This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2999939, IEEE
Transactions on Vehicular Technology
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 1

Computationally Efficient Stochastic Model

Predictive Controller for Battery Thermal
Management of Electric Vehicle
Seho Park and Changsun Ahn, Member, IEEE

 between cells of a lithium-ion battery cause unbalanced

Abstract— The performance and safety of batteries of electric discharging and aging [2]. Yang proposed the electro-thermal
vehicles deteriorate when the battery temperature is too low or model of a battery that includes the temperature-dependent
too high. The thermal management system regulating the battery internal parameters, such as resistance, open-circuit voltage,
temperature consumes considerable electric energy, particularly,
for cooling the battery. To maximize the vehicle driving range, the
and capacitance [3]. Wang reported that battery thermal
means of controlling the battery temperature should minimize the runaway could cause the battery to explode [4]. German used a
energy consumption. In this paper, stochastic model predictive battery electro-thermal model coupled with a traction model to
control is applied to the battery-cooling controller. Effective analyze the effects of temperature on the driving range of EVs
model predictive control requires a good but simple system model [5]. These studies showed that the driving range of EVs could
with proper estimation of near-future disturbances. The be underestimated or overestimated excessively unless the
components of the battery cooling system are modeled to
represent the energy-consumption and heat exchange mechanism
temperature effect is considered. Electric vehicles use a battery
with some assumptions for simplicity. The future information is for propulsion and regenerative braking. Accordingly, the
estimated using historical driving data in a stochastic sense. For battery always generates heat by the Joule effect while the
real-time implementation, an unequally spaced probability electric vehicle is driving. Therefore, battery thermal
distribution is introduced when designing a stochastic model of management is essential for electric vehicles.
future heat generation. The proposed control method shows Another issue in an electric vehicle is a driving range. Many
significantly lower energy consumption while maintaining an
acceptable temperature regulation performance compared to
drivers have anxiety regarding the driving range of electric
other temperature controllers, such as a thermostat-type vehicles [6, 7] because of the lower energy density of the power
controller and a model predictive controller with simply assumed source, longer refueling (charging) time, and less charging
future information. The proposed predictive controller shows infrastructure availability than those for conventional vehicles.
robust performance compared to typical controllers by utilizing Furthermore, the battery thermal management system
the stochastic estimation of future information. consumes energy, which could reduce the driving range of
electric vehicles [8-11]. Argonne National Laboratory showed
Index Terms— Battery Thermal Management, Cooling Cycle
Modeling, Stochastic Model Predictive Control, Unequally that the driving range of EVs might decrease by up to 40%
Spaced Probability Distribution. when the driver uses heating, ventilation, and air conditioning
systems [12]. Therefore, minimizing the energy consumption
I. INTRODUCTION of a battery thermal management system is essential for
extending the driving range of electric vehicles as well as for
T performance and safety of the battery of electrified
HE
vehicles are vulnerable to extreme temperature changes.
The battery properties, such as efficiency, durability, safety,
improving battery safety.
As the market share of electric vehicles increases, more
thermal runaway, and long-term performance, deteriorate when studies on a battery cooling system have been conducted. Most
the battery temperature is extremely low or high [1]. Many recent papers on battery cooling systems have been on system
researchers have studied the temperature-related characteristics design or mathematical models of battery cooling systems for
numerical simulation, such as design optimization [13], design
of a battery. Yang reported that the temperature differences
parameter analysis [14], energy consumption analysis [15, 16],
mathematical modeling [17], and simulator development
Manuscript received January 17, 2020; revised April 6, 2020; accepted June [18-20]. The main concerns in these studies were to develop
1, 2020. This work was supported in by the National Research Foundation of accurate models for numerical simulation not for controller
Korea funded by the Ministry of Science and ICT (No.
NRF-2019R1A2C1003103). (Corresponding author: Changsun Ahn) design nor minimization of energy use. Furthermore, those
Copyright (c) 2015 IEEE. Personal use of this material is permitted. models are too complex to be used in controller design.
However, permission to use this material for any other purposes must be Fewer papers have been published on the control of the
obtained from the IEEE by sending a request to [email protected].
Seho Park is with Pusan National University, Busan, 46241, South Korea
battery cooling system than on the design and performance
(e-mail: [email protected]) analysis. This is because the market share of electric vehicles
Changsun Ahn is with Pusan National University, Busan, 46241, South has been small, and most of the research effort focused on the
Korea (e-mail: [email protected])

0018-9545 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 27,2020 at 05:57:45 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2999939, IEEE
Transactions on Vehicular Technology
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 2

design and manufacture of a battery cooling system. Therefore, degrade the controller performance.
papers on the control of battery cooling systems are based Another issue of optimal control is the necessity of future
mostly on heuristic-based control methods because of their information. The controllers in most papers on optimal control
simplicity, e.g., rule-based control and fuzzy-based control of the battery cooling system assumed that future information
[21-23]. In practice, the most popular controller to regulate the could be obtained in advance using information technology. A
temperature of a system with a large specific heat capacity is a few studies on MPC for the battery cooling system simply
thermostat-type controller [24-26]. The performance of a assumed constant future information, which is called
thermostat-type controller is robust and its implementation is deterministic MPC. The performance of this method would be
straightforward. On the other hand, the controller may not degraded when many uncertainties exist. To consider such
minimize energy consumption. stochastic properties in building energy management and the
With advanced information technology, the energy power management of HEVs, stochastic MPC is used
consumption used for a battery thermal management system frequently and has shown robust and sub-optimal performances
might be reduced using a predictive controller based on optimal [42-44]. However, these advanced methods have not been used
control theory. The optimal control problem in the thermal for the battery-cooling problem, yet. Furthermore, any practical
management system is generally multi-objective optimization, methodology forecasting the future heat generation of the
e.g., regulation of battery temperature and energy consumption battery of electric vehicles has not yet been found in the
minimization. Optimal control methodologies, such as, literature.
dynamic programming (DP)[27, 28], Pontryagin’s minimum This paper proposes an MPC based battery-cooling
principle (PMP)[29, 30], and model predictive control controller designed with a simple but high-fidelity-cooling
(MPC)[31-33], that are widely used in other automotive control system model that can represent the component efficiencies and
applications, are as of yet not widely used in the control of the actuator characteristics with a practically implantable future
battery cooling system. Recently, some research groups have information prediction. To increase the model accuracy, a
published papers on optimal control for a battery cooling control-oriented model representing the component efficiency
system. These papers are on DP [34, 35], PMP [30], and MPC and heat transfer physics is described. A stochastic model
[36-40]. However, the number of papers is much smaller than predictive controller (SMPC) is designed to consider future
that on the control of hybrid electric vehicles, in which the information practically. One of the main challenges for SMPC
objective is to minimize fuel consumption. is obtaining stochastic future information or future heat
The optimal control methodologies achieved multiple generation. The stochastic data of heat generation is generated
objectives by defining explicit cost functions. Each paper using Autonomie®, electrified simulation software, with 90
clearly showed the effects of proposed control methods through driving cycles. To reduce the computation time for real-time
simulation. Nevertheless, the main concerns of the papers were implementation, we propose a stochastic model based on
on the methodologies, such as the controller structure, cost unequally spaced probability distributions in forecasting future
function, and efficient computation, not on a practical control heat generation. The proposed controller showed a significant
model or practical method to obtain future information. decrease in energy consumption compared to a conventional
The controllers in most papers on optimal control of the thermostat control approach or deterministic MPC.
battery cooling system were based on very simple or Furthermore, SMPC showed less variation in performance in
impractical battery cooling system models. For example, an air different driving cycles than that of a conventional thermostat
cooling system was used to cool down the battery [39], but it is control approach or deterministic MPC.
not generally used for regular-sized electric vehicles. In most This paper is organized as follows: Section II describes the
electric vehicles, a refrigeration cycle with a coolant cycle is battery-cooling system. Section III introduces a control model
used to regulate the battery temperature. Even in the studies for the MPC controller of the battery-cooling system. Section
considering the refrigeration cycle with a coolant cycle as the IV presents a stochastic MPC control algorithm and a
cooling system, the control models were too simplified. For probability distribution of the battery heat generation. In
example, the control model has one input (cooling rate) and one Section V, the performance of the proposed control algorithm is
state (battery temperature) only [35, 37, 41], the actuator validated. Finally, the conclusions are presented in Section VI.
efficiencies were neglected [34, 36, 37], and heat exchange
dynamics were not modeled [35-38, 41]. These models might II. BATTERY-COOLING SYSTEM
be sufficient to show the contributions in the controller In general, the battery-cooling system can be categorized by
structure, the cost function formulation, or the efficient solver the heat transfer medium and the way of cooling. The heat
but implementing the controller in the real cooling system is not transfer medium can be air or liquid. With air, the cooling
straightforward. The cooling rate is not directly controlled, but system structure is simple, and the manufacturing cost is low.
it is dynamic states controlled by the pump speed and the However, the cooling performance of an air-based system is
compressor speed. Furthermore, the battery temperature is relatively poor due to the low heat conductivity of air. On the
affected by the dynamic state, the coolant temperature. other hand, the liquid-based cooling system shows good
However, these models do not represent the energy performance because of the high heat conductivity of the liquid,
consumption of actuators, the dynamics of the cooling rate, and but it is expensive and requires additional structures, e.g., pipes,
the dynamics of the coolant temperature. Therefore, it might pumps, and liquid tanks.

