Advanced Grid-Forming Control of HVDC Systems

for Reliable Grid Restoration

Florian Mahr, Johann Jaeger
Chair of Electrical Energy Systems
Friedrich-Alexander-University of Erlangen-Nuremberg (FAU)
Erlangen, Germany
[email protected], [email protected]

Abstract—Grid restoration strategies as emergency response in [1]. Besides start-up modes of HVDC systems are discussed
gain increased attendance that arise from fundamental changes [2]. Different grid-forming HVDC control topologies with
inside electrical grids. Due to the integration of fluctuating energy external current limiting methods are analyzed in [3]. In [4]
sources, the reduction of installed inertia as well as natural
disasters a blackout might occur. High-voltage direct current system restoration using a VSC-HVDC connected offshore
(HVDC) systems open new possibilities to use electrical energy wind power plant as black-start unit is proposed.
from a far distance and break new ground during grid restoration In this paper a predictive grid-forming controller, that ex-
process through reliable power energy sources. In grid-forming tends the considerations and enhances the technical solutions
control mode special attention has to be paid on converter current above, is proposed. Following a holistic approach, limitation
limitation and handle unbalanced grid conditions.
In this paper a dynamic protection integrated model predictive of converter currents is included inside the control algorithm.
control (DPI-MPC) concept using a generalized state space If converter currents, which are not controlled in grid-forming
representation of a grid-connected modular multilevel converter mode inherently, would exceed limits, unsafe operation and
(MMC) is derived. It enables to reconstruct an AC-grid after grid disconnection might occur and consequently a restart
a blackout utilizing a VSC-HVDC system. With this approach, of the restoration process had to be initiated. Moreover a
it is possible to restore a balanced voltage system even under
unbalanced grid conditions and ensure functional protection of generalized state space model is developed to reconstruct a
the grid-forming unit. This concept guarantees a more safe balanced three-phase voltage system even under the presence
and reliable restoration process, which is validated in EMT of unbalanced sources and loads.
simulations using MATLAB/Simulink. The results demonstrate In Section II a generalized mathematical model of the grid-
that DPI-MPC and HVDC systems go together in a natural way connected MMC in grid-forming mode considering unbal-
offering novel possibilities during grid restoration process.
Index Terms—Predictive control, grid-forming, HVDC trans- anced conditions using state space representation is derived.
mission, grid restoration, balanced voltage, converter current Based on the comprehensive model of the converter, the
limitation DPI-MPC algorithm is presented in Section III in detail. To
investigate the proposed controller, simulation results proving
I. I NTRODUCTION dynamic and stable performance of DPI-MPC are presented
HVDC systems will play a significant role in future trans- in Section IV. Section V concludes the paper with a summary.
mission grids, especially because of their capability to trans-
port electrical energy over long distances with high efficiency.
II. S TATE SPACE MODELING
On-site mismatch of producers and loads can be overcome by
the installation of HVDC systems, e.g. by point-to-point links A. Equivalent circuit of the grid-forming MMC
or broad Multiterminal-HVDC-grids. VSC-HVDC technology In this paper, a VSC-HVDC station is considered as the
based on modular multilevel converters represent the most only grid-forming unit determining the frequency and the
promising power electronic technology to fulfill numerous phase angle. The three-phase double-star MMC is a circuit
demands from the AC- and DC-grid-side. The control algo- configuration to link an AC- and a DC-system that consists
rithm defines the behavior of the MMC together with passive of two arms per phase-leg. The DC-side is connected to the
components on the grid side. stars of the upper and lower arms, whereas the AC-side is
A blackout of the power grid is the worst-case scenario connected between the two arms of one phase leg. An inductor
which is followed by the restoration process. High power and a few hundred cells are series-connected per arm. These
HVDC stations offer new possibilities during this process. cells are controlled in a way to imitate a voltage source on a
For example auxiliary power of thermal power plants can be macroscopic point of view to adjust currents on the AC- and
provided by the HVDC station in early restoration stages to in- the DC-side and the circulating currents inside the converter
crease assured power and raise installed inertia by connection [5]. The upper (lower) arm of each phase-leg is represented
of further energy resources. by the index ’u’ (’l’).
In literature general investigations on HVDC systems as In grid-forming mode a capacitor bank at the point on
black start units in comparison to traditional units are found common coupling (PCC) is proposed to serve as a reliable

