Professional Documents
Culture Documents
Energy: Shradhdha Sarvaiya, Sachin Ganesh, Bin Xu
Energy: Shradhdha Sarvaiya, Sachin Ganesh, Bin Xu
Energy
journal homepage: www.elsevier.com/locate/energy
a r t i c l e i n f o a b s t r a c t
Article history: Hybrid Electric Vehicles (HEV) form an important category of the automotive segment and most
Received 19 November 2020 importantly fill the transition between internal combustion engines powered conventional vehicles and
Received in revised form electric motor-powered vehicles. One of the main propagandas and sales factor of HEV is improved fuel
5 April 2021
economy over conventional vehicles. Over the recent years, considering the longevity of HEV usage,
Accepted 6 April 2021
battery life evaluation has been brought to the forefront of research along with fuel economy and the key
Available online 17 April 2021
method to slow down the battery aging is through Energy Management Strategy (EMS). This research
paper presents the comparative analysis of battery life optimization with different control strategies in a
Keywords:
Hybrid electric vehicles
parallel hybrid vehicle. In the available research work, EMS considering battery aging is still lacking. This
Energy management strategy research work considers the impact of multiple parameters, including temperature and current on
Battery aging battery aging, providing more accurate battery life prediction. Four different control strategies are
analyzed including, Thermostat, Fuzzy logic, Adaptive Equivalent Consumption Minimization Strategy
(A-ECMS) and Q-learning considering battery aging. Results are compared concerning battery aging and
fuel economy. In this research, ECMS results show a 25% improved fuel economy compared to the rule-
based strategy. Also, a market cost-analysis is depicted to show the monetary savings for each of the
energy management strategy.
© 2021 Elsevier Ltd. All rights reserved.
https://doi.org/10.1016/j.energy.2021.120604
0360-5442/© 2021 Elsevier Ltd. All rights reserved.
S. Sarvaiya, S. Ganesh and B. Xu Energy 228 (2021) 120604
literature on HEV control strategies, most of the research work is battery capacity loss. The purpose of the analysis is to study the
focused on achieving the best fuel economy. In Ref. [3] the authors impact of different control strategies on vehicle fuel economy and
have compared a PHEV’s (Plug-in Hybrid Electric Vehicle) perfor- battery life. The contribution of this study is as follows.
mance in real world condition with control strategies including
PMP and genetic algorithm based fuzzy logic where, the perfor- (1) Four different control strategies, including the very widely
mance of the PHEV is evaluated with respect to fuel consumption used fuzzy logic controller, ECMS, Thermostat controller and
and emissions but, the effect on Lithium-ion (Li-ion) battery life Q-learning are discussed in this paper. Further, the models
which is the critical parameter performance for an HEV is not are validated with Environmental Protection Agency (EPA)
analyzed. In Ref. [4] the authors have developed a fuzzy logic mandated driving cycles including Urban Dynamometer
control strategy for the power split in a PHEV and the split is Driving Schedule (UDDS) and Highway Fuel Economy Test
decided by predicting and recognizing the traffic beforehand. The (HWFET) to simulate the on-road driving scenario in city and
underlying effect of this control strategy is reduced fuel con- highway driving conditions.
sumption and emissions. In Ref. [5] the authors have developed an (2) In this research work, a semi-empirical battery aging model
energy management strategy based on the genetic algorithm to is used for battery degradation estimation. This model is less
reduce fuel consumption and emissions. In Ref. [6] the authors have accurate compared to the electrochemical model but, it is
proposed a rule-based control strategy formulated over the energy computationally faster and easier to integrate with Battery
consumption over the trip and driving information. The perfor- management systems [10]. To estimate the battery capacity
mance of the controller is evaluated based on fuel consumption and loss more accurately, the temperature is considered as a
number of engine start-stop. In all these literature, Li-ion battery dynamic variable in each control strategy, eliminating the
performance or battery life is not considered in the energy man- constant temperature assumption found in the literature. A
agement strategy. Only in recent years, researchers have laid secondary control layer is added to the normal rule-based
emphasis on the importance of tailoring control strategies to and ECMS control strategies to show the in-depth impacts
improve battery life in HEVs. Therefore, this paper is focused on the of battery temperature control to decelerate aging.
