IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 38, NO.

4, JULY 2023 3279

Decentralized Robust Frequency Regulation of

Multi-terminal HVDC-linked Grids
Jun-Soo Kim , Student Member, IEEE, Jae-Young Park , Member, IEEE, Young-Jin Kim , Senior Member, IEEE,
and Oriol Gomis-Bellmunt , Fellow, IEEE

Abstract—This paper proposes a new strategy for optimal grid delivering offshore wind power to onshore power grids and
frequency regulation (FR) in an interconnected power system sharing it between interconnected regional grids [3], [4].
where regional ac grids and an offshore wind farm are linked via In HVDC technology, line-commutated converter (LCC) and
a multi-terminal high voltage direct-current (MTDC) network. In
the proposed strategy, decentralized H controllers are developed voltage source converter (VSC) have been widely used due to
to coordinate the operations of synchronous generators and MTDC their distinct characteristics. An LCC is characterized by, for
converters, thus achieving optimal power sharing of interconnected example, a high capacity for bulk power transmission, low power
ac grids and minimizing frequency deviations in each grid. To de- loss, and low installation cost [5]. However, it operates only as
velop the controllers, robust optimization problems are formulated a rectifier or an inverter that can control either dc voltage or
and solved using a dynamic model of the hybrid MTDC-linked grids
with model parameter uncertainty and decentralized control inputs dc current [6]. By contrast, a VSC can change the direction of
and outputs. The model orders of the controllers are then reduced dc current and independently control active and reactive power,
using a balanced truncation algorithm to eliminate unobservable although it has lower power rating and higher power loss even
and uncontrollable state variables while preserving their dominant when implemented in the form of modular multilevel converter
response characteris- tics. Sensitivity and eigenvalue analyses are [7]. In several projects, there is a recent interest in combining the
conducted focusing on the effects of grid measurements, parameter
uncertainty levels, and communication time delays. Comparative advantages of both LCC and VSC, developing hybrid schemes.
case studies are also carried out to verify that the proposed strategy An HVDC system can be further developed in the form of a
improves the effectiveness, stability, and robustness of real-time FR multi-terminal dc (MTDC) system [9], [10], [11], [12], [13],
in MTDC-linked grids under various conditions. [14], for example, where LCCs, grid-side VSCs (GSVSCs),
Index Terms—Grid frequency regulation, H infinity, hybrid and OWF-side VSCs (WFVSCs) are linked via a dc network
MTDC converters, offshore wind farm, optimal power sharing. to connect regional ac grids with OWFs. This enhances the
flexibility of inter-grid power delivery and sharing, thereby
facilitating the integration of large-scale OWFs with regional
I. INTRODUCTION load centers. However, due to the interconnection, each grid
USTAINABLE wind energy is critical to reducing fossil frequency is affected by power imbalances not only in the
S fuel emissions and mitigating global climate change [1].
In particular, the capacity of offshore wind farms (OWFs) is
corresponding grid but also in other grids [8], implying that
intermittent wind power causes freq- uency deviations in all
growing faster for many reasons, including a lack of suit- interconnected grids. This motivates the development of a proper
able onshore sites and better wind conditions of offshore sites strategy for power balancing in MTDC- linked grids, essentially
[2]. Meanwhile, high-voltage direct-current (HVDC) technol- requiring the coordinated control of regional power generation
ogy has been increasingly considered a promising solution for and ac-to-dc power transmission.
Several studies have been conducted on frequency regulation
(FR) in MTDC-linked grids. For example, in [9], VSCs were
Manuscript received 7 December 2021; revised 13 June 2022 and 8 August
2022; accepted 21 August 2022. Date of publication 24 August 2022; date of
droop-controlled to mitigate grid frequency deviations and dc
current version 22 June 2023. This work was supported by the National Research voltage variations resulting from wind power generation. In [10],
Foundation of Korea through Energy Cloud Research and Development Program communication-free coordination of VSC terminals was accom-
funded by the Ministry of Science, ICT under Grant NRF-2019M3F2A1073402.
Paper no. TPWRS-01886-2021. (Corresponding author: Young-Jin Kim.)
plished using a Vdc -f droop control scheme. In [11], adaptive
Jun-Soo Kim and Young-Jin Kim are with the Department of Electrical En- droop control of a VSC-MTDC system was discussed, wherein
gineering, Pohang University of Science and Technology (POSTECH), Pohang the Vdc -Idc -f characteristics of the system were analyzed to
37673, South Korea (e-mail: [email protected]; [email protected]).
Jae-Young Park is with the Korea Institute of Energy Research (KIER),
deter- mine the droop control gains. However, in [9], [10], [11],
Daejeon 34129, South Korea (e-mail: [email protected]). FR relied mainly on droop control or, equivalently, primary fre-
Oriol Gomis-Bellmunt is with the Centre d’Innovacio Tecnologica en quency control, rather than optimal secondary frequency control
Convertidors Estatics i Accionaments, Departament d’Enginyeria Electrica,
Universitat Politecnica de Catalunya, 08028 Barcelona, Spain (e-mail:
(SFC).
[email protected]). The optimal FR of MTDC-linked grids has been discussed in
Color versions of one or more figures in this article are available at recent studies (e.g., [12], [13], [14]). In [12], a model predictive
https://doi.org/10.1109/TPWRS.2022.3201316.
Digital Object Identifier 10.1109/TPWRS.2022.3201316
control (MPC) algorithm was adopted for the optimal power

0885-8950 © 2022 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: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on May 06,2024 at 05:50:37 UTC from IEEE Xplore. Restrictions apply.
 3280 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 38, NO. 4, JULY 2023

sharing in a VSC-MTDC system, wherein a discrete-time op- The main contributions of this paper are summarized below:
timization problem was formulated considering grid operating r To the best of our knowledge, this is the first study reporting
costs. In [13] and [14], the MPC-based power sharing was also the optimal robust FR of MTDC-linked grids using decen-
achieved to minimize the frequency deviations in the MTDC- tralized H controllers. Each H controller is designed
linked grids with only VSCs and LCCs, respectively. However, with reduced require- ment on inter-grid measurements and
in [12], [13], [14], hybrid MTDC systems were not considered. communications.
Moreover, in general, MPC is computationally intensive, be- r Hybrid MTDC-linked grids are modeled with decentral-
cause an optimization problem for MPC needs to be solved ized control inputs and outputs. For each grid, considering
directly at every time step. MPC is also likely to cause oscil- model parameter uncertainty, an optimization problem is
latory and even unstable operation of MTDC-linked grids due then formul- ated to determine the corresponding optimal
to communication time delays and model parameter uncertainty robust control gains.
[15]. Therefore, the optimal FR strategies discussed in [12], [13], r The model orders of the optimal decentralized H con-
[14] need to be further analyzed under practical conditions of trollers are reduced using a balanced truncation algorithm,
MTDC systems, for example, with regard to hybrid converters, facilitating practical applications of the proposed FR strat-
communications links, and system parameter estimates. egy.
When system parameters are uncertain, robust controllers can r The proposed strategy is comprehensively evaluated via
effectively reduce the effect of disturbance on system operation. comparison to conventional decentralized PI and H
For example, in [16], the robust dc voltage control of a VSC controller- based strategies with regard to effectiveness and
was achieved by establishing an H -mixed sensitivity function robustness against time-varying power imbalance, commu-
in a linear matrix inequality framework, wherein dc current nication failures and time delays, and different weighting
injection was set as a disturbance input. In [17], a systematic function models.
method was provided to design the H controller of a VSC for
optimal and robust control of active power and ac voltage. In [18] II. HYBRID MTDC-LINKED GRIDS
and [19], H2 and H controllers were discussed to optimally
Fig. 1 shows a schematic diagram of the proposed decen-
regulate the dc-link voltage and the converter internal energy of
tralized strategy for the optimal robust FR in MTDC-linked
an HVDC system, respectively. However, in [16], [17], [18],
grids, where regional ac grids and OWFs are interconnected
[19], robust control was applied to only a single converter;
with each other using hybrid MTDC converters (i.e., VSCs
the coordination with synchronous generators and the power
and LCCs). Each grid includes multiple synchronous generators
sharing between regional ac grids were not considered. In [19], a
(SGs), loads, and ac transmission lines, and each OWF contains
centralized optimization problem was formulated for centralized
several wind turbine generators (WTGs) using permanent mag-
and decentralized H control of the dc-link voltage in a modular
net synchro- nous generators (PMSGs). Note that for brevity,
multilevel converter when applied to a VSC-HVDC system. The
Fig. 1 illustrates an SG and a WTG. Moreover, for each WTG, a
centralized controller slightly outperformed the decentralized
back-to-back converter and its maximum-power-point-tracking
controller. However, in practice, such a centralized controller is
controller are modeled using simple equations for power balance
difficult to implement and apply for MTDC-linked grids due to
[35]. For each regional grid, a decentralized control structure is
computa- tional and communication overheads.
developed to establish the optimal control inputs for the local
This paper proposes a new decentralized strategy for optimal
controllers of the SGs and the MTDC converter and adjust
robust FR in regional ac grids that are interconnected via a hybrid
the generated and transmitted power in a coordinated manner,
MTDC system including LCCs and VSCs. For each ac grid, a
while receiving measurements from the corresponding grid and
decentralized H controller is developed for coordinated control
interconnected grids via communication systems with time de-
of generated and transmitted power, optimizing inter-grid power
lays [37], [38], [39]. Specifically, the decentralized controller k
sharing and thus minimizing frequency deviations. Specifically,
receives the measurements of ΔPgk , Δω rk , Δfk , ΔVmagk , ΔPk ,
for the proposed strategy, a dynamic model of hybrid MTDC-
and ΔVdck from grid k and the measurements of Δfj and ΔVdcj
linked grids is implemented with decentralized control inputs
from interconnected grids j  k, as shown in Fig. 1 and discussed
and outputs. Given the dynamic model, a robust optimization
in Appendix.
problem for the optimal FR is formulated considering model
parameter uncertainty. The solution to the problem leads to the
optimal gain of the decentralized H controller. A balanced trun- A. Modeling of Hybrid MTDC-Linked Grids
cation algorithm is then applied to reduce the controller model Without loss of generality, this paper considers three regional
order, thus facilitating its practical implementation and the ref- ac grids and one PMSG-type OWF, all of which are linked to
erence signal generation. Sensitivity and eigenvalue analyses a dc network via two GSVSCs, one LCC, and one WFVSC.
are conducted to verify grid frequency stability in the proposed Specifically, Grids 1 and 2 are interfaced using GSVSCs and
strategy, focusing on the effects of system parameters, parameter Grid 3 is connected with LCC at the points of common coupling
uncertainty, and communication time delay. Comparative case (PCCs): i.e., n = 2 and m = 3 in Fig. 1. The GSVSCs are equipped
studies are also carried out to verify that the proposed strategy with local P-Vdc droop controllers, so that power transmitted
improves the effectiveness and robustness in reducing frequency from Grids 1 and 2 are instantaneously adjusted according to
deviations under various grid conditions. variations in the corresponding dc terminal voltages. On the

