Electrical Power and Energy Systems 148 (2023) 109000

Contents lists available at ScienceDirect

International Journal of Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

Grid code analysis considering converter-based grid voltage support during

faults✩
Josep Fanals-Batllori a ,∗, Jie Song b , Marc Cheah-Mañé a,b , Eduardo Prieto-Araujo a,b ,
Oriol Gomis-Bellmunt a,b
a
eRoots Analytics SL, Barcelona 08039, Spain
bDepartament d’Enginyeria Elèctrica, Centre d’Innovació Tecnològica en Convertidors Estàtics i Accionaments (CITCEA-UPC), Universitat Politècnica de
Catalunya, Barcelona 08028, Spain

ARTICLE INFO ABSTRACT

Keywords: Modern power systems are increasingly integrated with power electronic converters, which allow precise
VSC control of active and reactive current injections within a certain range of operation. Power converters are
Grid-support usually required to provide grid voltage support service in case of fault. This poses the challenge to determine
Optimization
the current injection from a converter that maintains the grid voltage while not surpassing the converter’s
Asymmetrical fault
current limitation. Conventional grid codes simply set the linear relationship between the voltage droop and
Current saturation
the grid-support current, consequently not providing the optimized voltage support. In this paper, the grid
voltage support operation of power converters is analysed from an optimization perspective. A formulation
for the analysis of faults in a generic system is presented. Numerical cases have been studied with different
fault scenarios. The optimized solutions show the traditional grid code can be potentially improved in order to
maintain the grid voltage closer to the nominal value during the fault. In this line, two improved adaptations
of this grid code have been proposed. These two alternative grid codes have been found to be near-optimal
as they just differ slightly from the optimized results despite only injecting reactive power.

1. Introduction aspect is often referred to as low voltage ride through (LVRT) [8,9].
Nowadays renewable power plants have to carefully adjust their active
The rise in renewable energies has been implemented with the and reactive currents following the grid codes to provide grid volt-
inclusion of Voltage Source Converters (VSC) as means of coupling age support under disturbances, including short-circuit faults [10]. A
energetic resources to the grid while providing controllability of the traditional approach to raise the voltage is to inject reactive current
electrical magnitudes [1–3]. The high flexibility of VSC control enables proportionally to the voltage drop [10,11]. As for the analysis of faults,
advanced grid voltage support control, which could enhance the system it is often the case that voltages are decomposed into positive, negative,
performance during the fault and ensure a fast recovery after the fault and zero sequence values to deal with unbalanced conditions [12].
clearance. However, compared to conventional electrical machines, The influence of the sequence current components on the voltages can
VSCs cannot withstand overloads for a long period [4]. Such current be assessed. In this sense, each sequence current can be thoroughly
limitation modifies the operation modes of the VSCs and must be controlled by power converters as discussed in [11,13]. Even if this
considered in the power system steady-state analysis. paper is focused on the steady-state analysis, it is important to remark
Transmission System Operators (TSO) impose a set of rules, gener-
that other works have addressed the relevant topic of control design
ally known as grid codes (GC), that connected systems have to obey [5,
for enhanced fault ride-through considering balanced and unbalanced
6]. One of these specifications is related to the injection of powers from
voltage sags (see [14,15]).
VSCs to improve the voltages during voltage sags. Short-circuit faults
The provision of voltage support means rising the positive sequence
typically cause severe voltage sags, and as a consequence, converters
voltage close to the nominal value and decreasing the negative se-
have to provide support under these circumstances [7]. They may even
quence voltage so that it approaches zero. The majority of support
be forced to remain connected under strong fault conditions. This latter

✩ This work was supported in part by FEDER/Ministerio de Ciencia, Innovación y Universidades - Agencia Estatal de Investigación, Project PID2021-
124292OB-I00. Marc Cheah-Mañé and Eduardo Prieto-Araujo are also Serra Húnter Lecturers. Oriol Gomis-Bellmunt is also an ICREA Academia researcher.
∗ Corresponding author.
E-mail address: [email protected] (J. Fanals-Batllori).

https://doi.org/10.1016/j.ijepes.2023.109000
Received 22 June 2022; Received in revised form 10 December 2022; Accepted 24 January 2023
Available online 1 February 2023
0142-0615/© 2023 Elsevier Ltd. All rights reserved.
J. Fanals-Batllori et al. International Journal of Electrical Power and Energy Systems 148 (2023) 109000

strategies are compared and summarized in [16,17]. Grid codes gen- is derived, which provides fundamental information when proposing
erally require the injection of only reactive power [18,19]. This is future improvements in order to evolve towards more resilient grids.
because transmission networks are often considered to have an in- The optimized solution, the conventional grid code, and the modi-
ductive characteristic. Ref. [13] presents expressions to maximize the fied grid codes are tested for different systems and short-circuit faults.
positive sequence voltage, to minimize the negative sequence voltage, First, a basic system with a single converter constituting a two-bus grid
or to maximize the difference between both. The power injection is is studied. The chosen case studies cover multiple scenarios such as
given as a function of the voltage at the point of common coupling variations in the fault impedance, the resistive–inductive ratio of a line,
(PCC) and the connecting impedance to the grid. However, this is and the length of a power cable. This allows evaluating the performance
only intended to be used in a simple system with a single converter of the strategies under a wide range of scenarios. Then, the analysis
directly connected to the grid. Expressions of the same nature are is performed for a system derived from the IEEE 9-bus case with two
proposed in [20], where instead of solving the optimization problem, a converters to show that the proposed methodology can be extended to
control parameter is introduced. This takes various values to prioritize larger systems with several converters. This last case is complemented
the positive or the negative sequence voltages but does not guarantee with a dynamic simulation to validate the steady-state calculations
that the system is operating in an optimal state. The effect of varying and show how the waveforms of electrical magnitudes vary during the
this control parameter is studied in [21], although it is not computed short-circuit fault. In spite of this, it has to be clearly stated that the
with a systematic approach, but rather, manually. Ref. [22] proposes paper is centred on steady-state analysis, something not uncommon
a maximum allowed support (MAS) control scheme that could provide in power converters fault studies [29,30]. As a consequence, dynamic
the maximum voltage support and simultaneously satisfy the current transient stability issues are not studied. The interested reader can
limitations. This study does not explicitly indicate how to obtain the check Refs. [31–33], where this concern is examined.
current angle, and variations in input parameters are rather limited.
Another voltage support scheme is presented in [23], where the in- 2. Formulation
fluence that the active power and the resistive part of the impedance
connecting the VSC to the grid have on the injected currents has been
2.1. System modelling
neglected. In addition, positive and negative sequence grid voltage
values are imposed, which makes the obtention of the steady-state
VSCs are elements that interconnect AC and DC grids. As shown in
current values rather trivial. A variation of the grid code require-
Fig. 1, VSCs can be modelled following the so-called average model
ments is depicted in [24], where the authors found that it provided
[34]. The VSC has been assumed to be connected to the AC grid
better results than conventional grid codes by dynamically adjusting
through a filter denoted by 𝑍 𝑧 . The control of VSCs consists of ad-
the positive and negative sequence voltage references. However, the
justing the voltages to meet the current references through an inner
presented analysis is limited for a system with a purely inductive
current loop [35]. Fig. 2 displays the corresponding control scheme of
circuit. Maximization of the power injection is described in [5], al-
grid-following VSCs.
though the focus is placed on reducing the power oscillations during
unbalanced faults rather than improving the voltages. A comparison Power systems are likely to involve more than a single converter.
between methods that employ the negative sequence component to Therefore, the modelling is approached from a generalized perspective.
minimize power fluctuations or reduce the negative sequence voltage Fig. 3 presents the generic modelling for a system with 𝑛 converters.
is found in [25]. Still, the study case is limited to a single-converter Converters are active elements connected to a grid through impedances
two-bus system and the two strategies do not cause large differences of the form 𝑍 𝑣𝑘 . The grid has been split into a passive part, only formed
in most situations. Ref. [26] presents a methodology to determine the by impedances and denoted by its admittance matrix 𝐘𝐠 , and an active
active and reactive current injection in the positive and the negative part, modelled with a Thévenin equivalent constituted by 𝑉 𝑡 and 𝑍 𝑡 .
sequence. The technique relies on the estimation of the grid impedance. The connection point between the passive and the active part of the
No information is provided on how far the achieved voltage support grid is denoted as node 𝑔. Using a Norton equivalent would be valid as
is from the optimal, and multi-converter systems are unaddressed. well, and in fact, it simplifies the formulation since it does not create
The same caveats appear in [27], where a control scheme to flexibly an additional bus between 𝑉 𝑡 and 𝑍 𝑡 . Therefore, the mathematical
inject positive and negative sequence currents considering limitations model presented in this paper ends up using a Norton equivalent. The
is detailed. The IEEE Standard 2800–2022 provides some simple rules analysis of faults can be performed in the natural reference frame or in
to determine the negative sequence current injection, although they are symmetrical components as in [36]. For convenience, in this paper, the
very generic and there is no indication on their optimality [28]. circuits are modelled following the natural reference frame.
This work displays two main contributions. On the one hand, the VSCs are treated as current sources that inject currents of the form
paper describes two adaptations derived from traditional grid codes. of 𝐈𝐤 ∀𝑘 ∈ [1, … , 𝑛]. These current vectors, expressed in the natural
One prioritizes positive sequence current whereas the other prioritizes reference frame, are further developed as:
negative sequence current. However, the two strategies also provide [ ]𝑇
𝐈𝐤 = 𝐼 𝑎𝑘 , 𝐼 𝑏𝑘 , 𝐼 𝑐𝑘 , (1)
current in the non-prioritized sequence while considering the inherent
current limitations of VSCs. On the other hand, the paper identifies where 𝐼 𝑎𝑘 , 𝐼 𝑏𝑘 and 𝐼 𝑐𝑘 are the current phasors of phases 𝑎, 𝑏 and 𝑐
the optimized system equilibrium point in terms of voltage-support respectively. Since this paper assumes VSCs to be connected to the
considering current limitations during faults. From a voltage support system by only three wires (as this is the most typical configuration
perspective, the optimal state is where the positive sequence voltage is in transmission systems), there is no zero sequence current component.
as close as possible to the nominal value while the negative sequence Thus, the formulation of the problem also imposes:
voltage tends to zero. A comparison is made between the optimal
solution, the operation points given by traditional grid codes, and 𝐼 𝑎𝑘 + 𝐼 𝑏𝑘 + 𝐼 𝑐𝑘 = 0 ∀𝑘 ∈ [1, … , 𝑛]. (2)
the result of the adapted grid codes. Such a comparison serves as a The PCC is precisely where the voltages 𝐕𝐩𝐤 ought to be controlled
benchmark to determine how close to the ideal grid codes are. The two by the converter. Again, these vectors are constituted by the three phase
adapted grid codes have been found to provide near-optimal solutions.
voltages as:
Hence, they are generally superior to the conventional grid codes while [ ]𝑇
their implementation is arguably simple. Overall, the paper indicates 𝐕𝐩𝐤 = 𝑉 𝑎𝑝𝑘 , 𝑉 𝑏𝑝𝑘 , 𝑉 𝑐𝑝𝑘 , (3)
the preferable injected currents under diverse short-circuit fault condi-
tions. An assessment of the convenience of grid codes to support faults where 𝑉 𝑎𝑝𝑘 , 𝑉 𝑏𝑝𝑘 and 𝑉 𝑐𝑝𝑘 are the 𝑎, 𝑏, 𝑐 phase voltage phasors.