development of control strategies of HEVs, with significant (3) To overcome the drawbacks in gearshift, a quasi-static cost-
importance given to simultaneous optimization of fuel economy based gear controller is implemented and the improvements
and battery life. Also, a cost analysis is performed for all the control in fuel economy are analyzed.
strategies to operate the vehicle in Hybrid mode to give a better
idea of the economic impact of each of these strategies. The rest of the paper is organized as follows. The vehicle
Wang et al. [7] describe the damage accumulation model to modelling is described including vehicle layout in section 2. The
predict the battery life, which is a function of Temperature, DOD subsections in section 2 explain IC engine and electric motor-
(depth of discharge) and ampere-hour throughput of the battery. generator models, driver model and the transmission model. At
The relation derived between battery capacity loss and these pa- the end of section 2, a simulation model is described which is used
rameters is based on the tests conducted in the laboratory condi- in MATLAB and Simulink for simulation of the HEV. The battery
tions, where the test objects are subjected to a controlled model is explained in section 3, followed by the battery equivalent
environment and standard test profile. However, this scenario does circuit model and the battery aging models in the consecutive
not replicate the real-world driving conditions. In Ref. [8] the subsections. All the control strategies are discussed in Section 4,
optimization of battery life and fuel economy have been studied followed by the introduction of energy management strategies
through dynamic programming-based energy management strat- used to improvise the vehicle performance in terms of battery life
egy and the cost function of the strategy includes battery power and fuel economy. Section 5 explains the simulation results which,
consumption as a battery life optimization parameter. Similarly [9], compares each control strategy with the base-line strategy (Ther-
solves the multi-objective problem of enhancing battery life and mostat) and discusses the trade-off between the vehicle fuel
fuel-economy in an HEV and, the cost function in their dynamic economy and the battery life. The argument is supported with the
programming strategy includes battery replacement cost as one of battery capacity loss over 100 cycles and battery fuel consumption
the control parameters. However, the effect of temperature, which for better understanding of the trend. At the end of Section 5, cost
is a key parameter contributing to battery aging [10], is considered analysis is given for all the control strategy to operate the vehicle in
as a constant parameter in both literature which, is variable in the Hybrid mode. The paper ends with the conclusion in Section 6,
actual operating condition. In Ref. [11], where a supervisory control stating the results of this analysis.
strategy is proposed to solve the power-split optimization problem
in HEV, but the strategy considers only state-of-charge (SOC) 2. Vehicle model
variation to split the power, without taking the impact of temper-
ature on battery aging into consideration. Physical-chemical 2.1. Vehicle architecture
modelling provides a more accurate observation for battery life
degradation, based on solid electrolyte interface (SEI) layer growth. The analysis conducted in this paper is for a four-wheel drive
In Ref. [12], the electrochemical model-based degradation identi- PHEV with a total weight of 1636 kg. The vehicle powertrain
fication method is presented. This method requires detailed components are shown in Fig. 1. The front wheels are driven by the
knowledge of battery chemistry and longer time for computation. IC engine with peak power output of 63 kW and the rear wheels are
In the previous research [13], the proposed Q-learning strategy driven by the electric motor-generator pair producing peak power
considers optimizing only fuel consumption in a HEV, however, the of 22 kW. The engine is coupled to a five-speed automatic gearbox
battery degradation over the drive cycle is not considered. In with gear ratios: 5.5, 3.8, 2.2, 1.5 and 0.8. Between the gearbox and
summary, the battery degradation in HEV EMS is still lacking. the front wheels, there is a final differential with a drive ratio of
Different control strategies considering battery degradation are 4.13. In the rear, the electric machine is powered by a 9 kWh Li-ion
compared in this research work. The impact of each control strategy battery pack whose specifications are discussed in detail in section
is discussed in detail and the results are supported by the trend 3 and is coupled to the rear axle through a constant gearbox having
observed in the engine and electric machine efficiency operating a gear ratio of 6.45 with a transmission efficiency of 90%. The motor
points, battery SOC, operating temperature, fuel consumption and provides traction power to the vehicle during acceleration, and it
