CIGRE-745 2024 CIGRE Canada Conference & Exhibition

Winnipeg, MB, Oct. 28 – 31, 2024

An Improved Simulation-Based Approach to Compute Losses in Power Electronic

Converters – Application to Battery Energy Storage Converter Systems

D. KURANAGE1, S. FILIZADEH1, I. FERNANDO2

1
University of Manitoba
2
Manitoba Hydro

SUMMARY

Due to escalating integration of renewable resources, power electronic converters that tie such
resources to the grid have become an essential part of modern power systems. Accurate estimation of
these converters’ losses is crucial in the design of thermal management systems, selection of
components, and tuning of control systems. Prior to the actual implementation of a converter,
simulation-based methods are used to estimate the converter’s losses. While highly accurate, device-
level simulation models of semiconductor switches exist, they are neither readily available nor
particularly suited for use in power system simulators due to their heavy computational load.
Electromagnetic transient (EMT) simulators do not use physics-based switch models and are thus unable
to calculate switching losses. To bridge this gap, auxiliary EMT simulation-based loss calculation
methods are needed. This is particularly relevant for converters with a large number of switching devices
that require not only accurate but also fast and computationally efficient loss evaluation methods.
This paper presents an improved simulation-based loss calculation approach for power electronic
converters that is computationally efficient and accurate. This method uses simplified offline
calculations based upon steady state waveforms obtained from EMT simulations. For switching loss
calculations, nonlinear, yet computationally inexpensive, equations are used that produce results with
high accuracy. These features make the proposed loss estimation method particularly suited for
converters that employ many switching devices and operate at high frequencies, e.g., modular multilevel
converters (MMCs).
In this paper, firstly, comparisons of the loss estimations from the proposed switching loss calculation
equations and experimental results from existing literature are used to validate the accuracy of the
proposed method. Then, the method is demonstrated using an example case of an MMC (45 MW) with
embedded battery energy storage system (9.6 MW) for different converter switching frequencies. The
test system is modelled in PSCAD/EMTDC, and the loss calculation algorithm is implemented in
MATLAB. The example case proves that the proposed loss calculation approach is fast, easy to
implement and use, and accurate.

KEYWORDS
Switching losses, modular multilevel converters (MMC), embedded battery energy storage (BESS),
converter losses.

*[email protected]
1. INTRODUCTION

Power electronic (PE) converters play a vital role in the integration of distributed generation resources
in the grid and enhancement of power quality [1], [2]. The main components of a PE converter are the
semiconductor switches [3], which dictate, at least in part, the converter’s switching frequency, power
density, and efficiency [2]. To ensure safe operation of these switching devices, thermal management
systems of the converter should be carefully designed by considering the power losses [4]. Thus, precise
evaluation of converter losses is important not only to achieve high efficiency but also to improve
reliability.

Losses of switching devices include conduction losses, switching losses and off-state blocking losses.
At high switching frequencies, the switching losses become dominant [1]. However, calculation of
switching losses is complicated due to the small turn-on and turn-off transition time intervals of the
switching devices, which are typically in the range of few hundreds of nano seconds [4] and also due to
the irregular switching patterns of the PE converters. In [4], [5] and [6], authors have derived analytical
expressions to calculate switching losses considering the IGBT switching transient waveforms. This
approach is quite accurate, but it uses complex equations and causes a large computational burden. In
[7] it is assumed that the switching losses are linearly proportional to the product of current and voltage
at the switching instance to calculate losses of an MMC-based HVDC system; this is, however, not valid
for every operating point. Authors in [8] have obtained equations to calculate switching losses in an
MMC based on datasheet curves, which are simple and accurate. However, they have calculated the
currents and voltages of the converter with analytical equations and have assumed uniform distribution
of switching events during one cycle, which is not accurate.