2
J. Fanals-Batllori et al. International Journal of Electrical Power and Energy Systems 148 (2023) 109000

Fig. 1. Average model of a three-wire VSC connected to the grid.

Fig. 2. Control scheme of grid-following VSCs.

3
J. Fanals-Batllori et al. International Journal of Electrical Power and Energy Systems 148 (2023) 109000

Fig. 3. Single-phase representation of a complete system.

where matrices of the form 𝐘′ 𝐯𝐤 are 3 × 3𝑛𝑔 objects constituted by

zeros and only three 1∕𝑍 𝑣𝑘 elements. For instance, considering 𝑍 𝑣1 is
connected to the first bus of the grid 𝐘𝐠 , the corresponding 𝐘′ 𝐯𝟏 matrix
becomes:
⎛ 1 0 0 0 0 0 … 0 0 0⎞
⎜ 𝑍 𝑣1 ⎟
=⎜ 0 0⎟ .
′ 1
𝐘 𝐯𝟏
0 0 0 0 … 0 0 (6)
⎜ 𝑍 𝑣1
1
⎟
⎜ 0 0 0 0 0 … 0 0 0⎟
⎝ 𝑍 𝑣1 ⎠
The final goal of the modelling approach is the obtention of volt-
ages. They are computed from (5) by operating the product between
the inverse of the full admittance matrix and the currents vector. In
order to have a clearer comprehension of the voltages, they are even-
tually converted into symmetrical components by means of Fortescue’s
transformation [37]:
0 𝑎
⎛𝑉 𝑝𝑘 ⎞ ⎛1 1 1 ⎞ ⎛𝑉 𝑝𝑘 ⎞
⎜ +⎟ 1⎜ ⎟⎜ ⎟
Fig. 4. Overview of positive and negative sequence voltages and currents following
𝑉
⎜ 𝑝𝑘 ⎟ 3 ⎜1
= 𝑎 𝑎2 ⎟ ⎜𝑉 𝑏𝑝𝑘 ⎟ , (7)
⎜ −⎟ ⎜ ⎟⎜ ⎟
grid code rules. ⎝𝑉 𝑝𝑘 ⎠ ⎝1 𝑎2 𝑎 ⎠ ⎝𝑉 𝑐𝑝𝑘 ⎠
2𝜋
where 𝑎 = 𝑒𝑗 3 .
Voltages and currents are related via admittance submatrices of the
form 𝐘𝐯𝐤 , which in normal operating conditions, given that the system 2.2. VSC power injection
circuit is symmetrical, become:
While short-circuit faults on the AC side cause variations in the AC
⎛ 1 0 0 ⎞ magnitudes (such as voltages, for instance), they can also compromise
⎜ 𝑍 𝑣𝑘 ⎟ the DC side of the converter. Once a fault takes place, less active power
=⎜ 0 0 ⎟.
1
𝐘𝐯𝐤 (4)
⎜ 𝑍 𝑣𝑘
1
⎟ is transferred to the AC grid, the DC-bus capacitor absorbs extra power
⎜ 0 0 ⎟ and thus the DC-bus voltage suffers an increase. This phenomenon
⎝ 𝑍 𝑣𝑘 ⎠
may be problematic from the perspective of the DC-link integrity,
In case a fault occurs at the buses interconnected by 𝑍 𝑣𝑘 , elements and it can also jeopardize the operation of the renewable source of
1∕𝑍 𝑓 are added by observation, where 𝑍 𝑓 denotes the fault impedance. energy [38]. Several solutions to mitigate the above-presented issue
The admittance matrix 𝐘𝐠 has dimensions 3𝑛𝑔 × 3𝑛𝑔 , where 𝑛𝑔 is exist. A practical choice is to install a storage system in the DC-bus so
the number of buses found in this passive 𝐘𝐠 grid. The vector 𝐈𝐭 has a that it can absorb the extra active power and control the DC-bus voltage
length of 3 × 𝑛𝑔 and accounts for the injected currents into the grid 𝐘𝐠 . to a constant value [39]. Another possibility, perhaps more economical,
consists of placing a braking chopper in the DC-bus meant to be acti-
Consequently, the magnitudes depicted in Fig. 3 are finally related
vated under fault conditions [40,41]. Power is being dissipated in the
by:
braking resistor actively against the voltage fluctuations. An additional
⎛𝐈𝟏 ⎞ ⎛ 𝐘𝐯𝟏 𝟎 … 𝟎 −𝐘′ 𝐯𝟏 ⎞ ⎛𝐕𝐩𝟏 ⎞ alternative, significantly more rudimental, is to simply oversize the
⎜𝐈 ⎟ ⎜ 𝟎 𝐘𝐯𝟐 … 𝟎 −𝐘′ 𝐯𝟐 ⎟ ⎜𝐕𝐩𝟐 ⎟ DC-bus capacitor and therefore voltage variations are diminished. This
⎜ 𝟐⎟ ⎜ ⎟⎜ ⎟
⎜⋮⎟ = ⎜ ⋮ ⋮ ⋱ ⋮ ⋮ ⎟⎜ ⋮ ⎟, (5) paper focuses on the grid voltage support during the fault. Therefore,
⎜𝐈𝐧 ⎟ ⎜ 𝟎 𝟎 … 𝐘𝐯𝐧 −𝐘′ 𝐯𝐧 ⎟ ⎜𝐕𝐩𝐧 ⎟ the option with active power oscillation cancellation is not explicitly
⎜ ⎟ ⎜ ′𝐓 ⎟⎜ ⎟ explained.
⎝ 𝐈𝐭 ⎠ ⎝−𝐘 𝐯𝟏 −𝐘′𝐓𝐯𝟐 … −𝐘′𝐓𝐯𝐧 𝐘𝐠 ⎠ ⎝ 𝐕𝐠 ⎠