2
S. Sarvaiya, S. Ganesh and B. Xu Energy 228 (2021) 120604
2.2. Driver model parison of engine operating points for the two gearshift controllers.
The main reason for this improvement in fuel economy is with cost-
In this paper, the controller has been tuned for the UDDS and based control, the engine operating points are more concentrated
HWFET cycles to ensure versatility over both city and highway in the 2500 to 4500 RPM range, which is the optimum speed range
driving conditions. For the current model, the controller gains are for best fuel economy and on the maximum torque (high efficiency)
Kp ¼ 0.47, Ki ¼ 0.003 and Kd ¼ 0.07. The output of the controller is line (Fig. 3). Whereas, with the torque and speed controller, the
normalized in the range [-1,1] and alpha, beta values are calculated operating points are scattered over the entire map leading to a
according to the equation, greater operation in low efficiency region (Fig. 2) and increased fuel
consumption.
UðtÞcUðtÞ > 0
aðtÞ ¼ (2.2.1)
0cUðtÞ 0
0cUðtÞ 0
bðtÞ ¼ (2.2.2)
UðtÞcUðtÞ < 0
X
5
JðtÞ ¼ max twheel;i ðtÞ þ 1 m_ f ðtÞengine;i (2.3.1)
i¼1
3
S. Sarvaiya, S. Ganesh and B. Xu Energy 228 (2021) 120604
where, Vbatt is the battery voltage under load, VO is the battery open
circuit voltage, I is the overall battery current and R1 is the battery
internal resistance.
The cell open circuit voltage and internal resistance are calcu-
lated as a function of SOC and temperature. From the charge and
discharge characteristic curves provided by the manufacturer, the
voltage vs SOC points are collected. The coefficients are calculated
by polynomial curve fitting where the polynomial is,
For modelling the resistance of the battery pack at different In this paper, the cell temperature is calculated based on a
operating conditions, an equivalent circuit method has been used lumped parameter assumption [14]. The thermal model has two
to identify the circuit parameters, using a zeroth-order model. A heat transfer components, heat generated by the cell’s internal
simple representation of a zeroth-order model is shown in Fig. 4. resistance and the convectional heat loss between the cell and the
The battery voltage for the circuit shown in Fig. 4 is governed by the surroundings. The overall heat transfer equation is given as,
equation,
dTcell
mccell ¼ I 2 Rint þ AhðT Tamb Þ (3.4.1)
Vbatt ¼ VO IR1 (3.2.1) dt
Fig. 5. Cell Internal Resistance vs SOC at different temperatures, the overall resistance
Fig. 4. Equivalent circuit model for battery. reduces with the increasing temperature.
4
S. Sarvaiya, S. Ganesh and B. Xu Energy 228 (2021) 120604
temperature in K, I is the charge/discharge current through the cell condition with, temperature ¼ 25 C, Icnom ¼ 2.5 h1 and SOC ¼ 35%
(A), A is the surface area of the cell exposed to convection (m2), h is [10]. gðIc; T; SOC Þ is the total charge throughput through the bat-
the convective heat transfer coefficient (W/m2K) and Tamb is the tery under the actual operating condition with the actual value of Ic;
ambient environment temperature (303 K). T and SOC.