Clearly there are gaps that need to be filled to obtain an accurate, widely applicable, and computationally
efficient method for calculation of semiconductor losses. Accordingly, this paper presents a simulation-
based loss calculation approach, which utilizes equations based upon device datasheet curves to
calculate switching losses. The proposed loss calculation method is shown particularly for IGBTs due
to their wide range of application in high-power converters. Although the paper only focuses on
calculating semiconductor losses, the total converter losses include losses from transformer and other
passive elements that are not discussed in this paper.

Simplified switching loss calculation equations and their validation are described in Section 2. Section
3 explains the loss evaluation approach using an example case study and Section 4 concludes the paper.

2. EVALUATION OF LOSSES IN SWITCHING DEVICES

2.1 Losses in Switching Devices

Generally, a common type of switching pair consists of an IGBT with an anti-parallel diode to facilitate
bidirectional current conduction [2]. Conduction, switching, and off-state blocking losses are the main
losses in a switching device. Conduction losses are equal to the product of device current and its forward
saturation voltage, and off-state blocking losses are equal to the product of blocking voltage and leakage
current [4]. Both these losses may be readily and easily calculated. On the contrary, evaluation of
switching losses requires much more complicated calculations due to the short time intervals of
switching and the complicated voltage and current waveforms involved.

2.2 Evaluation of Switching Losses

When it comes to switching losses of an IGBT-diode switching pair, the turn-on (𝐸on ) and turn-off (𝐸off )
energy losses of the IGBT, and the reverse recovery (𝐸rr ) energy losses of the diode must be taken into
account. During turn-on, the diode can be assumed ideal due to its rapid turn-on time; thus, turn-on
energy losses of the diode can be neglected [9].

` 1
Switching losses mainly depend on voltage and current at the switching instant, junction temperature,
and gate circuit resistance (𝑅G ) [1], [8]. Typically, the device datasheets include the above-mentioned
energy loss values for one switching event under certain test conditions. In this paper, equations to
calculate switching losses are derived based upon these datasheet curves as follows:

𝐸on = 𝐸on,𝑇 0 C (𝐼c ) ∙ 𝐾𝑉 (1)

𝐸off = 𝐸off,𝑇0 C (𝐼c ) ∙ 𝐾𝑉 (2)
𝐸rr = 𝐸rr,𝑇 0 C (𝐼c ) ∙ 𝐾𝑉 (3)
𝑉
𝐾𝑉 = 𝑉 ce (4)
test

where 𝐸on,𝑇 0 C (𝐼c ), 𝐸off,𝑇 0 C (𝐼c ) and 𝐸rr,𝑇 0 C (𝐼c ) are the loss values corresponding to the collector current
at the switching event, and are obtained from the device datasheets for IGBT’s turn-on, turn-off, and
diode reverse recovery energy losses, respectively. 𝐾𝑉 is a correction factor for the voltage. 𝑉ce and 𝑉test
are collector-emitter voltages at the switching event and at the test condition in the device datasheet
respectively. Here, a linear relationship is assumed between the switching losses and the voltage. The
junction temperature is assumed to be the maximum value, for a worst-case estimation. Having observed
several datasheet curves for switching losses against 𝑅G values, it is concluded that the effect of 𝑅G on
switching losses is insignificant; hence a correction factor for 𝑅G is not included in the equations.

To further clarify the method of obtaining 𝐸on,𝑇 0 C (𝐼c ), 𝐸off,𝑇 0 C (𝐼c ) and 𝐸rr,𝑇 0 C (𝐼c ) from datasheets,
sample energy loss curves from Infineon IKW40T120 IGBT-diode module datasheet are shown in Fig.
1 (a). Mathematical functions are obtained by curve-fitting the datasheet traces at the selected
temperature as shown in Fig. 1 (b).

(a) (b)
Fig. 1. (a) Infineon IKW40T120 IGBT-diode module datasheet curves for switching energy losses (Junction
temperature = 1500C), (b) mathematical functions obtained by curve-fitting [10].