0018-9545 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 27,2020 at 05:57:45 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2999939, IEEE
Transactions on Vehicular Technology
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 3

TABLE I
COOLING SYSTEM PARAMETERS
Symbol Description Value
ρrfg Density of refrigerant 27.8 kg/m3
ρclnt Density of coolant 1069.5 kg/m3
mbatt Battery thermal mass 40 kg
Crfg Specific heat capacity of refrigerant 1117 J/kg/°C
Cclnt Specific heat capacity of coolant 3330 J/kg/°C
Cbatt Battery specific heat capacity 1350 J/kg/°C
Fig. 1. Schematic diagram of the battery cooling system. Vcomp Displacement volume of compressor 33ꞏ10-6 m3/rev

Another categorization is based on the way of cooling: a Vpump Displacement volume of pump 33ꞏ10-6 m3/rev
passive or an active cooling system [45, 46]. A passive cooling heva
Heat transfer coefficient between
1000 W/m2/°C
system uses only ambient air to cool the heat transfer medium evaporator and refrigerant/coolant
Heat transfer coefficient between the
or the cooling target. Therefore, the passive cooling system has hbatt
battery and the coolant
300 W/m2/°C
a limited range of control temperature determined by the Heat transfer sectional area between
ambient air temperature. On the other hand, the active cooling Aeva evaporator and refrigerant/coolant 0.3 m2
system can cool the heat transfer medium or the cooling target Heat transfer sectional area between the
Abatt battery and the coolant 1 m2
to temperatures lower than the ambient air temperature.
However, its manufacture is expensive because of the PR Compression ratio of the compressor 5
additional structures needed, such as a heat exchanger and hcomp,out Enthalpy at the outlet of compressor 284.3 kJ/kg
refrigeration cycle. The active cooling system can control the heva,out Enthalpy at the outlet of evaporator 250.9 kJ/kg
target temperature within a wider range of temperatures at the hcond,out Enthalpy at the outlet of condenser 131.7 kJ/kg
cost of energy consumption [47, 48]. A liquid-based active
pcomp,out Pressure at the outlet of compressor 1500 kPa
cooling system is used for electric vehicles because of the
considerable heat generation of the electric vehicle battery [49, pcomp,in Pressure at the inlet of compressor 300 kPa
50]. pcond,out Pressure at the outlet of condenser 1500 kPa
The usual active cooling system consists of a heat source, m rfg Mass flow rate of refrigerant
heat sink, and fluid delivery component, as shown in Fig. 1.
m clnt Mass flow rate of coolant
The battery is the heat source, and the ambient air is the heat
sink. The refrigeration cycle consists of a compressor, a
condenser, a thermal expansion valve, and an evaporator. The between the engine and the compressor, the electric vehicle
electric-motor driven compressor compresses the refrigerant uses electric motors to drive the compressor and the pump that
and the refrigerant gas heats up as it is pressurized. At the consume the most energy in the battery cooling system and the
condenser, heat exchange occurs between the refrigerant gas consumed energy for battery cooling comes from the battery
and the ambient air, and the refrigerant gas is condensed to a itself.
liquid. As a result, the condenser dissipates the heat of the A battery pack consists of battery cells. Each cell generates
refrigerant to the ambient air. The liquid refrigerant vaporizes heat due to internal resistance and entropy change. There are
by dropping its pressure at the thermal expansion valve. At the several battery cell arrangements and the battery thermal model
evaporator, the refrigerant receives the latent heat through a is affected by the arrangement. In the series arrangement [51],
phase change, and the coolant of the cooling cycle is cooled by the battery cells are arranged in a straight line and they are
transferring its heat to the refrigerant. The refrigeration cycle cooled in one direction, e.g., from left to right. The
operates when the battery cooling is required as well as the one-direction cooling makes temperature differences between
cabin cooling is required. Accordingly, the pump in the cooling cells because the cooling medium temperature gradually
cycle is necessary to make the battery cooling from cabin increases as it moves from inlet to outlet. In the parallel
cooling independent. arrangement [52], the battery cells are cooled from top to
The cooling cycle consists of an electric motor-driven pump, bottom or bottom to top. Therefore, they are cooled
coolant channels in the battery, and an evaporator shared with simultaneously, which makes the cell temperature uniformly.
the refrigeration cycle. The temperature of the coolant is Generally, the series/parallel arrangement [53] is applied to the
reduced through the heat exchange between the refrigerant and battery system to make the package compact. The temperature
coolant at the evaporator. The coolant passing through the uniformity can be improved through a short path of the inlet to
coolant channels cools the battery. The pump circulates the outlet compared to the series arrangement. In this research, the
coolant. battery is modeled as a lumped model without considering cell
In the battery cooling system, the power-consuming spacing and arrangement.
components are the pump and compressor. Unlike internal
combustion engine vehicles having a mechanical connection III. BATTERY-COOLING SYSTEM MODEL