Fig. 6 shows the severity factor map, which shows the relation of
severity factor function map concerning to the C-rate value and
3.5. Battery aging estimation
temperature. smap >1 shows the high stress zone. If the value of C-
rate and/or temperature is high enough then it raises the severity
Battery life is classified into two terms, calendar life and battery
factor value to high stress zone and the battery capacity loss in-
cycle life. In this research work, multiple control strategies are
creases. The reward function in Q learning strategy ensures that the
discussed to increase the battery cycle life. A Semi-empirical bat-
battery operation is in the low stress zone.
tery aging model from Refs. [11,13], is adopted for battery capacity
degradation calculation. This model is suitable for control appli-
cation where, the lower computation time is available, and the 4. Energy management strategies (EMS)
model required to be easily included with the battery management
system [10]. Battery life is a highly dynamic factor and is dependent 4.1. Thermostat strategy
on a wide variety of parameters including current, operating tem-
perature and the depth of discharge (DOD) [7]. The battery life The Thermostat strategy is a map-based strategy that considers
degradation is referred as battery capacity loss and the same is battery SOC limit to decide the electric machine torque mapped to
quantified by Eq. (3.5.1) [10]. it. It turns on the engine-generator unit when the battery SOC
approaches the lower limit and turns off the engine when the
Qbatt ð0Þ Qbatt ðr; AhÞ battery SOC approaches the upper limit. The upper and lower
Qloss ðr; AhÞ ¼ 100* (3.5.1)
Qbatt ð0Þ values of SOC are calculated based on battery internal resistance
[16]. In this research, a similar strategy is followed. Fig. 7 shows the
where r is the vector of aging factors (Ic; T; SOC) [15], Ah is the graphical representation of the Thermostat strategy followed. Fig. 7
ampere-hour throughput, which is the total amount of charging shows the Thermostat torque versus SOC map, which decides the
and discharging current through the battery during the operation. electric machine torque based on the current battery SOC. SOClow is
Qbatt ð0Þ is the initial capacity of the battery and Qloss ðr; AhÞ is the the battery SOC lower limit, SOChigh is the battery SOC upper limit,
capacity of the aged battery. The value of Qloss ðr; AhÞ is computed tEM;max and tEM;min is the maximum and minimum torque limits of
from Eq. (3.5.2), adopted from Ref. [10], motor and generator, respectively.
As represented in Fig. 7 the electric motor torque depends on
Qloss ðr; AhÞ ¼ sðrÞ*Ahz (3.5.2) the battery SOC value. When the motor torque is not sufficient to
meet the demanded torque, the engine operates and provides the
where z shows the dependency of the charge throughput on Qloss . remaining torque. In the traction region, when the SOC value is
Charge throughput is the amount of charging and discharging
above the higher limit, the motor provides the maximum torque.
current through the battery during the operation. sðrÞ is the Similarly, when there is a braking action and the battery SOC value
severity factor function which is expressed in Eq. (3.5.3) [10].
goes below the SOC low limit, the electric machine acts as a
generator and operates up to the highest torque limit.
Ea þ h:Ic
sðrÞ ¼ ða:SOC þ bÞ:exp (3.5.3)
Rg ð273:15 þ TÞ
4.2. Fuzzy logic controller
here, Ic is the C-rate which is defined as per Eq. (3.5.4)
This method is based on the utilization of fuzzy logic to split
jIj power between the engine and electric machine according to the
Ic ¼ (3.5.4)
Qbatt
5
S. Sarvaiya, S. Ganesh and B. Xu Energy 228 (2021) 120604
tEM ¼ 0:5*tEM; lim
tdmd ðtÞ > tEM; lim : tice ¼ tdmd tEM; lim (4.2.2)
tEM ¼ tdmd
tdmd ðtÞ < tEM; lim : (4.2.3)
tice ¼ 0
tEM ¼ 0:5*tEM;lim
tdmd ðtÞ > tEM;lim : tice ¼ tdmd 0:5*tEM; lim (4.2.4)
tEM ¼ 0:5*tdmd
tdmd ðtÞ < tEM; lim : (4.2.5)
tice ¼ 0:5*tdmd
Fig. 7. Thermostat electric motor torque map, based on battery SOC.