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on May 06,2024 at 05:50:37 UTC from IEEE Xplore. Restrictions apply.
 KIM et al.: DECENTRALIZED ROBUST FREQUENCY REGULATION OF MULTI-TERMINAL HVDC-LINKED GRIDS 3281

Fig. 1. A schematic diagram of the proposed decentralized strategy for optimal FR in MTDC-linked grids.

other hand, the LCC operates as a rectifier with a local controller T

r = [rT1 , rT2 , rT3 ] , wd = [ΔPL1 , ΔPL2 , ΔPL3 , ΔVW ]T ,
that transmits constant or time-varying power from the regional (7)
grid to the dc network. Note that the modeling and control
procedures, discussed in Sections II and III, can readily be Zdc = [Idc1 , Idc2 , Idc3 , Idc4 , Vdc ]T . (8)
applied to a general case with more than three regional grids.
The detailed procedure for (1)–(8) is provided in Appendix.
A dynamic model of the hybrid MTDC-linked grids is im-
In (6), Xk includes the frequency fk and dc terminal voltage Vdck
plemented by integrating the individual models of the ac grids,
for each grid k. Moreover, Xdc includes the dc currents Idc1–4
OWF, and dc network, as discussed in Appendix, as:
[see (8)] that flow through the dc transmission lines, as:
⎡ ⎤
Ẋ = AC · X + Br · r + Bw · wd + EC · Zdc , (1) 1 1 0 0
⎢ −1 0 1 0 ⎥
Y = C · X, (2) Idc1−4 = ⎢ ⎥
⎣ 0 −1 0 1 ⎦ · Xdc = Fdci · Xdc . (9)

where AC = diag(A1 , A2 , A3 , Aowf , Adc ), (3) 0 0 −1 −1

⎡ ⎤ ⎡ ⎤ Similarly, Vdc also can be extracted from X1–3 and Xowf
Br1 Bw1 T
⎢ B ⎥ ⎢ Bw2 ⎥ as Vdc = Fdcv · [XT1−3 XTowf ] . Then, Zdc can be represented
⎢ r2 ⎥ ⎢ ⎥
Br = ⎢⎢ ⎥
Br3 ⎥ , Bw = ⎢ ⎢ Bw3 ⎥, using X as:
⎥ 
⎣ O O O ⎦ ⎣ Bowf ⎦ Fdci
O O O O O O O Zdc = · X = Fdc · X. (10)
Fdcv
(4)
⎡ ⎤ Given (10), (1) can be expressed in a standard form as:
C1 O O
Ẋ = (AC + EC · Fdc ) · X + Br · r + Bw · wd
C=⎣ C2 O O⎦,
C3 O O = A · X + B r · r + B w · wd . (11)
EC = diag(E1 , E2 , E3 , Eowf , Edcv ), (5) When each grid reaches a steady state after disturbance, fk
T T
and Vdck should be restored to the nominal values. For the
X = [XT1 , XT2 , XT3 , XTowf , XTdc ] , Y = [Y1T , Y2T , Y3T ] , restoration, the MTDC-linked grid models (2) and (11) are
(6) augmented by combining the integrals of Δfk and ΔVdck for

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on May 06,2024 at 05:50:37 UTC from IEEE Xplore. Restrictions apply.
 3282 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 38, NO. 4, JULY 2023

k, as:

ẊE = AE · XE + BrEk · rk + [BwE BrEj ] · [wd rj ]T

= AE · XE + BrEk · rk + Bdrk · wdrk , ∀k and j = k.
(17)

III. OPTIMAL ROBUST FREQUENCY REGULATION

A. Design of an Optimal Decentralized H Controller
Fig. 2. A schematic diagram of the decentralized H controller for ac grid k.
Fig. 2 shows a closed-loop model of MTDC-linked grid k with
decentralized controller k. A general time-domain model of the
decentralized controller k can be established as:
all k with X and Y as:
ẊE = AE · XE + BrE · r + BwE · wd , (12) Ẋhk = Ahk · Xhk + Bhk · YTk

YE = CE · XE , (13) rk = Chk · Xhk + Dhk · YTk , (18)

with coefficient matrices augmented as: and, consequently, its frequency-domain model becomes:
 
A O Br rk = [Chk · (sI − Ahk )−1 · Bhk + Dhk ] · YTk . (19)
AE = , BrE = , 
As O O Kk (s)
 
Bw CO Specifically, for the optimal robust FR in grid k, the con-
BwE = , CE = , (14)
O O I troller k minimizes the target performance outputs Zek and
where inputs Zuk against disturbance dk . In the proposed strategy,
XE = [XT , Δfk dt, ΔVdck dt] , ∀k,
T
(15) Zek includes not only the weighted values of Δfk and Δfk
T dt but also the weighted values of ΔVdck and ΔVdck dt to
YE = [YT , Δfk dt, ΔVdck dt] , ∀k.
prevent excessive variations in dc terminal voltages during the
In (14), As is a sparse matrix with elements of ones that are optimal FR. Similarly, Zuk contains the weighted values of
ref
assigned to only the state variables Δfk and ΔVdck for all k in ΔPgk , ΔPkref , and ΔVdckref
when grid k is linked via a VSC.
ref
XE . For the case of an LCC, it consists of ΔPgk and ΔPkref .
Furthermore, We , Wu , and Wd are the weighting functions for
B. Decentralized Control Inputs and Outputs Zek , Zuk , and dk , respectively, as:
The proposed decentralized controllers are developed in the    
Zek We · C̄Ek O XE XE
form of output feedback controllers, as shown in Fig. 2, so that Zk = = = [C1 D12] ,
Zuk O Wu ·I rk rk
the optimal FR in MTDC-linked grids can readily be achieved
in practice using the measurement of YE , rather than the esti- and wdrk = Wd · dk , (20)
mation of XE . In (13), YE is decentralized to YTk for the design
where C̄E k is a sparse matrix with elements of ones to extract
of the proposed controller in grid k, discussed in Section III, as:
Δfk , ΔVdck , Δfk dt, and ΔVdck dt from XE . Note that in
T
T
YTk = [YEk T
YPj ] = CTk · XE (20), C1 and D12 are additionally defined for brief notation.
Moreover, in Fig. 2, Δk (s) reflects the uncertainty in the model
= [CEk CIk CPj ]T · XE , ∀k, and j = k. (16) parameters of grid k [20], [21]: i.e., Gk (s).(1 + Δk (s)), rather
than Gk (s). In time domain, Δk (s) affects the accuracy of AE ,
In (16), YEk and YPj are the measurements delivered from BrEk , Bdrk , and CTk [see (16) and (17)]. This implies that the
grids k and j, respectively, as shown in Fig. 2. For grid k, CEk and proposed decentralized controllers are developed considering
CIk are obtained by decentralizing the upper and lower parts, uncertainty in the model parameter estimates of SGs, converters,
respectively, of CE in (14): i.e., [C O] = [CE1 , ···, CEk , ···, and ac and dc transmission lines.
CEm ]T and [O I] = [CI1 , ···, CIk , ···, CIm ]T . Moreover, in (16), Given the weighting functions and parameter uncertainty, the
YPj is a sub-vector of YEj and CPj is the corresponding sub- model of the MTDC-linked grid with the decentralized control
matrix of CEj , because the design and activation of controller k inputs and outputs [i.e., (16) and (17)] can be completed as:
do not require all of the measurement data in YEj , as discussed ẊE = A1 · XE + B1 · dk + B2 · rk , (21)
in Section III-C.
Similarly, the controller outputs or, equivalently, the reference Zk = C1 · XE + D12 · rk , (22)
signals r in (11) can be decentralized to rk : i.e., BrE ·r = YTk = C2 · XE , (23)
Σk BrEk ·rk . The reference signals rj generated by controllers
j  k are regarded as disturbances for controller k. This implies where the notations are simplified as A1 = AE , B1 = Bdrk ·Wd ,
that (12) can be represented from the perspective of controller B2 = BrEk , and C2 = CTk . In frequency domain, (21) is