4
J. Fanals-Batllori et al. International Journal of Electrical Power and Energy Systems 148 (2023) 109000

Another problem is the appearance of second-order oscillations in is prioritized. It is assumed that VSCs employ all four degrees of
the DC-bus active power and voltage due to unbalanced faults [40]. freedom to contribute to the voltage support in positive and negative
According to the power instantaneous power theory, the active power sequences instead of regulating the power oscillations. Various options
in an unbalanced system is given by: of converter controls are not explicitly presented in this paper for the
sake of simplicity.
𝑝 = 𝑝 + 𝑝̃, (8)

where 𝑝 represents the constant active power and 𝑝̃ is the fluctuating 2.3. Optimization problem
power term. As described in [42], in a three-wire system the constant
term becomes: It can be typically assumed that positive sequence voltages have to
approach the nominal voltage while negative sequence voltages have to
𝑝 = 3(𝑉 + 𝐼 + cos(𝜃𝑣+ − 𝜃𝑖+ ) + 𝑉 − 𝐼 − cos(𝜃𝑣− − 𝜃𝑖− )), (9) be minimized. As previously stated, zero sequence components are not
taken into consideration as three-wire VSCs are unable to inject zero
where 𝜃𝑣+ , 𝜃𝑣− ,
and 𝜃𝑖+ 𝜃𝑖−
represent the angles of the positive and
sequence currents. Current saturation restrictions imposed by the VSC
negative sequence voltages and currents respectively. If (9) is split into
characteristics have to be considered in fault conditions. This applies
real and imaginary parts:
to each phase of each converter. Therefore, the generic optimization
𝑝 = 3(𝑉𝑟𝑒+ 𝐼𝑟𝑒
+ + +
+ 𝑉𝑖𝑚 𝐼𝑖𝑚 + 𝑉𝑟𝑒− 𝐼𝑟𝑒
− − −
+ 𝑉𝑖𝑚 𝐼𝑖𝑚 ), (10) problem is expressed as:
𝑛 [
∑ ]
where the subscripts of the form 𝑥𝑟𝑒 and 𝑥𝑖𝑚 are used to indicate the 𝜆+ +
|(1 − 𝑉𝑝𝑘 (𝐈))| + 𝜆− −
min
𝐈𝟏 ,…,𝐈𝐧 ∈𝐈
𝑓𝑜 = 𝑘 𝑘 |(0 − 𝑉𝑝𝑘 (𝐈))| ,
real and imaginary part of the magnitudes respectively. 𝑘=1 (20)
The fluctuating term follows an expression of the nature [42]: ( )
s.t. max 𝐼𝑘𝑎 , 𝐼𝑘𝑏 , 𝐼𝑘𝑐 ≤ 𝐼max,𝑘 ∀𝑘 ∈ [1, … , 𝑛] ,
𝑝̃ = 3(𝑉 + 𝐼 − cos(2𝜔𝑡 + 𝜃𝑣+ + 𝜃𝑖− ) + 𝑉 − 𝐼 + cos(2𝜔𝑡 + 𝜃𝑣− + 𝜃𝑖+ )), (11) where 𝑓𝑜 is the objective function, 𝐈𝟏 , … , 𝐈𝐧 respectively denote the
current injection from the 𝑛 converters, 𝐈 is the left-hand side vector
where 𝜔 is the grid frequency in rad/s and 𝑡 the time variable in s. If
in (5), 𝑘 identifies a given converter out of the total 𝑛 converters, 𝜆+ 𝑘
further developed, this oscillating power can be represented as the sum
and 𝜆− 𝑘
denote weighting factors for positive and negative voltage mag-
of two terms:
nitudes, 𝐼max,𝑘 is the maximum phase current allowed by the converter
+ − are the positive and negative sequence voltages of
𝑝̃ = 𝑝𝑐 cos(2𝜔𝑡) + 𝑝𝑠 sin(2𝜔𝑡), (12) 𝑘, and 𝑉𝑝𝑘 and 𝑉𝑝𝑘
the converter 𝑘 at the PCC. These voltages are expressed as a function
where 𝑝𝑐 denotes the cosine summand and 𝑝𝑠 the sine term. In turn, of the currents, as they are related in the natural reference frame by
they become: (5), and are eventually transformed to positive and negative sequence
𝑝𝑐 = 3(𝑉 + 𝐼 − (cos 𝜃𝑣+ cos 𝜃𝑖− − sin 𝜃𝑣+ sin 𝜃𝑖− ) using (7). Ideally, they should tend to 1.0 and 0.0 pu respectively.
It has been assumed that voltages at each phase of the converter do
+ 𝑉 − 𝐼 + (cos 𝜃𝑣− cos 𝜃𝑖+ − sin 𝜃𝑣− sin 𝜃𝑖+ )),
(13) not surpass the limitations, which seems a fair assumption considering
𝑝𝑠 = 3(𝑉 + 𝐼 − (sin 𝜃𝑣+ cos 𝜃𝑖− + cos 𝜃𝑣+ sin 𝜃𝑖− ) that AC voltages decrease substantially during faults. In any case, this
+ 𝑉 − 𝐼 + (sin 𝜃𝑣− cos 𝜃𝑖+ + cos 𝜃𝑣− sin 𝜃𝑖+ )). assumption is numerically validated in the case studies. The solution to
(20) is equivalent to the result of the strategy C shown in [13], although
Again, 𝑝𝑐 and 𝑝𝑠 can be divided into real and imaginary terms to yield:
it is not limited to a single converter. This solution is not meant to be
𝑝𝑐 = 3(𝑉𝑟𝑒+ 𝐼𝑟𝑒
− + −
− 𝑉𝑖𝑚 𝐼𝑖𝑚 + 𝑉𝑟𝑒− 𝐼𝑟𝑒
+ − +
− 𝑉𝑖𝑚 𝐼𝑖𝑚 ), implemented in practice; rather, it will act as a benchmark representing
+ −
(14) the ideal case.
𝑝𝑠 = 3(−𝑉𝑖𝑚 𝐼𝑟𝑒 − 𝑉𝑟𝑒+ 𝐼𝑖𝑚
− − +
− 𝑉𝑖𝑚 𝐼𝑟𝑒 − 𝑉𝑟𝑒− 𝐼𝑖𝑚
+
).
The expressions of 𝑝𝑐 and 𝑝𝑠 in (14) will be employed to quantify the 2.4. Conventional Grid Codes (GC)
magnitude of the oscillating power.
Similarly, the reactive power 𝑞 is formed by a constant term 𝑞 and In order to improve the voltage profile during faults, grid code con-
an oscillatory component 𝑞̃ [42]: trol rules often impose injection requirements for positive and negative
sequence currents proportional to the positive and negative sequence
𝑞 = 𝑞 + 𝑞̃. (15)
voltage drops respectively [10,45]. In the case of positive sequence, a
Each term is calculated as: generic piecewise function can be defined as:
𝑞 = 3(𝑉 + 𝐼 + sin(𝜃𝑣+ − 𝜃𝑖+ ) − 𝑉 − 𝐼 − sin(𝜃𝑣− − 𝜃𝑖− )), ⎧𝐼𝑘+ = 0 +
𝑉𝑝𝑘 +
≥ 𝑉high
𝑞̃ = 3(𝑉 − 𝐼 + sin(2𝜔𝑡 + 𝜃𝑣− + 𝜃𝑖+ ) (16) + + ⎪ + + + + + +
𝑓 (𝑉𝑝𝑘 ) ∶= ⎨𝐼𝑘 = 𝑘𝑝 (𝑉high − 𝑉𝑝𝑘 ) 𝑉low ≤ 𝑉𝑝𝑘 < 𝑉high (21)
− 𝑉 + 𝐼 − sin(2𝜔𝑡 + 𝜃𝑣+ + 𝜃𝑖− )). ⎪ + + +
⎩𝐼𝑘 = 𝐼max,𝑘 𝑉𝑝𝑘 < 𝑉low
The constant reactive power can be developed as: + +
where 𝑘𝑝 is a droop constant, 𝑉high and 𝑉low are the positive sequence
𝑞= + +
3(𝑉𝑖𝑚 𝐼𝑟𝑒 − 𝑉𝑟𝑒+ 𝐼𝑖𝑚
+ − −
− 𝑉𝑖𝑚 𝐼𝑟𝑒 + 𝑉𝑟𝑒− 𝐼𝑖𝑚
−
), (17) upper and lower voltage thresholds, and 𝐼𝑘+ stands for the absolute
value of the positive sequence current to be injected by the VSC. The
while the fluctuating component is equivalently expressed as: most basic of the four voltage support strategies analysed in this paper
𝑞̃ = 𝑞𝑐 cos(2𝜔𝑡) + 𝑞𝑠 sin(2𝜔𝑡), (18) will be denoted by GC. It is precisely the application of (21), i.e., only
positive sequence current is injected.
with 𝑞𝑐 and 𝑞𝑠 being: A similar proportionality can be established in the negative se-
𝑞𝑐 = 3(−𝑉𝑖𝑚 𝐼𝑟𝑒 − 𝑉𝑟𝑒+ 𝐼𝑖𝑚
+ − − − +
+ 𝑉𝑖𝑚 𝐼𝑟𝑒 + 𝑉𝑟𝑒− 𝐼𝑖𝑚
+
), quence:
(19)
𝑞𝑠 = 3(−𝑉𝑟𝑒+ 𝐼𝑟𝑒
− + −
+ 𝑉𝑖𝑚 𝐼𝑖𝑚 + 𝑉𝑟𝑒− 𝐼𝑟𝑒
+ − +
− 𝑉𝑖𝑚 𝐼𝑖𝑚 ). ⎧𝐼𝑘− = 0 − ≤𝑉−
𝑉𝑝𝑘 low
⎪
Such oscillating power during the unbalanced fault stands for a risk 𝑓 − (𝑉𝑝𝑘
−
) ∶= ⎨𝐼𝑘− = 𝑘𝑛 (𝑉𝑝𝑘
− −𝑉− ) 𝑉− <𝑉− ≤𝑉−
low low 𝑝𝑘 high
(22)
⎪ − − >𝑉−
to the DC voltage control [43]. One possible option is to use two ⎩𝐼𝑘 = 𝐼max,𝑘 𝑉𝑝𝑘 high
degrees of freedom from the converter to eliminate oscillating elements − and 𝑉 − are the
where 𝑘𝑛 is the negative sequence droop constant, 𝑉high
in active power [44]. However, since this paper focuses on steady- low
state analysis of AC grid codes, the voltage support operation of VSCs negative sequence upper and lower voltage thresholds, and 𝐼𝑘− stands

5
J. Fanals-Batllori et al. International Journal of Electrical Power and Energy Systems 148 (2023) 109000

Fig. 5. Single-phase representation of the single converter system under study.

that the nominal current of the converter could be exceeded due

to imbalances. Indeed, during extreme faults, the positive sequence
+
voltage could reach the lower threshold 𝑉low and the negative sequence
voltage could surpass the upper threshold 𝑉high − . Then, both 𝐼 + and
𝑘
−
𝐼𝑘 would be set at 𝐼max,𝑘 . Consequently, some of the currents in the
natural reference frame could become larger than 𝐼max,𝑘 . One way to
address this problem consists of injecting only positive or negative
sequence currents (the GC option is based on this idea). However, this
strategy is likely to be suboptimal if the maximum voltage-support
has to be provided. In case the voltages do not exceed these lower or
upper thresholds, the converter would still be capable of injecting some
current in the non-prioritized sequence up to reaching saturation.