2.3 Validation of Switching Loss Calculation Equations

To validate the accuracy of the above equations, loss estimations using this method for one switching
event are compared with the experimental results reported in [2]. The IGBT-diode module considered
is Infineon IKW40T120 with 1200 V and 40 A ratings. Fig. 2 illustrates the comparison of experimental
results with the calculated results. Fig. 2(a) shows that the fitted curves predict the measured losses with
adequate accuracy when the voltage is kept constant. The error in loss calculations increases with
increasing collector current. However, the error remains fairly small (10% or lower) for a wide range of
current values. Fig. 2(b) shows the impact of collector-emitter voltage variations on the accuracy of
losses while keeping the collector current constant. While voltage variations do have some impact on
the estimated losses, use of a linearly varying Kv does not seem to have significant impact on the

` 2
accuracy of estimations if the collector-emitter voltage remains close to the voltage at the test condition
and does not change drastically. This is particularly the case in MMC circuits wherein the SM voltage
is closely regulated around its nominal value. A significant advantage of this approach is that it only
requires voltage and current values at the switching event, which can be easily obtained from EMT
waveforms. Thus, the losses can be evaluated in an efficient and reasonably accurate manner.

(a) (b)
Fig. 2. Comparison of experimental and calculated results with (a) varying Ic and constant Vce = 600 V, and
(b) varying Vce and constant Ic = 40A.

3. CASE STUDY

To demonstrate the application of the proposed approach, semiconductor losses in an MMC with
embedded battery energy storage system (BESS) are calculated. Compared to a regular MMC,
evaluation of losses in an MMC with embedded BESS is significantly more complex, due to the
additional switches of the DC-DC converters that interface batteries with the submodule capacitor. The
problems are also exacerbated by the high switching frequency of the DC-DC converter. Thus, this
example case is selected to further explain the simplicity and efficiency of the proposed approach,
despite the complexity of the system of interest.

3.1 System Configuration

In this study, an MMC with embedded BESS is considered. The conventional 3-phase MMC consists
of three phase-legs, each with two stacks of submodules and two arm inductors as shown in Fig. 3(a).
The stack consists of full-bridge or half-bridge submodules. The MMC with embedded BESS has the
same converter structure as a regular MMC except for the arrangement of the submodule. Fig. 3(b) and
(c) illustrate the submodule in a regular MMC and in an MMC with embedded BESS, respectively. The
schematic diagram of the system under consideration is shown in Fig. 4. Every stack of the MMC in
this case study includes identical submodules as shown in Fig. 3(c). Detailed specifications of the system
are displayed in Table 1.

The nearest level control technique, which is a typical control technique in MMC applications, is used
to determine the number of submodules that need to be inserted from the upper and lower arms of the
MMC. The submodules to be inserted are decided by evaluating the sorted capacitor voltages and the
direction of the arm current. If the arm current is in the charging direction of the capacitor, sub-modules
with the lowest voltages are inserted or vice versa. Decoupled controller [11] is used to control the rms
voltage and the active power injection at the point of common coupling (PCC). To control the current
through the batteries, peak current control method [12] is used. The system including its controls is
simulated in PSCAD/EMTDC.

` 3
3.2 Simulation-Based Loss Calculation Approach

According to the loss calculation equations discussed in Section 2, current, voltage and on/off states of
the switching devices are required to evaluate the losses. In this paper, these details are obtained from
the steady state EMT waveforms. The losses are estimated after the EMT simulation run as an offline
loss calculation method. This makes this approach more convenient and computationally efficient than
other methods that rely on on-the-fly calculation of losses. To calculate the switching and conduction
losses, an algorithm is implemented in MATLAB considering the current flow and switching operations
in submodules. The current conduction paths for different switch operations in half-bridge and DC-DC
converter are shown in Fig. 5(a) and (b), respectively.

(b)

(a) (c)
Fig. 3. (a) Schematic diagram of a 3-phase MMC, (b) half-bridge submodule in a regular MMC, (c)
submodule for MMC with embedded BESS.

Fig. 4. Schematic diagram of the system for the case study.

Table 1. Specifications for the system in the case study.

