850 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 39, NO.

1, JANUARY 2024

A Comprehensive Review of Solving Selective

Harmonic Elimination Problem With
Algebraic Algorithms
Chenxu Wang , Qi Zhang , Student Member, IEEE, Wensheng Yu , and Kehu Yang , Member, IEEE

Abstract—Selective harmonic elimination pulsewidth modula- in our daily lives [1], [2], [3], [4], [5], [6], [7], [8], [9], [10], [11],
tion (SHEPWM) is an effective way to eliminate low-order har- [12], [13], [14]. However, high-power converters bring great
monics in high-power applications. However, one of the biggest challenges to energy conversion technology, especially to the
challenges of SHEPWM is to solve the selective harmonic elimi-
nation (SHE) equations, which are composed of some nonlinear modulation technology. High-power converters commonly op-
transcendental equations. Over the past few decades, algebraic erate at high voltage and current, leading to increased switching
algorithms have shown a considerable ability to solve SHE equa- losses in power semiconductor devices [15], [16], [17], [18],
tions, specifically for obtaining all exact solutions. Much research [19]. Furthermore, the high voltage and current also lead to
has been published about algebraic algorithms, struggling to solve
challenges related to electromagnetic interference (EMI), par-
more switching angles, solving different mathematic models of
SHEPWM, and so on. This article comprehensively reviews existing asitic capacitances, and voltage spikes. Therefore, high-power
algebraic algorithms, including elementary symmetric polynomi- converters have to utilize low switching frequency to reduce
als, power sums, Newton’s identities, resultant elimination method, power losses and to avoid EMI issues. However, the low switch-
Wu’s method, Gröbner-basis-based method, Chudnovsky algo- ing frequency can exacerbate harmonic distortion in the output
rithm, polynomial homotopy continuation algorithm, and real-time
waveform. The harmonics, especially the low-order harmonics,
implementation by algebraic algorithms. The principle operation
of these methods is summarized, and their performance is analyzed are harmful to the power system elements, and minimizing
in terms of execution time, solving ability, and applicability for harmonics for converters is crucial to comply with grid standards
different mathematical models. and prevent damage to sensitive connected devices [20], [21],
Index Terms—Algebraic algorithms, dc–ac conversion, high- [22], [23], [24], [25], [26], [27]. Thus, the challenge of dealing
power applications, inverters, renewable energy system, selective with low switching frequency and low-order harmonics is a
harmonic elimination (SHE). significant concern, and engineers must balance these seemingly
contradictory aspects when formulating high-power converter
I. INTRODUCTION designs.
ECENTLY, because of the rapid development of renew- Among the commonly used pulsewidth modulation (PWM)
R able energy generation and high-power motor-driven ap-
plications, high-power converters have played a significant role
technologies, the selective harmonic elimination (SHE) PWM
stands out as a remarkably effective method for mitigating low-
order harmonics, particularly when operating at exceedingly low
Manuscript received 6 June 2023; revised 29 August 2023; accepted 18 switching frequencies [28], [29], [30], [31], [32], [33], [34].
October 2023. Date of publication 24 October 2023; date of current version
6 December 2023. This work was supported in part by the National Natural Based on the principles of Fourier series expansion, selective
Science Foundation of China under Grants 61973307, 61936008, and 52121003, harmonic elimination pulsewidth modulation (SHEPWM) em-
in part by the 111 Project under Grant B21014, and in part by the Ordos Science ploys a mathematical model to represent the harmonic con-
and Technology Plan. Recommended for publication by Associate Editor Dorin
O Neacsu. (Corresponding author: Kehu Yang.) stituents of a PWM waveform through a set of nonlinear and
Chenxu Wang is with the School of Mechanical and Electrical Engineering, transcendental equations, known as SHE equations [35], [36].
China University of Mining and Technology (Beijing), Beijing 100083, China These equations dictate the switching angles necessary for gen-
(e-mail: [email protected]).
Qi Zhang is with the AAU Energy, Aalborg University, DK-9220 Aalborg, erating PWM drive signals. It is valuable to note that solving
Denmark (e-mail: [email protected]). SHE equations poses significant challenges due to their inherent
Wensheng Yu is with the School of Electronics Engineering, Beijing Uni- nonlinear and transcendental characteristics. Moreover, most
versity of Posts and Telecommunications, Beijing 100876, China (e-mail:
[email protected]). of the SHE equations yield multiple solutions that are often
Kehu Yang is with the School of Artificial Intelligence, China University of discontinuous [37], [38]. As a result, obtaining all solutions
Mining and Technology (Beijing), Beijing 100083, China, and also with the Key for SHE equations stands as a substantial hurdle within the
Laboratory of Intelligent Mining and Robotics, Ministry of Emergency Man-
agement of the People’s Republic of China, Inner Mongolia Research Institute SHEPWM strategy. Addressing this challenge ranks among the
of China University of Mining and Technology (Beijing), Ordos 017004, China foremost concerns associated with the practical implementation
(e-mail: [email protected]). of SHEPWM.
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/TPEL.2023.3327280. In recent years, many algorithms have been proposed
Digital Object Identifier 10.1109/TPEL.2023.3327280 to solve the SHE equations, which can be classified as

0885-8993 © 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: SHIV NADAR UNIVERSITY. Downloaded on March 26,2024 at 19:09:36 UTC from IEEE Xplore. Restrictions apply.
WANG et al.: COMPREHENSIVE REVIEW OF SOLVING SELECTIVE HARMONIC ELIMINATION PROBLEM WITH ALGEBRAIC ALGORITHMS 851

numerical, bioinspired intelligent, and algebraic algorithms.

Numerical algorithms can quickly obtain solutions because of
the fast convergence speed, but they highly rely on the guess
of initial values [39], [40], [41], [42], [43], [44], and it cannot
solve all solutions of SHE equations. Bioinspired intelligent al-
gorithms, such as genetic algorithm, particle swarm optimization
algorithm, and differential evolution algorithm, have been well
developed in recent years [45], [46], [47], [48], [49], [50], [51],
[52], [53], [54], [55], [56], [57]. Although bioinspired intelligent
algorithms can randomly choose initial values, but they lack
the support of the mathematical theory, and they are sensitive
to input parameters, so they are easy to fall into an optimal
local solution and cannot deal with the multisolution feature of
SHE equations. Furthermore, neither numerical algorithms nor
bioinspired intelligent algorithms cannot give a clear conclusion
on whether SHE equations have solutions or not under a certain
modulation index. When these algorithms fail to provide a final
result, the reason for the failure cannot be given by the algorithm
itself, maybe it is caused by the selection of initial values, or
maybe the parameters of the algorithm are unsuitable, or maybe
there are indeed no solutions for the SHE equations.
The algebraic method has developed rapidly in recent years
in solving SHE equations because of its outstanding advantages.
1) They can directly solve SHE equations without any initial
value.
2) All solutions under full modulation index can be obtained.
3) The solutions solved by algebraic algorithms are precise.
4) They can give explicit conclusion about whether the SHE
equations have a solution or not. Fig. 1. Classification of SHEPWM solving methods.
For example, the resultant elimination method [58], the Wu
method [59], the Gröbner-basis-theory-based method [37], and
the polynomial homotopy continuation method [60] are reported The rest of this article is organized as follows. Section II
to be used for solving SHE equations. These methods have summarizes and overviews the existing solving method of SHE
verified the powerful ability and outstanding advantages of equations. Then, in Section III, the main steps to solve SHE
algebraic methods in solving SHE equations. Besides, with the equations using algebraic algorithm are concluded. Section IV
development of computer algebra technology in recent years, focuses on the simplification of SHE equations with the alge-
more and more algebraic methods may be used to solve SHE braic algorithm, and Section V summarizes the principle and
equations. Moreover, algebraic methods also show powerful solving process of the commonly used algebraic algorithms. In
capabilities in the simplification of SHE equations, such as the Section VI, the real-time implementation methods of algebraic
elementary symmetric polynomial method [61] and Newton’s algorithms are summarized. In Section VII, the performance and
identities method [62]. With those methods, not only the solving comparisons of these methods are presented. In Section VIII, the
efficiency can be improved but also the maximum solvable experiential results of the SHEPWM based on algebraic methods
number of switching angles can be increased. Furthermore, it are given. Finally, Section IX concludes this article.
has been reported that some algebraic methods [63] can be
parallelized to reduce the time consumption. It is also valuable to II. OVERVIEW OF SHEPWM SOLVING METHODS
be pointed out that algebraic algorithms can be used for real-time To adopt the SHEPWM strategy for converters, SHE equa-
implementation of SHE modulation [64]. tions must be solved to generate a PWM drive signal. The
Consequently, it can be said that the algebraic-algorithm- reported solving methods in the literature can be divided into
based method has gradually formed a new research area for three families, namely, the numerical methods, the bioinspired
solving SHE equations. To give researchers and engineers a intelligent methods, and the algebraic methods, as shown in
clear and better understanding on the algebraic algorithms, this Fig. 1. In this section, those methods are summarized and com-
article provides a comprehensive review of algebraic algorithms, pared in a high viewpoint, with the advantages and limitations.
including the limitations and preconditions of these algorithms
for different SHE models and the algorithm’s complexity and
efficiency. This single article will be helpful to researchers A. Numerical Methods
and engineers who are working to improve the efficiency and As one of the most traditional approaches, numerical meth-
performance of inverters. ods employ iterative algorithms to compute solutions for SHE

Authorized licensed use limited to: SHIV NADAR UNIVERSITY. Downloaded on March 26,2024 at 19:09:36 UTC from IEEE Xplore. Restrictions apply.
852 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 39, NO. 1, JANUARY 2024

equations. Numerical methods are favored due to their ability to during a single iteration or over the entire computational cy-
quickly provide precise and accurate solutions for SHEPWM. cle. Algebraic algorithms first transform SHE equations into
However, their effectiveness heavily depends on the selection a high-order polynomial system. Afterward, some symbolic
of initial values. For many SHE equations, there is no universal computation theories can be applied to solve SHE equations.
method to choose these initial values, especially for multilevel Using symbols to represent variables is outstanding because of
converters. Typically, initial values required for numerical meth- its unrivaled accuracy, including the precision of every solu-
ods are determined through trial and error, often relying on tion and the capability to obtain all solutions of a nonlinear
experiential knowledge. Several predictive methods [40], [65] equation. Compared with numerical and bioinspired intelligence
have been proposed to estimate initial values for numerical algorithms, algebraic algorithms have a strong ability to provide
methods, but their capabilities are limited, and they can increase all feasible solutions of SHE equations with only one solving
the complexity of the solution process. Moreover, numerical procedure, without requiring any initial values. Obtaining all
methods tend to converge to local minimum solutions, thus solutions of SHE equations is crucial for achieving performance
often providing only one feasible solution for SHE equations. optimization in SHEPWM techniques, especially in high-power
Obtaining all possible solutions of SHE equations through nu- applications such as high-power motor drivers and high-power
merical methods is impossible. Recent literature has introduced grid-connected inverters. Currently, the most commonly used
improved numerical methods such as Walsh functions [39], algebraic algorithms for solving SHE equations include resultant
[66], [67], [68], homotopy algorithms [60], [69], [70], [71], and theory, the Wu method, and Gröbner basis theory [37]. It is worth
gradient optimization [72]. While these methods have enhanced noting that power sum theory and symmetric polynomial theory
performance to some extent, the challenge of accurately guess- are reported to simplify SHE equations, reducing their complex-
ing initial values remains a significant obstacle. ity. These simplification methods are decoupled from the main
solving process, meaning that they can also be applied to nu-
merical and bioinspired intelligence algorithms as preprocessing
B. Bioinspired Intelligent Algorithms steps to reduce computational complexity. Algebraic algorithms
Bioinspired intelligent algorithms [48], [73], [74], [75], [76], have also been reported for use in real-time implementations of
[77], [78], [79], [80] are a category of algorithms that draw SHEPWM. In summary, algebraic algorithms provide a pow-
inspiration from natural laws and phenomena, such as species erful tool for solving SHE equations. However, computational
migration, natural selection, ant colonies, human culture, honey- complexity and computational burden remain significant chal-
bee colonies, swarms of birds, and schools of fish. Compared to lenges. The following sections will provide detailed information
numerical methods, bioinspired intelligent algorithms stand out on the principles, implementation, advantages, and limitations
due to their minimal requirements for initial values. The initial of algebraic algorithms.
values for these algorithms have only a modest impact on the
search for optimized solutions, allowing them to be randomly se- III. MAIN STEPS TO SOLVE THE SHE PROBLEM WITH
lected or guessed. In contrast to initial values, objective functions ALGEBRAIC ALGORITHMS
play a pivotal role in influencing the performance of bioinspired
intelligent algorithms. These objective functions are typically A. Mathematic Model of SHE
designed by researchers and engineers based on their own expe- The principle of SHEPWM strategy is the Fourier series
riences, and as a result, the performance of these algorithms decomposition of the period PWM voltage waveform generated
can be influenced by the designer’s expertise. Furthermore, by a power electronics converter, as given by
bioinspired intelligent algorithms introduce many parameters,
∞

such as population size, activation functions, weighting factors,
allocation factors, etc. Achieving high performance with these V (t) = a0 + (ak cos(kθ) + bk sin(kθ)) . (1)
k=1
algorithms requires careful parameterization. However, param-
eterization is not a straightforward task because: 1) there is According to the different dc voltages of converters, the
no universally accepted method to determine these parameters symmetry feature of the output waveform, and the number of
and 2) there are numerous parameters, and they are interre- output levels, several mathematic models of SHE can be derived
lated. Thus, coordinating these parameters becomes essential from (1) [85], [86], [87]. First, the SHE model can be divided
for optimal performance. In summary, bioinspired intelligent into equal- and unequal-level models [88], [89], [90], [91] ac-
algorithms offer a unique approach to problem solving, char- cording to the dc voltage of converters. Second, the symmetry
acterized by their flexibility in handling initial values and their feature of the output waveform can determine quarter-symmetry,
dependence on well-designed objective functions and carefully half-symmetry [92], [93], [94], and nonsymmetry SHE models.
tuned parameters for achieving peak performance. Third, based on the level of the output waveform, the SHE model
can be classified as two-level [95], [96], [97] and multilevel [98],
[99], [100], [101], [102], [103], [104]. The commonly used
C. Algebraic Algorithms mathematic models and the corresponding output waveform
Algebraic algorithms [35], [38], [81], [82], [83], [84] are have been summarized in Fig. 2. In these mathematic models, ak
based on symbolic computation, in which all variables are represents the amplitude of the sine components of harmonic,
represented with their symbols, not specific values, either and bk represents the amplitude of the cosine components of