978-1-5386-7703-2/18/$31.00 ©2018 IEEE

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE MINAS GERAIS. Downloaded on April 25,2022 at 23:02:35 UTC from IEEE Xplore. Restrictions apply.
 + +
isd −σ1 ω 0 0 −σ2 0 0 0 isd σ2 0 0 0 0 0 0 0
© + ª © ª© + ª ©
σ2 0
ª © ª
­ isq ® ­ −ω −σ1 0 0 0 −σ2 0 0 ® ­ isq ® ­ 0 0 ® ­ 0 0 0 0 ®®
® ­ ® ® ­
+ +
­ ® ­
­ i− ® ­ 0 0 −σ −ω 0 0 −σ 0 ® ­ i − ® ­­ 0 0 σ2 0 ®® vsd ­ 0 0 0 0 ®® igd
­ sd ® ­ 1 2 ® ­ sd ® ­ ­
­ − ® ­ ®­ − ® ®© + ª ­ ®© + ª
d ­­ isq ®® ­­ 0 0 ω −σ1 0 0 0 −σ2 ®® ­­ isq ®® ­­ 0 0 0 σ2 ®® ­­ vsq ®® ­­ 0 0 0 0 ®® ­­ igq ®®
+ ®=­ + +­ − ®+­ − ® (4)
dt ­­ vgd ® ­ σ3 0 0 0 0 ω 0 0 ®® ­­ vgd ®® ­­ 0 0 0 0 ® ­ vsd ® ­ −σ3 0 0 0 ®® ­­ igd
® ­
®
­ + ® ­ ®­ + ® ­ ® − ® −
­ vgq ® ­ 0 σ3 0
­
­ ® ­ 0 −ω 0 0 0 ®® ­­ vgq ® ­ 0
® 0 0 0 ® « vsq ¬ ­ 0 −σ3 0
® ­ 0 ® « igq ¬
®
­ v− ® ­ 0 0 σ3 0 0 0 ® ­ − ® ­
0 −ω ® ­ vgd ® ­ 0 0 0 0 ® u
® |{z} ­
0 0 −σ 0 ® |{z}
­ gd ® ­ ­ 3 ® z
− −
« vgq ¬ « 0 0 0 σ3 0 0 ω 0 ¬ « vgq ¬ « 0 0 0 0 ¬ « 0 0 0 −σ3 ¬
|{z} | {z } | {z } | {z }
x Acont. Bcont. Econt.

C. Differential equations using DDSRF-components

iua
+ + +
vua Unbalanced three-phase signals without a zero sequence
component and harmonics can be described according to (3)
Lua id by superposition of positive (superscript ’+’) and negative
Rua sequence (superscript ’-’) components, where ω = 2π · f [6].
iga
Rsa Lsa isa +
vd cos (ωt + ϕ+ ) cos (ωt + ϕ− )
va 2π + ® ­ cos (ωt + 2π + ϕ− ) ®
© ª © ª
­ b ® = V̂ + · ­­ cos (ωt − 3 + ϕ ) ®® +V̂ − · ­­
­
Ld
©v ª
3 ®
®
« vc ¬
­ 2π +)
® ­ 2π ®
+ϕ )−
« cos (ωt + 3 + ϕ ¬
cos (ωt −
Rd « 3 ¬
Rla
(3)
Cga vgc
Lla The composition of positive and negative sequence compo-
nents can be separated and handled independently from each
+ + +
vla other using the principle of the decoupled double synchronous
reference frame (DDSRF). The space vector v, which draws
ila an ellipse in steady state on complex plane, is composed by
the space vector of the positive sequence v+ and the space
vector of the negative sequence v− . The projection of both
Fig. 1. Equivalent circuit of the MMC space vectors on the real and imaginary axis of the appropriate
coordinate system gives the DDSRF components [6].
voltage control target with constant parameters that ensures The state space description of the AC-side of the MMC
voltage robustness against current deviations throughout the using DDSRF components of currents and voltages is given
restoration process. Parallel sources and loads are modeled as in (4), where resistors, inductors and capacitors on the AC-
uncontrolled current sources. Besides the connection between side and inside the MMC are considered to be equal in
s +Rc
the MMC and the PCC is represented by a passive network. each phase. In addition, the abbreviations σ1 = 2R
2Ls +Lc , σ2 =
2Ls +Lc and σ3 = Cg are used.
2 1
The equivalent circuit is shown in Fig. 1.
The matrix C is applied to select the DDSRF components
B. Separation of AC-, inner converter- and DC-side of the capacitor voltages as output variables.
+
Applying Kirchhoffs voltage and current law and transform- isd
ing variables using the subsequent matrix equations (1) and © + ª
­ isq ®
(2), the AC-, the inner converter- and the DC-side can be +
­
−
®
vgd 0 0 0 0 1 0 0 0 ­­ isd ®®
seperated, where the index y ∈ {a, b, c} represents the phase © + ª © ª­ − ®
leg. ­ vgq ® ­ 0 0 0 0 0 1 0 0 ®® ­­ isq ®®
­ v − ® = ­­ 0 + (5)
­ ®
v 
sy 1 −1 1 vuy
   ­ gd ® 0 0 0 0 0 1 0 ®® ­­ vgd ®®
= · (1) −
« vgq ¬ « 0 0 0 0 0 0 0
­ + ®
1 ¬ ­­ vgq ®®
vcy 2 vly
 1 1   |{z} | {z }­ − ®
isy 1 2 −2 iuy ­ vgd ®
 