0018-9545 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 27,2020 at 05:57:45 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2999939, IEEE
Transactions on Vehicular Technology
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 4

For the controller design, the refrigeration and the cooling

cycles should be modeled to represent the corresponding
cooling mechanism sufficiently. Some studies used a too
simple model that did not represent the cooling mechanism and
energy loss mechanism. This is because the main objective of

Pressure [kPa]
the previous studies was to show the efficacy of the proposed
control methodologies. Furthermore, in many cases, the output
signals from the controller were indirect and not the actuator
commands, such as a cooling rate that cannot be controlled
directly. In this section, the battery-cooling system is modeled
as one representing a heat generation mechanism, a
refrigeration cycle, and a cooling cycle considering the physics
of heat exchange between fluids. The model parameters of the
electric system and thermal system were obtained from Fig. 2. Pressure and enthalpy diagram of R-134a refrigerant.
Autonomie® of Argonne National Laboratory [54], and the
parameters are listed in Table I.
A. Battery Thermal Model
The battery power consists of power for vehicle driving and
power for the battery cooling system as follows:

Pbatt  Pdriving  Pcooling , (1)

where Pbatt, Pdriving, and Pcooling are the total battery power,
power for vehicle driving, and power for battery-cooling,
respectively. The power for battery-cooling consists of the
electric power consumption of the compressor, Pcomp,elec, and
the electric power consumption of the pump, Ppump,elec:

Pcooling  Pcomp , elec  Ppump , elec , (2)

The battery current, Ibatt, can be expressed as Fig. 3. Component characteristics. (a) volumetric efficiency of compressor
and pump, (b) isentropic efficiency of compressor, (c) power conversion
efficiency of compressor and pump, (d) pump pressure drop
Voc  Voc2  4 Pbatt Rbatt
I batt  , (3)
2 Rbatt the energy in the battery-cooling system, and the compressor
consumes the most power. The compressor power is nonlinear
where Voc is the open-circuit voltage, and Rbatt is the internal and dependent on the speed, compression ratio, and mass flow
resistance of the battery, both are variables dependent on the rate. The compressor can be modeled as a static component if
battery state of charge [55]. the internal mass and the energy change at the compressor are
The heat generation of the battery is modeled as a lumped ignored [57, 58], as follows:
thermal mass model. The heat generation rate of a battery is
described as m rfg  Vcomp comp  comp   rfg  2 / 60, (5)

dVoc where ωcomp is the compressor speed in rpm. The volumetric

Q gen  I batt
2
Rbatt  I batt Tbatt , (4)
dTbatt efficiency of the compressor, ηcomp, can be expressed as a
function of the compression ratio of the compressor (PR) and
where Q̇ gen is a heat generation rate, and Tbatt is the battery the compressor speed, as shown in Fig. 3(a). PR is defined as
temperature. The first term on the right is caused by the internal follows:
resistance, and the second term is caused by a change in entropy
[56]. PR  pcomp , out pcomp , in , (6)
B. Refrigeration Cycle Model
where pcomp,out and pcomp,in are the pressure at the outlet and at the
The refrigeration cycle consists of a compressor, a condenser,
inlet of the compressor, respectively. The electric power
a thermal expansion valve, and an evaporator. Fig. 2 presents
consumption of the compressor is expressed as follows:
the pressure and enthalpy diagram of R-134a used in the
refrigeration cycle. The refrigeration cycle consumes most of

0018-9545 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 27,2020 at 05:57:45 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2999939, IEEE
Transactions on Vehicular Technology
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 5

Pcomp , mech m rfg ( hcomp , out  heva , out ) the coolant channels in the battery, respectively. The
Pcomp , elec  , Pcomp , mech  , (7) temperature at the outlet can be expressed as
c isen
hbatt Abatt

where Pcomp,mech, ηisen, and ηc are the mechanical power Tclnt, out  (Tclnt , in  Tbatt )e m clnt Cclnt
 Tbatt , (12)
delivered to the refrigerant by the compressor, the isentropic
efficiency of the compressor shown in Fig. 3(b), and the power
conversion efficiency of the compressor shown in Fig. 3(c), where Tbatt is the lumped battery temperature. The battery
respectively. thermal dynamics can be expressed as
At the condenser, the refrigerant gas exchanges its heat with
the ambient air (Q̇ c). Through heat exchange, the refrigerant gas 
Tbatt  Q gen  Q cool  m batt Cbatt . (13)
transforms into a liquid by releasing its latent heat (from
hcomp,out to hcond,out). The heat transfer rate between the The pump model is similar to the compressor model. The
refrigerant and ambient air at the condenser is as follows: electric motor can control the pump speed. The power
consumption of the pump is represented as follows:
Qc  m rfg (hcomp, out  hcond , out ). (8)
m clnt  V pump  pump   pump   clnt  2 / 60, (14)
The thermal expansion valve expands the liquid refrigerant,
and the refrigerant turns into a gas by passing through the valve. where ηpump the volumetric efficiency of the pump. The electric
The pressure of the refrigerant is lowered by as much as the power consumed by the pump is
pressure increase at the compressor as follows:
Ppump , mech m clnt
peva , in  pcond , out  pcomp , pcomp  pcomp , out  pcomp , in . (9) Ppump , elec  , Ppump , mech  p pump , (15)
p  clnt

The evaporator is a heat exchanger between the refrigerant of where Pcomp,mech is the mechanical power delivered to the
the refrigeration cycle and the coolant of the cooling cycle. coolant by the pump, ηp is the power conversion efficiency of
Heat exchange between refrigerant and coolant occurs by the pump shown in Fig. 3(c), and ∆ppump is the pump pressure
convection. The heat exchange process was modeled as a drop shown in Fig. 3(d).
parallel flow heat transfer model [59]. Therefore, the output
temperature of the evaporator can be formulated as follows: IV. CONTROLLER DESIGN