Parameter Value Parameter Value

DC voltage (Vdc) 45 kV No of SMs in a multivalve 20
AC grid voltage (Vsys) 20.9 kV Arm inductance 0.0045 H
SCR of AC grid 3∡ 800 SM capacitor voltage 2.25 kV
Ncon: Nsys 26.18:20.9 kV Submodule capacitor 8000 μF
Transformer rating 54 MVA Battery voltage 0.8 kV
Transformer leakage reactance 6% DC-DC converter inductance 0.1 H
Network side resistance (R) 0.561 Ω Network side inductance (L) 0.00845 H

` 4
In the switching waveform data, a change in value from 0 to 1 or vice versa in two consecutive time
steps indicates that there is a switching action between the time steps. By observing the direction of
current at this switching instance, turning on and turning off IGBT-diode can be identified. According
to this, the loss calculation program adds 𝐸on , 𝐸off or 𝐸rr . Fig. 6 illustrates the switching loss calculation
algorithm for the half-bridge submodule. The same procedure can be applied for the DC-DC converter.
The positive direction of the currents and other parameter notations are as shown in Fig. 3(c). To obtain
the power losses, energy losses are added for one second. Thus, 𝑖arm , 𝑣cap, and 𝑆1 in Fig. 6 are the arrays
of arm current, capacitor voltage and switching signal values for one second, respectively, obtained from
the steady state EMT waveforms. 𝑎 is the iteration and 𝑎end is the end iteration, which is equal to 1
divided by the simulation time step. When there is no switching action, conduction losses are calculated.
For that, the device current is multiplied by the forward voltage of IGBT or diode accordingly.

In this paper, the off-state blocking losses are assumed to be negligible considering almost zero leakage
current through the turned-off switching devices. The losses of all the IGBT-diode modules in a phase
leg are calculated by the algorithm over one second period and multiplied by 3 to obtain the total
semiconductor losses of the converter assuming an equal loss distribution among the three phases during
steady state. Rather than considering the losses in one submodule and multiplying it by the total number
of submodules in the converter, this method gives more accurate results.

(a) (b)
Fig. 5. Current conduction paths for different switching operations for (a) half-bridge, and (b) DC-DC
converter.

3.3 Analysis of Calculated Losses

The IGBT-diode module used for the study is ST1500GXH24 with 4.5 kV and 1.5 kA ratings. The
required datasheet parameters and derived equations for datasheet curves are shown in Table 2. The
losses are estimated for various operating points considering different DC-DC converter frequencies and
the effects of MMC’s circulating current suppression controller (CCSC). The details of the different
operating points are displayed in Table 3. 𝑉PCC,ref , 𝑃PCC,ref and 𝐼bat,ref are the reference values used in
the controllers for the rms voltage at PCC, active power injection at PCC, and battery current,
respectively.

Fig. 7 displays the actual switching and conduction loss values for the half-bridges and DC-DC
converters separately when DC-DC converter frequency is 1000 Hz and CCSC is not in operation to
demonstrate the ranges of the actual losses. It can be observed that when the real power output (𝑃PCC,ref)
is changed from 0.9 pu to 0.8 pu, both switching and conduction losses in the half-bridges reduce due
to the reduction of the arm current.

` 5
 Table 2. Parameters and equations from device datasheet required for loss estimation.

Datasheet parameters Equations fitted to the datasheet curves

IGBT saturation voltage (𝑉ce,sat) 3V 𝐸on,1250 C (𝐼c ) 0.0051𝐼c + 0.4885
Diode forward voltage (𝑉F) 3.2 V 𝐸off,1250 C (𝐼c ) 6 × 10−7 𝐼c2 + 0.0034𝐼c + 0.5916
Voltage at test condition (𝑉test ) 2.7 kV 𝐸rr,1250 C (𝐼c ) −1 × 10−7 𝐼c2 + 0.0021𝐼c + 0.5315

Start

Input iarm, vcap, S1,

aend, Vtest

a = 0, Eoff = 0,
Eon = 0, Err = 0
a = a +1
No If
(a < aend)
If
(S1(a) =1 No
Yes Diode or IGBT conducts.
&
If Calculate conduction losses.
No S1(a+1)=0)
(S1(a) =0
& D2 turns on. T1 turns off.
S1(a+1)=1) Yes C= Eoff,TºC(iarm(a )) × (vcap(a+1)/Vtest)
No Eoff = Eoff + C
If
(iarm(a)

Yes T2 turns on. D1 turns off.

