization with an equivalent fuel consumption metric. In particular, the vehicle CO2 emissions show the presence of an optimal condition varying the control strategy parameter i ,

2960

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 60, NO. 7, SEPTEMBER 2011

as well as a relatively large sweet-spot, where only marginal variations from the optimal condition occur. Conversely, a higher sensitivity to the control parameter i was observed on the battery SoE prole and, ultimately, the vehicle operating mode. Furthermore, the sensitivity to vehicle usage conditions and the tradeoff between fuel and electrical power consumption are dependent on the specic CO2 emissions associated with the electricity generation. In particular, a higher sensitivity was observed for the energy generation scenarios characterized by a low CO2 content. While this paper does not specically address real-time control developments, its insights are valuable when developing energy management strategies that can lead to more readily tunable algorithms that can address different objectives. In particular, the analysis presented in this paper can assist in addressing differences in electricity generations between different regions and countries, allowing for the development of energy management strategies that can achieve, for example, minimum CO2 emissions in the face of a different mix of electric power generation feedstocks. Given the increasing use of geographical information systems and navigation systems, which can lead to some degree of a priori knowledge of the vehicle trajectory, the results presented in this paper represent a step forward in understanding of the potential of formal optimization methods in guiding the design of real-time energy management strategies.

TABLE IV VEHICLE PARAMETERS USED IN (32)

TABLE V SUMMARY OF M ETRICS FOR THE D RIVING C YCLES C ONSIDERED IN THE S TUDY

A PPENDIX A brief description of the driving cycle characteristics considered in the validation study is reported here. A combination of regulatory and real-world driving cycles was used to validate the proposed supervisory energy management strategy. For each cycle, information on the distribution of vehicle velocity, the driving distance, and the energy demand at the wheel is provided. The energy demand at the wheel is computed based on the road load equation [25]
te te te

Ewheel =M
ti

dV dt

1 dt + a Cx Af 2

V 3 dt + Cr M g
ti ti

V dt (32)

where V is the vehicle velocity, M is the vehicle mass, a is the air density, Cx is the aerodynamic friction coefcient, Af is the vehicle frontal area, Cr is the tire rolling resistance coefcient, and g is the acceleration of gravity. The foregoing equation neglects the effects of the road grade. The vehicle parameters are listed in Table IV. Table V summarizes the most relevant metrics of the driving cycles. In particular, the distribution of the distance was chosen to ensure a driving length greater than the vehicle AER as well as a maximum distance that is representative of typical daily commuting trips.

STOCKAR et al.: ENERGY-OPTIMAL CONTROL OF PHEVs FOR REAL-WORLD DRIVING CYCLES

2961

TABLE V (Continued.) SUMMARY OF M ETRICS FOR THE D RIVING C YCLES C ONSIDERED IN THE S TUDY

R EFERENCES
[1] S. Stockar, P. Tulpule, V. Marano, and G. Rizzoni, Energy, economical and environmental analysis of plug-in hybrids electric vehicles based on common driving cycles, SAE Int. J. Engines, vol. 2, no. 2, pp. 467476, Mar. 2010. [2] S. Golbuff, Design and optimization of a plug-in hybrid electric vehicle, in Proc. SAE World Congress, SAE Technical Paper 2007-04-16, 2007. [3] A. Elgowainy, A. Burnham, M. Wang, J. Molburg, and A. Rousseau, Well-to-wheels energy use and greenhouse gas emissions of plug-in hybrid vehicles, in Proc. SAE World Congr., 2009. [4] E. Zgheib and D. Clodic, CO2 emission and energy reduction evaluations of plug-in hybrid vehicles, in Proc. SAE World Congr., 2009. [5] V. Freyermuth, E. Fallas, and A. Rousseau, Comparison of powertrain conguration for plug-in HEVs from a fuel economy perspective, SAE Int. J. Engines, vol. 1, no. 1, pp. 392398, Apr. 2008. [6] M. P. OKeefe and T. Markel, Dynamic programming applied to investigate energy management strategies for a plug-in HEV, presented at the 22nd Int. Battery, Hybrid Fuel Cell Electric Vehicle Symp. (EVS-22), Yokohama, Japan, 2006, NREL/CP-540-40376.