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on May 06,2024 at 05:50:37 UTC from IEEE Xplore. Restrictions apply.
 KIM et al.: DECENTRALIZED ROBUST FREQUENCY REGULATION OF MULTI-TERMINAL HVDC-LINKED GRIDS 3283

expressed as: From (19) and (31), the time-domain control gains Ahk , Bhk ,
−1 T Chk , and Dhk in (18) also can be obtained, which is omitted
XE = (sI − A1 ) · [B1 B2 ] · [dk rk ] . (24)
here.
Using (24), (22) and (23) are then equivalently represented in In each grid, the decentralized H controller operates as the
the frequency domain as: optimal secondary frequency controller that establishes extra
power references to restore the grid frequency to a nominal value
Zk = {C1 · (sI − A1 )−1 · [B1 B2 ] + [O D12 ]} · [dk rk ]T ,
[40]. The extra references do not disturb the references of the
(25)
droop controllers, because the optimization problem (27)–(29) is
and YTk = C2 · (sI − A1 )−1 · [B1 B2 ] · [dk rk ]T . (26) formulated using the dynamic model of the MTDC-linked grids
with the droop control. This also implies that the H control gain
In other words, for the optimal robust FR, decentralized Kk (s) inherently reflects the physical operating conditions and
controller k [see (18) and (19)] should be designed to generate limits of the SGs, MTDC converters, and ac and dc networks.
rk that minimizes Zk for all unknown dk in (25), given the It further enables Kk (s) to be deterministically obtained and
measurements YTk in (26). updated for changes in the grid operating conditions, unlike
In the proposed optimal FR strategy, the decentralized control conventional non-optimal strategies wherein control gains need
gain Kk (s) for grid k is determined to minimize the H -norm of to be tuned using a trial-and-error approach.
the transfer function TZkdk from dk to Zk by solving the robust
optimization problem:
B. Model Order Reduction of Decentralized H Controllers
minTZkdk (Kk (s))∞ < γ, (27)
In general, a dynamic model of MTDC-linked grids [i.e.,
where TZkdk (Kk (s))∞ = sup σ̄(TZkdk (jω)), (28) (12)–(15)] is implemented in high dimensions including several
ω
uncontrollable and unobservable variables particularly when the
numbers of SGs, VSCs, LCCs, and WTGs are large. For the case
TZkdk (s) = G11 (s) + G12 (s)Kk (s) of Grids 1–3, discussed in Section II-A, each decentralized H
controller is designed with a model order r of 87. The full-order
· (I − G22 (s)Kk (s))−1 G21 (s), (29a)
decentralized H controllers are difficult to implement in prac-
G11 (s) = C1 · (sI − A1 )−1 · B1 , (29b) tice due to high computational burden and high sensitivity to
measurement noises. In this paper, for the practical implementa-
G12 (s) = C1 · (sI − A1 )−1 · B2 + D12 , (29c) tion, a balanced truncation algorithm [24] is applied to reduce
the model orders of the decentralized H controllers while still
G21 (s) = C2 · (sI − A1 )−1 · B1 , (29d)
preserving their dominant response characteristics. Specifically,
and G22 (s) = C2 · (sI − A1 )−1 · B2 . (29e) the optimal control gains Ahk , Bhk , and Chk of the full-order
H controller k are first converted to Ãh k = T-1 Ahk T, B̃h k =
In (27) and (28), TZkdk can be obtained as (29) by substituting
T-1 Bhk , and C̃h k = Chk T using the balancing transformation
(19) into (24)–(26). Moreover, σ̄ represents a maximum singular
matrix T, as:
value of TZkdk and γ is the performance level. In this study, a
 
γ-iteration method is adopted to solve the optimization problem Ahk,11 Ahk,12 Bhk,1
(27)–(29), wherein γ is repeatedly updated to a smaller value Ãhk = , B̃hk = , (32)
Ahk,21 Ahk,22 Bhk,2
until convergence [22]. For each value of γ, there exists a stable  
H controller that satisfies (27) if and only if the three conditions C̃hk = Chk,1 Chk,2 . (33)
hold [23] as:
In (32) and (33), the matrix decomposition can be adaptively
X∞ = −A1 AT1 + CT1 C1 (γ −2 B1 BT1 − B2 BT2 ) ≥ 0, (30a) achieved considering the cumulative energy and number of Han-
kel singular values (HSVs). The reduced-order H controller k
Y∞ = −AT1 A1 + B1 BT1 (γ −2 CT1 C1 − CT2 C2 ) ≥ 0, (30b)
can then be determined as:
and ρ(X∞ Y∞ ) < γ 2 . (30c) ˙
X̃ hk = Ahk,11 · X̃hk + Bhk,1 · YTk
In (30c), ρ(X Y ) is the largest eigenvalue of X Y . When
γ converges, the optimal control gain Kk (s) is then determined rk = Chk,1 · Xhk + Dhk · YTk . (34)
using (30), as:
 Fig. 3(a) shows the HSVs of the full-order and reduced-order
A∞ −Z∞ L∞ models of the decentralized H controllers for the MTDC-
Kk (s) = , (31a)
F∞ 0 linked grids. The HSVs were obtained for the step response to
the load demand variation by 0.1 pu in each grid, given the test
where A∞ = A1 + γ −2 B1 BT1 X∞ + B2 F∞ + Z∞ L∞ C2 , conditions discussed in Section IV-A and IV-C. Fig. 3(b) shows
(31b) the corresponding cumulative energy curves. The results of the
F∞ = −BT2 X∞ , L∞ = −Y∞ CT2 , HSV analysis indicate that the energy magnitudes of the three
decentralized H controllers are still higher than 99.9% of the
−1
and Z∞ = (I − γ −2 Y∞ X∞ ) . (31c) total energy when their model orders are reduced to r = 26, 29,

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on May 06,2024 at 05:50:37 UTC from IEEE Xplore. Restrictions apply.
 3284 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 38, NO. 4, JULY 2023

(a)

(a)

(b)

(b)

(c)

Fig. 3. (a) Hankel singular values and (b) cumulative energy curves for the
optimal decentralized H controllers.