Authorized licensed use limited to: SHIV NADAR UNIVERSITY. Downloaded on March 26,2024 at 19:09:36 UTC from IEEE Xplore. Restrictions apply.
WANG et al.: COMPREHENSIVE REVIEW OF SOLVING SELECTIVE HARMONIC ELIMINATION PROBLEM WITH ALGEBRAIC ALGORITHMS 853

Fig. 2. Commonly used mathematic models of SHE.

harmonic. For the quarter-symmetry SHE models, the dc com-

ponent, even harmonics, and sine components are automatically
eliminated due to the symmetry, so only the cosine term needs
to be controlled. For the half-symmetry models, the dc com-
ponent and even harmonics are eliminated, so both the sine
and cosine terms need to be controlled in the half-symmetry
models. Besides, Vdc represents the value of dc power supply.
In some actual industrial applications, the dc voltage of power
supply is different and even time varying, so the unequal-level
SHE models need to be used. θ is the switching angle, which is
determined by two factors: one is the state of the edge, that is,
rising or falling, and the other is the value of θ. These two factors
of θ determine different SHE waveforms, commonly referred to
as the switching pattern. Since only one switching pattern is
possible for a two-level waveform, only one solution exists for
Fig. 3. Procedure of solving SHE problems by algebraic algorithms.
the two-level SHE models, but usually multiple solutions exist
for the multilevel SHE models [102], [103], [104]. No matter
which SHE models are listed in Fig. 2, all switching patterns 1) Step 1—The Algebraization of the Cosine Form SHE
can be obtained within one solving process by using algebraic Equations: Taking the mathematic model of three-phase
algorithms. quarter-wave (QW)-symmetry multilevel SHE as example, the
SHE equations can be expanded as
⎧
B. Procedure of Solving SHE Equations With Algebraic
⎪cos(θ1 ) + cos(θ2 ) + cos(θ3 ) + · · · + cos(θn ) = m
⎪
⎪
Algorithms ⎪
⎪
⎨cos(5θ1 ) + cos(5θ2 ) + cos(5θ3 ) + · · · + cos(5θn ) = 0
⎪
In this section, the relatively unified solving process for SHE cos(7θ1 ) + cos(7θ2 ) + cos(7θ3 ) + · · · + cos(7θn ) = 0
equations is extracted from the literature on many algebraic ⎪
⎪ ..
⎪
⎪
algorithms so that readers can have an overall grasp of the ⎪
⎪ .
⎩
solving process and make it easier to evaluate the algorithm’s cos(kθ1 ) + cos(kθ2 ) + cos(kθ3 ) + · · · + cos(kθn ) = 0
performance later. Generally, the standard procedure for solving (2)
SHE problems with algebraic algorithms contains three main where k is the order of the eliminated harmonic; m is the
steps, as shown in Fig. 3. modulation index, which defines the ratio of the fundamental

Authorized licensed use limited to: SHIV NADAR UNIVERSITY. Downloaded on March 26,2024 at 19:09:36 UTC from IEEE Xplore. Restrictions apply.
854 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 39, NO. 1, JANUARY 2024

amplitude b1 and the dc-link voltage Vdc equation plus a set of univariate linear equations
⎧
πb1 ⎪
⎪ g1 (en ) = am em n + am−1 en
m−1
+ · · · + a 1 en = 0
m= . (3) ⎪
⎪
4Vdc ⎪ g (e
⎨ 2 n−1 n , e ) = b e
1 n−1 + f 1 n) = 0
(e
g3 (en−2 , en ) = b2 en−2 + f2 (en ) = 0 (8)
It can be seen that the SHE equations are all cosine functions in ⎪
⎪..
⎪
switching angles, which cannot be directly solved by algebraic ⎪
⎪ .
⎩
algorithms. As the multiple angle cosine and the Chebyshev gn−1 (e2 , en ) = bn−2 e2 + fn−2 (en ) = 0
polynomial [105] have the following relationship, (2) can be where if the modulation index is preset, am , . . . , a1 and
converted to an algebraic polynomial system: b1 , . . . , bn−2 are integers, and f1 , f2 , . . . , fn−2 are univariate
cos(kθi ) = Tk (cos(θi )). (4) polynomials in en .
Once the SHE equations are converted to the form of (8), the
Let xi = cos(θi ); then, all the cosine terms in (2) can be subsequent solving procedure is relatively straightforward. One
converted to polynomials by using the following iterations: thing that should be noted is that the exact form of (8) is different
⎧
⎪ 2 according to the algebraic algorithm used. For example, if the
⎨T1 (x) = x, T2 (x) = 2x − 1
⎪
resultant elimination method is used, (8) is the form of triangular
..
. . (5) algebraic polynomial.
⎪
⎪
⎩T (x) = 2xT (x) − T (x)
k k−1 k−2
IV. SIMPLIFICATION METHODS
Then, the original cosine-form SHE equations are converted
to the following polynomial equations: The polynomial system (6) is composed of symmetric poly-
⎧  nomials, which means that when any two variables within a
⎪
⎪ f1 (x) = ni=1 x
i − m = 0 polynomial are exchanged, the polynomial remains unaltered.
⎪
⎨f5 (x) = ni=1 5xi − 20x3i + 16x5i = 0
  Based on this feature, several symmetric polynomial theories
⎪ f7 (x) = ni=1 −7xi + 56x3i − 112x5i + 64x7i = 0 . can be employed to simplify (6) into a reduced polynomial
⎪
⎪..
⎩ system, thereby improving the efficiency and ability of alge-
.
(6) braic algorithms. This section summarizes the principles and
It can be seen that the algebraic-form SHE equations (6) are properties of the existing simplification methods, including the
with multivariables and a high degree of these variables. The elementary-symmetric-polynomial-based-method [83], power
algebraic algorithms can directly solve (6), but the computa- sums [106], and Newton’s identities-based method [62].
tional complexity is exceptionally high. Generally, the degree of
(6) can be reduced before using the algebraic algorithm. More A. Elementary Symmetric Polynomials
than one simplification method has been proposed in recent The definition of the elementary symmetric polynomials is
years [62], [83], [106]. The elementary symmetric polynomial given in (10), which describes the relation between the original
is the most commonly used in the existent simplification meth- variables x and the elementary symmetric polynomials e. The
ods [81], [82], [83], so take it as an example here. principle of elementary symmetric polynomials indicates that
2) Step 2—The Simplification of the Algebraic Polynomial any symmetric polynomial can be represented as the form of
System: Based on the symmetry feature of (6), its degrees can elementary symmetric polynomials. Due to the symmetrical
be reduced by substituting the variables xi with the elementary property of the algebraic polynomial system (6), it can be
symmetric polynomials ei . Then, the reduced polynomial sys- represented as the form of elementary symmetric polynomials
tem can be obtained as follows: with e as shown in (7), so that (7) is relatively simple to be
⎧
⎪
⎪ f1 (e) = e1 − m = 0 solved due to the reduced degree. After getting the results of
⎪
⎨f5 (e) = 5e1 − 20e31 + 16e51 + · · · − 80e2 e3 = 0 e, the original variables x can be solved by constructing a
⎪ f7 (e) = −7e1 + 56e31 − 112e51 + · · · − 448e2 e5 = 0 (7) univariate high-order polynomial (9). It is not hard to see that
⎪
⎪..
⎩ the elementary symmetric polynomials e are the coefficients of
. a univariate high-order polynomial with roots x1 , x2 , . . . , xn
where e1 –en have an equivalent relation with x1 –xn , called f (x) = xn − e1 xn−1 + · · · + (−1)n−1 en−1 x1 + (−1)n en
the elementary symmetric polynomials, and it will be detailed
described in next section. = (x − x1 )(x − x2 ) · · · (x − xn ) (9)
⎧
As the modulation index m is known, by substituting e1 = m ⎪ e1 = x 1 + x 2 + · · · + x n
⎪
⎪
into the following equations, it can be seen that the degrees of ⎪
⎨ 2 = x1 x2 + x1 x3 + · · · + xn−1 xn
⎪ e
the unsolved variables are greatly reduced, which dramatically e3 = x1 x2 x3 + x1 x2 x4 + · · · + xn−2 xn−1 xn . (10)
accelerates the solving procedure. ⎪
⎪ .
⎪
⎪..
3) Step 3—The Triangularization of the Reduced Polynomial ⎪
⎩
System: Finally, the algebraic algorithms can be applied to en = x 1 x 2 · · · x n
solve the reduced polynomial system (7). For example, if the Then, (7) will be solved by the subsequent algebraic meth-
Gröbner-basis-based method [37] is used, the following form ods, and the original variables x1 , x2 , . . . , xn can be solved
can be obtained, which are composed of a high-order univariate from e by constructing the univariate high-order polynomial (9)

Authorized licensed use limited to: SHIV NADAR UNIVERSITY. Downloaded on March 26,2024 at 19:09:36 UTC from IEEE Xplore. Restrictions apply.
WANG et al.: COMPREHENSIVE REVIEW OF SOLVING SELECTIVE HARMONIC ELIMINATION PROBLEM WITH ALGEBRAIC ALGORITHMS 855