Case 2 : Traction case; SOCBatt ðtÞ < SOCLow
4.3. Thermostat strategy with high voltage battery pack and special
DC-DC converter 0 tice tice;max (4.4.1.6)
The cycle life of a battery is majorly affected by the charging and 0 uice uice;max (4.4.1.7)
discharging of the battery with high amount of current [7]. If the
current passing through the battery is limited, the aging of the
tEM;min tEM tEM;max (4.4.1.8)
battery can be controlled. The same concept is used in this control
strategy to reduce battery aging. To reduce the current throughput
through the battery, the battery pack arrangement is redesigned. 0 uEM uEM;max (4.4.1.9)
The total number of cells in the battery are 1170. The number of
series cells used in this battery pack is 130 and the number of Pbatt;min Pbatt Pbatt;max (4.4.1.10)
parallel cells is 9. It is assumed that the nominal output voltage of
the battery is around 440 V which is step down by the DC-DC where, m_ ice andm_ batt;eq are the instantaneous engine fuel con-
converter at around 220 V to meet the induction motor input sumption and motor equivalent fuel consumption factors. The en-
requirement. A bidirectional buck type of DC-DC converter is used. gine instantaneous fuel consumption is calculated through the
For a buck converter, the output voltage is lower than the input interpolation of the engine fuel map (Fig. 2). The motor equivalent
voltage and, the output current is higher than the input current. In fuel consumption is calculated using Eq. (4.4.1.3). where SCECMS is
EV and HEV, the primary function of the DC-DC converter is to the equivalent fuel consumption factor which accounts for the
boost/reduce the voltage of different component in a vehicle amount of fuel needed to replenish the electrical energy consumed,
depending on the load type and system requirement [20]. In this fSOC is the correction factor for maintaining battery SOC governed
case, the scope of the discussion is limited to the impact of a step- by Eq. (4.4.1.11) and Pbatt is the battery power requirement corre-
down DC-DC converter on battery life. The performance of DC-DC sponding to the motor operating point at time instant t.
converter is limited by its efficiency, electromagnetic interference, The ECMS employed in this paper is tailored to be adaptive like
and stress on internal switches with a higher conversion ratio. DC- the one discussed in Ref. [23]. The controller calculates the battery
DC converter with higher power capacity adds additional weight to cost based on the motor power and battery operating SOC as shown
the vehicle [21]. It is assumed that the performance of DC-DC in Eq. (4.4.1.3). The SOC correction factor fSOC is written as,
converter used here is a function of the converter efficiency, and
the efficiency reduces at higher voltage and increases at higher !3
current at the given voltage. SOCtþ1 SOCdesired
fsocðtþ1Þ ¼ 1 2 (4.4.1.11)
SOChigh SOClow
Tðf Tðf
JðtÞ ¼ _
meq ðtÞ dt ¼ m_ ice ðtÞ þ m_ batt;eq ðtÞ dt (4.4.1.1)
0 0
where,
7
S. Sarvaiya, S. Ganesh and B. Xu Energy 228 (2021) 120604
R ¼ wf m_ ICE þ m_ ESS 1 wf s þ 1 (4.5.1)
Fig. 9. Effect of including fsoc factor on battery pack SOC in the ECMS cost function.
where m_ ICE is the fuel rate of the engine and m_ ESS is the equivalent
fuel rate of battery (i.e., energy storage system (ESS)), s is battery
Therefore, the battery is discharged up to 50% SOC and then the
severity factor, wf is the weight of fuel consumption, ð1 wf Þ is the
controller goes into charge sustaining mode, maintaining the SOC
level constant. weight of battery severity factor (i.e., battery aging speed). The less
the fuel rate and battery severity factor, the greater the reward is.