(a) (b)

(d)

(e)

Fig. 4. Gains of the optimal decentralized H controllers for grids k =1, 2,

and 3: (a) Bhk and (b) Dhk for all k and j  k.

and 13, respectively. This implies that the reduced-order mod- Fig. 5. Closed-loop system eigenvalues for variations in (a) Lf ranging from
0.03 mH to 1.50 mH, (b) Trh from 4 s to 10 s, (c) Ls from 0.01 pu to 0.30 pu,
els of the H controllers can successfully reflect the dynamic (d) Δ(s) from 0% to 50%, and (e) Td from 0 ms to 60 ms.
operating characteristics of the original, 87th -order models. In
other words, the reduction of the model orders to 26, 29, and
13, respectively, only marginally affects the FR performance,
while effectively mitigating the computational burden. Note that conditions. It can be seen that the elements in Bhk and Dhk
r can be set to smaller values, considering the trade-off between have values close to zero for all measurements from grid j 
control performance and implementation complexity. k, apart from the measurements of Δfj and ΔVdcj . In other
words, Xhk and rk in each grid k are mainly affected by the
measurements of Δfj and ΔVdcj for grid j  k. This demonstrates
C. Sensitivity and Eigenvalue Analyses
that the decentralized H controller for grid k can be effectively
The closed-loop system, shown in Fig. 2, was analyzed with designed only using the measurements of Δfj and ΔVdcj for grid
respect to the gain sensitivity and stability. Note that the anal- j  k, thus mitigating the requirement for inter-grid measurement
ysis was conducted using reduced-order models of the three and communication systems and facilitating wide applications
decentralized H controllers, discussed in Section III-B. Speci- of the proposed strategy in practice.
fically, Fig. 4 shows the optimal gains of the H controllers In addition, Fig. 5 shows the eigenvalues of the closed-loop
for the GSVSC- and LCC-interfaced grids (i.e., k = 1, 2, and system with changes in system parameters, model uncertainty
3) parti- cularly with regard to the measurements from the level, and communication time delay. In Fig. 5(a)–(c), the
corresponding grid and the interconnected grids: i.e., Bhk and complex-conjugate eigenvalues move closer to the imaginary
T
Dhk for YTk = [YT Ek YT Pj ] in Fig. 2 and (18). It represents axis, implying increased overshoot and settling time, as the
the extent to which the measurements YTk affect the controller converter phase inductance Lf , the turbine time constant Trh , and
states Xhk and the reference signals rk under disturbed grid the stator inductance Ls increase from 0.03 mH to 1.50 mH, from

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on May 06,2024 at 05:50:37 UTC from IEEE Xplore. Restrictions apply.
 KIM et al.: DECENTRALIZED ROBUST FREQUENCY REGULATION OF MULTI-TERMINAL HVDC-LINKED GRIDS 3285

4 s to 10 s, and from 0.01 pu to 0.30 pu, respectively. However, TABLE I

SYSTEM PARAMETERS FOR CASE STUDIES
all eigenvalues are still placed in the left-hand half plane (LHP),
verifying that the proposed decentralized H controllers ensure
grid frequency stability for large variations in the electrical and
electromechanical dynamics of the MTDC-linked grids.
Analogously, Fig. 5(d) shows the loci of the eigenvalues
when the uncertainty level Δ(s) of the system model parameters
increases from 0% to 50%. The dominant eigenvalues are still
located in the LHP for a large value of Δ(s) (i.e., up to Δ(s) =
47%). This confirms that the proposed H controllers can secure
the stable operation of MTDC-linked grids even when there is
relatively large uncertainty in the system parameter estimates.
Moreover, Fig. 5(e) shows the eigenvalue analysis for an increase
in Td from 0 ms to 60 ms; for brevity, the delay is assumed to
be constant and identical for all decentralized H controllers.
Given Td , (18) changes to:
Ẋhk (t) = Ahk · Xhk (t) + Bhk · YTk (t − Td )
rk (t) = Chk · Xhk (t) + Dhk · YTk (t − Td ). (35)
Therefore, in the frequency domain, (19) is replaced by:
rk (s) = Kk (s) · e−Td s · YTk (s), (36)
where e–Tds can be approximated as:
Td2 s2 − 6Td s + 12
e−Td s ≈ . (37)
Td2 s2 + 6Td s + 12
In Fig. 5(e), with an increase in Td , the complex-conjugate
eigenvalues move toward the imaginary axis. When Td increases
to approximately 51 ms, the eigenvalues cross the axis and are
placed on the right-hand half plane (RHP). It was reported in [25]
that the time delay in common fiber optic communications for
(a) (b)
application to MTDC systems ranges between 1 ms and 10 ms.
Thus, the proposed H controllers can successfully guarantee
the stability and robustness of the closed-loop system in practice.

IV. CASE STUDIES AND RESULTS

A. Test System and Simulation Conditions
For comparative case studies, a dynamic model of the hybrid Fig. 6. Continuous variations in the (a) loads ΔPL1–3 and (b) wind speed
MTDC-linked grids was implemented in MATLAB/SIMULINK ΔVw .
(or, simply, SIMULINK). Each ac grid consisted mainly of SGs,
transmission lines, regional loads, and a GSVSC or LCC station
TABLE II
for the interface with the MTDC system. In SIMULINK, the hybrid FEATURES OF THE PROPOSED AND CONVENTIONAL STRATEGIES
MTDC converter was modeled using average circuit models with
inner-loop and outer-loop feedback controllers in a dq frame.
The dc transmission line between the MTDC converters was
then modeled using a pi-section model, rather than a T-section
model, due to its higher modeling accuracy [41]. An PMSG-type
OWF model was also implemented connected to the dc network
via a WFVSC. Table I shows the model parameters of the SGs,
MTDC converters, ac and dc networks, and PMSG-type OWF,
as well as the corresponding control gains at the local level [26], Given the test bed parameters and operating conditions, the
[27], [28]. Fig. 6 shows the load demand variations ΔPLk for proposed decentralized H controllers were developed with no
all grids k, reflecting the scaled-down RegD signals [29] over a loop shaping (i.e., unity weighting factors) [20] and analyzed
period of 200 s. It also shows the intermittent wind speed for the in comparison with conventional decentralized controllers [19].
OWF. Table II shows the main features of the proposed (Case 1)

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on May 06,2024 at 05:50:37 UTC from IEEE Xplore. Restrictions apply.
 3286 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 38, NO. 4, JULY 2023

(a) (b) (a)

(b)

(c)

Fig. 7. Comparisons of the step responses of the small-signal model and the
comprehensive SIMULINK model of the MTDC-linked grids to ΔPL1 = 0.1 pu
and ΔPL2 = 0.15 pu: (a) f1 , (b) f2 , (c) f3 , and (d) Vdc .

and conventional strategies (Case 2 and 3). In Case 2, a de- (d)

centralized PI controller was adopted to regulate the regional
grid frequency deviations without the coordination between
SGs and MTDC converters [30]. In Case 3, a centralized op-
timization problem was formulated to determine the decen-
tralized control gains. Specifically, the centralized H control
gains were first obtained from the solution to the optimiza-
tion problem and, consequently, the decentralized controller for Fig. 8. Step responses of the proposed and conventional strategies: (a) f1 , (b)
grid k was implemented by arbitrarily setting the off-diagonal f2 , (c) f3 , and (d) Vdc .
elements of the centralized gains to zero, apart from the el-
ements corresponding to YEk and YPj for j  k, shown in
(15). This leads the decentralized control for Case 3 to become TABLE III
COMPARISONS OF STEP RESPONSE TEST RESULTS
non-optimal.

B. Validating the Dynamic Model of MTDC-Linked Grids

Fig. 7 shows comparisons between the responses of two mod-
els of the MTDC-linked grids to a step increase in ΔPL1 = 0.1 pu
and ΔPL2 = 0.15 pu at t = 45 s: i.e., the small-signal model and
the comprehensive SIMULINK model, discussed in Sections II
and IV-A, respectively. The small-signal model was developed
C. Comparisons for Stepwise Load Variations
in a linear time-invariant form, whereas the SIMULINK model
was implemented in a nonlinear time-varying form. Note that Fig. 8 shows f1–3 and Vdc for the step responses of the MTDC-
the comparisons were conducted only for MTDC-linked grids linked grids to load variations ΔPLk , which increased by 0.3 pu
without the proposed controller. In Fig. 8, the dynamic responses for all grids k at t = 10 s over a 10 s period. Table III lists
of the two models were very similar to each other for each the corresponding numerical results. In the proposed strategy
profile of the grid frequencies f1–3 and the average dc voltage (i.e., Case 1), the frequency deviations became smaller for all
Vdc for all MTDC terminals. This indicates good consistency grids k than in the conventional strategies (i.e., Cases 2 and 3).
between the two models even when relatively large variations Specifically, Case 1 led to a decrease in the sum of the maximum
in load demands occur in the MTDC-linked grids under normal frequency deviations (i.e., Σk |Δfk |max ) by 75.1% and 11.1%,
conditions. It hence validates the accuracy of the results obtained compared to Cases 2 and 3, respectively. Moreover, in Case 1,
by applying the proposed and conventional controllers to the f1–3 were restored back to the nominal values more rapidly, and
small-signal dynamic model. Moreover, Fig. 7(a)–(c) shows had smaller overshoot responses than in Cases 2 and 3. Case
that in each grid, the frequency fell off and then reached to 1 also resulted in smaller |ΔVdc |max by 17.7% and 44.1% than
a steady state value with a large overshoot and settling time. Cases 2 and 3, respectively, as shown in Fig. 8(d). In other words,
This confirms the motivation of this study on the coordinated the proposed decentralized H controllers were more effective
FR of MTDC-linked grids. Note that f3 was maintained at in reducing the frequency deviations in all the MTDC-linked
60 Hz in the steady state, because the LCC operated with a grids while further mitigating the dc voltage variations. The
constant power reference, unlike the GSVSCs with P-Vdc droop improve- ment was made only using the measurements of Δfj
controllers. and ΔVdcj , rather than all the measurements in grid j  k, thus

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on May 06,2024 at 05:50:37 UTC from IEEE Xplore. Restrictions apply.
 KIM et al.: DECENTRALIZED ROBUST FREQUENCY REGULATION OF MULTI-TERMINAL HVDC-LINKED GRIDS 3287

(a) (b) (a)

(b)

(c)

(d)
Fig. 9. Step responses of the (a) generated and (b) transmitted power for the
proposed and conventional strategies.