2.5. Improved Grid Codes (GCP and GCN)

The two adaptations of the conventional grid code, which are a

central contribution of this paper, basically consist of determining the
available current margin. The two direct prioritizations are covered.
One strategy is to have the positive sequence current following (21)
and injecting the remaining maximum allowed current in the negative
sequence. This methodology will be commonly referred to as GCP.
The other option, abbreviated as GCN, prioritizes the negative
sequence. It obeys (22) and analogously injects the maximum positive
sequence current that respects the limits. During all situations, the
converters work under saturated conditions (current limits reached) as
the non-prioritized current is set to the allowed maximum.
Also, only reactive power is injected in both GCP and GCN. It is
generally accepted that since transmission systems are mainly induc-
tive, injecting reactive current has a larger impact on the voltages
than injecting active current. Hence, leaving aside stability issues, it is
beneficial to prioritize reactive currents. The procedure to determine
the non-prioritized current is the main challenge in GCP and GCN.
Thus, it is depicted in detail below.
Consider a generic distribution of positive and negative sequence
voltages in the complex plane, as in Fig. 4. Phase currents are related
to sequence components as follows:

⎛𝐼 𝑎 ⎞ ⎛1 1 1 ⎞ ⎛𝐼0 ⎞
⎜ 𝐼 𝑏 ⎟ = ⎜1 𝑎2 𝑎 ⎟ ⎜𝐼 + ⎟ , (23)
⎜ 𝑐⎟ ⎜ ⎟⎜ ⎟
⎝ 𝐼 ⎠ ⎝1 𝑎 𝑎2 ⎠ ⎝ 𝐼 − ⎠
Fig. 6. Influence of the currents for a balanced fault with a varying fault impedance,
one-converter case.
where 𝐼 0 is zero as the three-wire VSC is incapable of injecting it. The
𝑘 index that identifies a given VSC has been omitted to alleviate the
notation. Explicitly, (23) becomes:
for the absolute value of the negative sequence current to be injected
by the VSC. ⎧𝐼 𝑎 = 𝐼 + + 𝐼 −
⎪ 𝑏 + −
Constants 𝑘𝑝 and 𝑘𝑛 are set as fixed quantities [11]. The un- ⎨𝐼 = 𝑎2 𝐼 + 𝑎𝐼 (24)
favourable consequence of applying (21) and (22) simultaneously is ⎪ 𝑐 + 2 −
⎩𝐼 = 𝑎𝐼 + 𝑎 𝐼

6
J. Fanals-Batllori et al. International Journal of Electrical Power and Energy Systems 148 (2023) 109000

The goal is to determine the maximum current that respects the Table 1
Generic grid code parameters.
converter limits. It is known beforehand that not all the three phases
Parameter Value Parameter Value
of a particular converter will operate with a saturated current; most
+ +
likely, only one phase will reach saturation. The phase currents in (24) 𝑉high 0.9 𝑉low 0.4
− −
+ , 𝐼 + , 𝐼 − and 𝑉high 0.6 𝑉low 0.1
are squared in order to express them as functions of 𝐼𝑟𝑒 𝑖𝑚 𝑟𝑒
− 𝑘𝑝 2.0 𝑘𝑛 2.0
𝐼𝑖𝑚 :

⎧
⎪𝐼max
2 + + 𝐼 − )2 + (𝐼 + + 𝐼 − )2
≥ (𝐼𝑟𝑒 𝑟𝑒 𝑖𝑚 𝑖𝑚
⎪ √ √
⎪𝐼 2 ≥ (− 1 𝐼 + + 3 𝐼 + − 1 𝐼 − − 3 𝐼 − )2
⎪ max 2 𝑟𝑒 2 𝑖𝑚 2 𝑟𝑒 2 𝑖𝑚
⎪ √ √ In essence, Algorithms 1 and 2 compute one sequence current
𝑔 ∶= ⎨ +(− 2 𝐼𝑖𝑚 − 2 𝐼𝑟𝑒 − 2 𝐼𝑖𝑚 + 23 𝐼𝑟𝑒
1 + 3 + 1 − − )2 (25)
⎪ with the basic grid codes (positive and negative respectively). The
√ √
⎪𝐼 2 ≥ (− 1 𝐼 + − 3 𝐼 + − 1 𝐼 − + 3 𝐼 − )2 other sequence current (negative and positive respectively) is found by
⎪ max 2 𝑟𝑒 2 𝑖𝑚 2 𝑟𝑒 2 𝑖𝑚
⎪ √ √ solving the 𝑔 function shown in (25), which is implemented with a 𝑠𝑜𝑙𝑣𝑒
3 + 3 −
⎪ +( 2 𝐼𝑟𝑒 − 2 𝐼𝑖𝑚 − 2 𝐼𝑟𝑒 − 21 𝐼𝑖𝑚
1 + − )2
function in Algorithms 1 and 2. Since there would be infinite solutions,
⎩
it is forced that at least in one phase the current magnitude is equal
It is also convenient to define the angles of the positive and negative to 𝐼max . In this sense, one inequality is turned into an equality. This
sequence currents, which are denoted by 𝜃𝑖+ and 𝜃𝑖− respectively: is done in an iterative fashion until a solution that validates all three
{ inequalities is found. The algorithm also makes sure that the resulting
𝜃𝑖+ = ∠(𝑉 +
𝑝
) − 𝜋2
(26) current does not surpass the one obtained from (22) for GCP or from
𝜃𝑖− = ∠(𝑉 −
𝑝
) + 𝜋2 (21) for GCN.
where ∠(⋅) indicates the function that extracts the angle of a complex
magnitude. By shifting the current angle ± 𝜋2 rad with respect to the 3. Single converter case study
corresponding voltages 𝑉 + 𝑝
and 𝑉 −𝑝
, only reactive power is injected.
As transmission systems are mostly inductive, providing only reactive
The analysis is first performed considering a one-converter case
power is expected to boost the voltages magnitude. This is verified in
study as the one depicted in Fig. 5. Unless noted otherwise, the cor-
the results section where the suitability of these adaptations is assessed
responding baseline parameters are indicated in Table 2. It has been
by means of varying the angle of the impedances.
+ , 𝐼 + , 𝐼 − and 𝐼 − is
decided to model a transmission system rather than a distribution grid
A procedure to find the four components 𝐼𝑟𝑒 𝑖𝑚 𝑟𝑒 𝑖𝑚 since the operation under faults is much more critical there. A fault
presented. Contrarily to traditional grid codes where the grid-support
is forced at the grid-equivalent terminals. The impedances that model
current is only specified in one sequence, the algorithms that follow
the fault are set accordingly to the type of fault, i.e., balanced or
provide the currents in both positive and negative sequences. First, Al-
unbalanced. Most of the case studies covered in the paper correspond to
gorithm 1 contains the methodology to be employed when the positive
unbalanced faults as they have a higher probability of occurrence [46].
sequence is prioritized (GCP).
The goal is to improve the voltage 𝑉 𝑝1 by injecting the optimal 𝐼 𝑎1 , 𝐼 𝑏1
and 𝐼 𝑐1 currents. Three parametric studies are covered:
Algorithm 1 Current calculation for the GCP strategy.
Input: 𝑉𝑝+ , 𝑉𝑝− , 𝜃𝑖+ , 𝜃𝑖− , 𝐼max 1. The fault impedance experiences variations. This case is explic-
Output: 𝐼𝑟𝑒 +, 𝐼+ , 𝐼−, 𝐼− itly described in order to exemplify the formulation.
𝑖𝑚 𝑟𝑒 𝑖𝑚
1: 𝐼 + ← 𝑓 + (𝑉𝑝+ ); {use (21)} 2. The ratio 𝑅1 ∕𝑋1 , which stands for the proportion between the
2: 𝐼 − ← 𝑓 − (𝑉𝑝− ); {use (22)} resistive and the inductive parts that compose the impedance
− , 𝐼 − ) ← 𝑠𝑜𝑙𝑣𝑒(𝑔) subject to ∠(𝐼 − + 𝑗𝐼 − ) = 𝜃 − and (𝐼 𝑎 = 𝐼 𝑍 𝑣1 , takes a range of values. In any case the absolute value of
3: (𝐼𝑟𝑒 𝑖𝑚 𝑟𝑒 𝑖𝑚 𝑖 max or
𝐼 𝑏 = 𝐼max or 𝐼 𝑐 = 𝐼max ); the impedance 𝑍 𝑣1 is kept constant.
4: 𝐼 − ← min(𝐼 − , |𝐼𝑟𝑒 − + 𝑗𝐼 − |);
𝑖𝑚 3. The length of a hypothetical submarine cable is increased. As it
+ +
+ +
5: 𝐼𝑟𝑒 ← ℜ(𝐼 ∠𝜃𝑖 ); 𝐼𝑖𝑚 ← ℑ(𝐼 + ∠𝜃𝑖+ ); is represented by a 𝜋 line model, both the magnitude and the
− ← ℜ(𝐼 − ∠𝜃 − ); 𝐼 − ← ℑ(𝐼 − ∠𝜃 − );
6: 𝐼𝑟𝑒 𝑖 𝑖𝑚 𝑖 angle of the impedance change.
7: return 𝐼𝑟𝑒 +, 𝐼+ , 𝐼−, 𝐼− .
𝑖𝑚 𝑟𝑒 𝑖𝑚
Regarding the grid codes, the values of the constants are gathered
in Table 1. They remain the same for all studies. Both the weighting
Similarly, Algorithm 2 details the procedure to follow in case the
factors 𝜆+
1
and 𝜆−
1
have been set to the same value to attribute the same
negative sequence is prioritized (GCN).
importance to each sequence, and take the same values as in [47]. The
optimal results can change significantly in case one sequence is prior-
Algorithm 2 Current calculation for the GCN strategy. itized. The results obtained in this paper are computed with Python
Input: 𝑉𝑝+ , 𝑉𝑝− , 𝜃𝑖+ , 𝜃𝑖− , 𝐼max 3.9.1 using the Mystic package, a highly constrained non-convex opti-
Output: 𝐼𝑟𝑒 +, 𝐼+ , 𝐼−, 𝐼− mization framework [48,49]. A differential global optimization solver
𝑖𝑚 𝑟𝑒 𝑖𝑚
1: 𝐼 + ← 𝑓 + (𝑉𝑝+ ); {use (21)} has been employed with a relative precision up to 1⋅10−6 . Nevertheless,
2: 𝐼 − ← 𝑓 − (𝑉𝑝− ); {use (22)} other software capable of handling complex convex problems can be
+ , 𝐼 + ) ← 𝑠𝑜𝑙𝑣𝑒(𝑔) subject to ∠(𝐼 + + 𝑗𝐼 + ) = 𝜃 + and (𝐼 𝑎 = 𝐼 adopted as well.
3: (𝐼𝑟𝑒 𝑖𝑚 𝑟𝑒 𝑖𝑚 𝑖 max or
𝐼 𝑏 = 𝐼max or 𝐼 𝑐 = 𝐼max );
4: 𝐼 + ← min(𝐼 + , |𝐼𝑟𝑒 + + 𝑗𝐼 + |);
𝑖𝑚
+ ← ℜ(𝐼 + ∠𝜃 + ); 𝐼 + ← ℑ(𝐼 + ∠𝜃 + );
3.1. Fault impedance variation analysis
5: 𝐼𝑟𝑒 𝑖 𝑖𝑚 𝑖
− ← ℜ(𝐼 − ∠𝜃 − ); 𝐼 − ← ℑ(𝐼 − ∠𝜃 − );
6: 𝐼𝑟𝑒 𝑖 𝑖𝑚 𝑖
7: return 𝐼𝑟𝑒 +, 𝐼+ , 𝐼−, 𝐼− . For a single converter system as the one depicted in Fig. 5, the fault
𝑖𝑚 𝑟𝑒 𝑖𝑚
impedance connected to the bus at voltage 𝑉 𝑓 represents the event of