Tclnt , eva , out  ATclnt , eva , in  (1  A)Trfg , eva , in , A. Stochastic Model Predictive Control
The control objective of the controller is to minimize energy
Trfg , eva , out  (1  B )Tclnt , eva , in  BTrfg , eva , in ,
consumption while satisfying all the physical constraints, such
k2  k1e  ( k1  k2 ) k  k e  ( k1  k2 ) as the limits of the compressor speed, pump speed, and battery
A , B 1 2 , (10)
temperature. The problem is a multi-state, constrained, and
k1  k2 k1  k2
nonlinear dynamic optimal control problem. The disturbance of
heva Aeva h A
k1  , k2  eva eva , the control problem is the heat generation of the battery (Q̇ gen).
m rfg Crfg m clnt Cclnt The states are the battery temperature (Tbatt) and coolant
temperature (Tclnt). The control variables are the compressor
where Tclnt,eva,out is the coolant temperature at the outlet of the speed (ωcomp) and pump speed (ωpump).
evaporator, Trfg,eva,out is the refrigerant temperature at the outlet Controllers based on MPC can be categorized into two
of the evaporator, Tclnt,eva,in is the coolant temperature at the inlet groups according to the way of addressing future information:
of the evaporator, and Trfg,eva,in is the refrigerant temperature at deterministic MPC (DMPC) and stochastic MPC (SMPC).
the inlet of the evaporator. DMPC is usually referred to as frozen-time MPC. Frozen-time
MPC assumes that the disturbance is unchanged within the
C. Cooling Cycle Model
prediction horizon and it minimizes the cost function in the
The cooling cycle consists of coolant channels in the battery, prediction horizon [60]. Therefore, frozen-time MPC can be
a pump, and an evaporator shared with the refrigeration cycle. implemented easily, but the performance is strongly dependent
The heat transfer rate of the coolant channels can be formulated on the driving cycle.
using a uniform wall temperature model [59], as follows: In contrast, the SMPC minimizes the expected value of the
given cost function in the prediction horizon. Therefore, it
Q cool  m clnt Cclnt (Tclnt , in  Tclnt , out ), (11) requires a probability distribution of future information [43]. In
SMPC for the battery cooling system, the future information is
where Q̇cool is the heat transfer rate of the coolant channels; heat generation of the battery. This probability distribution is
Tclnt,in and Tclnt,out are the temperatures at the inlet and outlet of the critical input to a controller based on SMPC. Generally,

0018-9545 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 27,2020 at 05:57:45 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2999939, IEEE
Transactions on Vehicular Technology
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 6

historical data with an assumption of a Markov process is used

to achieve the probability distribution of future information.
The SMPC problem for a battery-cooling system is defined
to minimize the energy consumption with inequality constraints
of the battery temperature and component operations. Based on
the probability distribution of future battery heat generation,
SMPC determines the compressor speed and the pump speed
that minimizes the expected energy consumption while
regulating the battery temperature within the range. The cost
function is formulated as follows:

 N p 1  Fig. 4. Probability function of the vehicle states. Pr[vk, ak]= f1(vk, ak)
J  E   Pcomp (k )  Ppump (k )   (Tbatt ( N p )  Tdes )2  , (16)
Qgen
 k 0 

where Np is the number of the prediction horizon that is five

(five seconds); Pcomp(k) and Ppump(k) are the electric power
consumption of the compressor and pump at the kth time step,
respectively; and α is a weighting factor. The last term of the
cost function constrains the battery temperature deviation from
the desired temperature. The equality constraint of the SMPC is
the system dynamics described in Section III and expressed as

x ( k  1)  f ( x ( k ), u ( k ), d ( k )), (17)

where x(k) is the state, u(k) is the control input, d(k) is the
disturbance, which are represented as follows:

 T (k )   comp (k )  
x(k )   batt  , u(k )    , d (k )  Qgen (k ). (18)
 Tclnt (k )   pump (k ) 
Fig. 5. Probability function of the battery heat generation. Pr[Q̇ gen,k, vk, ak] =
The inequality constraints are as follows: f2(Q̇ gen,k, vk, ak)

Tbatt , min  Tbatt ( k )  Tbatt , max , B. Probability Distribution of Battery Heat Generation
Tclnt , min  Tclnt ( k )  Tclnt , max , The probability distribution of battery heat generation can be
comp , min  comp ( k )  comp , max , (19) acquired from real driving data or simulation data. To obtain
the probability distribution of heat generation, an electric
 pump , min   pump ( k )   pump , max , vehicle with a battery-cooling system on 90 different driving
Pbatt , min  Pbatt ( k )  Pbatt , max , cycles was simulated using Autonomie® software.
Battery heat generation is strongly correlated with the
where min and max refer to the minimum and maximum values, vehicle velocity and acceleration. The probability function of
respectively. the vehicle velocity and acceleration was extracted from the
The MPC control algorithm is based on a receding horizon simulation result and named as f1(v, a), as shown in Fig. 4.
mechanism. The optimal control sequences in the prediction The probability function of battery heat generation was also
horizon are solved and the first control input is used. Stochastic extracted and named as f2(Q̇ gen, v, a), as shown in Fig. 5. One
MPC requires the statistical characteristics of future can observe that the heat generation correlates to vehicle
information. The most crucial future information is battery heat. velocity and acceleration. As the velocity increases, the most
The battery heat generation is highly correlated to the states of probable heat generation increases because higher-speed
the vehicle longitudinal dynamics, such as the speed and driving requires more power to overcome the higher
acceleration. Therefore, we need the correlation between the aerodynamic drag forces. As the acceleration increases, more
heat generation and vehicle states, and the correlation between power is more likely because the battery has to provide more
the future vehicle states and the current vehicle states. The power for traction. As the deceleration increases, higher power
correlations are expressed in the form of probability density is also more likely because the battery has to absorb more
functions and they are used to calculate the expected cost (16). energy from the regenerative braking. Given the vehicle
velocity and acceleration, the probability distribution of the

0018-9545 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 27,2020 at 05:57:45 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2999939, IEEE
Transactions on Vehicular Technology
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 7

Fig. 8. Probability distributions of Pr[Q̇ gen,k+i| vk=90km/h, ak=−0.5m/s2]

Fig. 6. Probability function of the vehicle state. Pr [vk+1, vk, a] = f3(vk+1, vk, ak)

Fig. 9. Equally spaced probability distribution of Pr[Q̇ gen,k+1| vk=30km/h,

ak=1.5m/s2] v.s. unequally spaced probability distribution of Pr[Q̇ gen,k+1|
vk=30km/h, ak=1.5m/s2]

Pr  vk 1 , ak 1 | vk , ak 
 Pr  vk 1 | vk , ak   Pr  a k  1 | vk , ak  (21)
f 3 (vk 1 , vk , ak ) f 4 ( ak 1 , vk , ak )
 .
 f3 (vk 1 , vk , ak )
vk 1
 f 4 (ak 1 , vk , ak )
ak 1

Given the vehicle velocity and acceleration of the kth time

step, the probability of heat generation of the k+1th time step
can be derived using (20) and (21) as follows:

Pr  Q gen , k 1 | vk , ak 

 Pr[Q gen, k 1 | vk 1 , ak 1 ] Pr[vk 1 , ak 1 | vk , ak ]

xk 1
(22)
 ,
Fig. 7. Probability function of the vehicle state. Pr[ak+1, vk, ak] = f4(ak+1, vk, ak) Pr  vk , ak 
current heat generation, Q̇ gen,k, can be expressed as follows:
Using a similar derivation, the probability of heat generation of
the later time steps, Pr[Q̇ gen,k+i| vk, ak] where i is a positive
f 2 (Q gen , k , vk , ak )
Pr  Q gen , k | vk , ak   , (20) integer, can be derived. Fig. 8 provides some examples of such

Q gen
f (Q , v , a )
2 gen k k probability.
C. Unequally Spaced Probability Distribution of Battery Heat
where v is the vehicle velocity and a is the vehicle acceleration. Generation
The future states of vehicle dynamics are also estimated in a In the probability distributions, the heat generation is equally
stochastic sense. Given the current states, the probability of the spaced into 31 values from 0 kW to 3kW. The number of
future states was extracted, as shown in Fig. 6 and Fig. 7. The discretized values of the stochastic variable is critical to the
velocity and acceleration of one-second future are highly computation time of the stochastic MPC. A large number of
correlated with the current velocity and acceleration. We discretized values can represent the overall shape of the
assumed that vk and ak are independent, and the vehicle states probability distribution very well but they require considerable
are in Markov process. The probability distributions are computing time. A small number of discretized values can
expressed as follows: reduce the computing time, but they may not represent the
probability distribution sufficiently. To represent the

0018-9545 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 27,2020 at 05:57:45 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2999939, IEEE
Transactions on Vehicular Technology
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 8

Fig. 10. Test cycle for controller validation (modified UDDS cycle).