The 1 added at the end of R is to make the reward a positive
4.4.2. ECMS with integrated gear shift control
number. The key equation in the Q-learning is the update equation
An attempt was made to improve the supervisory energy
as follows.
management control through the integration of gearshift and po-
wer split controllers into one common ECMS cost function by
building the cost matrix into a 2-D vector instead of a 1-D power
split vector. The second dimension of the cost matrix accounts for
the selection of optimum gear ratio for the gearbox coupled to the
Q ðst ; at Þ ¼ Q ðst ; at Þ þ a R þ gmaxQ ðstþ1 ; ai Þ Q ðst ; at Þ
engine. This will introduce an additional degree of freedom to the i
controller allowing more optimum powertrain operating point (4.5.2)
selection based on vehicle torque demand.
The updated cost function can be written as where Q ðst ; at Þ is the Q value at state and action pair ðst ; at Þ,
subscript t is the time step, a is the learning rate, g is the discount
Tðf Tðf factor, maxQ ðstþ1 ; ai Þ is the maximum Q value at state stþ1 among
Jt ¼ m_ eq ðtÞ dt ¼ m_ ice ðtÞ þ m_ batt;eq ðtÞ dt (4.4.2.1) i
all the action-state Q values. The term R þ gmaxQ ðstþ1 ; ai Þ is the
0 0 i
observed Q value at state and action pair ðst ;at Þ, whereas Q ðst ; at Þ is
where m_ ice ðtÞis f ðtice ; gear ratioÞ the estimated Q value. More information about the tabular Q-
learning parameter setup and implementation process can be
found in reference [4]. The procedures of Q-learning update during
the simulation are summarized as follows.
4.5. Q-learning
1) Initialize the Q value table with zeros.
Q-learning is a model-free reinforcement learning algorithm. 2) Run the vehicle simulation in one driving cycle (i.e., one itera-
Compared to Thermostat and rule-based strategies, Q-learning is an tion). During the simulation, the action corresponding to the
optimization-based strategy, which generally has higher compu- largest Q value at a given state is selected 92% of the time, and
tation cost and produces better fuel economy. Compared to ECMS, random action is taken at remaining 8% of the time.
Q-learning is model-free and long-horizon optimization. ECMS is a 3) Update the Q value table based on Eq. (6) from the first time-
model-based strategy because it requires engine and electric motor step to the last time-step using the data from the step number 2.
models when calculating the minimum fuel consumption. How- 4) Repeat step number 2 and step number 3 until sum of reward
ever, Q-learning does not need those models and only needs to does not increases any more.
interact with the actual vehicle or vehicle plant model (in this
study). In the vehicle interaction process, Q-learning takes the real- The Q value look-up table is a 4D table. The four inputs are three
time vehicle speed, torque demand, gear number signals and de- states and one action. The one output is the power split ratio be-
termines the optimal power split between engine and electric tween the engine and electric motor. State discretization is 5 and
motor. Additionally, ECMS only find the minimum fuel consump- action discretization is 20. The states and action discretization in-
tion for one-time step and does not consider the long-term future, formation are summarized in Table 1. The sum of rewards for each
while Q-learning utilizes Bellman Equation and considers future in iteration is shown in Fig. 10. It is observed that the Q-learning
the optimization process. In this study, a tabular Q-learning converges after 50 iterations.
8
S. Sarvaiya, S. Ganesh and B. Xu Energy 228 (2021) 120604
Table 1
State and action discretization and resolution data for Q-learning.