TABLE IV
COMPARISONS OF CONTINUOUS RESPONSE TEST RESULTS

Fig. 10. Responses to the continuous variations in ΔPL1–3 and ΔVW for the
proposed and conventional strategies: (a) f1 , (b) f2 , (c) f3 , and (d) Vdc .

(a) (b) (c) (d)

lessening the requirement on inter-grid sensing and communi-

cation systems.
In addition, Fig. 9(a) shows the profiles of the total power Fig. 11. Failures of the communication links among the MTDC-linked grids:
generation in each grid for the stepwise load variations. Case 1 (a) no, (b) single-link, and (c) double-link failures and (d) no communication.
led to only moderate variations in the generated power. This is
also the case for the power transmitted by the MTDC converters,
as shown in Fig. 9(b). The comparison results demonstrate 3, respectively. Moreover, in Case 1, ΔVdc,rms was reduced by
that the improved performance of Case 1 is attributable to the 66.7% and 75.0%, compared to Cases 2 and 3. The case study
optimal, coordinated power sharing among the MTDC-linked results consistently verify that the proposed decentralized H
grids through the dc transmission network in transient and steady controllers more effectively improve the FR of the MTDC-linked
states, rather than using any extra power source and via excessive grids while further suppressing the dc network voltage variations
variations in the generated and transmitted power. through optimal, coordinated sharing of the time-varying power
outputs of the SGs and OWF.
D. Comparisons for Continuous Load and Wind Variations
The proposed FR strategy was also tested with regard to E. Effects of Communication Failures and Time Delays
continuous variations in the grid load demands ΔPL1–3 and Additional case studies were conducted to evaluate the per-
wind speed ΔVw over a time period of 200 s, as shown in formances of the decentralized H controllers in Cases 1 and
Fig. 6. Fig. 10 shows f1–3 and Vdc for the responses of the 3, particularly, under different conditions of the communication
MTDC-linked grids to ΔPL1–3 and ΔVw , and Table IV lists the systems for the MTDC-linked grids. Note that Case 2 requires
numerical results as root-mean-square (rms) values. Each rms no inter-grid measurements and communications, so the effects
value was calculated as {Σl Δfk (l)2 /L}1/2 and {Σl ΔVdc (l)2 /L}1/2 , of different communication conditions were not analyzed here.
where l is the index of the measurement samples and L is the Fig. 11 represents five different conditions of the inter-grid
total number of samples. Case 1 led to significantly smaller communication systems: i.e., no failure, single-link and double-
variations in the frequencies and dc terminal voltages for all link failures, and no active communication. When the communi-
the grids than Cases 2 and 3. Specifically, the sum of Δf1–3,rms cation between grids k and j fails, the decentralized H controller
in Case 1 was 92.5% and 62.5% smaller than those in Case 2 and for grid k generates the optimal reference signals rk without

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on May 06,2024 at 05:50:37 UTC from IEEE Xplore. Restrictions apply.
 3288 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 38, NO. 4, JULY 2023

(a)
(a)

(b)

(b)

Fig. 13. Comparison between the proposed and conventional H controllers

(i.e., Cases 1 and 3) for Td ranging from 20 ms to 50 ms: (a) f2 and (b) f3 .

TABLE VI
COMPARISONS OF COMMUNICATION DELAY TEST RESULTS

Fig. 12. Step responses of f1 and Vdc for the (a) proposed and (b) conventional
H controllers (Cases 1 and 3) when the inter-grid communication links fail.

TABLE V
COMPARISONS OF COMMUNICATION FAILURE TEST RESULTS were directly obtained by solving the decentralized optimization
problems (27)–(29). This stands in contrast to Case 3 where the
non-optimal gains were acquired from the centralized optimiza-
tion problem. In other words, the inter-grid communication
failure more affects the non-optimal controllers, thus leading
to larger and longer overshoot responses of f1–3 and Vdc . The
case study results confirm that the proposed decentralized H
controllers are still more effective and robust in the optimal FR
against the events of inter-grid communication failure.
receiving the Δfj and ΔVdcj from grid j  k. Fig. 12 shows f1 and In addition, Fig. 13 compares the step responses of f2 and f3 for
Vdc for the step responses of the MTDC-linked grids to ΔPLk variations in ΔPL1–3 by ±0.1 pu over every 40 s between Cases
= 0.1 pu for all k at t = 10 s, and Table V lists the corresponding 1 and 3, when increasing Td gradually from 20 ms to 50 ms.
numerical results. In Fig. 12(a), Case 1 enabled the smaller Table VI lists the corresponding numerical results. Note that for
deviation of f1 (and also f2 and f3 ) and the faster recovery back brevity, Td was assumed to be the same for all the decentralized
to the nominal value with smaller overshoot responses for all the H controllers, as discussed in Section III-C. As Td increased,
communication conditions than Case 3 in Fig. 12(b). Case 3 led to larger Σk fk,rms than Case 1. Moreover, in Case
Fig. 12(a) shows that in Case 1, it took approximately 3.0 s 3, Td = 50 ms led to unstable, large oscillations of f2 and f3 ;
for the frequency recovery, and the recovery time is relatively this was also the case for f1 . It was mainly because in Case 3,
smaller than in the existing SFC schemes for the real-world the centralized H control gains were simply divided for the
SG-only grids [40], but similar to the case of the previous decentralized controllers by setting the off-diagonal elements
SFC schemes for the power-electronics-interfaced grids (e.g., to zero, which reduced the maximum allowable value of the
[12] and [17]). The improved performance of Case 1 is mainly communication time delay. In other words, the eigenvalues
because the optimal gains for the decentralized H controllers moved across the imaginary axis from the LHP to the RHP

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on May 06,2024 at 05:50:37 UTC from IEEE Xplore. Restrictions apply.
 KIM et al.: DECENTRALIZED ROBUST FREQUENCY REGULATION OF MULTI-TERMINAL HVDC-LINKED GRIDS 3289

(a) (b) (a) (b)

Fig. 14. (a) |Δf1 |max and (b) (|Δf2 |max + |Δf3 |max )/2 for ΔPL1 (t = 5 s)
= 0.3 pu in Case 1.

Fig. 15. Comparison between the proposed and conventional H controllers

(i.e., Cases 1 and 3) for different weighting functions: (a) LPF and (b) BPF.
when Td increased to 50 ms. By contrast, Case 1 still ensured
stable, smaller variations in f2 and f3 for all Td ranging up to
50 ms, although the damped oscillations occurred for large Td .
TABLE VII
The comparison results prove that the proposed strategy still COMPARISONS FOR NO MEASUREMENT OF ROTOR SPEEDS
further enhances the performance and stability of the FR in the
MTDC-linked grids when there are relatively large time delays
in the communication systems.