Cumulative Energy

Cumulative Energy
Consumption [kJ]

Consumption [kJ]
Fig. 12. Simulation results of the modified UDDS cycle. (a) SMPC with a
simple model (b) SMPC with a high-fidelity model
Fig. 11. (a) SMPC with a simple model vs. (b) SMPC with a high-fidelity
model 1000
Controller Cooling Command
Actual Cooling Rate
800
probability distribution effectively with a small number of
discretized values, this paper proposes an unequally spaced 600

probability distribution, as shown in Fig. 9. 400

The values for the unequally distributed probability were
200
selected as the mean values of the groups of probabilities
clustered by the k-means clustering. SMPC with the unequally 0
1740 1750 1760 1770 1780 1790 1800 1810 1820
spaced probability distribution can reduce the computing time Time [s]
Fig. 13. Controller cooling command and actual cooling rate of the simple
significantly with little performance loss. The following model result from 1735 to 1820 seconds.
chapter presents the performance and computing time of SMPC
with the unequally spaced probability distribution. For validation, a vehicle was simulated with two controllers
on the modified UDDS cycle shown in Fig. 10. Fig. 11 presents
V. CONTROLLER VALIDATION the schematic diagrams of controllers. An optimal command
The performance of the proposed controller was validated allocator based on DP was located because SMPC with a
with respect to three contributions: effect of the model fidelity, simple model does not generate actuator commands
effect of SMPC, and effect of the use of an unequally spaced (compressor speed and pump speed). Therefore, the system in
probability distribution. Fig. 11(a) runs only in off-line mode. Fig. 11(b) presents the
structure of the proposed controller.
A. Effect of Model Fidelity Figs. 12 and 13 present the simulation results. The
To validate the effect of model fidelity on the performance of operations of the compressors, the most energy-consuming
the controller, SMPC with a simple model and SMPC with the component, differ according to the control methods, as shown
high-fidelity model were simulated on the same driving cycle. in Fig. 12. The compressor with the proposed controller
Many studies used a simple model to control the operated less frequently than that with the simple model.
battery-cooling system and showed the performance of the Furthermore, the compressor with the proposed controller ran
proposed control methodologies. The assumptions of the mostly at 5000 rpm, where the compressor is efficient, whereas
research using a simple model were as follows. The cooling rate the compressor with the simple model ran at a variety of speeds.
can be controlled directly. The power consumption is As a result, the controller with the high-fidelity model
proportional to the battery-cooling rate, and the component consumed much less energy on the battery-cooling system than
efficiencies of the fluid cycle and the heat exchange physics can the controller using the simple model. This is because the
be ignored. The high-fidelity model introduced in Chapter III proposed controller took the component efficiencies into
does not rely on such assumptions. account and the controller with the simple model frequently

0018-9545 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 27,2020 at 05:57:45 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2999939, IEEE
Transactions on Vehicular Technology
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 9

Temperature [°C]

Temperature [°C]

Temperature [°C]

Temperature [°C]
Battery

Battery

Battery

Battery
Fig. 14. Simulation results of the modified UDDS. (a) thermostat-type controller (b) DMPC (c) SMPC (d) DP

commanded a cooling rate that could not be met with the where Tthres,upper is the upper threshold and Tthres,lower is the lower
current battery and coolant temperatures whatever the pump or threshold. The operating rules of thermostat controller are
compressors worked, as shown in Fig. 13. determined by tuning parameters and the controller
performance could vary according to the parameters. The
B. Effect of Stochastic Future Information
controller operating rules were tuned as 2000 rpm and 3000
Another factor that affects the performance of the MPC rpm for pump and compressor respectively through several
controller is the means of forecasting future information. Most series of simulations.
MPC controllers for a battery-cooling system assumed that the
future information is given or assumed to be constant. The 2) Reference Controller 2: MPC with Deterministic Future
proposed controller, SMPC, was validated with a vehicle Information
simulated on a modified UDDS cycle. To demonstrate the The MPC with deterministic future information or
superiority of the proposed controller, a simple feedback-based deterministic MPC (DMPC) assumed the future information to
controller and a controller based on simply assumed future be a constant value in the prediction horizon. In the
information were also simulated on the same driving cycle. For battery-cooling problem, the required future information is the
comparison, the globally optimal control behavior was also level of heat generation in the battery, and DMPC assumes that
achieved by dynamic programming. future heat generation remains the same as the averaged heat
generation of the near past. The averaged heat generation can
1) Reference Controller 1: Thermostat Controller be calculated by the moving average filter, as follows:
A thermostat controller is widely used in industry. The

 
controller has a simple structure, and its implementation is
straightforward. The thermostat rule for the pump and Q gen ( k )  Q gen ( k  1)  Q gen ( k )  Q gen ( k  n) / n, (24)
compressor speeds is as follows:
where n is the corresponding number of applied average
 2000 rpm, Tbatt ( k )  Tthres , upper moving filters and Q gen is the average battery heat generation.
 pump ( k )   ,
 0 rpm, Tbatt ( k )  Tthres , lower
(23) The cost function of DMPC is defined as follows:
 3000 rpm, Tbatt ( k )  Tthres , upper
comp ( k )   , N p 1
 0 rpm, Tbatt ( k )  Tthres , lower
J 
k 0
Pcomp ( k )  Ppump ( k )   (Tbatt ( N p )  Tbatt , des ) 2 . (25)

0018-9545 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 27,2020 at 05:57:45 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2999939, IEEE
Transactions on Vehicular Technology
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 10

Temperature [°C]

Temperature [°C]

Temperature [°C]

Temperature [°C]
Battery

Battery

Battery

Battery
Fig. 15. Simulation results of the modified India Highway. (a) thermostat-type controller (b) DMPC (c) SMPC (d) DP

Fig. 16. Simulation results of the modified CUEDC. (a) thermostat-type controller (b) DMPC (c) SMPC (d) DP
Temperature [°C]

Temperature [°C]

Temperature [°C]

Temperature [°C]
Battery

Battery

Battery

Battery

Fig. 17. Simulation results of the modified LA92. (a) thermostat-type controller (b) DMPC (c) SMPC (d) DP

N 1

where Np is the size of the prediction horizon. The variables in J   Pcomp (k )  Ppump ( k )   (Tbatt ( N )  Tbatt , des ) 2 , (26)
k 0
the cost function are the same in the cost function of the SMPC
controller. The equality and inequality constraints are also the
where N is the number of time steps of a given driving cycle.
same. However, the DMPC minimizes the deterministic cost
(25), not the expected value of the cost (16).
4) Simulation
The four controllers (thermostat controller, DMPC, SMPC,
3) Reference Controller 3: Dynamic Programming
and DP) were simulated with the same vehicle on the modified
DP is usually used to compute the globally optimal control
UDDS cycle. Fig. 14 presents the simulation results. The
behaviors and the optimal cost. However, it is not implemented
energy consumption of the thermostat controller was 322.57 kJ.
in real-time due to its non-causal nature. The cost function of
The upper and lower thresholds of the thermostat controller
DP was defined, as follows:
were 34 °C and 32.5 °C, respectively. Unlike the other
controllers based on optimization methods, the thermostat
controller uses only feedback of the battery-cooling system.