with the known coefficients of e1 , e2 , . . . , en . By using the TABLE I

COMPARISON OF THE DEGREE OF f13 (x), f13 (e), AND f13 (p)
elementary-symmetric-polynomial-based method, the degree of
the polynomial system (6) is significantly reduced, and the
computational ability of the subsequent algebraic algorithms is
improved.
However, simplifying the polynomial system by this method
is a nonlinear process, which has a considerable computation
burden with the increasing switching angles. In the symbolic
computing software Mathematica, this simplification process Similar to the elementary-symmetric-polynomial-based
can be done by calling the SymmetricReduction command, and method, the original variables x can be solved by construct-
the maximum number of switching angles that can be solved ing the univariate polynomial (9), with coefficients e derived
by Mathematica is 8. Besides, this method requires that SHE from the results of p through the application of (13). Both
equations be symmetric polynomials, which can only be used to the elementary-symmetric-polynomial-based method and the
simplify equal-level SHE models. Newton’s identities-based method can reduce the degree of the
SHE equations. The degree comparison among the original SHE
B. Power Sum and Newton’s Identities equations (6) and the reduced SHE equations (7) and (14) is
Except for the elementary-symmetric-polynomial-based presented in Table I. In the case of SHE equations with five
method, (6) can also be transformed into the form of power sum switching angles taken as an example, the highest degree of the
symmetric polynomials. The definition of power sum symmetric original SHE equations is 13. When the elementary-symmetric-
polynomials can be described as polynomial-based method is applied, the highest degree is re-
n duced to 6. When the Newton-identity-based method is used, the
pk = xki . (11) highest degree can be reduced to 4. Besides, Newton-identity-
i=1 based method has a relatively simple solving process, which only
Obviously, (6) can be linearly transformed into the form of involves the multiplications and additions, so the computational
power sum symmetric polynomial as (12). However, it can be burden is much less than the elementary-symmetric-polynomial-
seen that (12) is undetermined as its variables are always more based method [62].
than the equations, so it cannot be solved directly
⎧
⎪
⎪ f1 (p) = p1 − m = 0 V. ALGEBRAIC ALGORITHMS
⎪
⎨f5 (p) = 5p1 − 20p3 + 16p5 = 0
f7 (p) = −7p1 + 56p3 − 112p5 + 64p7 = 0 . (12) To obtain the multiple solutions and switching patterns with
⎪
⎪ high accuracy, many algebraic algorithms have been explored
⎪..
⎩
. and improved to solve SHE equations during the past 20 years.
Newton’s identities describe the relation between power sum One kind of algebraic algorithm is to solve polynomials based on
symmetric polynomials and elementary symmetric polynomials. computational symbolic mathematics, aiming to eliminate vari-
Based on the Newton’s identities, a balance can be achieved ables and get a triangular polynomial system like (8) based on
between the number of equations and variables in (12), leading different elimination principles, including the resultant elimina-
to its representation as follows [62]: tion method [35], the Wu method [107], and the Gröbner-basis-
⎧ based method [37]. Besides, there are other kinds of algebraic
⎪
⎪ p1 = e1
⎪
⎪ algorithm for solving SHE equations, including the polynomial
⎪
⎪ p2 = e1 p1 − 2e2
⎪
⎪ homotopy continuation algorithm [60] and the Chudnovsky
⎪
⎪ p3 = e1 p2 − e2 p1 + 3e3
⎪
⎪ algorithm [108]. These methods can solve all the exact solutions
⎨.
.. without any initial value, but they rely on some mathematical
.
⎪
⎪ transformation or numerical operation instead of the principle
⎪pn = e1 pn−2 − e2 pn−3 + · · · + (−1) (n − 1)en−1
⎪ n−2
⎪
⎪ of eliminating variables in the solving process.
⎪
⎪ ..
⎪
⎪
⎪
⎪
.
⎩
pi = e1 pi−1 + e2 pi−2 + · · · + (−1)n+1 en pi−n A. Resultant Elimination Method
(13) The resultant elimination theory to solve SHE equations was
Therefore, the undetermined polynomial system (12) can be first proposed by John Chiasson, which has been applied to both
transformed into a balanced polynomial system (14) based on two-level and multilevel converters [83], [109]. After the SHE
the theory of Newton’s identities. Compared with the original equations consisting of n switching angles are transformed into a
polynomial system (6), the degree of (14) is significantly reduced polynomial system with n variables, the first step of the resultant
⎧
⎪
⎪ f1 (p) = p1 − m = 0 elimination method is integrating the polynomial system into a
⎪
⎪
⎨f5 (p) = −20p3 + 40 2 40
3 p1 p3 − · · · + 3 p3 p2 + 5p1 = 0
form where only one variable is seen as the unknown and all other
⎪ f7 (p) = 56p3 − 280 2 112 2
3 p1 p3 + · · · + 3 p3 p2 − 7p1 = 0
. variables are seen as coefficients. Denote f1 (x1 , x2 , . . . , xn )
⎪
⎪ and f2 (x1 , x2 , . . . , xn ) as two multivariate polynomials of the
⎪
⎩..
. polynomial system F = {f1 , f2 , . . . , fn }, and their maximum
(14) degree of xn are k1 and k2 , respectively; then, f1 and f2 can be
Authorized licensed use limited to: SHIV NADAR UNIVERSITY. Downloaded on March 26,2024 at 19:09:36 UTC from IEEE Xplore. Restrictions apply.
856 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 39, NO. 1, JANUARY 2024

TABLE II
PROCEDURE AND DEVELOPMENT OF THE RESULTANT ELIMINATION METHOD

rewritten as follows:
Algorithm 1: Resultant Elimination Method.
f1 (x1 , x2 , . . . , xn ) = 0 1: Eliminating variable xn by
f2 (x1 , x2 , . . . , xn ) = 0 xn = m − x1 − · · · − xn−1 , according to f1 (x) in (6).
A (x1 , x2 , . . . , xn−1 ) · xn = 0 2: while n = 1 do
⇒ (15)
B (x1 , x2 , . . . , xn−1 ) · xn = 0 3: Choosing arbitrary every two polynomials in the
remainder of (6);
where xn is a vector consisting of monomials generated by
4: Computing the Sylvester matrix of the two
xn , x2n , . . . , xkn , k is k1 or k2
polynomials;
A = [a1 (x1 , x2 , . . . , xn−1 ) , . . . , ak1 (x1 , x2 , . . . , xn−1 )] 5: One variable can be eliminated, n − 2 variables
B = [b1 (x1 , x2 , . . . , xn−1 ) , . . . , bk2 (x1 , x2 , . . . , xn−1 )] . remaining;
(16) 6: end while
7: A triangular polynomial system can be obtained, solve
Then, the resultant elimination method point out the two poly- the univariate higher polynomial with x1 at first;
nomials have common zero only if the determinant of Sylvester 8: Solving the solution of x2 , . . . , xn from the triangular
matrix equals zero. The Sylvester matrix of f1 and f2 is defined polynomial system by taking the solved variables back.
as (17), where the size of the matrix is (k1 + k2 ) × (k1 + k2 ).
Besides, the resultant of f1 and f2 is defined as (18)
⎡ ⎤
A 0 ··· 0 B 0 ··· 0 the intermediate expression will swell in polynomial expan-
⎢0 A ··· 0 0 B ··· 0⎥ sion, which leads to the exhaustion of the computer’s physical
⎢ ⎥
S=⎢. .. . . . .. .. . . .⎥ (17) memory.
⎣ .. . . .. . . . .. ⎦
To solve the problem of intermediate expression swell, Yang
0 0 ··· A 0 0 ··· B
et al. [63] proposed the polynomial interpolation method to
Res(f1 , f2 ) = det(S). (18) convert the symbolic operation to the numerical operation. The
main idea of this method is to substitute all variables in the
For the polynomial system F with n variables, it is necessary
S matrix with integers or rational numbers and then obtain
to select n − 1 groups of binary polynomials and then solve
the final result through multivariate interpolation. Take a uni-
n − 1 Sylvester matrices to eliminate one variable. The proce-
variate polynomial g(x) as an example to clarify the princi-
dure of the resultant elimination method continuously eliminates
ple of the polynomial interpolation method at first. Denote
variables by constructing and solving the Sylvester matrix until
g(x) = hm xm + hm−1 xm−1 + · · · + h0 , and then, the coeffi-
there is only one variable left so that the polynomial system is
cients of g(x) can be easily determined by solving the linear
transformed into a triangular set form. Therefore, the principle
equations (19)
of the resultant elimination method is relatively easy, and the ⎡ 0 ⎤
solving process of this algorithm is shown in Algorithm 1, and a x0 x10 · · · xm 0 ⎡ ⎤ ⎡ ⎤
summary of the application of this method is provided in Table II. ⎢ x01 x11 · · · xm ⎥ h0 y0
⎢ 1 ⎥
However, the nonzero terms in the S matrix are all sym- ⎢ .. .. ⎥ ⎢
.. ⎢ 1 ⎥ ⎢ y1 ⎥
h ⎥ ⎢
⎥
⎢ . . ⎥
⎢ . ⎥⎢⎣
.. ⎥ = ⎢ .. ⎥
⎦ ⎣
(19)
bolic polynomials, and the dimension of the S matrix increases ⎢ . .. .. ⎥ . . ⎦
rapidly along with the number of eliminated variables. There- ⎣ . . . . ⎦ hm ym
fore, in the process of solving the determinant of the S matrix, x0m x1m · · · xm m

Authorized licensed use limited to: SHIV NADAR UNIVERSITY. Downloaded on March 26,2024 at 19:09:36 UTC from IEEE Xplore. Restrictions apply.
WANG et al.: COMPREHENSIVE REVIEW OF SOLVING SELECTIVE HARMONIC ELIMINATION PROBLEM WITH ALGEBRAIC ALGORITHMS 857

where x0 –xm are m + 1 interpolation points, which can always the Wu method was first applied to solve SHE problems in
guarantee the uniqueness of g(x), and the x matrix is a Vander- engineering applications [59]. The Wu method’s main idea is to
monde matrix, which is defined as V here. transform the polynomial system (6) into a triangular character-
However, the S matrix is composed of multivariate polyno- istic set, and the solutions of the characteristic set are equal to the
mials. Taking (17) as an example to explain the determinant of original polynomial system. The essence of the Wu method is the
the S matrix solved by the polynomial interpolation method, the same as the resultant elimination method, which is transforming
whole solving process can be divided into four steps. the problem of solving multivariate nonlinear equations into
1) Step 1: The total degree of variables needs to be evaluated the problem of solving linear triangular equations. The main
to determine the number of interpolation points. A simple difference between the two methods is the principle and process
way is to add all the degree of variables in each column or of eliminating variables. According to the Wu method, the
row, and then, the total number of interpolation points can primary step is to solve the characteristic set, and the solving
be defined as (20), where n − 1 is the number of variables process of the characteristic set is briefly introduced as follows.
in (17), and di is the degree of xi . Also, there is another Given the polynomial system F = {f1 , f2 , . . ., fn }, it is
way to get the number of interpolation points; the detailed grouped by class, i.e., by the maximum subscript of variables,
principle can refer to [63] and the same kind belongs to the same group. Then, in each

n−1 group, a triangular group P is constructed by choosing a polyno-
D= (di + 1), 1 < m < n. (20) mial with the lowest power of the primary variable xi . Usually,
i=1 P is not an ascending set, and it can be restructured into an
2) Step 2: Determine the di + 1 interpolation points for each ascending set by calculating the remainder of polynomials in
variable xi , (i = 1, 2, . . . , n − 1), and the corresponding P to expand polynomials. Taking the SHE equations with three
Vandermonde matrix Vi can be generated by the interpola- switching angles, the detailed calculation of the characteristic
tion points x0i , x1i , . . . , xdi
i . Then, the vector y in (19) can
set is described follows.
be solved by calculating the determinant of the S matrix First, selecting a basic set P1 in a given polynomial group,
on each multivariate interpolation points. then, calculating the remainder of P1 for each polynomial in F1 ,
3) Step 3: After the matrices V and y are known, the coef- all the nonzero remainder is recorded as R1 .
ficient matrix h can be uniquely determined by solving Second, adding all the polynomials of R1 to polynomial
(19); for the multivariate polynomials, V and y can be group F1 can get a new polynomial group F2 = {F1 , R1 }. Then,
defined as selecting a basic set P2 from F2 , calculating the remainder of P2
for each polynomial in F2 , all the nonzero remainder is recorded
V = V 1 ⊗ V 2 ⊗ · · · ⊗ Vn−1
as R2 .
y = [y 1 , y 2 , . . . , y D ] (21) Third, repeat these steps inductively. After finite repetition
where ⊗ represents the Kronecker product of two matri- of elimination steps, the polynomial group Fk and the basic
ces. set Pk are obtained, so that any polynomial in Fk has zero
4) Step 4: Since the coefficient matrix of h is the successive remainder of Pk . If Pk is a contradictory ascending set (a
Kronecker product of Vandermonde matrices, (19) can polynomial is a nonzero constant), polynomial group F = 0 is
be solved by the extended n-dimensional Björck–Pereyr unsolvable; if Pk is not a contradictory ascending set, then Pk
algorithm [110], which can be implemented on a parallel is the characteristic set of F . The simple algorithm procedure is
computing system. Once the coefficient matrix h is solved, shown in Algorithm 2.
the final resultant polynomial can be obtained as

D C. Gröbner-Basis-Based Method
Res(f1 , f2 ) = hi B i (22)
The Gröbner basis theory was proposed by an Austrian math-
i=1
ematician Bruno Buchberger in 1965 [117] and first applied
where Bi is a basis that consists of monomials generated
to solve SHE problems in 2015 [37]. Like the aforementioned
by x1 , x2 , . . . , xn , which are arranged in ascending order
methods, the Gröbner basis theory also provides nonnumerical
in pure dictionary order.
solutions for nonlinear algebraic polynomial systems in the way
Due to the mutual independence between the interpolation
of iterative variable elimination.
points, the calculation of the function values of the interpolation
However, the Gröbner basis theory is much more complex
points can be easily distributed to multicore systems for parallel
than the resultant elimination method and the Wu method, so
calculating [63]. Therefore, this algorithm can solve the problem
that some necessary principles will be indicated intuitively here.
of intermediate expression swell and improve the efficiency of
I is an ideal of a polynomial system F , if I satisfies the
the resultant elimination method by the parallel implementation.
following three conditions.
1) 0 ∈ I.
B. Wu Method 2) If f, g ∈ I, then f + g ∈ I.
The Wu method was proposed by Wu Wentsun in the 1970s 3) If f ∈ I, h ∈ F , then hf ∈ I.
and has been used in theorem machine proving, mathematical Based on the Hilbert’s basis theorem, I can be generated by fi-
science, system science, and so on [114], [115], [116]. In 2005, nite number of polynomials p = (p1 , p2 , . . . , pm ) ⊂ F , denoted

Authorized licensed use limited to: SHIV NADAR UNIVERSITY. Downloaded on March 26,2024 at 19:09:36 UTC from IEEE Xplore. Restrictions apply.
858 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 39, NO. 1, JANUARY 2024

Algorithm 2: Wu Method. Algorithm 3: Gröbner Basis.