F. Effects of Dynamic Responses of SGs and VSCs

For Case 1, Fig. 14 shows |Δf1 |max and (|Δf2 |max +
|Δf3 |max )/2 due to stepwise load variation ΔPL1 (t = 5+ s) =
0.3 pu when Tg and Kp range from 0.01 s to 0.10 s and from
0.1 s to 4.0 s, respectively. The SGs and VSCs respond slowly
as Tg increases and Kp decreases, respectively. In general, as Tg Similarly, Fig. 15(b) shows the comparison results when BPFs
increases, |Δfk |max increases for all k, verifying that the test were adopted with a bandwidth B from 102 rad/s to 105 rad/s.
bed and the H controllers successfully reflect the dynamic For all different weighting functions, Case 1 led to considerably
responses and practical operating conditions of SGs. In addition, smaller Σk Δfk,rms and ΔVdc,rms than Case 3. This verified
the dynamic responses of the VSCs affect the FR, because they the improved effectiveness and robustness of the proposed FR
transfer the power outputs of the SGs throughout the MTDC- strategy when applied to practical grids with different local
linked grids. Moreover, Fig. 14 reveals that the fast dynamic controllers of the SGs and MTDC converters and different
responses of the VSCs mitigate the frequency deviation in the characteristics of the input measurement data and disturbances.
corresponding grid (i.e., Grid 1), but increase the deviations In the test bed (see Fig. 1), Cases 1–3 were implemented using
in other grids (i.e., Grids 2 and 3). In other words, the VSCs the measurements of both the generator rotor speeds ω r and the
conflict with each other. The case study results verify the need grid frequencies f at the PCCs of the MTDC converters. This
for coordinated control of both SGs and VSCs, wherein the aimed to further improve the FR performance for all Cases 1–3,
optimal amounts of power to be generated and transferred are because ω r and f differ from each other in transient state due to
simultaneously determined, with consideration of the different the distant locations between the SGs and MTDC converters and
dynamic responses and operating conditions of SGs and VSCs. the large inertia moments of the SGs [42] and [43]. Table VII
Note that for all Tg and Kp , Case 1 still led to significantly smaller shows that even when ω r was not measured, Case 1 still led to
|Δf1, 2, 3 |max than Case 2 wherein no coordination among the smaller Σk Δfk,rms and ΔVdc,rms than Cases 2 and 3, confirming
SGs and VSCs was achieved. the robustness and applicability of the proposed FR strategy.
For further comparisons of Cases 1−3, another test bed was
implemented using a CIGRE benchmark system (i.e., CIGRE B4
G. Performance Under Various Conditions DC Grid Test System) with modifications [44] and [45]. Briefly,
The continuous response tests in Section IV-D were repeated the test bed consists of three regional ac grids with three GSVSCs
to analyze the FR performance of the proposed strategy under and one PMSG-type OWF with a WFVSC: i.e., n = 3 and m =
various conditions. In the proposed and conventional H control 0. The length of each dc transmission line is 200 km and the
strategies, the only hyper-parameters are the weighting functions GSVSC for Grid 1 is not directly connected with the WFVSC.
Wu , We , and Wd for the system inputs rk , outputs YEk , and Given the test bed, Table VIII shows the test results of Cases 1−3
disturbances dk , respectively (see Fig. 2). Fig. 15(a) shows for the continuous variations in ΔPL1–3 and ΔVw . In Case 1,
comparisons between Cases 1 and 3 when all the weighting Σk Δfk,rms and ΔVdc,rms were still smaller than in Cases 2 and
functions were modeled as LPFs with a cutoff frequency ω c 3. This consistently verifies the effectiveness and applicability
ranging from 102 rad/s to 105 rad/s, rather than the unity function. of the proposed H controllers in real-world MTDC systems.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on May 06,2024 at 05:50:37 UTC from IEEE Xplore. Restrictions apply.
 3290 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 38, NO. 4, JULY 2023

TABLE VIII included in each grid (i.e., 30 SGs in total), whereas it was
COMPARISONS FOR THE CIGRE BENCHMARK TEST BED
42.17 s for Case 3. Moreover, Case 1 became more time efficient
for a higher number of the SGs, compared to Case 3. The case
study results verified the computational efficiency and hence
scalability of the proposed decentralized H controllers. Note
that in Cases 1 and 3, the H controllers were designed offline
and then executed online for the real-time FR, as in Case 2.
The computational efficiency of Case 2 was not discussed here,
because the PI control gains were determined using a trial-and-
TABLE IX
COMPARISONS WITH A CONSENSUS ALGORITHM-BASED STRATEGY
error approach, rather than the optimization problem solving.

V. CONCLUSION
This paper proposes a new strategy for optimal FR in MTDC-
linked grids via coordinated, decentralized control of SGs and
hybrid converters. A dynamic model of hybrid MTDC- linked
grids was implemented considering the decentralized control
inputs and outputs. Given the dynamic model, a robust opti-
mization problem was formulated to develop a decentralized
H controller for each grid, considering the inter-grid communi-
cation systems and the model parameter uncertainty. A balanced
truncation algorithm was applied to reduce the model orders
of the decentralized H controllers, while still preserving their
dominant response characteristics. Sensitivity and eigenvalue
analyses were then conducted, focusing on the effects of inter-
grid measurement, system parameter, uncertainty level, and
communication time delay. The results of comparative case
studies verified that the proposed strategy was more effective in
Fig. 16. Computation time required to solve the optimization problems of reducing the frequency deviations in all MTDC-linked grids than
Cases 1 and 3 for an increase in the number of the SGs in each MTDC-linked
grid from 1 to 10. conventional strategies, while maintaining smaller variations in
dc network voltages. Moreover, compared to the conventional
H control strategy, the proposed strategy also improved the
The performance of the proposed optimal strategy was also effectiveness, stability, and robustness in the optimal FR of
analyzed in comparison with the conventional non-optimal strat- hybrid MTDC-linked grids under various conditions.
egy using a consensus algorithm (i.e., Case 4) [9]. Table IX
shows the test results of Cases 1 and 4 for the continuous APPENDIX
variations in ΔPL1–3 and ΔVw . Case 1 led to the significant A. Modeling of AC Grids with Hybrid Converter Interfaces
reductions of Σk Δfk,rms and ΔVdc,rms by 91.2% and 88.7%,
respectively, compared to Case 4. The improved performance of A regional ac grid includes SGs with reheat steam turbines.
Case 1 is attributable to the fact that the dynamic response capa- For each SG, a synchronous machine is modelled using the 6th -
bilities of the SGs and MTDC converters could be fully exploited order differential equations, discussed in [31]. A governor and
and optimally coordinated for fast and accurate compensation exciter are modeled using the 1st -order transfer functions, based
for the power imbalances in all grids k. By contrast, the control on [32]. The dynamic model of an SG can then be established
gains in Case 4 (and also in Case 2) need to be tuned using a as:
trial-and-error approach and hence there is a high possibility that
Ẋsg = Asg · Xsg + Bsg · rsg + Esg · Ibus , (A1)
the fast operating capabilities of the MTDC converters are not
 
effectively exploited. where Xsg = [ΔE q , ΔE d , Δφ1d , Δφ2q , Δδr , Δωr ,
In addition, the computational efficiency of the proposed
strategy was evaluated while gradually increasing the number Δxgov1 , Δxgov2 , Δxtur1 , Δxtur2 , ΔVf d ]T ,
of the SGs in each MTDC-linked grid. Specifically, Fig. 16 T
rsg = [ΔPgref , ΔEtref ] , and Ibus = ΔIsdq . (A2)
shows the computation time required to solve each decentralized
optimization problem [i.e., (27)–(29) for k = 1, 2, and 3] for Case Furthermore, in each grid, a phase-locked loop (PLL) is used
1. It also shows the computation time to solve the centralized to measure the grid frequency at the PCC of an MTDC converter
optimization problem [19], as discussed in Section IV-A. In Case and calculate the dq values of the PCC voltages and currents [33].
1, the optimal H control gains were determined within shorter Considering the PLL operation, an ac network is modeled as:
computation time than in Case 3. The maximum computation
time for Case 1 was estimated as 22.99 s when 10 SGs were Ẋac = Aac · Xac + Eac1 · Vac + Eac2 · Iac + Bacw · wac

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on May 06,2024 at 05:50:37 UTC from IEEE Xplore. Restrictions apply.
 KIM et al.: DECENTRALIZED ROBUST FREQUENCY REGULATION OF MULTI-TERMINAL HVDC-LINKED GRIDS 3291