7
J. Fanals-Batllori et al. International Journal of Electrical Power and Energy Systems 148 (2023) 109000

Fig. 8. Influence of the currents for the line-to-ground fault with a varying 𝑅1 ∕𝑋1
ratio and a fault impedance 𝑍 𝑎𝑔 = 0.01, one-converter case.
Fig. 7. Influence of the currents for a line-to-line fault with a varying fault impedance,
one-converter case.

𝐘𝐟 is constituted by the admittances that intervene in the fault and has

short-circuit in this case. Explicitly, voltages and currents are related the same size as 𝐘𝐯 . This way, the admittance matrices are defined as:
by:
𝑌 𝑣1 0 0 −𝑌 𝑣1 0 0
𝑎
⎛ ⎞
0 𝑌 𝑣1 0 0 −𝑌 𝑣1 0
𝑎 ⎛𝑉 𝑝1 ⎞ ⎜ ⎟
⎛𝐼 1 ⎞
𝐘𝐯 = ⎜ ⎟
0 0 𝑌 𝑣1 0 0 −𝑌 𝑣1
⎜𝐼 𝑏 ⎟ ⎜ 𝑏 ⎟ (28)
⎜𝑉 𝑝1 ⎟ ⎜ −𝑌 𝑣1 0 0 𝑌 𝑣1 + 𝑌 𝑡 0 0 ⎟
⎜ 1⎟ ⎜ 𝑐 ⎟ ⎜ 0 −𝑌 𝑣1 0 0 𝑌 𝑣1 + 𝑌 𝑡 0 ⎟
⎜𝐼 𝑐 ⎟ ( ) ⎜𝑉 𝑝1 ⎟ ⎝ 0 0 −𝑌 𝑣1 0 0 𝑌 𝑣1 + 𝑌 𝑡 ⎠
⎜ 1⎟ = 𝐘𝐯 + 𝐘𝐟 ⎜ 𝑎 ⎟ , (27)
⎜𝐼 𝑎 ⎟ ⎜𝑉 𝑓 ⎟
⎜ 𝑡⎟ where 𝑌 𝑣1 = 1∕𝑍 𝑣1 and 𝑌 𝑡 = 1∕𝑍 𝑡 .
⎜𝐼 𝑏𝑡 ⎟ ⎜ 𝑏⎟
⎜ 𝑐⎟ ⎜𝑉 𝑓 ⎟ On the other hand, the fault admittance matrix 𝐘𝐟 is fully dependent
⎝𝐼 ⎠ ⎜ 𝑐⎟
𝑡 ⎝𝑉 𝑓 ⎠ on the type and depth of the fault. All its elements are zero except
where the admittance matrix has been splitted into two parts: 𝐘𝐯 for a few entries that contain the fault admittance connected to their
represents the non-faulted admittance matrix of the system, whereas respective bus. The optimization problem for the one case converter,

8
J. Fanals-Batllori et al. International Journal of Electrical Power and Energy Systems 148 (2023) 109000

Fig. 9. Single-phase representation of the single converter connected to a system with a cable.

Table 2
System parameters for the one-converter case.
Parameter Value Parameter Value
𝑉𝑡 1.00 𝐼max 1.00
𝑍 𝑣1 0.01 + 𝑗0.05 𝑍𝑡 0.01 + 𝑗0.1
𝜆+1 1.00 𝜆−1 1.00

always the smallest. In other words, the positive and negative sequence
voltages tend to be closer to the references of 1 and 0 pu respectively
Fig. 10. Thévenin voltage and impedance depending on the cable distance. (see (29)). This optimal solution is achieved by distributing the currents
between the real and imaginary parts. The real part of the current is
the so-called active current, whereas the imaginary part represents the
particularized from (20), reads: reactive current. It has to be noted that the phase of the positive and
negative currents is referred to their corresponding positive and nega-
min 𝑓𝑜 = 𝜆+
1
+
|(1 − 𝑉𝑝1 (𝐈))| + 𝜆−
1
−
|(0 − 𝑉𝑝1 (𝐈))| ,
𝐈𝟏 ∈𝐈
(29) tive voltages. In this case, the imaginary current remains slightly larger
s.t. max(𝐼1𝑎 , 𝐼1𝑏 , 𝐼1𝑐 ) ≤ 𝐼max,1 , than the real current (in absolute value). This is the main difference
between OPT and the grid code implementations found in GCP, GCN,
where 𝐈𝟏 contains the currents 𝐼 𝑎1 , 𝐼 𝑏1 , 𝐼 𝑐1 for which the optimization and GC. These three strategies require the converter to employ its full
problem is actually solved for, and 𝐈 is the left-hand side vector in (27). capability on the imaginary positive sequence current. Therefore, their
Because of the nature of the three-wire converter, it is also imposed that objective functions and voltage profiles become identical in this case
the sum of the three-phase currents becomes zero. with a balanced fault.
The procedure to solve the optimization problem is the following: Significantly different results are obtained in the case of a line-
to-line fault, as shown in Fig. 7. This unbalanced fault causes the
• Initialize the admittance matrices 𝐘𝐯 as in (28), and then, con- optimal currents to be almost zero in the positive sequence, so the
struct 𝐘𝐟 . majority of the current capability is employed in the negative sequence.
• Currents are initialized to a random array of values. Severe faults require a large real current (in absolute value), whereas
• The Mystic package is called to solve the optimization problem in the case of less extreme faults, the imaginary current tends to the
stated in (29). maximum, i.e, 1.0. Even though the active current is kept at zero for
• Positive and negative sequence voltages are evaluated to deter- GCP, GCN, and GC, their reactive currents vary significantly across the
mine the optimality of the solution by means of (7). range of 𝑍𝑓 values. GCP tends to approach the results provided by
• In the case of sweeping a range of 𝑚 scenarios with different OPT for large 𝑍𝑓 , whereas GCN evolves in the contrary direction. This
fault impedance values, the above-presented steps are repeated finding was not forced, but rather, it happens to be the case that GCP
𝑚 times. closely follows the optimal trajectory.
• Currents are transformed to positive and negative sequence values For small purely resistive fault impedances, GCP prioritizes injecting
since the final goal is to evaluate their values in this frame of current in the positive sequence whereas GCN does the same for the
reference. negative sequence, as expected. As a consequence, positive sequence
voltages in the case of GCP are initially superior to the ones obtained
Solving the problem with the aforementioned steps for a balanced with GCN, and the contrary applies to the negative sequence voltages.
fault yields the results shown in Fig. 6, where impedance 𝑍 𝑓 denotes The situation is reversed around 𝑍𝑓 ≈ 0.1. Since the fault is not
the fault impedance connected to each phase, which is considered to that severe, much of the current capability is employed in the non-
be fully resistive in this case. The results suggest that the optimal case prioritized sequence in order to reach saturation. In the end, there is
(OPT) is the preferred one, as its associated objective function 𝑓𝑜 is an almost constant minimal difference between OPT and GCP, as their