1: Suppose polynomials system is F0 , the number of 1: Suppose polynomials system is f1 , f2 , . . ., fm and the
variables is n, N = n, and RN −n is a nonzero number of variables is m.
polynomial. 2: while m ≥ 1 do
2: while RN −n = 0 do 3: I ≤ f1 , f2 , . . ., fm represents the ideal generated by
3: Select the polynomial with the lowest order in FN −n f1 , f2 , . . ., fm ;
as the basic set PN −n , and the remaining 4: Find the lowest common multiplier of fi and fj
(i = j), that is xr1 1 x2 · · · xn ;
r2 rn
polynomials called FN −n+1 ;
4: Computing the remainder of FN −n+1 divided by 5: Cancel out the highest terms of fi and fj (i = j),
PN −n , and the remainder called RN −n+1 ; that is
5: if The variables of RN −n+1 are n − 1 then xr1 xr2 · · · xrn xr1 xr2 · · · xrn
s(f1 , f2 ) = 1 2 n
fi − 1 2 n
fj ;
6: Go back to step 4 LT (fi ) LT (fj )
7: else 6: Find the remainder of the polynomial s(fi , fj )
8: Go back to step 5 I
divided by the ideal I, that is .
9: end if s(fi , fj )
10: n=n−1 1
7: if = 0 then
11: end while s(fi , fj )
12: If PN −n is not a contradictory ascending set, PN −n is 1
8: set = fm+1 , then get a new set
the characteristic set of F0 . s(fi , fj )
F = {f1 , f2 , . . ., fm , fm+1 }, go back to 2,
m = m + 1.
as I = p, and p is called a basis of I. The Gröbner basis is a kind 9: else
of basis with this nice property, and it always can be constructed 10: Go back to step 2, m = m − 1.
by using some specific algorithms. A Gröbner basis for an ideal 11: end if
I is a finite set of polynomials g = (g1 , g2 , . . . , gn ), if 12: end while
LT (g1 ), LT (g2 ), . . . , LT (gn ) = LT (f ) : f ∈ I (23)
where LT denotes the leading term of a polynomial. However,
the Gröbner basis of the algebraic polynomial system is not polynomials and several linear equations which any mathematic
unique, which is decided by choice of the monomial order. To software can quickly solve.
solve the SHE equations, the order must be under the pure
lexicographical monomial so that the SHE equations can be D. Chudnovsky Algorithm
transformed into a triangular form
⎧ ⎧ In [108], [118], [119], and [120], a kind of algebraic method
⎪
⎪ f1 (x1 , x2 , . . . , xn ) = 0 ⎪
⎪ g1 (xn ) = 0 for solving SHE equations was proposed, and in [120], this
⎪
⎨f2 (x1 , x2 , . . . , xn ) = 0 ⎪
⎨g2 (xn−1 , xn ) = 0 method is named Chudnovsky algorithm. The main properties
. =⇒ . .
⎪
⎪.. ⎪
⎪.. of the Chudnovsky algorithm different from the aforementioned
⎪
⎩ ⎪
⎩ algorithms are as follows. 1) the computing speed of this method
fn (x1 , x2 , . . . , xn ) = 0 gn (x1 , x2 . . . , xn ) = 0
(24) is much faster than them; 2) this method can only be used to solve
single-phase SHE equations, which means the eliminated har-
Therefore, the core problem of this method is how to solve monics must be all odd harmonics; and 3) this method is based on
the Gröbner basis. It can be done using the famous Buchberger mathematic formula transformations rather than computational
algorithm and some improved versions, and the detailed solving symbolic mathematics.
process is given in Algorithm 3. As the implementation of According to the prior knowledge introduced in Section III,
this algorithm requires solid knowledge of the polynomials and SHE equations can be transformed into the form of power sums,
professional programming skills, it is unrealistic to develop the as described in (12), which describes the power sum form of
algorithm by engineers. Some commercial symbolic computing three-phase SHE equations. However, (12) is an undetermined
software, such as M aple and M athematica, provide some polynomial system because the number of variables is more
functions to compute the Gröbner basis, and it is user-friendly than the number of equations. Actually, the power sum form
for the engineers. The simple algorithm procedure is shown in of single-phase SHE equations is balanced, and all odd power
Algorithm 3. Besides, the development of the Gröbner basis sums can be directly obtained. For example, the power sum form
method in the SHE problems is summarized in Table III. of single-phase SHE equations with three switching angles can
The essence of the three algorithms is to solve polynomial be written as (25). Obviously, the value of p1 , p3 , and p5 can be
systems with the elimination method and then get a triangu- easily solved if the modulation index is given
lar polynomial system. However, the elimination principle is ⎧
entirely different, resulting in different computational abilities ⎨f1 (p) = p1 = m
and efficiency. Once the SHE equations have been converted f3 (p) = 4p3 − 3p1 = 0 . (25)
⎩
to their equivalent triangular form like (8), there are univariate f5 (p) = 16p5 − 20p3 + 5p1 = 0

Authorized licensed use limited to: SHIV NADAR UNIVERSITY. Downloaded on March 26,2024 at 19:09:36 UTC from IEEE Xplore. Restrictions apply.
WANG et al.: COMPREHENSIVE REVIEW OF SOLVING SELECTIVE HARMONIC ELIMINATION PROBLEM WITH ALGEBRAIC ALGORITHMS 859

TABLE III
PROCEDURE AND DEVELOPMENT OF THE WU METHOD AND THE GRÖBNER-BASIS-BASED METHOD

The right-hand side of (28) can be expanded based on Taylor

Algorithm 4: Chudnovsky Algorithm.
series to obtain a polynomial with odd powers sums p as coef-
1: Solve all the odd power sums p2i−1 from (25); ficients. All coefficients of f (x) can be solved by matching the
−2p2i−1
2: Compute v2i = 0, v2i−1 = , and the detailed coefficients on both sides of (28). Therefore, the third step is to
2i − 1
definition of v can refer to [120]; construct the matching equations, which can be expressed as
3: Compute gi (0 ≤ i ≤ 2n) from f (x) = (−1)n f (−x)G(1/x) (29)
1 i
g 0 = ev 0 , g i = kvk gi−k ; where G(x) is also a polynomial and its coefficients gi can be
i k=1 solved from (28). Once f (x) is determined, the results of x can
4: ⎡Construct T oeplitz ⎤ ⎡system
⎤ ⎡ ⎤
gn · · · g1 p̃1 gn+1 be easily solved. The detailed solving process can refer to [108]
⎢ .. .. . ⎥⎢ . ⎥ ⎢ . ⎥ and [120], and the deduced calculation formula has been roughly
⎣ . . .. ⎦ ⎣ .. ⎦ = − ⎣ .. ⎦ and set given in Algorithm 4.
g2n−1 · · · gn p̃n g2n The Chudnovsky algorithm efficiently solves the SHE equa-
pi = (−1)i p̃i ; tion by primarily employing mathematical formula conversions
5: Solve pi from the T oeplitz system and construct the rather than extensive symbolic operations. Consequently, the
univariate polynomial (9) to get the final results xi ; Chudnovsky algorithm minimizes computational time and mem-
6: Solve the switching angles by using αi = arccos(xi ). ory usage, expanding the range of solvable switching angles
compared to the aforementioned algebraic methods. Besides, as
shown in (25), the value of all odd power sums p1 , p3 , p5 , . . .
Based on the principle of power sums and Newton’s identities, must be calculated so that the subsequent functions can be
if all power sums p are known, the result of the original variables constructed. However, in the three-phase SHE equations, the
x can be solved by constructing the univariate high-order (9). odd power sums cannot be directly obtained. As shown in the
The Chudnovsky algorithm can obtain f (x) by a series of math- three-phase original algebraic equations (12), there are only
ematical transformations with fast speed. The transformation p1 , p5 , p7 , . . . in the equations. Therefore, the Chudnovsky algo-
process contains three important steps: the first step is getting the rithm can only be used to solve the single-phase SHE equations.
logarithmic derivative of f (x), which can be finally expressed
as E. Polynomial Homotopy Continuation Algorithm
∞ ∞
f (x)  pn n  pn Although the homotopy algorithm is a kind of numerical algo-
= = + (26)
f (x) n=0
xn+1 x n=1 xn+1 rithm, it can overcome the local convergence characteristics of
numerical methods [71]. When F (x) is a polynomial system, the
and then, integrating (26) and f (x) can be written as follows:
 ∞  homotopy algorithm has its particular form, theory, algorithm,
 pn and software package and is closely related to the theory of
f (x) = x exp −
n
. (27) algebraic geometry.
n=1
nxn
Denote F (x) = {f1 , f2 , . . . , fn } as a polynomial system in
(x)
The second step is to construct ff(−x) to eliminate all unknown which fi are polynomials in variables x1 , x2 , . . . , xn . The poly-
even power sums p. nomial homotopy continuation algorithm operates in two main
⎛ ⎞ stages. In the first stage, a start system called the homotopy
∞
f (x) pn ⎠
= (−1)n exp ⎝−2 n
(28) mapping equation is constructed as
f (−x) n=1,n◦dd
nx H(x, t) = c(1 − t)F (x) + tG(x) = 0 (30)

Authorized licensed use limited to: SHIV NADAR UNIVERSITY. Downloaded on March 26,2024 at 19:09:36 UTC from IEEE Xplore. Restrictions apply.
860 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 39, NO. 1, JANUARY 2024

Fig. 5. Solve n nonoverlapping intervals by Sturm’s theorem.

switching angles is limited due to the sampled modulation in-

dexes. To overcome this problem, two aforementioned algebraic
algorithms were proposed to be implemented in real time.

A. Algebra-Numerical Hybrid Algorithm

Fig. 4. Schematic diagram of the polynomial homotopy continuation algo- In [64], an algebra-numerical hybrid algorithm is used to
rithm.
realize SHE in real time, in which the solving process is divided
into the offline stage and the online stage. In the offline stage, the
where G(x) = H(x, 0) is the start system and F (x) = H(x, 1) SHE equations are simplified by using the elementary symmetric
is the target system that needs to be solved, c is a random complex polynomials at first. Then, the Gröbner basis with undetermined
number, and t is the homotopy parameter that is moving from 0 parameters of the reduced polynomial system (7) is computed so
to 1. In order to find all solutions for F (x), the solution count that (7) can be converted to a group of univariate linear equations.
of the start system G(x) cannot be less than F (x). Denote E is Taking the single-phase SHE equations with four switching
the total degree of F (x): angles as an example, the Gröbner basis with undetermined
 n parameters has been described as follows, which needs to be
E= deg(fi ) (31) stored in the microcontroller:
⎧
i=1
⎪
⎪ e1 − m = 0
where deg(fi ) is the degree of the polynomial fi of F (x). Based ⎪
⎨b e + c = 0
1 2 1
on Bézout’s theorem, if F (x) has finite solutions in the complex (32)
⎪
⎪ b e
2 3 + c 2 =0
field, then the number of the solutions is no more than E, so, the ⎪
⎩
start system G(x) should also have E solutions which originate b3 e 4 + c 3 = 0
E paths connect to the E solutions of F (x). However, Bézout’s where
theorem gives the upper bound of the number of solutions. In
b1 = 448m4 − 1680m2 + 1260
order to evaluate an exact number of solutions, some linear
reduction method or the nonlinear reduction method can be used c1 = −192m6 + 1008m4 − 1680m2 + 945
to reduce the target system and give a smaller E [71].
In the second stage, as t moves from 0 to 1, numerical con- b2 = 672m4 − 2520m2 + 1890
tinuation methods trace the paths that originate at the solutions c2 = −64m7 + 504m5 − 1260m3 + 945m
of the start system toward the solutions of the target system,
and the solving process is shown in Fig. 4. The target system b3 = 26880m4 − 100800m2 + 75600
F (x) can be solved based on the known and smooth solutions of
c3 = −256m8 + 2880m6 − 10080m4 + 12600m2 − 4725.
G(x), where the nonsingular isolated solutions are the required
solutions. In the online stage, ei in (32) can be directly solved according
During the past several decades, many applications have to the input modulation index m, so that the univariate (9) can be
adopted the polynomial homotopy algorithm (PHA) to solve obtained. Therefore, the problem is converted to how to solve (9)
engineering problems, and the PHCpack was developed by in real time. Yang et al. [64] proposed to use Sturm’s theorem to
Dr. Jan Verschelde to solve the polynomial system [121]. The determine the intervals of distinct real roots of (9) and solve these
construction of the start system and the homotopy mapping equa- real roots by using the bisection method. Sturm’s theorem can
tion, the numerical continuation, or path-following algorithms divide n real roots into n nonoverlapping intervals, as shown in
are all included in this package. Therefore, it is very user-friendly Fig. 5, which is completed by constructing the following Sturm
for us to solve the SHE equations. chain:
s0 (x) = f (x) = 0, s1 (x) = f (x) = 0
VI. REAL-TIME IMPLEMENTATION OF ALGEBRAIC METHODS
s2 (x) = −rem(s0 (x), s1 (x)) = 0
As the SHE equations are challenging to solve by algebraic
algorithms in real time, the most common way to implement ···
these algebraic algorithms is still on offline and storing the sn (x) = −rem(sn−2 (x), sn−1 (x)) = 0
results in a lookup table. However, the lookup table method
costs lots of on-chip memories, and the accuracy of the output sn+1 (x) = −rem(sn−1 (x), sn (x)) = 0 (33)