0018-9545 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 27,2020 at 05:57:45 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2999939, IEEE
Transactions on Vehicular Technology
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 11

Fig. 18. Energy consumptions of battery cooling system with different control
methods on seven driving cycles (energy consumptions are normalized by the
consumption of DP), (a) energy consumptions, (b) average and standard
deviation of energy consumptions.

Hence, it shows higher energy consumption as well as more

significant battery temperature deviations than the other
controllers, as shown in Fig. 14(a). The energy consumption of Fig. 19. Effect of unequally spaced probability on SMPC implementation, (a)
performance (energy consumption), (b) computing time
DMPC was 186.34 kJ. The DMPC controller simply forecasts
the future heat generation using moving average filter and To validate the effects of the unequally spaced probability
assumed the future heat generation remained as constant value distribution in SMPC implementation, the results were
until the end of the prediction horizon. The DMPC controllers compared with those of two types of MPCs. The first was the
outperformed the thermostat controller, as shown in Fig. 14(b). SMPC with an equally spaced probability distribution, where
The globally optimal energy consumption calculated by DP the future heat generation was discretized into 31 equally
was 125.95 kJ, as shown in Fig. 14(d). DP requires the entire spaced values. The second was the DMPC. Furthermore, three
future heat generation of the driving cycle and thus DP was levels of discretization were also compared to examine the
applied to battery-cooling system offline. It was used as the optimal level of discretization because the performance and
benchmark performance. The proposed controller, SMPC, computing time are strongly dependent on how many discrete
showed an energy consumption of 146.82 kJ, which is much values are used to represent the probability distribution. The
less than that of DMPC and reasonably close to the optimal performance variations according to the level of discretization
consumption. This is because that the stochastic model of are shown in Fig. 19. As fewer numbers are used for
future heat generation represents the driving patterns better discretization, the performance degrades but not proportionally.
than the simply assumed and deterministic future heat For the computing time, as fewer numbers are used for
generation in DMPC. Figs. 15, 16, and 17 show other three discretization, the computing time is reduced, as shown in Fig.
driving cycles validation results. The controller validation 19. From the two plots, we concluded that the probability
results showed similar performances of modified UDDS cycle. distribution with three discrete values are optimum because it
Another advantage of SMPC over deterministic controllers showed comparable performance as the conventional SMPC
is its robust performance. To evaluate the robust performance but with a very low computing load. These results show that the
of the proposed controller, the four controllers were simulated proposed controller, SMPC with an unequally spaced
on seven different driving cycles, such as urban, suburban, probability distribution, is as powerful as regular SMPC that is
highway, and so on. Fig. 18(a) shows the energy consumption too slow, and as fast as DMPC that is less powerful than SMPC.
for the different driving cycles. The energy consumption of the
SMPC controller was lower than that of the DMPC controller VI. CONCLUSION
and thermostat controllers. Fig. 18(b) shows the average and
standard deviation of energy consumption. The SMPC This paper presented a computationally efficient stochastic
controller outperformed the DMPC controller and the model predictive controller for a battery cooling system. The
thermostat controller as well as showing smaller variations in optimal control-based algorithm was required to minimize the
performance for diverse driving cycles. energy usage in the battery cooling system while regulating the
battery temperature within the desired range. Furthermore, a
C. Effect of an Unequally Spaced Probability Distribution stochastic model of future disturbance was used for robust

0018-9545 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 27,2020 at 05:57:45 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2999939, IEEE
Transactions on Vehicular Technology
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 12