Yes A= Eon,TºC(iarm(a+1 )) × (vcap(a)/Vtest)
Eon = Eon + A
If
No B= Err,TºC(iarm(a )) × (vcap(a+1)/Vtest)
(iarm(a)
Err = Err + B

Yes T1 turns on. D2 turns off.

A= Eon,TºC(iarm(a+1 )) × (vcap(a)/Vtest)
D1 turns on. T2 turns off. Eon = Eon + A
C = Eoff,TºC(iarm(a )) × (vcap(a+1)/Vtest) B= Err,TºC(iarm(a )) × (vcap(a+1)/Vtest)
Eoff = Eoff + C Err = Err + B

End

Fig. 6. Switching loss calculation algorithm for the half-bridge submodule.

Table 3. Details of different operating points in the study.

Operating 𝑉PCC,ref 𝑃PCC,ref 𝐼bat,ref Operating 𝑉PCC,ref 𝑃PCC,ref 𝐼bat,ref

point (pu) (pu) (A) point (pu) (pu) (A)
1 1 0.9 0 4 1 0.8 0
2 1 0.9 100 5 1 0.8 100
3 1 0.9 -100 6 1 0.8 -100

` 6
The loss analysis for different DC-DC converter frequencies is shown in Fig. 8. The loss percentage is
calculated by taking the percentage of total losses over 𝑃PCC,ref (in MWs) at the considered operating
point. As seen, the frequency of the DC-DC converter does not affect the losses in the half-bridges;
however, it affects the losses in the DC-DC converters. When it comes to operating points 1, 2, and 3,
the only difference is the battery current. When it is in the positive direction, battery is supplying part
of the required power for the AC side. Thus, the power coming from the DC side is reduced. When it is
in the negative direction, battery is charging so that the power for both the AC side and the batteries
comes from the DC side. This is reflected in the arm currents for these three operating points as shown
in Fig. 9. Due to this difference in arm currents, losses in half-bridges for the operating points 1, 2, and
3 vary as in Fig. 8(a). The same reasoning explains the loss differences in operating points 4, 5, and 6
in Fig. 8(a). With regard to DC-DC converter losses at a certain frequency, operating points 2 and 3
have almost the same losses since the same amount of current flows through the switching devices and
operating point 1 has lower losses than those two because of zero battery current. The same applies to
the loss variations in operating points 4, 5, and 6 in Fig. 8(b).

(a) (b)
Fig. 7. Actual switching and conduction losses when DC-DC converter frequency = 1000 Hz and CCSC is
not in action, (a) losses in the half-bridges, (b) losses in the DC-DC converters.

(a) (b) (c)

Fig. 8. Loss analysis for different DC-DC converter frequencies (CCSC is not in operation) (a) Loss
percentages of the half-bridges, (b) Loss percentages of DC-DC converters, (c) Total loss percentages.

Fig. 9. Arm currents for 1 (Ibat = 0 A), 2 (Ibat = 100 A) and 3 (Ibat = -100 A) operating points when DC-DC
converter frequency = 2000 Hz and CCSC is not in operation.

` 7
The effect of CCSC operation on losses is illustrated in Fig. 10. As depicted, the losses in the DC-DC
converters are not affected by the CCSC operation. However, losses in the half-bridges are lower when
CCSC is in action, since it reduces the arm current by eliminating the second harmonic component from
it. The overall loss comparison demonstrates that the losses in the MMC with embedded BESS alter
approximately between 1.3-2.5 % depending on the operating point, DC-DC converter frequency, and
the CCSC operation. In comparison to the losses in DC-DC converters, losses in half-bridges are much
higher due to the high arm current, which has a peak value of about 10 times that of the battery current.