Authorized licensed use limited to: SHIV NADAR UNIVERSITY. Downloaded on March 26,2024 at 19:09:36 UTC from IEEE Xplore. Restrictions apply.
WANG et al.: COMPREHENSIVE REVIEW OF SOLVING SELECTIVE HARMONIC ELIMINATION PROBLEM WITH ALGEBRAIC ALGORITHMS 861

Fig. 6. Overall control system based on the two real-time implementation algorithms. (a) Algebra-numerical hybrid algorithm. (b) Chudnovsky algorithm real-time
implementation.

where rem represents the remainder of the long division of two index is undetermined, the computational burden of solving the
polynomials. Suppose any interval (a1 , a2 ) and f (a1 )f (a2 ) = Gröbner basis significantly increases so that only the three-phase
0; then, the number of sign change of (33) in a1 and a2 is SHE equation with few switching angles can be solved by the
t(a1 ) and t(a2 ), respectively. Sturm’s theorem indicates that algebra-numerical hybrid method. The Chudnovsky algorithm,
f (x) has t(a1 ) − t(a2 ) real roots within (a1 , a2 ). The process due to the limitation of the principle, cannot be used to solve
of dividing the intervals is to find intervals (ai , ai+1 ) until the three-phase SHE equations no matter online or offline.
t(ai ) − t(ai+1 ) = 1, which means there is one real root in Therefore, these two real-time implementation methods are both
(ai , ai+1 ). After finite times division of intervals, the n real roots used to solve the single-phase SHE problem. Nevertheless, the
of the univariate equation can be isolated into n nonoverlapping computing speed of the Chudnovsky algorithm is faster than the
intervals like Fig. 5. Finally, all n roots can quickly be found in algebra-numerical hybrid method. The main reason is that there
these determined intervals using the bisection method or other is an iterative process of the method proposed in [64] when using
iterative methods. This method has been implemented in an Sturm’s theorem and numerical methods. Besides, limited by a
ARM Cortex-M4-based STM32F407 to control the multilevel huge computation burden of solving Gröbner basis, the solvable
inverters, and the execution time is around 0.3 ms, satisfying the number of switching angles of the method proposed in [64] is
requirement of the control system. much less than the Chudnovsky algorithm.

VII. PERFORMANCE COMPARISON AND DISCUSSION

B. Chudnovsky Algorithm Real-Time Implementation
In the literature, each work claimed the superiority of their
In [120], the Chudnovsky algorithm was proposed to con-
proposed method on various grounds and aspects. Thus, it is
trol the two-level inverters in real time. As the above section
difficult to assess the true performance of any algorithm from the
described, the Chudnovsky algorithm only involves some con-
existing literature. Therefore, the performance of the algebraic
versions of mathematical formulas. Like the real-time method
algorithms mentioned above will be compared and discussed
proposed in [64], the Chudnovsky algorithm is also to transform
in this section. For a fair comparison of the performance of
the SHE equations into a univariate high-order equation (9). The
algebraic algorithms, the computer with a 2.2-GHz quad-core
coefficients of f (x) can be quickly solved using the deduced
i7-2720QM CPU and 8-GB RAM is used to solve the switching
formulas of the Chudnovsky algorithm. Then, the roots of the
angles by these algorithms. The resultant elimination method,
univariate high-order equation f (x) can be obtained by using a
the Wu method, the Gröbner-basis-based method, and the Chud-
technique similar to Descartes’ sign rule. This method redefines
novsky algorithm are all implemented in the software M aple21.
f (x) with a substitution of x = cos(ϕ) and x = −cos(ϕ), where
As all of these methods are based on complete mathematical
ϕ is the reference voltage angles, which is a period symmetric
theory, the accuracy of switching angles solved by these methods
function within 0 ∼ 2π. Therefore, when f (cos(ϕ)) equals zero,
is equal to zero. Hence, the performance of these methods can be
a root of the polynomial is detected. To obtain the switching
assessed based on their capability to solve switching angles, the
command in real time, a technique similar to Descartes’ sign
computational complexity of algorithms, and the applicability
rule is proposed, and the detailed analysis can refer to [120].
to various SHE models.
These two real-time implementation methods are used to
control the single-phase inverters, and the overall control system
A. Applicability for Different SHE Models
for these two methods is shown in Fig. 6. The algebra-numerical
hybrid can handle the three-phase inverters in theory because As summarized in Fig. 2, the mathematical model of SHE
the Gröbner basis of the three-phase SHE equations with a given can be divided into different categories. According to the dc
modulation index can be solved. However, when the modulation supply of inverters, SHE models can be divided into equal-

Authorized licensed use limited to: SHIV NADAR UNIVERSITY. Downloaded on March 26,2024 at 19:09:36 UTC from IEEE Xplore. Restrictions apply.
862 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 39, NO. 1, JANUARY 2024

TABLE IV
APPLICABILITY AND COMPLEXITY OF ALGEBRAIC ALGORITHMS FOR DIFFERENT SHE MODELS

and unequal-level models. According to the output waveform TABLE V

TIME COMPARISON BETWEEN THE TWO SIMPLIFICATION METHODS
of inverters, SHE models can be divided into quarter-symmetry
and half-symmetry. Besides, the difference between single- and
three-phase systems leads to whether there are triple harmon-
ics in the SHE equations. Based on the divergence of these
mathematical models, the complexity of solving corresponding
SHE equations is quite different. Therefore, the solving ability
of these mentioned algebraic methods is related to the mathe-
matical models. The applicability of these algebraic algorithms
for different SHE mathematical models has been summarized
in Table IV. It can be seen from Table IV that both the two
simplification methods, the elementary-symmetric-polynomial-
based method (ESP), and the Newton-identity-based method
(NI), are constrained to handle only the quarter-symmetry SHE of switching angles. To facilitate a comprehensive comparison,
models. Besides, the resultant elimination algorithm (REA), steps 2 and 3 mentioned in Section III, which involve the simpli-
Wu’s algorithm (WA), and the Gröbner basis algorithm (GA) fication process and the triangularization process, are compared
have the same properties that excels in solving quarter-symmetry and discussed.
models. With the half-symmetry models and the unequal-level First, in the simplification process, the comparison is pre-
models lacking symmetrical property, they cannot be simplified sented between the elementary-symmetric-polynomial-based
by ESP and NI. Although REA, WA, and GA can theoretically method and the Newton-identity-based method in terms of
solve the half-symmetry models and unequal-level models, the computation time and the number of solvable switching angles.
number of solvable switching angles is notably limited. For Table V shows the comparison results of the two simplifica-
example, GA exhibits higher computational capacity compared tion methods, where tESP represents the computing time of
to REA and WA. However, up to a maximum of three switching the elementary symmetric polynomial based method and tNI
angles can be solved using GA for the equal-level half-symmetry represents the computing time of the Newton-identity-based
SHE models. Regarding the Chudnovsky algorithm, while it method. It can be seen that the execution time of the Newton-
proves effective in solving single-phase SHE models with rel- identity-based method is much less than elementary-symmetric-
ative ease, its applicability is limited when dealing with other polynomial-based method. Actually, when the number of
types of models. The PHA does not directly solve the original switching angles exceeds eight, the elementary-symmetric-
SHE equations, whereas it constructs a start system and then polynomial-based method totally fails to give the final results
approximates the original solutions by using the continuation or due to the huge computing burden, but the Newton’s identities
path-following methods. Therefore, the PHA has the strongest can solve more switching angles. Besides, it can be seen that the
applicability among the above algorithms. single-phase SHE equations are quite easier to solve than the
three-phase SHE equations.
Then, in the triangularization process, the resultant elimina-
B. Computational Complexity tion method, the Wu method, and the Gröbner basis method are
As the algebraic algorithms are based on complete mathe- used to solve the quarter-symmetry SHE models. Table VI shows
matical theory, these methods can obtain all the exact solutions. the comparison results of the execution time and the solvable
However, the efficiency of the these methods is quite different. number of switching angles, where tRE , tWu , and tGB represent
In this section, the computational complexity of the aforemen- the execution time of the resultant elimination method, the
tioned algebraic algorithms will be compared and discussed Wu method, and the Gröbner-basis-based method, respectively.
mainly in the aspect of computing time and the solvable number It can be seen that the efficiency of the Gröbner-basis-based

Authorized licensed use limited to: SHIV NADAR UNIVERSITY. Downloaded on March 26,2024 at 19:09:36 UTC from IEEE Xplore. Restrictions apply.
WANG et al.: COMPREHENSIVE REVIEW OF SOLVING SELECTIVE HARMONIC ELIMINATION PROBLEM WITH ALGEBRAIC ALGORITHMS 863

Fig. 7. Experimental results of the single-phase CHB. (a) Three-level phase voltage of the CHB with 20 switching angles. (b) Three-level phase voltage of the
CHB with 15 switching angles. (c) Three-level phase voltage of the CHB with 20 switching angles. (d) FFT result of the output voltage (a). (e) FFT result of the
output voltage (b). (f) FFT result of the output voltage (c).

TABLE VI VIII. EXPERIMENTAL RESULTS

COMPUTING TIME OF ALGEBRAIC ALGORITHMS
As any algebraic algorithms can derive the exact solutions for
SHE equations, the solutions obtained are the same regardless
of the algorithm applied. Therefore, the experimental results
of the SHEPWM solved by algebraic algorithms will remain
uniform. This section chooses single-phase SHE equations with
15 and 20 switching angles and three-phase SHE equations
with 12 switching angles as example to verify the correctness
and effectiveness of algebraic algorithms. The experimental
study is established on five-, seven-, and nine-level cascaded
H-bridge inverters, in which the IRFP250N MOSFETs are used as
method is higher than that of the Wu method and the resultant switching devices, the ADum1400 is used as the isolator, and the
elimination method. The Gröbner basis method offers solutions STM32F407 are used as the controller to generate the SHEPWM
for eight switching angles in single-phase SHE equations and driven signal. The dc power supply of each H-bridge is set to
five switching angles in three-level SHE equations, whereas 30 V. The experiments on both single- and three-phase inverters
the resultant elimination method and the Wu method provide have been carried out, and the results validate the effectiveness
solutions for only three and four switching angles, respectively. of the switching angles computed by the algebraic methods.
Besides, to compare the computational complexity of the In the single-phase experiments, the results of switching
whole solving process, the resultant elimination method, the angles are solved by the Chudnovsky algorithm proposed
Wu method, and the Gröbner basis method combined with in [120]. As mentioned above, the Chudnovsky algorithm has
the two simplification methods are compared in terms of the the strongest ability to solve single-phase SHE equations. In the
execution time and the solvable number of switching angles. In experimental case of single phase as shown in Fig. 7, 15 and 20
the Table VII, it can be seen that the efficiency of these algebraic switching angles are randomly selected to be verified. As shown
algorithms has been improved compared with Table VI, and the in the fast Fourier transform (FFT) results, for 20 switching
solvable number of switching angles has been increased by using angles, all the harmonics before 41st are precisely eliminated.
the simplification methods. Besides, the solving ability of these Besides, for 15 switching angles, all the harmonics before 31st
algorithms in the Gröbner-basis-based method is greater than are eliminated very well.
that in the resultant elimination method and the Wu method. In the three-phase experiments, the results of switching
This is different from the results in Table VI that the solving angles are solved by the composed algorithm based on the
ability of the Wu method is greater than the ability of the resul- Newton-identity- and Gröbner-basis-based methods proposed
tant elimination method. Therefore, by using the simplification in [62], which has the strongest ability to solve the three-phase
methods, the resultant elimination method has more improved SHE equations. Fig. 8 shows the experimental results of the
than the Wu method. three-phase SHE equations with 12 switching angles when the

Authorized licensed use limited to: SHIV NADAR UNIVERSITY. Downloaded on March 26,2024 at 19:09:36 UTC from IEEE Xplore. Restrictions apply.
864 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 39, NO. 1, JANUARY 2024

TABLE VII
COMPUTING TIME OF THE RESULTANT ELIMINATION METHOD, THE WU METHOD, AND THE GRÖBNER-BASIS-BASED METHOD WITH SIMPLIFICATION METHOD
FOR QUARTER-SYMMETRY SHE EQUATIONS