performance of the model predictive controller. These of a hybrid electric vehicle," Energy Conversion and Management, vol.
129, pp. 250-261, 2016.
approaches resulted in a stochastic model predictive controller. [11] J. R. M. D. Reyes, R. V. Parsons, and R. Hoemsen, "Winter happens:
However, the critical issue of the stochastic model predictive The effect of ambient temperature on the travel range of electric
control is the high computing load. We proposed an unequally vehicles," IEEE Transactions on Vehicular Technology, vol. 65, no. 6,
pp. 4016-4022, 2016.
spaced probability distribution model for SMPC [12] M. A. Jeffers, L. Chaney, and J. P. Rugh, "Climate control load reduction
implementation. The new model can represent the overall strategies for electric drive vehicles in warm weather," SAE Technical
probability distribution with the probabilities of three discrete Paper, 2015. Available: https://doi.org/10.4271/2015-01-0355
disturbance values. This approach reduced the computing load [13] A. Jarrett and I. Y. Kim, "Design optimization of electric vehicle battery
cooling plates for thermal performance," Journal of Power Sources, vol.
by 91% compared to the regular SMPC with only 4% of 196, no. 23, pp. 10359-10368, 2011.
performance reduction. The unequally spaced probability [14] J. Zhao, Z. Rao, Y. Huo, X. Liu, and Y. Li, "Thermal management of
distribution model could be a good choice for those wishing to cylindrical power battery module for extending the life of new energy
electric vehicles," Applied Thermal Engineering, vol. 85, pp. 33-43,
apply SMPC to control applications but are reluctant to because 2015.
of the excessive computing demand. The performance [15] Q. Zhang, Y. Meng, C. Greiner, C. Soto, W. Schwartz, and M. Jennings,
(55~85% of the globally optimal performance) of optimal "Air conditioning system performance and vehicle fuel economy
trade-offs for a hybrid electric vehicle," SAE Technical Paper, 2017.
controllers for the battery cooling system was not as good as the Available: https://doi.org/10.4271/2017-01-0171
optimal controllers for the power management of hybrid [16] T. W. Kim, M. Y. Ryu, S. S. Park, K. H. Kim, H. Y. Kim, and S. J. Park,
electric vehicles (80~95% of the globally optimal "Performance estimation of new vehicle thermal management system for
an unmanned military ground vehicle using integrated simulation
performance). The optimal controllers for battery cooling model," Int J Automot Technol, vol. 20, no. 3, pp. 429-444, 2019.
system could be improved further (possibly up to 95%, which is [17] S. Shojaei, S. Robinson, C. Chatham, A. McGordon, and J. Marco,
the performance of the most successful HEV power "Modelling the electric air conditioning system in a commercially
available vehicle for energy management optimisation," SAE Technical
management controller reported thus far) by better forecasting Paper, 2015. Available: https://doi.org/10.4271/2015-01-0331
of future disturbances, more accurate approximation of the [18] T. Kiss and J. Lustbader, "Comparison of the accuracy and speed of
terminal cost, and a computationally more efficient optimizer transient mobile A/C system simulation models," SAE Int. J. Passeng.
for a longer prediction horizon with a smaller computing load. Cars - Mech. Syst., vol. 7, no. 2, pp. 739-754, 2014.
[19] T. Kiss, L. Chaney, and J. Meyer, "A new automotive air conditioning
The proposed control method showed significant system simulation tool developed in MATLAB/Simulink," SAE Int. J.
performance improvement with a minimum increase of Passeng. Cars - Mech. Syst., vol. 6, no. 2, pp. 826-840, 2013.
computing load. The next research goal would be increasing the [20] J. Choi, K. Bae, J. Wi, S. Ahn, and H. Kim, "Hydraulic control system
design for a PHEV considering motor thermal management," Int J
control performance up to the performance level of the HEV Automot Technol, vol. 19, no. 1, pp. 147-157, 2018.
power management controller and conducting experimental [21] B. S. K. K. Ibrahim et al., "Fuzzy-based temperature and humidity
validation that could lead to be used as a controller in control for HVAC of electric vehicle," Procedia Engineering, vol. 41,
pp. 904-910, 2012.
production electric vehicles. [22] G. Titov and J. A. Lustbader, "Modeling control strategies and range
impacts for electric vehicle integrated thermal management systems with
REFERENCES MATLAB/Simulink," SAE Technical Paper, 2017. Available:
https://doi.org/10.4271/2017-01-0191
[1] G. Xia, L. Cao, and G. Bi, "A review on battery thermal management in [23] A. Yokoyama, T. Osaka, Y. Imanishi, and S. sekiya, "Thermal
electric vehicle application," Journal of Power Sources, vol. 367, pp. management system for electric vehicles," SAE Int. J. Mater. Manuf.,
90-105, 2017. vol. 4, no. 1, pp. 1277-1285, 2011.
[2] N. Yang, X. Zhang, B. Shang, and G. Li, "Unbalanced discharging and [24] X. Yan, J. Fleming, and R. Lot, "A/C energy management and vehicle
aging due to temperature differences among the cells in a lithium-ion cabin thermal comfort control," IEEE Transactions on Vehicular
battery pack with parallel combination," Journal of Power Sources, vol. Technology, vol. 67, no. 11, pp. 11238-11242, 2018.
306, pp. 733-741, 2016. [25] E. Vrettos, C. Ziras, and G. Andersson, "Fast and reliable primary
[3] Z. Yang, D. Patil, and B. Fahimi, "Electrothermal modeling of frequency reserves from refrigerators with decentralized stochastic
lithium-ion batteries for electric vehicles," IEEE Transactions on control," IEEE Transactions on Power Systems, vol. 32, no. 4, pp.
Vehicular Technology, vol. 68, no. 1, pp. 170-179, 2019. 2924-2941, 2017.
[4] Q. Wang, P. Ping, X. Zhao, G. Chu, J. Sun, and C. Chen, "Thermal [26] R. Kamyar and M. M. Peet, "Optimal thermostat programming for
runaway caused fire and explosion of lithium ion battery," Journal of time-of-use and demand charges with thermal energy storage and
Power Sources, vol. 208, pp. 210-224, 2012. optimal pricing for regulated utilities," IEEE Transactions on Power
[5] R. German, S. Shili, A. Desreveaux, A. Sari, P. Venet, and A. Systems, vol. 32, no. 4, pp. 2714-2723, 2017.
Bouscayrol, "Dynamical coupling of a battery electro-thermal model and [27] H. Pei, X. Hu, Y. Yang, H. Peng, L. Hu, and X. Lin, "Designing
the traction model of an EV for driving range simulation," IEEE multi-mode power split hybrid electric vehicles using the hierarchical
Transactions on Vehicular Technology, pp. 328-337, 2019. topological graph theory," IEEE Transactions on Vehicular Technology,
[6] T. Franke, I. Neumann, F. Buhler, P. Cocron, and J. F. Krems, pp. 1-1, 2020.
"Experiencing range in an electric vehicle: Understanding psychological [28] A. M. Ali, A. Ghanbar, and D. Söffker, "Optimal control of multi-source
barriers," Applied Psychology, vol. 61, no. 3, pp. 368-391, 2012. electric vehicles in real time using advisory dynamic programming,"
[7] Q. Yuan, W. Hao, H. Su, G. Bing, X. Gui, and A. Safikhani, IEEE Transactions on Vehicular Technology, vol. 68, no. 11, pp.
"Investigation on range anxiety and safety buffer of battery electric 10394-10405, 2019.
vehicle drivers," Journal of Advanced Transportation, p. 11, 2018. [29] W. Lee, H. Jeoung, D. Park, and N. Kim, "An adaptive concept of
[8] H. S. Hamut, I. Dincer, and G. F. Naterer, "Analysis and optimization of PMP-based control for saving operating costs of extended-range electric
hybrid electric vehicle thermal management systems," Journal of Power vehicles," IEEE Transactions on Vehicular Technology, vol. 68, no. 12,
Sources, vol. 247, pp. 643-654, 2014. pp. 11505-11512, 2019.
[9] E. Samadani, R. Fraser, and M. Fowler, "Evaluation of air conditioning [30] B. Nguyen, R. German, J. P. F. Trovão, and A. Bouscayrol, "Real-time
impact on the electric vehicle range and li-ion battery life," SAE energy management of battery/supercapacitor electric vehicles based on
Technical Paper, 2014. Available: https://doi.org/10.4271/2014-01-1853 an adaptation of Pontryagin's minimum principle," IEEE Transactions
[10] L. Horrein, A. Bouscayrol, Y. Cheng, C. Dumand, G. Colin, and Y. on Vehicular Technology, vol. 68, no. 1, pp. 203-212, 2019.
Chamaillard, "Influence of the heating system on the fuel consumption

0018-9545 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 27,2020 at 05:57:45 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TVT.2020.2999939, IEEE
Transactions on Vehicular Technology
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 13