T  T level controllers can regulate dq-axis currents of the MTDC

where Vac = [ΔEtd , ΔEtq , Iac = ΔIcd , ΔIcq ] ,
converters for the optimal power sharing with the SGs.
and wac = ΔPL ., (A3)
In an MTDC system, a VSC includes inner and outer con- B. Modeling of a PMSG-type OWF and a DC Network
troller loops with a P-Vdc droop function. For an LCC, dc A PMSG-type OWF is modeled based on [35] and [36], as:
terminal voltage or current is controlled to transmit constant
Ẋowf = Aowf · Xowf + Bowf · wowf + Eowf · Idc4 , (B1)
or time-varying power from an ac grid to a dc network. The
dynamic models of a VSC and an LCC are given [33], [34] as: T
dq
where Xowf = [Δωe , Δθe , ΔIm , ΔVdc , Δmowf , Δndq
owf ] , ,
Ẋvsc = Avsc · Xvsc + Bvsc · rvsc + Ebus · Vbus + Edc · Idc , and wowf = ΔVw .
(B2)
(A4) In addition, a dc transmission network is characterized using
a pi-section model. The corresponding dynamics is linearized
Ẋlcc = Alcc · Xlcc + Blcc · rlcc + Ebus · Vbus + Edc · Idc , as:
(A5)
ΔI˙dc,ij = (−Rdc,ij ΔIdc,ij + (ΔVdci − ΔVdcj ))/Ldc,ij ,
dq T
where Xvsc = [ΔIcdq , ΔVdc , Δndq vsc , Δmvsc ] , (B3)
T
Xlcc = [ΔIcdq , ΔVcdq , ΔIdco , ΔVdc , Δmlcc , Δθlcc , Δnlcc ] , where ΔIdc,ij and ΔVdci are the incremental dc current from
ref
rvsc = [ΔP ref , ΔVdc ref
, ΔVmag
T
] , and rlcc = ΔP ref . bus i to j and the incremental dc voltage at bus i, respectively.
Moreover, Rdc,ij and Ldc,ij are the dc resistance and inductance.
(A6)
According to the configuration, the dynamic model of a dc
By combining (A1)–(A6), the dynamic model of ac grid k is network is given as:
represented as:
Ẋdc = Adc · Xdc + Edcv · Vdc , (B4)
Ẋk = Ak1 · Xk + Bk · Uk + [EI EV ] · [Ik Vk ]T + Ek · Idc ,
(A7) where Xdc = [ΔIdc12 , ΔIdc13 , ΔIdc23 , ΔIdc34 ]T
(B5)
and Vdc = [ΔVdc1 , ΔVdc2 , ΔVdc3 , ΔVdc4 ]T .
where Ik = [ITbus ITac ], and Vk = T
[Vbus T
Vac ]. (A8)
REFERENCES
Moreover, [Ik Vk ]T in (A7) can be equivalently expressed as:
[1] K. Bell et al., “Economic and technical criteria for designing future
[Ik Vk ]T = [FI · Xk FV · Xk ]T , (A9) off–shore HVDC grids,” in Proc. Power Energy Soc. Innov. Smart Grid
Technol. Conf. Eur., Gothenburg, Sweden, 2010, pp. 1–8.
where FI and FV are sparse matrices to extract Ik and Vk from [2] S. M. Muyeen, R. Takahashi, and J. Tamura, “Operation and control of
Xk . Substitution of (A9) into (A7) gives: HVDC–connected offshore wind farm,” IEEE Trans. Sustain. Energy,
vol. 1, no. 1, pp. 30–37, Apr. 2010.
Ẋk = Ak · Xk + Bk · Uk + Ek · Idc , (A10) [3] N. M. Kirby et al., “HVDC transmission for large offshore wind farms,”
Power Eng. J., vol. 16, no. 3, pp. 135–141, Jun. 2002.
where Ak = Ak1 + EI ·FI + EV ·FV and Uk = [rk wdk ]T . Using [4] T. Ackermann, “Transmission systems for offshore wind farms,” IEEE
Power Eng. Rev., vol. 22, no. 12, pp. 23–27, Dec. 2002.
(A10), the dynamic model of ac grid k is completed as: [5] C. Li, P. Zhan, J. Wen, M. Yao, N. Li, and W. -J. Lee, “Offshore wind
farm integration and frequency support control utilizing hybrid multi-
Ẋk = Ak · Xk + Brk · rk + Bwk · wdk + Ek · Idc , (A11) terminal HVDC transmission,” IEEE Trans. Ind. Appl., vol. 50, no. 4,
pp. 2788–2797, Jul./Aug. 2014.
Yk = Ck · Xk , (A12) [6] D. Kwon et al., “Modeling and analysis of an LCC HVDC system using
DC voltage control to improve transient response and short–term power
where the states, inputs, and outputs are arranged as: transfer capability,” IEEE Trans. Power Del., vol. 33, no. 4, pp. 1922–1933,
 Aug. 2018.
T
[XTsg , XTac , XTvsc ] , for k = 1 and 2, [7] B. R. Andersen, L. Xu, P. J. Horton, and P. Cartwright, “Topologies for
Xk = T (A13) VSC transmission,” Power Eng. J., vol. 16, no. 3, pp. 142–150, 2002.
[XTsg , XTac , XTlcc ] , for k = 3, [8] B. Silva, C. L. Moreira, L. Seca, Y. Phulpin, and J. A. Pecas Lopes, “Pro-
 vision of inertial and primary frequency control services using offshore
ref ref T
[ΔPgk , ΔPkref , ΔVdck ] , for k = 1 and 2, multiterminal HVDC networks,” IEEE Trans. Sustain. Energy, vol. 3, no. 4,
rk = ref ref T
(A14) pp. 800–808, Oct. 2012.
[ΔPgk , ΔPk ] , for k = 3, [9] A. Kirakosyan, E. F. El-Saadany, M. S. E. Moursi, and K. Al Hosani, “DC
voltage regulation and frequency support in pilot voltage droop controlled
wdk = ΔPLk , ∀k, (A15) multi terminal HVDC systems,” IEEE Trans. Power Del., vol. 33, no. 3,
pp. 1153–1164, Jun. 2018.
and Yk = [ΔPgk , Δωrk , Δfk , ΔVmagk , ΔVdck , ΔPk ]T , ∀k. [10] O. D. Adeuyi, M. Cheah–Mane, J. Liang, and N. Jenkins, “Fast frequency
response from offshore multi–terminal VSC–HVDC schemes,” IEEE
(A16) Trans. Power Del., vol. 32, no. 6, pp. 2442–2452, Dec. 2017.
[11] W. Wang, Y. Li, Y. Cao, U. Hager, and C. Rehtanz, “Adaptive droop control
In (A16), the references for the exciter and PCC voltage of VSC–MTDC system for frequency support and power sharing,” IEEE
variations (i.e., ΔEtref and ΔVmag
ref
) are set to zero, because this Trans. Power Syst., vol. 33, no. 2, pp. 1264–1274, Mar. 2018.
paper focuses on optimal FR. In the proposed and conventional [12] Y. Jia, K. Meng, C. Sun, L. Yuan, and Z. Dong, “Economic–driven
frequency regulation in multi–terminal HVDC systems: A cooperative dis-
H control strategies, ΔVmag still need to be measured for the tributed approach,” IEEE Trans. Power Syst., vol. 35, no. 3, pp. 2245–2255,
feedback into the H controllers, so that the device- and grid- May 2020.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on May 06,2024 at 05:50:37 UTC from IEEE Xplore. Restrictions apply.
 3292 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 38, NO. 4, JULY 2023