Fig. 8. Experimental results of the three-phase CHB with 12 switching angles under m = 0.8. (a) Three-level output phase voltage. (b) Seven-level output phase
voltage. (c) Nine-level output phase voltage. (d) FFT result of the output voltage (a). (e) FFT result of the output voltage (b). (f) FFT result of the output voltage (c).

modulation index is 0.8. Since multiple solutions can be solved article. Finally, a detailed comparison and evaluation of these
by algebraic algorithms, three groups’ solutions are randomly algebraic algorithms are given, and the experimental results
selected to verify the correctness of the reviewed methods. It verify the correctness of the solutions obtained by algebraic
can be seen that the harmonics before 37th (except for triple algorithms.
harmonics) have been precisely eliminated. In the whole solving procedure of algebraic algorithms, the
performance of simplification methods and algebraic methods
determines the solving efficiency and ability for SHE equations.
IX. CONCLUSION At present, the development of algebraic algorithms is relatively
The SHE technique is an important modulation method aim- complete and mature. However, some future works are still
ing to solve the harmonic elimination problem for converters. reserve to be promoted.
The biggest challenge for using the SHE technique is to solve 1) Simplification methods: The existing simplification meth-
the nonlinear and transcendental SHE equations. Algebraic algo- ods are all based on the principle of symmetric polyno-
rithms are powerful methods in solving SHE equations, which mial, which requires that SHE equations must be QW
can find all exact solutions without requirements on choosing symmetry. No research has ever discussed whether there
initial values. This review article comprehensive explored the are other ways to equivalent transform SHE equations into
existing algebraic algorithms, including the principle and the a reduced polynomials system or not. Therefore, further
solving process of these methods. In this article, two simplifica- research still need to be explored in the simplification
tion methods for reducing the degree of SHE equations and five methods.
algebraic methods for solving the SHE equations are introduced. 2) Efficiency improvement of algebraic methods: The most
Besides, the efficiency improvement of the algebraic method and algebraic methods, such as the Wu method and the Gröb-
the real-time implementation of SHE are also contained in this ner basis-based method, are all based on the universal

Authorized licensed use limited to: SHIV NADAR UNIVERSITY. Downloaded on March 26,2024 at 19:09:36 UTC from IEEE Xplore. Restrictions apply.
WANG et al.: COMPREHENSIVE REVIEW OF SOLVING SELECTIVE HARMONIC ELIMINATION PROBLEM WITH ALGEBRAIC ALGORITHMS 865

algorithm packed in software, which may limit the best [11] L. Wang, C.-S. Lam, and M.-C. Wong, “Analysis, control, and design
performance of these methods. To improve the efficiency of a hybrid grid-connected inverter for renewable energy generation with
power quality conditioning,” IEEE Trans. Power Electron., vol. 33, no. 8,
of solving SHE equations, further attempts can be made to pp. 6755–6768, Aug. 2018.
design the targeted algorithms for SHE equations and get [12] Z. Pan and R. A. Bkayrat, “Modular motor/converter system topology
rid of redundant procedures of the universal algorithms. with redundancy for high-speed, high-power motor applications,” IEEE
Trans. Power Electron., vol. 25, no. 2, pp. 408–416, Feb. 2010.
3) Real-time implementation: Although there are two real- [13] V. Yaramasu, B. Wu, P. C. Sen, S. Kouro, and M. Narimani, “High-
time methods that can be implemented by algebraic al- power wind energy conversion systems: State-of-the-art and emerging
gorithms, they have only been applied to solve single- technologies,” Proc. IEEE, vol. 103, no. 5, pp. 740–788, May 2015.
[14] E. Beser, B. Arifoglu, S. Camur, and E. K. Beser, “A grid-connected pho-
phase SHE equations. However, most industrial applica- tovoltaic power conversion system with single-phase multilevel inverter,”
tions are three-phase systems, such as motor drive and Sol. Energy, vol. 84, no. 12, pp. 2056–2067, 2010.
grid-connected converters. Based on the existing real-time [15] S. Agalar and Y. A. Kaplan, “Power quality improvement using
STS and DVR in wind energy system,” Renewable Energy, vol. 118,
methods, the Chudnovsky algorithm cannot solve three- pp. 1031–1040, 2018.
phase SHE equations because of the limitation of the [16] C. S. Edrington, O. Vodyakho, M. Steurer, S. Azongha, F. Fleming,
principle; however, the hybrid method proposed in [64] and M. Krishnamurthy, “Power semiconductor loss evaluation in voltage
source IGBT converters for three-phase induction motor drives,” in Proc.
is limited by the computational burden, which may be IEEE Veh. Power Propulsion Conf., 2009, pp. 1434–1439.
improved by the promotion of the above two points. [17] S. Yin et al., “Evaluation of power loss and efficiency for 50 kW SiC high
Therefore, further work can be conducted on this aspect. power density converter,” in Proc. Asian Conf. Energy, Power Transp.
Electrific., 2016, pp. 1–6.
4) Optimal SHE switching pattern: Algebraic algorithms can [18] A. K. Sadigh, V. Dargahi, and K. A. Corzine, “Analytical determina-
solve all the possible switching patterns for SHE, which tion of conduction and switching power losses in flying-capacitor-based
have been discussed in many literature works [62], [81], active neutral-point-clamped multilevel converter,” IEEE Trans. Power
Electron., vol. 31, no. 8, pp. 5473–5494, Aug. 2016.
[102], [103], [104]. However, there is the absence of [19] C. Liu et al., “Hybrid SiC-Si dc–ac topology: SHEPWM Si-IGBT master
discussion about how to utilize these switching patterns. unit handling high power integrated with partial-power SiC-MOSFET
Therefore, a generalized method to obtain the optimal slave unit improving performance,” IEEE Trans. Power Electron., vol. 37,
no. 3, pp. 3085–3098, Mar. 2022.
switching pattern can be further conducted based on the [20] B. Basta and W. Morsi, “Low and high order harmonic distortion in the
requirements of industrial applications, such as the con- presence of fast charging stations,” Int. J. Elect. Power Energy Syst.,
ducting time of each switching tube, the value of total vol. 126, 2021, Art. no. 106557.
[21] M. Gray and W. Morsi, “New power quantities definition for low and high
harmonic distortion, and the balance of capacitor voltage. order harmonic distortion,” Electr. Power Syst. Res., vol. 119, pp. 11–18,
2015.
[22] A. Matusiak, J. Borkowski, and J. Mroczka, “Noniterative method for fre-
REFERENCES quency estimation based on interpolated DFT with low-order harmonics
[1] S. A. Azmi, G. P. Adam, K. H. Ahmed, S. J. Finney, and B. W. Williams, elimination,” Measurement, vol. 196, 2022, Art. no. 111241.
“Grid interfacing of multimegawatt photovoltaic inverters,” IEEE Trans. [23] S. Bhalerao, S. Pawar, and N. Patwardhan, “Design and implementation
Power Electron., vol. 28, no. 6, pp. 2770–2784, Jun. 2013. of bipolar SHE-PWM single phase inverter for lower order harmonic
[2] I. P. Panapakidis, D. N. Sarafianos, and M. C. Alexiadis, “Comparative suppression,” in Proc. 3rd IEEE Int. Conf. Recent Trends Electron., Inf.
analysis of different grid-independent hybrid power generation systems Commun. Technol., 2018, pp. 1066–1071.
for a residential load,” Renewable Sustain. Energy Rev., vol. 16, no. 1, [24] Y. Kou and F. Xiong, “Influence of phase angle for low-order harmonics
pp. 551–563, 2012. on vibration of an induction traction motor,” in Proc. IEEE 5th Int. Elect.
[3] M. N. Hamidi, D. Ishak, M. A. A. M. Zainuri, and C. A. Ooi, “An Energy Conf., 2022, pp. 3736–3740.
asymmetrical multilevel inverter with optimum number of components [25] M. Boby A. R. S., K. Gopakumar, L. Umanand, F. Blaabjerg, and S.
based on new basic structure for photovoltaic renewable energy system,” Bhattacharya, “A low-order harmonic elimination scheme for induction
Sol. Energy, vol. 204, pp. 13–25, 2020. motor drives using a multilevel octadecagonal space vector structure
[4] R. Mahalakshmi and K. S. Thampatty, “Grid connected multilevel in- with a single dc source,” IEEE Trans. Power Electron., vol. 33, no. 3,
verter for renewable energy applications,” Procedia Technol., vol. 21, pp. 2430–2437, Mar. 2018.
pp. 636–642, 2015. [26] B. Guan and S. Doki, “A current harmonic minimum PWM for three-
[5] N. Al Sayari, R. Chilipi, and M. Barara, “An adaptive control algorithm level converters aiming at the low-frequency fluctuation minimum of
for grid-interfacing inverters in renewable energy based distributed gen- neutral-point potential,” IEEE Trans. Ind. Electron., vol. 66, no. 5,
eration systems,” Energy Convers. Manage., vol. 111, pp. 443–452, 2016. pp. 3380–3390, May 2019.
[6] A. Lamkaddem and K. Kassmi, “Design and realization of a new structure [27] K. Haghdar and H. A. Shayanfar, “Selective harmonic elimination with
of dc/dc and dc/ac converters dedicated to renewable energies,” in Proc. optimal dc sources in multilevel inverters using generalized pattern
Int. Conf. Elect. Inf. Technol., 2020, pp. 1–6. search,” IEEE Trans. Ind. Informat., vol. 14, no. 7, pp. 3124–3131,
[7] S.-A. Amamra, K. Meghriche, A. Cherifi, and B. Francois, “Multilevel Jul. 2018.
inverter topology for renewable energy grid integration,” IEEE Trans. [28] G. Konstantinou, M. Ciobotaru, and V. Agelidis, “Selective harmonic
Ind. Electron., vol. 64, no. 11, pp. 8855–8866, Nov. 2017. elimination pulse-width modulation of modular multilevel converters,”
[8] R. R. Karasani, V. B. Borghate, P. M. Meshram, H. M. Suryawanshi, and IET Power Electron., vol. 6, no. 1, pp. 96–107, 2013.
S. Sabyasachi, “A three-phase hybrid cascaded modular multilevel in- [29] J. Wells, B. Nee, P. Chapman, and P. Krein, “Selective harmonic control:
verter for renewable energy environment,” IEEE Trans. Power Electron., A general problem formulation and selected solutions,” IEEE Trans.
vol. 32, no. 2, pp. 1070–1087, Feb. 2017. Power Electron., vol. 20, no. 6, pp. 1337–1345, Nov. 2005.
[9] F. Wu, X. Li, F. Feng, and H. B. Gooi, “Multi-topology-mode grid- [30] H. Zhou, Y. W. Li, N. R. Zargari, Z. Cheng, R. Ni, and Y. Zhang, “Selective
connected inverter to improve comprehensive performance of renewable harmonic compensation (SHC) PWM for grid-interfacing high-power
energy source generation system,” IEEE Trans. Power Electron., vol. 32, converters,” IEEE Trans. Power Electron., vol. 29, no. 3, pp. 1118–1127,
no. 5, pp. 3623–3633, May 2017. Mar. 2014.
[10] M. A. Hosseinzadeh, M. Sarebanzadeh, E. Babaei, M. Rivera, and P. [31] A. Mishra, P. Tripathi, and K. Chatterjee, “A review of harmonic elimina-
Wheeler, “A switched-dc source sub-module multilevel inverter topol- tion techniques in grid connected doubly fed induction generator based
ogy for renewable energy source applications,” IEEE Access, vol. 9, wind energy system,” Renewable Sustain. Energy Rev., vol. 89, pp. 1–15,
pp. 135964–135982, 2021. 2018.