(a) (b) (c)

Fig. 10. Loss analysis with and without CCSC operation (DC-DC converter frequency = 1000 Hz), (a) Loss
percentages of the half-bridges, (b) Loss percentages of DC-DC converters, (c) Total loss percentages.

4. CONCLUSIONS

The paper presented an improved simulation-based loss calculation approach to calculate semiconductor
losses in PE converters. Simplified yet accurate non-linear equations were presented and validated for
switching loss calculations. An offline switching and conduction loss calculation algorithm was
implemented, which utilized steady state EMT waveforms to calculate semiconductor losses in an MMC
with embedded BESS. Loss analysis was done for various operating points considering different DC-
DC converter frequencies and CCSC operation. Total converter losses increase by approximately 0.25-
0.5% depending on the operating point, when the DC-DC converter frequency is increased from 1000
Hz to 2000 Hz. When the CCSC is in operation, the total loss percentages decrease by around 0.3%,
since it eliminates the second order component from the arm current. Overall converter losses vary
between about 1.3-2.5% depending on the operating conditions. The proposed approach in the paper is
straightforward, efficient, and accurate and can be effectively used to estimate semiconductor losses in
PE converters that employ a large number of switches.

BIBLIOGRAPHY

[1] S. Abbas, A. Salemnia, M. Hamzeh, “Applicable Method for Average Switching Loss Calculation
in Power Electronic Converters,” Journal of Power Electronics, vol. 17, no. 4, pp. 1097- 1108,
2017.
[2] Y. Xu, C. Ngai, A. Ghosh, D. Muthumuni, “An Electrical Transient Model of IGBT-Diode
Switching Cell for Power Semiconductor Loss Estimation in Electromagnetic Transient
Simulation,” IEEE Trans. on Power Electronics, vol. 35, no. 3, pp. 2979-2989, 2020.
[3] R. Ramshaw, Power Electronics Semiconductor Switches, Boston: Springer, 1993.
[4] A. Rajapakse, A.M. Gole, R.P. Jayasinghe, “An Improved Representation of FACTS Controller
Semiconductor Losses in EMTP-Type Programs Using Accurate Loss-power Injection into
Network Solutions,” IEEE Transactions on Power Delivery, vol. 24, pp. 381-389, 2009.
[5] A. Rajapakse, A. M. Gole, P. L. Wilson, “Approximate Loss Formulae for Estimation of IGBT
Switching Losses Through EMTP-Type Simulations,” in Proc. International Conference of Power
Systems Transients, Montreal, Canada, 2005.

` 8
[6] A. D. Rajapakse, A. M. Gole, P. L. Wilson, “Electromagnetic Transient Simulation Models for
Accurate Representation of Switching Losses and Thermal Performance in Power Electronic
Systems,” IEEE Transactions on Power Delivery, vol. 20, pp. 319-327, 2005.
[7] J. Li, X. Zhao, Q. Song, H. Rao, S. Xu, M. Chen, “Loss Calculation Method and Loss
Characteristic Analysis of MMC Based VSC-HVDC System,” in Proc. IEEE International
Symposium on Industrial Electronics, Taipei, Taiwan, 2013.
[8] S. Rodrigues, A. Papadopoulos, E. Kontos, T. Todorcevic, P. Bauer, “Steady-State Loss Model of
Half-Bridge Modular Multilevel Converters,” IEEE Transactions on Industrial Applications, vol.
52, no. 3, pp. 2415-2425, 2016.
[9] N. Mohan, T.M. Undeland, W. P. Robbins, Power Electronics- Converters, Applications and
Design (3rd Edition), John Wiley & Sons, 2003.
[10] Infineon Technologies, “TrenchStop Series,” IKW40T120 datasheet, Mar.2013.
[11] N. M. Herath, Operation, Control, and Simulation of Modular Multilevel Converters with
Embedded Energy Storage, M. Sc. Thesis, 2019.
[12] R. W. Erickson, D. Maksimovic, Fundamentals of Power Electronics, Boulder, USA: Springer,
1997.

` 9