9
J. Fanals-Batllori et al. International Journal of Electrical Power and Energy Systems 148 (2023) 109000

Table 3
Cable parameters for the one-converter case.
Parameter Value Units
𝑍𝑠 6.674 ⋅ 10−5 + 𝑗2.597 ⋅ 10−4 pu/km
𝑍𝑝 −𝑗77.372 pu km

all the range of values. Most of the current capability is dedicated to

+ −.
𝐼𝑖𝑚 and 𝐼𝑖𝑚
Differences among the objective functions 𝑓𝑜 of OPT and the rest of
the options are more significant for a large 𝑅1 ∕𝑋1 value since it has
been established that GCP, GCN, and GC cannot inject any active cur-
rent. In particular, the OPT option achieves a larger positive sequence
voltage than both GCP and GC for all ranges of values. Regarding
the negative sequence voltage, OPT is usually the best choice despite
having placed the same importance on the positive and the negative
sequence voltages. GC does not make any effort to minimize 𝑉 − , and
this is why it remains significantly superior to the other strategies. The
price to pay for an optimized solution is the presence of large active
power oscillating terms 𝑝𝑐 and 𝑝𝑠 . Despite the fact that GCP prioritizes
active current, as the fault is not severe and it has been imposed
that the converter has to operate under saturated conditions, the GCP
strategy injects more current in the negative sequence than in the
positive sequence. The contrary applies to the GCN option. Therefore,
GCN yields larger 𝑉 + and 𝑉 − voltages compared to GCP. In the end,
the objective function is practically the same for GCP and GCN. The
positive aspect about employing GCP and GCN is that given that power
systems are mainly inductive (i.e., 𝑅∕𝑋 ratios tend to be small) [50],
the injection of zero active power is not that crucial, as shown in Fig. 8
for 𝑅1 ∕𝑋1 < 1.

3.3. Cable length variation analysis

Fig. 9 represents the third parametric study regarding single con-

verter systems, which deals with a hypothetical submarine cable mod-
elled with its 𝜋 equivalent. The data selected to model the cable are
extracted from [51] and adapted to per-unit values, which are shown
in Table 3. The base values are a power of 100 MVA and a voltage of
220 kV. The followed approach to analyse the system is identical to
the previous ones, except for the reduction to a Thévenin equivalent
of the set of elements including the cable and the grid. With this
reduction, the system in Fig. 9 takes the form of the system in Fig. 5.
The corresponding Thévenin equivalent is defined by:

⎧𝑉 ′ = 𝑉 𝑍𝑝𝑍𝑝
⎪ 𝑡 𝑡 2𝑍 𝑍 +𝑍 𝑍 +𝑍 𝑍 +𝑍 𝑍
𝑡 𝑝 𝑝 𝑠 𝑝 𝑝 𝑡 𝑠
⎨ (30)
𝑍 𝑍 𝑍 +𝑍 𝑍 𝑍 +𝑍 𝑍 𝑍
⎪𝑍 ′ = 𝑝 𝑝 𝑡 𝑝 𝑝 𝑠 𝑝 𝑠 𝑡
⎩ 𝑡 2𝑍 𝑍 +𝑍 𝑍 +𝑍 𝑍 +𝑍 𝑍
𝑝 𝑡 𝑠 𝑝 𝑠 𝑡 𝑝 𝑝
Fig. 11. Influence of the currents for a line-to-line fault with 𝑍 𝑎𝑏 = 0.1 and a varying
cable distance, one-converter case. Similarly as before, if a Norton equivalent is preferred, the current
injection 𝐼 ′𝑡 is simply equal to 𝑉 ′𝑡 ∕𝑍 ′𝑡 .
The analysis performed considers a cable with a varying length,
objective functions 𝑓𝑜 are nearly the same. However, the oscillating up to 100 km. Its impact on the Thévenin equivalent parameters is
power terms 𝑝𝑐 and 𝑝𝑠 are more prominent for OPT and GCP than for shown in Fig. 10. The voltage experiences a subtle increase, while the
GCN and GC. The fundamental grid code GC is usually the suboptimal equivalent impedance at 100 km almost doubles the one at 0 km.
+ The four strategies are again evaluated, as depicted in Fig. 11. A
choice. Even though it specifies a similar 𝐼𝑖𝑚 current compared to GCP,
the negative sequence voltage is not minimized since by definition line-to-line fault between phases 𝑎 and 𝑏 with an impedance of 0.1 pu
𝐼 − = 0. takes place. Increasing the cable distance causes the negative sequence
voltage to grow, while the positive sequence voltage also tends to grow.
3.2. 𝑅1 ∕𝑋1 variation analysis This phenomenon mostly differs from the profiles obtained in Figs. 7
and 8, where both voltages either simultaneously approached the ob-
The 𝑅1 ∕𝑋1 variation analysis performs a sweep for a range of jective value (1 and 0 respectively), or distanced from it. Consequently,
a varying angle of the 𝑍 𝑣1 impedance, while the absolute value is in the cases of the grid code implementation, larger distances imply a
+
preserved. The goal of this study is to determine how this affects the lower 𝐼𝑖𝑚 current, while the less convenient growing negative sequence
distribution of currents between the real and the imaginary part. The voltages force an increment in 𝐼𝑖𝑚 − . Most of the current capability is

results in Fig. 8 for the OPT case show that despite an increase in the precisely devoted to this negative sequence imaginary current. The
resistive part of the impedance, the currents remain nearly constant for OPT option achieves a more favourable objective function value than

10
J. Fanals-Batllori et al. International Journal of Electrical Power and Energy Systems 148 (2023) 109000

Fig. 12. Simplistic single-phase representation of the two-converter IEEE 9-bus system under study.