[7] J. Gonder and T. Markel, Energy management strategies for plugin hybrid electric vehicles, in Proc. SAE World Congr., Detroit, MI, Apr. 1619, 2007, pp. 15. [8] S. J. Moura, H. K. Fathy, D. S. Callaway, and J. L. Stein, A stochastic optimal control approach for power management in plug-in hybrid electric vehicles, IEEE Trans. Control Syst. Technol., vol. 19, no. 3, pp. 545555, May 2011. [9] A. Simpson, Cost-benet analysis of plug-in hybrid electric vehicle technology, presented at the 22nd Int. Battery, Hybrid Fuel Cell Electric Vehicle Symp. Exhibition, Yokohama, Japan, 2006, NREL/CP-540-40485. [10] Q. Gong, Y. Li, and Z. Peng, Trip-based optimal power management of plug-in hybrid electric vehicles, IEEE Trans. Veh. Technol., vol. 57, no. 6, pp. 33933401, Nov. 2008. [11] D. Karbowski, A. Rousseau, S. Pagerit, and P. Sharer, Plug-in vehicle control strategy: From global optimization to real-time application, in Proc. 22nd Battery, Hybrid Fuel Cell EVS, 2006. [12] P. Sharerm, A. Rousseau, D. Karbowski, and S. Pagerit, Plug-in hybrid electric vehicle control strategy: Comparison between EV and chargedepleting options, in Proc. SAE World Congr., SAE Technical Paper 2008-01-0460, 2008. [13] I. Kolmanovsky, I. Siverguina, and B. Lygoe, Optimization of powertrain operating policy for feasibility assessment and calibration: Stochastic dynamic programming approach, in Proc. Amer. Control Conf., 2002, pp. 14251430. [14] S. J. Moura, D. S. Callaway, H. K. Fathy, and J. L. Stein, Tradeoffs between battery energy capacity and stochastic optimal power management in plug-in hybrid electric vehicles, J. Power Sources, vol. 195, no. 9, pp. 29792988, 2010. [15] P. Tulpule, V. Marano, and G. Rizzoni, Effects of different PHEV control strategies on vehicle performance, in Proc. Amer. Control Conf., St. Louis, MO, Jun. 2009, pp. 39503955. [16] V. Marano, P. Tulpule, S. Stockar, S. Onori, and G. Rizzoni, Comparative study of different control strategies for plug-in hybrid electric vehicles, in Proc. 9th Int. Conf. Engines Veh., Capri, Italy, SAE Technical Paper 2009-24-0071, Sep. 2009. [17] P. Tulpule, S. Stockar, V. Marano, and G. Rizzoni, Optimality assessment of equivalent consumption minimization strategy for PHEV applications, in Proc. ASME Dyn. Syst. Control Conf., Hollywood, CA, 2009, pp. 265272. [18] P. Tulpule, V. Marano, and G. Rizzoni, Energy management for plugin hybrid electric vehicles using equivalent consumption minimization strategy, Int. J. Elect. Hybrid Veh., vol. 2, no. 4, pp. 329350, 2010. [19] C. Zhang, A. Vahidi, X. Li, and D. Essenmacher, Role of trip information preview in fuel economy of plug-in hybrid vehicles, in Proc. ASME, 2009, pp. 253258. [20] C. Zhang and A. Vahidi, Real-time optimal control of plug-in hybrid vehicles with trip preview, in Proc. ACC, 2010, pp. 69176922. [21] K. Koprubasi, Modeling and control of hybrid-electric vehicle for drivability and fuel economy improvements, Ph.D. dissertation, Ohio State Univ., Columbus, OH, 2008. [22] Center for Automotive Research, Final Design Vehicle Tech. Specs., The Ohio State University, Challenge X Team, Columbus, OH, Fall, Tech. Rep. 2006. [23] M. Arnett, K. Bayar, C. Coburn, Y. Guezennec, K. Koprubasi, S. Midlam-Mohler, K. Sevel, M. Shakiba-Herfeh, and G. Rizzoni, Cleaner diesel using model-based design and advanced aftertreatment in a student competition vehicle, in Proc. SAE World Congr., Detroit, MI, SAE Technical Paper 2008-01-0868, 2008. [24] Y. Hu, S. Yurkovich, Y. Guezennec, and R. Bornatico, Model-based calibration for battery characterization in HEV applications, in Proc. Amer. Control Conf., Seattle, WA, June 11-13, 2008, pp. 318325. [25] L. Guzzella and A. Sciarretta, Vehicle Propulsion System: Introduction to Modeling and Optimization. Berlin, Germany: Springer-Verlag, 2007. [26] S. Stockar, V. Marano, G. Rizzoni, and L. Guzzella, Optimal control for plug-in hybrid electric vehicles applications, in Proc. Amer. Control Conf., Baltimore, MD, Jun. 2010, pp. 50245030. [27] L. Serrao, A comparative analysis of energy management strategies for hybrid electric vehicles, Ph.D. dissertation, Ohio State Univ., Columbus, OH, 2009. [28] Y. Hu, B. J. Yurkovich, S. Yurkovich, and Y. Guezennec, Electro-thermal battery modeling and identication for automotive applications, in Proc. ASME Dyn. Syst. Control Conf., Hollywood, CA, Oct. 12-14, 2009, pp. 233240. [29] A. D. Filippi, S. Stockar, M. Canova, S. Onori, and Y. Guezennec, Model-based life estimation of Li-Ion batteries in PHEVs using largescale vehicle simulations: An introductory study, in Proc. 6th IEEE Veh. Power Propulsion Conf., 2010, pp. 16.