[31] S. Park and C. Ahn, "Power management controller for a hybrid electric [53] N. Yang, X. Zhang, G. Li, and D. Hua, "Assessment of the forced
vehicle with predicted future acceleration," IEEE Transactions on air-cooling performance for cylindrical lithium-ion battery packs: A
Vehicular Technology, vol. 68, no. 11, pp. 10477-10488, 2019. comparative analysis between aligned and staggered cell arrangements,"
[32] S. Zhang, Y. Luo, K. Li, and V. Li, "Real-time energy-efficient control Applied Thermal Engineering, vol. 80, pp. 55-65, 2015.
for fully electric vehicles based on an explicit model predictive control [54] S. Halbach, P. Sharer, S. Pagerit, A. P. Rousseau, and C. Folkerts,
method," IEEE Transactions on Vehicular Technology, vol. 67, no. 6, "Model architecture, methods, and interfaces for efficient math-based
pp. 4693-4701, 2018. design and simulation of automotive control systems," SAE Technical
[33] F. Xu and T. Shen, "Look-ahead prediction-based real-time optimal Paper, 2010. Available: https://doi.org/10.4271/2010-01-0241
energy management for connected HEVs," IEEE Transactions on [55] V. H. Johnson, "Battery performance models in advisor," Journal of
Vehicular Technology, vol. 69, no. 3, pp. 2537-2551, 2020. Power Sources, vol. 110, no. 2, pp. 321-329, 2002.
[34] C. Zhu, F. Lu, H. Zhang, J. Sun, and C. C. Mi, "A real-time battery [56] D. Bernardi, E. Pawlikowski, and J. Newman, "A general energy balance
thermal management strategy for connected and automated hybrid for battery systems," Journal of The Electrochemical Society, vol. 132,
electric vehicles (CAHEVs) based on iterative dynamic programming," no. 1, pp. 5-12, 1985.
IEEE Transactions on Vehicular Technology, vol. 67, no. 9, pp. [57] J. M. S. Jabardo, W. G. Mamani, and M. R. Ianella, "Modeling and
8077-8084, 2018. experimental evaluation of an automotive air conditioning system with a
[35] Y. Masoudi, A. Mozaffari, and N. L. Azad, "Battery thermal variable capacity compressor," International Journal of Refrigeration,
management of electric vehicles: An optimal control approach," in vol. 25, no. 8, pp. 1157-1172, 2002.
ASME Dynamic Systems and Control Conference, 2015. [58] Y. Chen, N. P. Halm, J. E. Braun, and E. A. Groll, "Mathematical
[36] J. Lopez-Sanz et al., "Nonlinear model predictive control for thermal modeling of scroll compressors — part ii: Overall scroll compressor
management in plug-in hybrid electric vehicles," IEEE Transactions on modeling," International Journal of Refrigeration, vol. 25, no. 6, pp.
Vehicular Technology, vol. 66, no. 5, pp. 3632-3644, 2017. 751-764, 2002.
[37] M. R. Amini, J. Sun, and I. Kolmanovsky, "Two-layer model predictive [59] T. L. Bergman, F. P. Incropera, D. P. DeWitt, and A. S. Lavine,
battery thermal and energy management optimization for connected and Fundamentals of heat and mass transfer. Wiley, 2011.
automated electric vehicles," in IEEE Conference on Decision and [60] S. D. Cairano, W. Liang, I. V. Kolmanovsky, M. L. Kuang, and A. M.
Control, 2018, pp. 6976-6981. Phillips, "Power smoothing energy management and its application to a
[38] J. Lopez-Sanz et al., "Thermal management in plug-in hybrid electric series hybrid powertrain," IEEE Transactions on Control Systems
vehicles: A real-time nonlinear model predictive control Technology, vol. 21, no. 6, pp. 2091-2103, 2013.
implementation," IEEE Transactions on Vehicular Technology, vol. 66,
no. 9, pp. 7751-7760, 2017.
[39] Y. Masoudi and N. L. Azad, "MPC-based battery thermal management
controller for plug-in hybrid electric vehicles," in American Control
Conference, 2017, pp. 4365-4370.
[40] S. Park and C. Ahn, "Stochastic model predictive controller for battery
thermal management of electric vehicles," in 2019 IEEE Vehicle Power
and Propulsion Conference (VPPC), 2019, pp. 1-5.
[41] S. Bauer, A. Suchaneck, and F. Puente León, "Thermal and energy
battery management optimization in electric vehicles using Pontryagin's
maximum principle," Journal of Power Sources, vol. 246, pp. 808-818, Seho Park received the B.S. and M.S.
2014. degrees from Pusan National University,
[42] Y. Ma, J. Matuško, and F. Borrelli, "Stochastic model predictive control
for building HVAC systems: Complexity and conservatism," IEEE
Busan, South Korea, in 2015 and 2019,
Transactions on Control Systems Technology, vol. 23, no. 1, pp. respectively, in mechanical engineering.
101-116, 2015. He is currently pursuing the Ph.D.
[43] S. D. Cairano, D. Bernardini, A. Bemporad, and I. V. Kolmanovsky, degree in mechanical engineering from
"Stochastic MPC with learning for driver-predictive vehicle control and
its application to HEV energy management," IEEE Transactions on
Pusan National University, Busan, South
Control Systems Technology, vol. 22, no. 3, pp. 1018-1031, 2014. Korea. His research interests include
[44] S. Xie, X. Hu, Z. Xin, and L. Li, "Time-efficient stochastic model automotive system modeling and optimal control, and
predictive energy management for a plug-in hybrid electric bus with an autonomous vehicle control. Recently, he has focused on
adaptive reference state-of-charge advisory," IEEE Transactions on
Vehicular Technology, vol. 67, no. 7, pp. 5671-5682, 2018.
optimal control for thermal management and power
[45] C. Alaoui, "Passive/active BTMS for EV lithium-ion batteries," IEEE management of electrified vehicles.
Transactions on Vehicular Technology, vol. 67, no. 5, pp. 3709-3719,
2018. Changsun Ahn received the B.S. and
[46] R. Sabbah, R. Kizilel, J. R. Selman, and S. Al-Hallaj, "Active M.S. degrees from Seoul National
(air-cooled) vs. Passive (phase change material) thermal management of
high power lithium-ion packs: Limitation of temperature rise and
University, Seoul, South Korea, in 1999
uniformity of temperature distribution," Journal of Power Sources, vol. and 2005, respectively, and the Ph.D.
182, no. 2, pp. 630-638, 2008. degree from the University of Michigan,
[47] A. Pesaran, "Battery thermal management in EVs and HEVs: Issues and Ann Arbor, in 2011, all in mechanical
solutions," in Advanced Automotive Battery Conference, 2001.
[48] D. Chen, J. Jiang, G.-H. Kim, C. Yang, and A. Pesaran, "Comparison of
engineering.
different cooling methods for lithium ion battery cells," Applied Thermal He is currently an Associate Professor at
Engineering, vol. 94, pp. 846-854, 2016. Pusan National University, South Korea.
[49] P. Zhou, "Electric vehicle thermal management system," U.S. Patent His research interests include the fields of automotive
7789176, Sep. 7, 2010.
[50] M. D. Nemesh and W. Ibri, "Battery thermal system for vehicle," U.S.
control/estimation. Recently, he has focused on autonomous
Patent 8215432, Jul. 10, 2012. vehicle control and hybrid vehicle control.
[51] T. Wang, K. J. Tseng, J. Zhao, and Z. Wei, "Thermal investigation of
lithium-ion battery module with different cell arrangement structures and
forced air-cooling strategies," Applied Energy, vol. 134, pp. 229-238,
2014.
[52] H. Sun and R. Dixon, "Development of cooling strategy for an air cooled
lithium-ion battery pack," Journal of Power Sources, vol. 272, pp.
404-414, 2014.

0018-9545 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Exeter. Downloaded on June 27,2020 at 05:57:45 UTC from IEEE Xplore. Restrictions apply.