the rest by keeping the positive sequence currents practically constant, As can be deduced from Fig. 12, a short-circuit fault is provoked
− (in absolute value), and decreasing 𝐼 − , which is the
increasing 𝐼𝑟𝑒 in bus 8. Fig. 13 shows the most representative results for this line-
𝑖𝑚
inverse trend followed by GCP and GCN. Again, 𝑉 − is large for the GC to-line short-circuit fault with different fault impedances, where the
option and consequently there is a non-negligible margin between the plotted voltages correspond to the positive and negative sequence of
objective function 𝑓𝑜 of GC and the other strategies. Nonetheless, the 𝑉 𝑝1 and 𝑉 𝑝2 . The OPT choice prioritizes 𝐼 + , mainly its imaginary part.
oscillating power elements 𝑝𝑐 and 𝑝𝑠 for GC are significantly inferior This helps to obtain a higher 𝑉 + . Contrarily, the suggested grid codes
than the ones obtained for OPT, GCP and GCN. GCP and GCN inject a larger 𝐼 − current, which of course, is purely
imaginary. Their associated 𝑉 + values are inferior to the ones obtained
4. Two-converter case study with OPT, yet their 𝑉 − voltage is also smaller. As a result, there is
little difference in the objective function, which makes GCP and GCN
This Section presents a two-converter case in order to spot the near-optimal strategies. The decision to choose between GCP and GCN
may be based on other factors such as the limitations the sequence
differences in the distribution of currents between both converters, and
components impose on protective relays. On the other hand, the basic
at the same time, evaluate the feasibility of the suggested strategies.
grid code GC injects a small current as the fault is not extremely severe.
Fig. 12 shows a single-phase representation of the system under study.
Only about 25% of the full current capability of the converters is
It corresponds to an adaptation of the IEEE 9-bus system where the
employed. Thus, there are relevant differences in the objective function
generators in buses 2 and 3 have been replaced by VSCs. PQ loads
between GC and the other options.
were originally connected to buses 4 to 9; they have been disconnected
A dynamic simulation has been performed in order to validate
during the fault for protection. Bus 1 remains the slack bus. The
the results in this two-converter scenario. A purely resistive fault
connections between the converters and the slack bus with respect to
impedance of 0.01 p.u. has been considered. Fig. 14 shows the evo-
the boxed grid are achieved through transformers which are modelled
lution of the voltages for a fault at 𝑡 = 0.5 s. The steady-state voltages
by a single impedance. This boxed grid represents the passive part of
calculated with the proposed methodology are plotted in black dashed
the grid. Its internal connections are modelled with an equivalent 𝜋
lines, and their values are annotated as well. It can be concluded that
circuit. All data have been extracted from [52]. the steady-state results are correct since the dynamic simulation tends
Analogous to the one-converter situation, the optimization problem to converge to the same operating points. Matlab/Simulink has been
becomes: the chosen tool to generate the presented dynamic results.
min 𝑓𝑜 = 𝜆+
1
+
|(1 − 𝑉𝑝1 (𝐈))| + 𝜆−
1
−
|(0 − 𝑉𝑝1 (𝐈))| As a complement, Fig. 15 displays the dynamics regarding the
𝐈𝟏 ,𝐈𝟐 ∈𝐈
active and reactive powers generated by both converters. Under normal
+ 𝜆+
2
+
|(1 − 𝑉𝑝2 (𝐈))| + 𝜆−
2
−
|(0 − 𝑉𝑝2 (𝐈))|, situations, the active power is a non-zero positive value indicating the
(31)
s.t. max(𝐼1𝑎 , 𝐼1𝑏 , 𝐼1𝑐 ) ≤ 𝐼max,1 , injection of power from the hypothetical renewable sources the VSCs
could be connected to. Once the fault occurs, the constant active power
max(𝐼2𝑎 , 𝐼2𝑏 , 𝐼2𝑐 ) ≤ 𝐼max,2 ,
𝑝 goes to zero for all strategies. In spite of that, oscillations appear
where 𝐈𝟏 and 𝐈𝟐 are respectively the currents injected by converters 1 due to the unbalanced conditions the system is facing. These active
and 2, 𝐈 symbolizes all current injections as in the left-hand side of (5), power oscillations are larger for GCP and GCN compared to GC, an
and all 𝜆 have been set to 1.0. The application of the GC, GCP and GCN unsurprising result considering that their current references are also
strategies is the same as before, in the sense that the injection of current significantly larger. Neglecting the losses in the semiconductors of the
𝐈𝟏 only depends on 𝑉 𝑝1 , and 𝐈𝟐 is expressed as a function of 𝑉 𝑝2 . converters, the DC bus power waveform would coincide with the ones

11
J. Fanals-Batllori et al. International Journal of Electrical Power and Energy Systems 148 (2023) 109000

Fig. 14. Dynamic validation to check the obtained steady-state results for 𝑍 𝑓 = 0.01
in the IEEE 9-bus system.

Fig. 13. Influence of the currents for the line-to-line fault with a varying fault
impedance in bus 8 of the IEEE 9-bus system.

displayed in Fig. 15. On the other side, the reactive power is initialized
at zero yet it increases to provide voltage-support. The average value
is positive and strong fluctuations are present.
The instantaneous phase voltages 𝑣 and currents 𝑖 for GC, GCP
and GCN are shown respectively in Figs. 16, 17, 18. Differences in
the waveforms between GCP and GCN are almost imperceptible. The
faulted phases 𝑎 and 𝑏 present similar voltage magnitudes whereas
phase 𝑐 suffers an increase in voltage. The root-mean-square (RMS)
currents injected by the VSCs are bounded between −1 and 1 in steady-
state, just as expected considering 𝐼max = 1. The strategy GC only
injects positive sequence current, hence the balanced waveform. The
magnitudes of these currents do not reach 𝐼max due to the fault severity Fig. 15. Evolution of the active and reactive powers injected by the VSCs for 𝑍 𝑓 = 0.01
in the IEEE 9-bus system.
not being extreme.

12
J. Fanals-Batllori et al. International Journal of Electrical Power and Energy Systems 148 (2023) 109000

Fig. 16. Evolution of the voltage and currents waveforms for GC corresponding to the Fig. 18. Evolution of the voltage and currents waveforms for GCN corresponding to
VSCs for 𝑍 𝑓 = 0.01 in the IEEE 9-bus system. the VSCs for 𝑍 𝑓 = 0.01 in the IEEE 9-bus system.

is possible to improve the voltage-support with arguably simple rules

where only reactive current/power is injected from power converters.
The optimal situation is not suitable to be implemented in reality as
it requires knowledge about the fault impedance and a high compu-
tational effort. However, GCP and GCN can be easily implemented in
reality since they are of the same nature as traditional grid codes. Due
to the ease of implementation and the near-optimal voltage support, the
adapted grid codes GCP and GCN are considered the most convenient
choices.

CRediT authorship contribution statement

Josep Fanals-Batllori: Methodology, Software, Writing – original

draft. Jie Song: Investigation, Methodology, Writing – review & edit-
ing. Marc Cheah-Mañé: Methodology, Writing – review & editing.
Eduardo Prieto-Araujo: Conceptualization, Writing – review & edit-
ing. Oriol Gomis-Bellmunt: Conceptualization, Writing – review &
editing, Supervision.

Fig. 17. Evolution of the voltage and currents waveforms for GCP corresponding to Declaration of competing interest
the VSCs for 𝑍 𝑓 = 0.01 in the IEEE 9-bus system.
The authors declare that they have no known competing finan-
cial interests or personal relationships that could have appeared to
5. Conclusion influence the work reported in this paper.

Data availability
The system operation points during the fault achieved with con-
ventional grid codes have been compared with the optimal solution in
The numerical data have been indicated along the document in
systems that integrate converters considering limitations. In the optimal
some cases, and in others, the references containing them have been
solution, the positive and negative sequence voltages approach as much
cited.
as possible the nominal voltage and a zero voltage respectively. Two
strategies have been derived to respectively prioritize positive (GCP) References
and negative (GCN) currents in saturated conditions, therefore max-
imizing the voltage-support. These two strategies are superior to the [1] Pan Donghua, Wang Xiongfei, Liu Fangcheng, Shi Rongliang. Transient stability
basic grid code. Although there are certain differences in the currents of voltage-source converters with grid-forming control: A design-oriented study.
for GCP and GCN, both of them provide almost the same objective IEEE J Emerg Sel Top Power Electron 2019;8(2):1019–33.
[2] Karimi Houshang, Karimi-Ghartemani Masoud, Sheshyekani Keyhan. Robust con-
function value. Several parameters have been swept, such as the fault trol of three-phase voltage source converters under unbalanced grid conditions.
impedance, the angle of the interconnecting impedance, and the length IEEE Trans Power Electron 2019;34(11):11278–89.
of a cable. Generally, the distribution of currents derived from the [3] Li Chang, Cao Yijia, Yang Yaqian, Wang Lei, Blaabjerg Frede, Dragicevic Tomis-
optimization differs considerably from the currents injected following lav. Impedance-based method for DC stability of VSC-HVDC system with VSG
control. Int J Electr Power Energy Syst 2021;130:106975.
the conventional grid code. However, the proposed approaches GCP [4] Abdou AF, Abu-Siada Ahmed, Pota HR. Improving the low voltage ride through
and GCN offer an objective function value that is usually closer from of doubly fed induction generator during intermittent voltage source converter
the optimal than from the basic grid code. This work concludes that it faults. J Renew Sustain Energy 2013;5(4):043110.