2962

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 60, NO. 7, SEPTEMBER 2011

[30] G. Paganelli, S. Delprat, T. Guerra, J. Rimaux, and J. Santin, Equivalent consumption minimization strategy for parallel hybrid powertrains, in Proc. VTC, 2002, pp. 20762081. [31] A. Sciarretta, M. Back, and L. Guzzella, Optimal control of parallel hybrid electric vehicles, IEEE Trans. Control Syst. Technol., vol. 12, no. 3, pp. 352363, May 2004. [32] A. Sciarretta and L. Guzzella, Control of hybrid electric vehiclesa survey of optimal energy-management strategies, IEEE Control Syst. Mag., vol. 27, no. 2, pp. 6670, Mar. 2007. [33] G. Paganelli, G. Ercole, A. Brahma, Y. Guezennec, and G. Rizzoni, A general formulation for the instantaneous control of the power split in charge-sustaining hybrid electric vehicles, in Proc. 5th Int. Symp. Adv. Veh. Control, Ann Arbor, MI, 2000. [34] P. Pisu and G. Rizzoni, A supervisory control strategy for series hybrid electric vehicles with two energy storage systems, in Proc. IEEE Conf. Veh. Power Propulsion, Sep. 79, 2005, pp. 6572. [35] H. Geering, Optimal Control with Engineering Applications. Berlin, Germany: Springer-Verlag, 2007. [36] L. S. Pontryagin, V. G. Boltyanskii, R. V. Gamkrelidze, and E. F. Mishchenko, The Mathematical Theory of Optimal Processes. New York: Interscience, 1962. [37] L. Serrao, S. Onori, and G. Rizzoni, ECMS as a realization of the Pontryagins minimum principle for HEV control, in Proc. Amer. Control Conf., St. Louis, MO, Jun. 2009, pp. 39643969. [38] T. Schneider, Transportation efciency through electric drives and the power grid, plug-in hybrid electric vehicles: towards energy independence, Capitol Hill Forum, Rome, Italy, Jul. 2007. [39] Greet 1.8b, Argonne Nat. Lab., Argonne, IL, 2008. [40] V. Marano, S. Onori, Y. Guezennec, G. Rizzoni, and N. Madella, Lithium-ion batteries life estimation for plug-in hybrid electric vehicles, in Proc. IEEE VPPC, 2009, pp. 536543. [41] A. Rousseau, S. Pagerit, and D. Gao, Plug-in hybrid electric vehicle control strategy parameter optimization, in Proc. 23rd Int. Battery, Hybrid Fuel Cell Elect. Veh. Symp. Exhib., 2007, pp. 114. [42] S. Midlam-Mohler, S. Ewing, V. Marano, Y. Guezennec, and G. Rizzoni, PHEV eet data collection and analysis, in Proc. Veh. Power Propulsion Conf., Dearborn, MI, Sep. 711, 2009, pp. 12051210. [43] U.S. Energy Information Administration (EIA), World net electricity generation by type (billion kilowatthours), Energy Inf. Admin.EIA, Washington, DC, Tech. Rep., 2005. [44] Energy Information Administration, Annual energy outlook 2007 with projections to 2030, Dept. Energy, Washington, DC, Tech. Rep. DOE/EIA-0383(2007), 2007. [45] Federal Ofce for the Environment, Quanto CO2 si produce consumando un kilowattora di energia elettrica in svizzera? Divisione Clima, Ufcio Federale dellAmbiente, Bern, Switzerland, 2008. [46] French Environment and Energy Management Agency, Note de cadrage sur le contenu CO2 du kWh par usage en france, Agence de lEnvironmement et de la Matrise de lEnergie, Paris, France, 2007. [47] J. Schuberth and H. Kaschenz, Elektrische waermepumpen, eine erneuerbare energie? Umweltbundesamt, Dessau, Germany, 2008. [48] K. R. Genung, C. Lawrence, T. Markel, C. S. Hinds, R. McGill, R. E. Ziegler, D. E. Smith, D. Brooks, S. W. Hadley, R. L. Smith, D. L. Greene, H. Wiegman, and V. Marano, Plug-in hybrid electric vehicle, value proposition study, Phase 1, Task 3, Technical requirements and procedure for evaluation of one scenario, Oak Ridge Nat. Lab., Oak Ridge, TN, Tech. Rep., Final Rep. DE-AC05-00OR22725, Jan. 2009.