Fig. 10. Sum of Q-learning algorithm rewards over 150 iterations. Fig. 13 shows the comparative analysis of the quasi-static plot of
electric machine efficiency with the operating points; Fig. 13(a) and
(b) shows the electric machine efficiency map with the operating
5. Simulation and results points for Thermostat strategy and rule-based strategy, respec-
tively. As mentioned in case 1b (Section 4.2), if the demand torque
5.1. Comparative analysis of Thermostat and Thermostat with a is higher than ±50Nm or if the temperature exceeds the threshold
high voltage battery pack and a special DC-DC converter strategy value, the electric machine operation is limited to half of the
maximum/minimum torque it can provide. This behavior can be
Fig. 11 shows the comparative analysis of control strategies seen in Fig. 13 (b) for the rule-based strategy. This leads to lower
Thermostat and, Thermostat with a high voltage battery pack. As
shown in Fig. 11 (b), the C-rate in the high voltage battery pack is
slightly less, which results in slightly lower battery temperature
Fig. 12. Comparative analysis of Thermostat and Thermostat with special DC-DC
converter strategy, (a) Engine torque (b) Electric machine torque (c) Engine output
Fig. 11. Comparative analysis of Thermostat and Thermostat with special DC-DC energy and (d) Electric machine output energy comparison. Compared to Thermostat,
converter strategy, (a) SOC, (b) C-rate and (c) Temperature comparison. Thermostat Thermostat with special DC-DC converter strategy has overall lower motor torque
with special DC-DC converter strategy has lower C-rate, lower temperature rise and leading to the lower electrical energy consumption and, higher IC engine torque
lower SOC variation compared to Thermostat strategy. leading to higher fuel energy consumption.
9
S. Sarvaiya, S. Ganesh and B. Xu Energy 228 (2021) 120604
Fig. 13. Electric machine efficiency map with operating points, (a) Thermostat, and (b) rule-based strategy. The rule-based strategy has more operating points in the higher ef-
ficiency region compared to Thermostat strategy.
battery capacity loss and hence, improves the battery life. In the operating points with Thermostat strategy in Fig. 14(a) and,
case of Thermostat, the motor has the operating points all over the Fig. 14(b) shows the engine map with operating points with rule-
map (Fig. 13(a)). Also, from the efficiency point of view, the rule- based strategy. The rule-based strategy has more operating points
based strategy has more points in the high-efficiency region on maximum torque line in Fig. 14(b), compared to Thermostat.
compared to Thermostat. Since the Thermostat strategy is map- This shows that the engine is providing the additional torque to
based, the motor output torque is proportional to the existing meet the demanded torque when the motor operation is limited to
battery SOC level which limits the motor from operating in the ±50Nm. This leads to the higher fuel-economy with the rule-based
high-efficiency zone. Whereas, for rule-based strategy, the motor strategy. In Thermostat strategy, most of the engine operating
continues to supply until the battery SOC is beyond the battery SOC points are in the low-efficiency zone compared to the rule-based
lower limit. This allows the motor to operate in comparatively high- strategy. This is because the motor operation with Thermostat is
efficiency zones. not limited to ±50Nm like rule-based strategy. With Thermostat,
Fig. 14 shows the engine efficiency plot with the engine the motor provides the maximum amount of demanded torque
Fig. 14. Engine efficiency map with the operating points, (a) Thermostat, and (b) rule-based strategy. The rule-based strategy has more operating points on the maximum torque
line compared to Thermostat strategy.
10
S. Sarvaiya, S. Ganesh and B. Xu Energy 228 (2021) 120604
Fig. 17. Comparison of Thermostat and ECMS strategies in terms of power split be-
tween engine and electric machine, (a) UDDS Velocity Profile, (b) ICE Instantaneous
torque request, (c) ICE instantaneous fuel consumption, (d) Electric machine torque
request. ECMS avoids peaks in the IC engine torque curve by finding optimal power
split and thereby reduces overall fuel consumption over the drive cycle.
11
S. Sarvaiya, S. Ganesh and B. Xu Energy 228 (2021) 120604
Fig. 20. (a) Engine Operating points with normal ECMS (b) Engine Operating points with ECMS integrated with gearshift controller. ECMS with gearshift strategy can operate the
engine at comparatively lower speeds and hence lower fuel consumption.
5.6. Comparative analysis of fuel economy and battery life for all
the control strategies
1
EPA combined fuel economy ¼ 0:55
(5.6.1)
FEUDDS þ FE0:45
HWFET
Fig. 22. Electric machine efficiency map with operating points, (a) Thermostat strategy, (b) Q-learning strategy. Q-Learning strategy operates the electric motor at higher efficiency
points compared to Thermostat strategy.