13
J. Fanals-Batllori et al. International Journal of Electrical Power and Energy Systems 148 (2023) 109000

[5] Todorović Ivan, Grabić Stevan, Ivanović Zoran. Grid-connected converter active [27] Taul Mads Graungaard, Wang Xiongfei, Davari Pooya, Blaabjerg Frede. Current
and reactive power production maximization with respect to current limitations reference generation based on next-generation grid code requirements of grid-
during grid faults. Int J Electr Power Energy Syst 2018;101:311–22. tied converters during asymmetrical faults. IEEE J Emerg Sel Top Power Electron
[6] Etxegarai Agurtzane, Eguia Pablo, Torres Esther, Iturregi Araitz, Valverde Victor. 2019;8(4):3784–97.
Review of grid connection requirements for generation assets in weak power [28] IEEE standard for interconnection and interoperability of inverter-based resources
grids. Renew Sustain Energy Rev 2015;41:1501–14. (IBRs) interconnecting with associated transmission electric power systems. IEEE
[7] Energinet DK. Technical regulation 3.2. 5 for wind power plants above 11 kw. Power and Energy Society; 2022.
Tech. Rep., 2016. [29] Dağ Bülent, Boynueğri Ali Rıfat, Ateş Yavuz, Karakaş Arif, Nadar Abdullah,
[8] Niu Lele, Wang Xiao, Wu Linlin, Yan Feng, Xu Man. Review of low Uzunoğlu Mehmet. Static modeling of microgrids for load flow and fault analysis.
voltage ride-through technology of doubly-fed induction generator. J Eng IEEE Trans Power Syst 2016;32(3):1990–2000.
2019;2019(16):3106–8. [30] Song Jie, Cheah-Mane Marc, Prieto-Araujo Eduardo, Gomis-Bellmunt Oriol.
[9] Zarei Seyed Fariborz, Mokhtari Hossein, Ghasemi Mohammad Amin, Blaab- Short-circuit analysis of ac distribution systems dominated by voltage source
jerg Frede. Reinforcing fault ride through capability of grid forming voltage converters considering converter limitations. IEEE Trans Smart Grid 2021.
source converters using an enhanced voltage control scheme. IEEE Trans Power [31] Wu Tianhao, Jiang Qirong, Huang Meng, Xie Xiaorong. Synchronization stability
Deliv 2018;34(5):1827–42. of grid-following converters governed by saturation nonlinearities. IEEE Trans
[10] Mohseni Mansour, Islam Syed M. Review of international grid codes for wind Power Syst 2022.
power integration: Diversity, technology and a case for global standard. Renew [32] Huang Sen, Yao Jun, Pei Jinxin, Chen Shiyue, Luo Yi, Chen Zhaoyang. Transient
Sustain Energy Rev 2012;16(6):3876–90. synchronization stability improvement control strategy for grid-connected VSC
[11] Haddadi Aboutaleb, Kocar Ilhan, Mahseredjian Jean, Karaagac Ulas, Faran- under symmetrical grid fault. IEEE Trans Power Electron 2021;37(5):4957–61.
tatos Evangelos. Negative sequence quantities-based protection under inverter- [33] Zhuang Kehao, Xin Huanhai, Hu Pengfei, Wang Zijun. Current saturation analysis
based resources challenges and impact of the German grid code. Electr Power and anti-windup control design of grid-forming voltage source converter. IEEE
Syst Res 2020;188:106573. Trans Energy Convers 2022.
[12] Wieserman Laura, McDermott TE. Fault current and overvoltage calculations for [34] Peralta Jaime, Saad Hani, Dennetière Sébastien, Mahseredjian Jean. Dynamic
inverter-based generation using symmetrical components. In: 2014 IEEE energy performance of average-value models for multi-terminal VSC-HVDC systems. In:
conversion congress and exposition. ECCE, IEEE; 2014, p. 2619–24. 2012 IEEE power and energy society general meeting. IEEE; 2012, p. 1–8.
[13] Camacho Antonio, Castilla Miguel, Miret Jaume, de Vicuña Luis García, Guz- [35] Andani Majid Taheri, Pourgharibshahi Hamed, Ramezani Zahra,
man Ramon. Positive and negative sequence control strategies to maximize the Zargarzadeh Hassan. Controller design for voltage-source converter using
voltage support in resistive–inductive grids during grid faults. IEEE Trans Power LQG/LTR. In: 2018 IEEE texas power and energy conference. TPEC, IEEE; 2018,
Electron 2017;33(6):5362–73. p. 1–6.
[14] Mortazavian Shahed, Shabestary Masoud M, Mohamed Yasser Abdel-Rady I. [36] Taul Mads Graungaard, Golestan Saeed, Wang Xiongfei, Davari Pooya, Blaab-
Analysis and dynamic performance improvement of grid-connected voltage– jerg Frede. Modeling of converter synchronization stability under grid faults:
source converters under unbalanced network conditions. IEEE Trans Power The general case. IEEE J Emerg Sel Top Power Electron 2020.
Electron 2016;32(10):8134–49. [37] Fortescue Charles L. Method of symmetrical co-ordinates applied to the solution
[15] Abbassi Rabeh, Marrouchi Sahbi, Saidi Salem, Abbassi Abdelkader, of polyphase networks. Trans Am Inst Electr Eng 1918;37(2):1027–140.
Chebbi Souad. Optimal energy management strategy and novel control [38] Errouissi Rachid, Al-Durra Ahmed. Disturbance-observer-based control for dual-
approach for DPGSs under unbalanced grid faults. J Circuits Syst Comput stage grid-tied photovoltaic system under unbalanced grid voltages. IEEE Trans
2019;28(04):1950057. Ind Electron 2018;66(11):8925–36.
[16] Shabestary Masoud M, Mortazavian Shahed, Mohamed Yasser I. Overview [39] Saadat Nima, Choi San Shing, Vilathgamuwa D Mahinda. A statistical eval-
of voltage support strategies in grid-connected VSCs under unbalanced grid uation of the capability of distributed renewable generator-energy-storage
faults considering LVRT and HVRT requirements. In: 2018 IEEE international system in providing load low-voltage ride-through. IEEE Trans Power Deliv
conference on smart energy grid engineering. SEGE, IEEE; 2018, p. 145–9. 2014;30(3):1128–36.
[17] Jia Jundi, Yang Guangya, Nielsen Arne Hejde. A review on grid-connected [40] Nasiri Mojtaba, Arzani Ali. Robust control scheme for the braking chopper of
converter control for short-circuit power provision under grid unbalanced faults. PMSG-based wind turbines–A comparative assessment. Int J Electr Power Energy
IEEE Trans Power Deliv 2017;33(2):649–61. Syst 2022;134:107322.
[18] de España Red Eléctrica. Información sobre implementación de códigos de red [41] Al-Shetwi Ali Q, Sujod Muhamad Zahim, Blaabjerg Frede. Low voltage
de conexión. textos de los códigos de red de conexió europeos. Reglamento ride-through capability control for single-stage inverter-based grid-connected
2016/631. 2016, URL https://www.esios.ree.es/es/pagina/codigos-red-conexion. photovoltaic power plant. Sol Energy 2018;159:665–81.
[19] Paspatis Alexandros G, Konstantopoulos George C. Voltage support under grid [42] Akagi Hirofumi, Watanabe Edson Hirokazu, Aredes Mauricio. Instantaneous
faults with inherent current limitation for three-phase droop-controlled inverters. power theory and applications to power conditioning. John Wiley and Sons;
Energies 2019;12(6):997. 2017.
[20] Camacho Antonio, Castilla Miguel, Miret Jaume, Vasquez Juan C, Alarcon- [43] Nasiri Mojtaba, Mohammadi Reza. Peak current limitation for grid side inverter
Gallo Eduardo. Flexible voltage support control for three-phase distributed by limited active power in PMSG-based wind turbines during different grid faults.
generation inverters under grid fault. IEEE Trans Ind Electron 2012;60(4):1429– IEEE Trans Sustain Energy 2016;8(1):3–12.
41. [44] Nasiri Mojtaba, Arzani Ali, McCormack Sarah J. A simple and effective grid-
[21] Guo Xiaoqiang, Zhang Xue, Wang Baocheng, Wu Weiyang, Guerrero Josep M. supporting low voltage ride-through scheme for single-stage photovoltaic power
Asymmetrical grid fault ride-through strategy of three-phase grid-connected plants. Sol Energy 2022;232:248–62.
inverter considering network impedance impact in low-voltage grid. IEEE Trans [45] Grid National. The grid code, issue 6, revision 1. 2021, https://www.
Power Electron 2013;29(3):1064–8. nationalgrideso.com/document/162271/download.
[22] Shabestary Masoud M, Mohamed Yasser Abdel-Rady I. An analytical method to [46] Moschakis MN, Hatziargyriou ND. Analytical calculation and stochastic
obtain maximum allowable grid support by using grid-connected converters. IEEE assessment of voltage sags. IEEE Trans Power Deliv 2006;21(3):1727–34.
Trans Sustain Energy 2016;7(4):1558–71. [47] Mohseni Mansour, Masoum Mohammad AS, Islam Syed. Emergency control of
[23] Dai Zhiyong, Lin Hui, Yin Hang, Qiu Yanan. A novel method for voltage support DFIG-based wind turbines to meet new European grid code requirements. In:
control under unbalanced grid faults and grid harmonic voltage disturbances. IET ISGT 2011. IEEE; 2011, p. 1–6.
Power Electron 2015;8(8):1377–85. [48] McKerns Michael M, Strand Leif, Sullivan Tim, Fang Alta, Aivazis Michael AG.
[24] Shabestary Masoud M, Mohamed Yasser Abdel-Rady I. Asymmetrical ride- Building a framework for predictive science. 2012, arXiv preprint arXiv:1202.
through and grid support in converter-interfaced DG units under unbalanced 1056.
conditions. IEEE Trans Ind Electron 2018;66(2):1130–41. [49] McKerns M, Hung P, Aivazis M. Mystic: highly-constrained non-convex
[25] Nejabatkhah Farzam, Li Yun Wei, Wu Bin. Control strategies of three-phase optimization and UQ. 2009.
distributed generation inverters for grid unbalanced voltage compensation. IEEE [50] Bian Zhipeng, Xu Zheng. Fault ride-through capability enhancement strategy
Trans Power Electron 2015;31(7):5228–41. for VSC-HVDC systems supplying for passive industrial installations. IEEE Trans
[26] Hasan Shamim, Agarwal Vivek. An unconstrained voltage support scheme for Power Deliv 2016;31(4):1673–82.
distributed generation connected to resistive-inductive grid under unbalanced [51] Cheah Marc. Offshore wind integration through high voltage direct current
conditions. IEEE Trans Ind Appl 2021;57(4):4253–62. systems (Ph.D. thesis), Cardiff University; 2017.
[52] PSCAD. Knowledge base. IEEE 09 bus system. 2021, https://www.pscad.com/
knowledge-base/article/25.

14