[13] M. Mehrabankhomartash, M. Saeedifard, and S. Grijalva, “Model predic- [36] L. Wang and M. Sa-Nguyen Thi, “Stability enhancement of a PMSG–based
tive control based AC line overload alleviation by using multi–terminal DC offshore wind farm fed to a multi–machine system through an LCC–HVDC
grids,” IEEE Trans. Power Syst., vol. 35, no. 1, pp. 177–187, Jan. 2020. link,” IEEE Trans. Power Syst., vol. 28, no. 3, pp. 3327–3334, Aug. 2013.
[14] P. M. Namara, R. R. Negenborn, B. De Schutter, and G. Lightbody, [37] Y. Wang et al., “Decentralized H infinity controller design for large-scale
“Optimal coordination of a multiple HVDC link system using centralized civil structures,” Earthq. Eng. Struct. Dyn., vol. 38, no. 3, pp. 377–401,
and distributed control,” IEEE Trans. Control Syst. Technol., vol. 21, no. 2, 2009.
pp. 302–314, Mar. 2013. [38] Q. Li, F. X. Chen, M. Chen, J. M. Guerrero, and D. Abbott, “Agent-based
[15] B. -C. Jeong, D. -H. Kwon, J. -Y. Park, Y. -J. Kim, and O. Gomis- decentralized control method for islanded microgrids,” IEEE Trans. Smart
Bellmunt, “Optimal secondary control to suppress voltage fluctuations Grid, vol. 7, no. 2, pp. 637–649, Mar. 2016.
in an HVDC-linked wind farm grid,” IEEE Trans. Power Syst., vol. 37, [39] N. Yorino, Y. Zoka, M. Watanabe, and T. Kurushima, “An optimal au-
no. 4, pp. 2563–2577, Jul. 2022. tonomous decentralized control method for voltage control devices by
[16] A. Banerjee et al., “A novel explicit disturbance model–based robust using a multi-agent system,” IEEE Trans. Power Syst., vol. 30, no. 5,
damping of inter–area oscillations through MTDC grids embedded in pp. 2225–2233, Sep. 2015.
AC systems,” IEEE Trans. Power Del., vol. 33, no. 4, pp. 1864–1874, [40] Y. Rebours, D. Kirschen, M. Trotignon, and S. Rossignol, “A survey
Aug. 2018. of frequency and voltage control ancillary services—Part I: Technical
[17] L. Huang, H. Xin, and F. Döfler, “H –control of grid–connected con- features,” IEEE Trans. Power Syst., vol. 22, no. 1, pp. 350–357, Feb. 2007.
verters: Design, objectives and decentralized stability certificates,” IEEE [41] Z. Xiaodong, T. Nengling, and S. T. James, “A transient harmonic current
Trans. Smart Grid, vol. 11, no. 5, pp. 3805–3816, Sep. 2020. protection scheme for HVDC transmission line,” IEEE Trans. Power Del.,
[18] E. Sánchez–Sánchez, D. Groß, E. Prieto-Araujo, F. Dörfler, and O. Gomis- vol. 27, no. 4, pp. 2278–2285, Oct. 2012.
Bellmunt, “Optimal multivariable MMC energy–based control for dc volt- [42] A. D. Paquette and D. M. Divan, “Virtual impedance current limiting for
age regulation in HVDC applications,” IEEE Trans. Power Del., vol. 35, inverters in microgrids with synchronous generators,” IEEE Trans. Ind.
no. 2, pp. 999–1009, Apr. 2020. Appl., vol. 51, no. 2, pp. 1630–1638, Mar./Apr. 2015.
[19] S. D. Tavakoli, S. Fekriasl, E. Prieto-Araujo, J. Beerten, and O. Gomis- [43] K. Shi, W. Song, H. Ge, P. Xu, Y. Yang, and F. Blaabjerg, “Transient
Bellmunt, “Optimal H infinity control design for MMC–based HVDC analysis of microgrids with parallel synchronous generators and virtual
links,” IEEE Trans. Power Del., vol. 37, no. 2, pp. 786–797, Apr. synchronous generators,” IEEE Trans. Energy Convers., vol. 35, no. 1,
2022. pp. 95–105, Mar. 2020.
[20] D. Babazadeh and D. V. Hertem, “Study of centralized and distributed [44] T. K. Vrana, S. Dennetire, Y. Yang, J. Jardini, D. Jovcic, and H. Saad, “The
coordination of power injection in multi–TSO HVDC grid with large CIGRE B4 dc grid test system,” Electra, no. 270, pp. 10–19, Oct. 2013.
offshore wind integration,” Elect. Power Syst. Res., vol. 136, pp. 281–288, [45] H. Saad et al., “Modular multilevel converter models for electromag-
Jul. 2016. netic transients,” IEEE Trans. Power Del., vol. 29, no. 3, pp. 1481–1489,
[21] M. Aragüés-Peñalba et al., “Droop control for loss minimization in HVDC Jun. 2014.
multi-terminal transmission systems for large offshore wind farms,” Elect.
Power Syst. Res., vol. 112, pp. 48–55, Jun. 2014.
[22] K. Glover and J. C. Doyle, “State–space formulae for all stabilizing Jun-Soo Kim (Student Member, IEEE) received the B.S. degree in electrical
controllers that satisfy an H1–norm bound and relations to relations to engineering from Kyoungpook National University, Daegu, South Korea, in
risk sensitivity,” Syst. Control Lett., vol. 11, no. 3, pp. 167–172, 1988. 2019, and the M.S. degree in electrical engineering in 2021 from the Pohang
[23] J. Doyle, K. Glover, P. P. Khargonekar, and B. A. Francis, “State–space University of Science and Technology, Pohang, South Korea, where he is
solutions to standard H2 and H control problems,” IEEE Trans. Automat. currently working toward the Ph.D. degree in electrical engineering. His research
Control, vol. 34, no. 8, pp. 831–847, Aug. 1989. interests include power system control, renewable energy sources, and HVDC
[24] B. Moore, “Principal component analysis in linear systems: Controllabil- systems.
ity, observability, and model reduction,” IEEE Trans. Automat. Control,
vol. 26, no. 1, pp. 17–32, Feb. 1981.
[25] Y. Pipelzadeh, B. Chaudhuri, and T. C. Green, “Inertial response from Jae-Young Park (Member, IEEE) received the B.S. degree (Hons.) in electrical
remote offshore wind farms connected through VSC–HVDC links: A engineering from Konkuk University, Seoul, South Korea, in 2015, and the Ph.D.
communication–less scheme,” in Proc. IEEE Power Energy Soc. Gen. degree in convergence IT engineering from the Pohang University of Science and
Meeting, 2012, pp. 1–6. Technology, Pohang, South Korea, in 2022. He is currently a Senior Researcher
[26] T. Sugiyama, T. Nishiwaki, S. Takeda, and S. Abe, “Measure- with the Korea Institute of Energy Research, Daejeon, South Korea. His research
ments of synchronous machine parameters under operating condition,” interests include self-healing power systems, renewable energy resources, and
IEEE Trans. Power App. Syst., vol. PAS-101, no. 4, pp. 895–904, grid-forming power converters.
Apr. 1982.
[27] L. Wang and M. N. Thi, “Stability enhancement of a PMSG-based offshore
wind farm fed to a multi-machine system through an LCC-HVDC link,” Young-Jin Kim (Senior Member, IEEE) received the B.S. and M.S. degrees
IEEE Trans. Power Syst., vol. 28, no. 3, pp. 3327–3334, Aug. 2013. in electrical engineering from Seoul National University, Seoul, South Korea,
[28] K. Zhou et al., “Topologies and control of VSC-HVDC systems for grid in 2007 and 2010, respectively, and the Ph.D. degree in electrical engineering
connection of largescale off-shore wind farms,” in Proc. Int. Conf. Elect. from the Massachusetts Institute of Technology, Cambridge, MA, USA, in 2015.
Mach. Syst., Wuhan, China, 2008, pp. 2357–2361. From 2007 to 2011, he was with Korea Electric Power Corporation as a Power
[29] Normalized Signal Test, PJM, 2019. [Online]. Available: https://www.pjm. Transmission and Distribution System Engineer. He was also a Visiting Scholar
com/-/media/markets-ops/ancillary/reg-d.ashx with the Catalonia Institute for Energy Research in 2014, and a Postdoctoral
[30] M. Raza, E. Prieto-Araujo, and O. Gomis-Bellmunt, “Small-signal stabil- Researcher with the Center for Energy, Environmental, and Economic Systems
ity analysis of offshore ac network having multiple VSC-HVDC systems,” Analysis, Energy Systems Division, Argonne National Laboratory from 2015
IEEE Trans. Power Del., vol. 33, no. 2, pp. 830–839, Apr. 2018. to 2016. He joined the Faculty with the Pohang University of Science and
[31] P. W. Sauer and M. A. Pai, Power System Dynamics and Stability. Upper Technology, where he is currently an Associate Professor with the Department
Saddle River, NJ, USA: Prentice–Hall, 1998, pp. 95–99. of Electrical Engineering. His research interests include distributed generators,
[32] P. Kundur, Power System Stability and Control. New York, NY, USA: renewable energy resources, and smart buildings.
McGraw–Hill, 1994, pp. 341–444.
[33] J. Z. Zhou, H. Ding, S. Fan, Y. Zhang, and A. M. Gole, “Impact of
short–circuit ratio and phase–locked–loop parameters on the small–signal Oriol Gomis-Bellmunt (Fellow, IEEE) received the Degree in industrial en-
behavior of a VSC–HVDC converter,” IEEE Trans. Power Del., vol. 29, gineering from the School of Industrial Engineering of Barcelona, Technical
no. 5, pp. 2287–2296, Oct. 2014. University of Catalonia (UPC), Barcelona, Spain, in 2001, and the Ph.D. degree
[34] S. V. Bozhko, R. Blasco-Gimenez, R. Li, J. C. Clare, and G. M. Asher, in electrical engineering from the UPC in 2007. In 1999, he joined Engitrol S.L.,
“Control of offshore DFIG–based wind farm grid with line–commutated where he was a Project Engineer with Automation and Control Industry. Since
HVDC connection,” IEEE Trans. Energy Convers., vol. 22, no. 1, 2004, he has been with Electrical Engineering Department, UPC, where he is
pp. 71–78, Mar. 2007. a Professor and participates with CITCEA-UPC Research Group. Since 2020,
[35] R. Vijayapriya, P. Raja, and M. P. Selvan, “A modified active power he has been an ICREA Academia Researcher. His research interests include the
control scheme for enhanced operation of PMSG-based WGs,” IEEE fields linked with electrical machines, power electronics, and renewable energy
Trans. Sustain. Energy, vol. 9, no. 2, pp. 630–638, Apr. 2018. integration in power systems.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on May 06,2024 at 05:50:37 UTC from IEEE Xplore. Restrictions apply.