Authorized licensed use limited to: SHIV NADAR UNIVERSITY. Downloaded on March 26,2024 at 19:09:36 UTC from IEEE Xplore. Restrictions apply.
866 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 39, NO. 1, JANUARY 2024

[32] A. M. Amjad, K. Mehran, S. Gadoue, and F. Blaabjerg, “Online harmonic [52] K. Shen et al., “Elimination of harmonics in a modular multilevel
elimination pulse width modulation technique for modular multilevel converter using particle swarm optimization-based staircase modulation
cascade converter,” Int. J. Elect. Power Energy Syst., vol. 133, 2021, strategy,” IEEE Trans. Ind. Electron., vol. 61, no. 10, pp. 5311–5322,
Art. no. 107242. Oct. 2014.
[33] S. R. Pulikanti, M. S. A. Dahidah, and V. G. Agelidis, “Voltage balancing [53] K. P. Panda and G. Panda, “Application of swarm optimisation-based
control of three-level active NPC converter using SHE-PWM,” IEEE modified algorithm for selective harmonic elimination in reduced
Trans. Power Del., vol. 26, no. 1, pp. 258–267, Jan. 2011. switch count multilevel inverter,” IET Power Electron., vol. 11, no. 8,
[34] M. G. Cimoroni, M. Tinari, C. Buccella, and C. Cecati, “A high efficiency pp. 1472–1482, 2018.
selective harmonic elimination technique for multilevel converters,” in [54] A. Routray, R. K. Singh, and R. Mahanty, “Harmonic reduction in hybrid
Proc. Int. Symp. Power Electron., Elect. Drives, Autom. Motion, 2018, cascaded multilevel inverter using modified grey wolf optimization,”
pp. 673–677. IEEE Trans. Ind. Appl., vol. 56, no. 2, pp. 1827–1838, Mar.–Apr. 2020.
[35] J. Chiasson, L. Tolbert, K. McKenzie, and Z. Du, “A unified ap- [55] M. T. Yaqoob, M. K. Rahmat, and S. M. M. Maharum, “Modified teaching
proach to solving the harmonic elimination equations in multilevel learning based optimization for selective harmonic elimination in multi-
converters,” IEEE Trans. Power Electron., vol. 19, no. 2, pp. 478–490, level inverters,” Ain Shams Eng. J., vol. 13, no. 5, 2022, Art. no. 101714.
Mar. 2004. [56] E. Deniz, O. Aydogmus, and Z. Aydogmus, “Implementation of
[36] K. Yang et al., “Unified selective harmonic elimination for multilevel ANN-based selective harmonic elimination PWM using hybrid genetic
converters,” IEEE Trans. Power Electron., vol. 32, no. 2, pp. 1579–1590, algorithm-based optimization,” Measurement, vol. 85, pp. 32–42, 2016.
Feb. 2017. [57] Z. Salam, A. M. Amjad, and A. Majed, “Using differential evolution
[37] K. Yang, Z. Yuan, R. Yuan, W. Yu, J. Yuan, and J. Wang, “A Groebner to solve the harmonic elimination pulse width modulation for five level
bases theory-based method for selective harmonic elimination,” IEEE cascaded multilevel voltage source inverter,” in Proc. 1st Int. Conf. Artif.
Trans. Power Electron., vol. 30, no. 12, pp. 6581–6592, Dec. 2015. Intell., Model. Simul., 2013, pp. 43–48.
[38] J. Chiasson, L. Tolbert, K. McKenzie, and Z. Du, “A complete solution to [58] J. Chiasson, L. Tolbert, K. McKenzie, and Z. Du, “Control of a multilevel
the harmonic elimination problem,” IEEE Trans. Power Electron., vol. 19, converter using resultant theory,” IEEE Trans. Control Syst. Technol.,
no. 2, pp. 491–499, Mar. 2004. vol. 11, no. 3, pp. 345–354, May 2003.
[39] T.-J. Liang, R. O’Connell, and R. Hoft, “Inverter harmonic reduction [59] C. Zheng and B. Zhang, “Application of Wu method to harmonic elimi-
using Walsh function harmonic elimination method,” IEEE Trans. Power nation techniques,” Proc. CSEE, vol. 25, pp. 40–45, 2005.
Electron., vol. 12, no. 6, pp. 971–982, Nov. 1997. [60] K. Yang, D. Lu, X. Kuang, Z. Yuan, and W. Yu, “Harmonic elimination
[40] J. Sun and H. Grotstollen, “Solving nonlinear equations for selective for multilevel converters with unequal dc levels by using the polynomial
harmonic eliminated PWM using predicted initial values,” in Proc. homotopy continuation algorithm,” in Proc. 35th Chin. Control Conf.,
Int. Conf. Ind. Electron., Control, Instrum., Autom., 1992, vol. 1, 2016, pp. 9969–9973.
pp. 259–264. [61] J. Chiasson, L. Tolbert, K. McKenzie, and Z. Du, “Elimination of harmon-
[41] V. Agelidis, A. Balouktsis, and I. Balouktsis, “On applying a min- ics in a multilevel converter using the theory of symmetric polynomials
imization technique to the harmonic elimination PWM control: The and resultants,” in 42nd IEEE Int. Conf. Decis. Control, 2003, vol. 4,
bipolar waveform,” IEEE Power Electron. Lett., vol. 2, no. 2, pp. 41–44, pp. 3507–3512.
Jun. 2004. [62] C. Wang, Q. Zhang, D. Chen, Z. Li, W. Yu, and K. Yang, “Application of
[42] V. Agelidis, A. Balouktsis, I. Balouktsis, and C. Cossar, “Multiple sets of newton identities in solving selective harmonic elimination problem with
solutions for harmonic elimination PWM bipolar waveforms: Analysis algebraic algorithms,” IEEE Trans. Emerg. Sel. Topics Power Electron.,
and experimental verification,” IEEE Trans. Power Electron., vol. 21, vol. 10, no. 5, pp. 5870–5881, Oct. 2022.
no. 2, pp. 415–421, Mar. 2006. [63] K. Yang, L. Chen, J. Zhang, J. Hao, and W. Yu, “Parallel resultant elim-
[43] V. G. Agelidis, A. I. Balouktsis, and C. Cossar, “On attaining the multiple ination algorithm to solve the selective harmonic elimination problem,”
solutions of selective harmonic elimination PWM three-level waveforms IET Power Electron., vol. 9, no. 1, pp. 71–80, 2016.
through function minimization,” IEEE Trans. Ind. Electron., vol. 55, [64] K. Yang et al., “Real-time switching angle computation for selec-
no. 3, pp. 996–1004, Mar. 2008. tive harmonic control,” IEEE Trans. Power Electron., vol. 34, no. 8,
[44] A. Kumar, D. Chatterjee, and A. Dasgupta, “Harmonic mitigation of cas- pp. 8201–8212, Aug. 2019.
caded multilevel inverter with non equal dc sources using hybrid Newton [65] P. Kujan, “Numerical iterative methods solving three-phase multilevel
Raphson method,” in Proc. 4th Int. Conf. Power, Control Embedded Syst., voltage converter,” in Proc. IEEE Int. Symp. Comput.-Aided Control Syst.
2017, pp. 1–5. Des., 2010, pp. 1801–1806.
[45] M. H. Etesami, N. Farokhnia, and S. H. Fathi, “Colonial competitive [66] T. Liang and R. Hoft, “Walsh function method of harmonic elimination,”
algorithm development toward harmonic minimization in multilevel in Proc. 8th Annu. Appl. Power Electron. Conf. Expo., 1993, pp. 847–853.
inverters,” IEEE Trans. Ind. Informat., vol. 11, no. 2, pp. 459–466, [67] F. Swift and A. Kamberis, “A new Walsh domain technique of harmonic
Apr. 2015. elimination and voltage control in pulse-width modulated inverters,”
[46] M. A. Memon, S. Mekhilef, M. Mubin, and M. Aamir, “Selective har- IEEE Trans. Power Electron., vol. 8, no. 2, pp. 170–185, Apr. 1993.
monic elimination in inverters using bio-inspired intelligent algorithms [68] C. Zheng, B. Zhang, and D. Qiu, “Solving switching angles for the
for renewable energy conversion applications: A review,” Renewable inverter’s selected harmonic elimination technique with Walsh function,”
Sustain. Energy Rev., vol. 82, pp. 2235–2253, 2018. in Proc. Int. Conf. Elect. Mach. Syst., 2005, vol. 2, pp. 1366–1370.
[47] S. S. Lee, B. Chu, N. R. N. Idris, H. H. Goh, and Y. E. Heng, [69] T. Kato, “Sequential homotopy-based computation of multiple solutions
“Switched-battery boost-multilevel inverter with ga optimized SHEPWM for selected harmonic elimination in PWM inverters,” IEEE Trans.
for standalone application,” IEEE Trans. Ind. Electron., vol. 63, no. 4, Circuits Syst. I: Fundam. Theory Appl., vol. 46, no. 5, pp. 586–593,
pp. 2133–2142, Apr. 2016. May 1999.
[48] M. A. Memon, M. D. Siddique, S. Mekhilef, and M. Mubin, “Asyn- [70] G. Hosseini Aghdam, “Optimised active harmonic elimination technique
chronous particle swarm optimization-genetic algorithm (APSO-GA) for three-level T-type inverters,” IET Power Electron., vol. 6, no. 3,
based selective harmonic elimination in a cascaded H-bridge multilevel pp. 425–433, 2013.
inverter,” IEEE Trans. Ind. Electron., vol. 69, no. 2, pp. 1477–1487, [71] J. Verschelde, “Homotopy continuation methods for solving polynomial
Feb. 2022. systems,” Ph.D. dissertation, Dept. Comput. Sci., Univ. Leuven, Leuven,
[49] M. A. Memon, S. Mekhilef, and M. Mubin, “Selective harmonic elimi- Belgium, 1996.
nation in multilevel inverter using hybrid APSO algorithm,” IET Power [72] G. T. Son et al., “Improved PD-PWM for minimizing harmonics of
Electron., vol. 11, no. 10, pp. 1673–1680, 2018. multilevel converter using gradient optimization,” in Proc. IEEE Power
[50] K. Ren, F. He, Z. Li, and K. Yang, “Parallel differential evolution algo- Energy Soc. Gen. Meeting, 2014, pp. 1–5.
rithm accelerated by graphics processing unit for harmonic minimization [73] K. El-Naggar and T. H. Abdelhamid, “Selective harmonic elimination
in power converters,” in Proc. IEEE Energy Convers. Congr. Expo., 2021, of new family of multilevel inverters using genetic algorithms,” Energy
pp. 2888–2893. Convers. Manage., vol. 49, no. 1, pp. 89–95, 2008.
[51] A. Chatterjee, A. Rastogi, R. Rastogi, A. Saini, and S. K. Sahoo, “Selec- [74] H. Taghizadeh and M. T. Hagh, “Harmonic elimination of cascade
tive harmonic elimination of cascaded H-bridge multilevel inverter using multilevel inverters with nonequal dc sources using particle swarm op-
genetic algorithm,” in Proc. Innov. Power Adv. Comput. Technol., 2017, timization,” IEEE Trans. Ind. Electron., vol. 57, no. 11, pp. 3678–3684,
pp. 1–4. Nov. 2010.

Authorized licensed use limited to: SHIV NADAR UNIVERSITY. Downloaded on March 26,2024 at 19:09:36 UTC from IEEE Xplore. Restrictions apply.
WANG et al.: COMPREHENSIVE REVIEW OF SOLVING SELECTIVE HARMONIC ELIMINATION PROBLEM WITH ALGEBRAIC ALGORITHMS 867