Fig. 23. Engine efficiency map with operating points, (a) Thermostat (b) Q-learning strategy. Q-Learning strategy operates the engine at comparatively higher efficiency operating
points.
deciding the fuel economy and battery life. However, in the high- The rule-based strategy has the second lowest capacity loss in
way driving conditions, there is very little for the controller to work Fig. 27 and lower fuel consumption compared to Thermostat. This
on, as the vehicle does not brake much leading to the lower can be attributed to a comparatively more sophisticated control
regeneration of the energy. strategy. The rule-based strategy restricts the motor operation
Figs. 26 and 27 show the comparison of the battery capacity loss when the battery operating temperature and the torque request to
with 100 UDDS cycles for all the control strategies. From Fig. 26, it the electric machine cross the threshold value. This allows intelli-
can be inferred that the Q-learning strategy has the highest battery gent control over the battery capacity loss and hence, the battery
capacity loss whereas the high voltage battery pack strategy has the capacity loss is the least compared to ECMS, ECMS with gear con-
least capacity loss. As shown in the figure, the capacity loss is trol, Thermostat, and Q-learning strategies. When the battery SOC
closely related to the Ah-throughput. On the contrary, the fuel is higher than the SOC lower limit, the electric machine is given
consumption of the high voltage battery pack is high. priority for operation over the engine, which ensures higher fuel
14
S. Sarvaiya, S. Ganesh and B. Xu Energy 228 (2021) 120604
Fig. 24. Comparison of Thermostat and Q-learning strategy, (a) ICE torque, (b) Electric
Fig. 26. Comparison of battery capacity loss for all the control strategies with 100
machine torque, and (c) instantaneous fuel consumption. The overall ICE torque with
UDDS cycles (the strategies with the highest and the lowest battery capacity losses are
Q-learning strategy is lower compared to the Thermostat strategy, leading to lower fuel
pointed out). The red dotted box is zoomed in Fig. 27. Q-learning has the highest
consumption with Q-learning strategy.
battery capacity loss, and Thermostat with the high voltage battery pack has the lowest
battery capacity loss. (For interpretation of the references to colour in this figure
legend, the reader is referred to the Web version of this article.)
Table 2
Comparison of EPA combined Fuel Economy (FE) and battery capacity loss (%) values for different strategies over a period of 100 drive cycles of UDDS and HWFET.
Strategy Thermostat Rule-based Thermostat with special DC-DC converter ECMS no gear ECMS with gear Q-Learning
15
S. Sarvaiya, S. Ganesh and B. Xu Energy 228 (2021) 120604
6. Conclusion
Table 3
Cost analysis for the strategies discussed based on fuel economy, battery capacity loss and additional architecture cost for 50,000 miles. The last column shows the overall
operating cost for 50,000 miles and Q-learning has achieved the lowest operating cost among all the strategies.
16
S. Sarvaiya, S. Ganesh and B. Xu Energy 228 (2021) 120604
References
17
S. Sarvaiya, S. Ganesh and B. Xu Energy 228 (2021) 120604
VPPC.2005.1554534. [24] Duffner F, Wentker M, Greenwood M, Leker J. Battery cost modeling: a review
[23] Onori S, Serrao L, Rizzoni G. Adaptive equivalent consumption minimization and directions for future research. Renew Sustain Energy Rev
strategy for hybrid electric vehicles. ASME 2010 Dynamic Systems and Control 2020;127(April):109872. https://doi.org/10.1016/j.rser.2020.109872.
Conference, DSCC2010 2010;1:499e505. https://doi.org/10.1115/DSCC2010- [25] US Gasoline and Diesel Retail Prices. https://www.eia.gov/dnav/pet/pet_pri_
4211. gnd_dcus_nus_w.htm. [Accessed 19 September 2020].
18