Vincenzo Marano received the B.S./M.S. and Ph.D. degrees in mechanical engineering from the University of Salerno, Salerno, Italy, in 2003 and 2007, respectively. He is currently a Senior Research Associate with the Center for Automotive Research (CAR), which is an interdisciplinary research center at The Ohio State University (OSU), Columbus. He conducts research in the area of plug-in electric vehicles (PEVs), energy storage, energy management, PEV interaction with the power grid, macroeconomics, and energy policy/regulations. He also serves as program manager of SMART@CAR, which is an OSU CAR collaborative research program with the participation of major automotive original equipment manufacturers and electric power companies. Dr. Marano is a member of the American Society of Mechanical Engineers and the Society of Automotive Engineers.

Marcello Canova received the Diploma di Laurea (cum laude) and Ph.D. degrees in mechanical engineering from the University of Parma, Parma, Italy, in 2002 and 2006, respectively. He is currently an Assistant Professor of mechanical engineering with The Ohio State University, Columbus, where he is afliated with the Center for Automotive Research. He conducts research in the broad area of uid and thermal sciences and energy conversion systems, with emphasis on modeling, optimization, and dynamic systems and control problems associated with these areas. His research interests are in future ground vehicle propulsion systems, including advanced engines, hybrid drivetrains, and energy storage systems.

Giorgio Rizzoni (F04) received the B.S., M.S., and Ph.D. degrees in electrical and computer engineering from the University of Michigan, Ann Arbor. He is a Ford Motor Company Chair in electromechanical systems, a Professor of mechanical engineering and electrical and computer engineering, and the Director of the Center for Automotive Research (an interdisciplinary research center supporting some 30 full-time staff and 50 graduate students), The Ohio State University, Columbus. His research interests include dynamics, control, and diagnosis of automotive systems, with emphasis on hybrid and electric vehicles and on energy conversion and storage systems. Dr. Rizzoni is a Fellow of the Society of Automotive Engineers.

Stephanie Stockar received the B.S. and M.S. degrees in mechanical engineering from the Swiss Federal Institute of Technology (ETH) Zurich, Zurich, Switzerland, in 2007 and 2009, respectively. She is currently working toward the Ph.D. degree in mechanical and aerospace engineering with the Ohio State University, Columbus. Her research interests include dynamic systems and control problems related to automotive powertrain systems and hybrid vehicles.

Lino Guzzella (F11) was born in Zurich, Switzerland, in 1957. He received the Diploma in mechanical engineering and the Dr.Sc.Techn. degree in control engineering from the Swiss Federal Institute of Technology (ETH) Zurich in 1981 and 1986, respectively. From 1987 to 1989, he was with the R&D Department, Sulzer Bros., Winterthur, Switzerland. From 1989 to 1991, he was an Assistant Professor of automatic control with the Electrical Engineering Department, ETH Zurich. He then joined Hilti R&D, Schaan, Lichtenstein, where he was the Head of the Mechatronics Department from 1992 to 1993. He is currently a Professor of thermotronics with the Department of Mechanical and Process Engineering, ETH Zurich. His research interests are the modeling of dynamic systems and nonlinear and robust control and the application of these ideas to thermal and, particularly, automotive systems.

You might also like