[75] F. Filho, L. M. Tolbert, Y. Cao, and B. Ozpineci, “Real time selective [97] M. Dahidah, G. Konstantinou, N. Flourentzou, and V. Agelidis, “On
harmonic minimization for multilevel inverters connected to solar panels comparing the symmetrical and non-symmetrical selective harmonic
using artificial neural network angle generation,” in Proc. IEEE Energy elimination pulse-width modulation technique for two-level three-phase
Convers. Congr. Expo., 2010, pp. 594–598. voltage source converters,” IET Power Electron., vol. 3, pp. 829–842,
[76] E. Butun, T. Erfidan, and S. Urgun, “Improved power factor in a low-cost 2010.
PWM single phase inverter using genetic algorithms,” Energy Convers. [98] A. Ramesh and H. H. Sait, “RETRACTED: An approach towards
Manage., vol. 47, no. 11, pp. 1597–1609, 2006. selective harmonic elimination switching pattern of cascade switched
[77] K. Ganesan, K. Barathi, P. Chandrasekar, and D. Balaji, “Selective har- capacitor twenty nine-level inverter using artificial bee colony algorithm,”
monic elimination of cascaded multilevel inverter using bat algorithm,” Microprocessors Microsyst., vol. 79, 2020, Art. no. 103292.
Procedia Technol., vol. 21, pp. 651–657, 2015. [99] M. S. Dahidah and V. G. Agelidis, “Single-carrier sinusoidal PWM-
[78] J. Hao, G. Zhang, K. Yang, M. Wu, Y. Zheng, and W. Hu, “Online unified equivalent selective harmonic elimination for a five-level voltage source
solution for selective harmonic elimination based on stochastic configu- converter,” Electr. Power Syst. Res., vol. 78, no. 11, pp. 1826–1836, 2008.
ration network and Levenberg–Marquardt algorithm,” IEEE Trans. Ind. [100] V. Castiglia, R. Miceli, G. Schettino, M. G. Cimoroni, C. Buccella, and
Electron., vol. 69, no. 10, pp. 10724–10734, Oct. 2022. C. Cecati, “Selective harmonic elimination in a 5-level single phase
[79] M. Sadoughi, A. Zakerian, A. Pourdadashnia, and M. Farhadi-Kangarlu, converter with FPGA based controller,” in Proc. 5th Int. Symp. Environ.-
“Selective harmonic elimination PWM for cascaded H-bridge multilevel Friendly Energies Appl., 2018, pp. 1–6.
inverter with wide output voltage range using PSO algorithm,” in Proc. [101] M. Wu, K. Wang, K. Yang, G. Konstantinou, Y. W. Li, and Y. Li, “Unified
IEEE Texas Power Energy Conf., 2021, pp. 1–6. selective harmonic elimination control for four-level hybrid-clamped in-
[80] S. Padmanaban, C. Dhanamjayulu, and B. Khan, “Artificial neural net- verters,” IEEE Trans. Power Electron., vol. 35, no. 11, pp. 11488–11501,
work and Newton Raphson (ANN-NR) algorithm based selective har- Nov. 2020.
monic elimination in cascaded multilevel inverter for PV applications,” [102] M. Wu, H. Tian, Y. W. Li, G. Konstantinou, and K. Yang, “A composite
IEEE Access, vol. 9, pp. 75058–75070, 2021. selective harmonic elimination model predictive control for seven-level
[81] K. Yang, Q. Zhang, R. Yuan, W. Yu, J. Yuan, and J. Wang, “Selective hybrid-clamped inverters with optimal switching patterns,” IEEE Trans.
harmonic elimination with Groebner bases and symmetric polynomials,” Power Electron., vol. 36, no. 1, pp. 274–284, Jan. 2021.
IEEE Trans. Power Electron., vol. 31, no. 4, pp. 2742–2752, Apr. 2016. [103] A. L. Kouzou, A. Krama, S. S. Refaat, and H. Abu-Rub, “Selective
[82] K. Yang, Q. Zhang, R. Yuan, W. Yu, and J. Wang, “Harmonic elimination harmonic elimination PWM for a cascaded multi-level inverter,” in Proc.
for multilevel converter with Groebner bases and symmetric polynomi- IEEE Int. Conf. Ind. Technol., 2020, pp. 1145–1150.
als,” in Proc. IEEE Energy Convers. Congr. Expo., 2015, pp. 689–694. [104] J. Lago and M. L. Heldwein, “Generalized synchronous optimal pulse
[83] J. Chiasson, L. Tolbert, K. McKenzie, and Z. Du, “Elimination of harmon- width modulation for multilevel inverters,” IEEE Trans. Power Electron.,
ics in a multilevel converter using the theory of symmetric polynomials vol. 32, no. 8, pp. 6297–6307, Aug. 2017.
and resultants,” IEEE Trans. Control Syst. Technol., vol. 13, no. 2, [105] F. Pérez-Acosta and P. González-Vera, “A note on quadrature formulas
pp. 216–223, Mar. 2005. for the Chebyshev weight function of the first kind,” Appl. Math. Lett.,
[84] K. Yang, X. Lan, Q. Zhang, and X. Tang, “Unified selective harmonic vol. 7, no. 3, pp. 97–100, 1994.
elimination for cascaded H-bridge asymmetric multilevel inverter,” IEEE [106] J. N. Chiasson, L. M. Tolbert, Z. Du, and K. J. McKenzie, “The use of
Trans. Emerg. Sel. Topics Power Electron., vol. 6, no. 4, pp. 2138–2146, power sums to solve the harmonic elimination equations for multilevel
Dec. 2018. converters,” EPE J., vol. 15, no. 1, pp. 19–27, Feb. 2005.
[85] N. Rai and S. Chakravorty, “A review on the generalized formulations [107] C. Zheng, B. Zhang, and D. Qiu, “Solving selective harmonic elimination
for selective harmonic elimination (SHE-PWM) strategy,” in Proc. IEEE polynomials of inverters using the theory of symmetric polynomials
First Int. Conf. Smart Technol. Power, Energy Control, 2020, pp. 1–6. and Wu method,” Control Theory Appl., vol. 24, no. 3, pp. 361–365,
[86] D. Ahmadi, K. Zou, C. Li, Y. Huang, and J. Wang, “A universal selective Jun. 2007.
harmonic elimination method for high-power inverters,” IEEE Trans. [108] D. Czarkowski, D. Chudnovsky, and I. Selesnick, “Solving the optimal
Power Electron., vol. 26, no. 10, pp. 2743–2752, Oct. 2011. PWM problem for single-phase inverters,” IEEE Trans. Circuits Syst. I:
[87] M. Ahmed et al., “General mathematical solution for selective harmonic Fundam. Theory Appl., vol. 49, no. 4, pp. 465–475, Apr. 2002.
elimination,” IEEE Trans. Emerg. Sel. Topics Power Electron., vol. 8, [109] J. Chiasson, L. Tolbert, K. McKenzie, and Z. Du, “A complete solution
no. 4, pp. 4440–4456, Dec. 2020. to the harmonic elimination problem,” in Proc. 18th Annu. IEEE Appl.
[88] C. Buccella, C. Cecati, and M. G. Cimoroni, “She formulation for five Power Electron. Conf. Expo., 2003, pp. 596–602.
level inverters with unequal dc sources,” in Proc. IEEE Int. Conf. Ind. [110] L. Chen and Z. Zeng, “Parallel computation of determinants of matri-
Technol., 2015, pp. 1167–1172. ces with multivariate polynomial entries,” Sci. China Inf. Sci., vol. 56,
[89] K. Yang, X. Tang, Q. Zhang, and W. Yu, “Unified selective harmonic pp. 1–16, 2013.
elimination for fundamental frequency modulated multilevel converter [111] J. Chiasson, L. Tolbert, K. McKenzie, and Z. Du, “Real-time computer
with unequal dc levels,” in Proc. 42nd Annu. Conf. IEEE Ind. Electron. control of a multilevel converter using the mathematical theory of resul-
Soc., 2016, pp. 3623–3628. tants,” Math. Comput. Simul., vol. 63, no. 3, pp. 197–208, 2003.
[90] V. J. Manohar, I. S. Rani, and K. V. Ramana, “SHE controlled CHB 7-level [112] J. Chiasson, L. Tolbert, K. McKenzie, and Z. Du, “Eliminating harmonics
inverter with equal & unequal dc sources using C-genetic algorithm,” in in a multilevel converter using resultant theory,” in Proc. IEEE 33rd Annu.
Proc. Int. Conf. Commun. Electron. Syst., 2016, pp. 1–6. IEEE Power Electron. Spec. Conf. Proc., 2002, vol. 2, pp. 503–508.
[91] K. Yang, H. Li, Y. Huang, Q. Zhang, and G. Zhao, “GPU-accelerated [113] L. Tolbert, J. Chiasson, K. McKenzie, and Z. Du, “Elimination of har-
particle swarm optimization for selective harmonic elimination in multi- monics in a multilevel converter with nonequal dc sources,” in Proc. 18th
level converters with unequal dc levels,” in Proc. 43rd Annu. Conf. IEEE Annu. IEEE Appl. Power Electron. Conf. Expo., 2003, pp. 589–595.
Ind. Electron. Soc., 2017, pp. 1186–1191. [114] D. Kapur and J. L. Mundy, “Wu’s method and its application to perspec-
[92] A. Pérez-Basante, S. Ceballos, G. Konstantinou, J. Pou, I. Kortabarria, tive viewing,” Artif. Intell., vol. 37, no. 1, pp. 15–36, 1988.
and I. M. d. Alegría, “A universal formulation for multilevel selective- [115] C. Chen, G. Cao, and X. Zhao, “Mechanized mathematics—Wu elim-
harmonic-eliminated PWM with half-wave symmetry,” IEEE Trans. ination method applied in power system,” Proc. CSEE, vol. 18, no. 1,
Power Electron., vol. 34, no. 1, pp. 943–957, Jan. 2019. pp. 51–56, 1998.
[93] W. Fei, X. Du, and B. Wu, “A generalized half-wave symmetry SHE- [116] X. Zhao and W. Zhang, “Research of geometry language formal-
PWM formulation for multilevel voltage inverters,” IEEE Trans. Ind. ization based on the Wu-method,” Procedia Environ. Sci., vol. 12,
Electron., vol. 57, no. 9, pp. 3030–3038, Sep. 2010. pp. 1335–1340, 2012.
[94] M. Aleenejad, H. Mahmoudi, and R. Ahmadi, “Multifault tolerance [117] D. A. Cox, J. Little, and D. O’Shea, Ideals, Varieties, and Algorithms:
strategy for three-phase multilevel converters based on a half-wave sym- An Introduction to Computational Algebraic Geometry and Commutative
metrical selective harmonic elimination technique,” IEEE Trans. Power Algebra, ser. Undergraduate Texts in Mathematics. New York, NY, USA:
Electron., vol. 32, no. 10, pp. 7980–7989, Oct. 2017. Springer, 2013.
[95] G. Ghosh, A. K. Basu, and S. Seal, “Finding optimum switching angles [118] P. Kujan, M. Šebek, and M. Hromčík, “New special system of polynomial
using genetic algorithm for SHE-PWM two-level inverter,” in Proc. 3rd equations for 3-phase multilevel voltage converter derived from the
Int. Conf. Electron., Mater. Eng. Nano-Technol., 2019, pp. 1–7. PWM problem for 1-phase inverters,” in Proc. Eur. Control Conf., 2007,
[96] G. Konstantinou and V. Agelidis, “On re-examining symmetry of two- pp. 4568–4575.
level selective harmonic elimination PWM: Novel formulations, solu- [119] P. Kujan, M. Hromčík, and M. Šebek, “Complete fast analytical solution
tions and performance evaluation,” Electr. Power Syst. Res., vol. 108, of the optimal odd single-phase multilevel problem,” IEEE Trans. Ind.
pp. 185–197, 2014. Electron., vol. 57, no. 7, pp. 2382–2397, Jul. 2010.
Authorized licensed use limited to: SHIV NADAR UNIVERSITY. Downloaded on March 26,2024 at 19:09:36 UTC from IEEE Xplore. Restrictions apply.
868 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 39, NO. 1, JANUARY 2024

[120] A. Janabi, B. Wang, and D. Czarkowski, “Generalized Chudnovsky al- Wensheng Yu received the Ph.D. degree in dynamics
gorithm for real-time PWM selective harmonic elimination/modulation: and control from Peking University, Beijing, China,
Two-level VSI example,” IEEE Trans. Power Electron., vol. 35, no. 5, in 1998.
pp. 5437–5446, May 2020. In 1998, he joined the Institute of Automation,
[121] J. Verschelde, “Algorithm 795: PHCpack: A general-purpose solver for Chinese Academy of Sciences, Beijing, where he be-
polynomial systems by homotopy continuation,” ACM Trans. Math. came a Professor in 2004. He was a Visiting Research
Softw., vol. 25, no. 2, pp. 251–276, 1999. Fellow with the University of Melbourne, Melbourne,
VIC, Australia, from 2001 to 2002. He was a Profes-
sor with the Shanghai Key Laboratory of Trustworthy
Computing, East China Normal University, Shanghai,
Chenxu Wang received the B.Eng. degree in au- China, from 2009 to 2014. He is currently a Professor
tomation from the Qingdao University of Technology, with the School of Electronic Engineering, Beijing University of Posts and
Qingdao, China, in 2017. She is currently working Telecommunications, Beijing. He has authored or coauthored numerous articles
toward the Ph.D. degree in electrical engineering with in journals and conference proceedings, which span the fundamental fields of
the School of Mechanical and Electrical Engineering, control theory and control engineering and applied mathematics. His current
China University of Mining and Technology (Bei- research interests include robust and optimal control, adaptive filter theory and
jing), Beijing, China. linear estimation, control theory and control engineering, robotics and control,
Her main research interests include modeling and complex systems and control, and signal processing and communications.
control of high-power converters, model-predictive
control of power electronics converters, selective har-
monic elimination, and the applications of computer
algebra in power electronics. Kehu Yang (Member, IEEE) received the B.S. degree
in electrical engineering from Northwestern Poly-
technical University, Xi’an, China, in 2003, and the
Ph.D. degree in control theory and control engi-
neering from the Institute of Automation, Chinese
Qi Zhang (Student Member, IEEE) received the Academy of Sciences, Beijing, China, in 2009.
B.Eng. and M.Sc. degrees in electrical engineering From 2013 to 2014, he was a Postdoctoral Re-
from the China University of Mining and Technology search Fellow with the Department of Electrical and
(Beijing), Beijing, China, in 2015 and 2018, respec- Computer Engineering, The Ohio State University,
tively, and the Ph.D. degree in energy technology Columbus, OH, USA, working on selective harmonic
from Aalborg University, Aalborg, Denmark, in 2023. elimination technology and its applications in high-
He is currently a Postdoctoral Researcher with power converter and electric drives. He is currently a Professor with the China
AAU Energy, Aalborg University. His main research University of Mining and Technology (Beijing), Beijing, and also with the
interests include modeling, control, fault analysis, Key Laboratory of Intelligent Mining and Robotics, Ministry of Emergency
and transient stability of power-electronic-dominated Management of the People’s Republic of China, Inner Mongolia Research
power systems; control of power electronic convert- Institute of China University of Mining and Technology (Beijing), Ordos, China.
ers; selective harmonic elimination; HVDC transmission system; and HVDC His research interests include harmonic control technologies and modeling and
tapping technology. control methods for high-power converters.

Authorized licensed use limited to: SHIV NADAR UNIVERSITY. Downloaded on March 26,2024 at 19:09:36 UTC from IEEE Xplore. Restrictions apply.