Ccont.
= · (2) y
−
icy 2 1 1 ily « vgq ¬

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE MINAS GERAIS. Downloaded on April 25,2022 at 23:02:35 UTC from IEEE Xplore. Restrictions apply.
 III. DYNAMIC PROTECTION INTEGRATED MPC B. Prediction of system trajectories and optimization
Subsequent the design and implementation of the dynamic Based on the current system state, the trajectories of all state
protection integrated model predictive control (DPI-MPC) is variables are predicted with respect to the input variables and
derived based on the state space formulation in (4). Because within the prediction horizon np . It represents the number of
of the high number of possible switching states of the MMC samples in the future that are calculated. The control horizon
circuit that provoke very high computational effort using finite nc correspond to the number of changes of the input variables
control set model predictive control (FCS-MPC), a continuous that are performed inside the prediction horizon [7]. Effects
control set model predictive controller with underlaying mod- due to unknown disturbance variables can not be taken into
ulator according to [7] is proposed. Fig. 2 shows the principle account.
of DPI-MPC.
A recursive law can be obtained for calculating the predicted
state variables at t = k · TS for future samples t = (k + i) · TS .
isy,max
DPI-MPC
i−1
Õ
x(k + i | k) = Ai x(k) + Ai−j−1 Bu(k + j) (12)
j=0

x Predictive controller for Dynamic u with i = 1... np and u(k + i) = u(k + nc − 1) for i ≥ nc .
positive sequence current limiter
The predicted output variables can be obtained using
isy,min Predictive controller for Dynamic y(k + i | k) = Cx(k + i | k). (13)
negative sequence current limiter
To calculate input variables a cost function J is defined.
np
Õ
Fig. 2. Principle of DPI-MPC
J= (w(k + i) − y(k + i)) Q (w(k + i) − y(k + i))
i=1
(14)
nÕc −1
In state of the art control schemes, cascaded controllers + ∆u(k + i)R∆u(k + i)
for positive and negative sequence are used. Both consist of i=0
an outer voltage feedback control loop and an inner current
where ∆u(k + i) = u(k + i) − u(k).
feedback control loop. Current limitation circuits are added
artificially to the control loop. By contrast DPI-MPC includes Minimization of the cost function under receding horizon
the controllers for the positive and negative sequence, as policy with respect to the constraints (15) and (16) leads to
well as an intuitive phase current limitation scheme inside optimal input variables that affect the reduction of the control
the control algorithm. This natural way of incorporation of error.
constraints, that are taken into account when minimizing the x(k + i | k) ≤ xmax (15)
objective function, is a unique characteristic of the MPC
x(k + i | k) ≥ xmin (16)
scheme. Thus limit violations can be predicted and retaliatory
action can be initiated before limits are reached. C. Converter current limitation

A. System discretization The DPI concept includes the dynamic change of the
The first step for implementing DPI-MPC is to apply zero- maximum and minimum values of the converter currents isy
order hold discretization with a sampling period TS according that represent a subset of the state vector x. It is defined that
to (6)-(9). isy,max = −isy,min . (17)
A=e Acont. ·TS
(6)
The maximum value of the converter currents has to be
∫ TS 
shared upon their DDSRF components in an intuitive way.
B= eAcont. ·τ dτ · Bcont. (7)
0 Due to the high relevance of a safe and stable grid restoration
C = Ccont. (8) process, a conservative rule for current limitation is proposed:
∫ TS  r r
 2  2  2  2
E= eAcont. ·τ dτ · Econt. (9) + +
isd + isq + −
isd + isq− ≤ isy,max (18)
0
The differential equations from (4) and (4) in continuous-time Hence, the maximum converter current is shared between
domain turn to difference equations in discrete-time domain the positive and the negative sequence based on the unbalance
with t = k · TS , where k ∈ N. factor k1 of the disturbance currents.
x(k + 1) = Ax(k) + Bu(k) + Ez(k) (10) Iˆg−
k1 B (19)
y(k) = Cx(k) (11) Iˆg+

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE MINAS GERAIS. Downloaded on April 25,2022 at 23:02:35 UTC from IEEE Xplore. Restrictions apply.
 To overcome dominant positive and negative disturbance IV. S IMULATION RESULTS
currents, the factor k 2 is proposed: In order to validate the proposed DPI-MPC concept, EMT

k 1, for k 1 ≤ 1 (20a) simulations in MATLAB/Simulink in grid restoration scenar-
k2 B ios utilizing a VSC-HVDC station with a fixed frequency of
1/k1, for k 1 > 1 (20b)
50 Hz have been performed. The simulation parameters are
According to the current value of k2 , the maximum values listed in Table I.
of the magnitudes of the space vectors of the converter currents A. Voltage ramp-up
can be calculated.
In the first scenario the ramp-up of grid voltages without
|i+s |max = (1 − k2 ) · isy,max
!
(21) disturbance currents is performed. The desired balanced phase-
! to-phase voltage of 380 kV is reached within 10 ms while the
|i−s |max = k2 · isy,max (22)
current overshoot is limited in a ’smooth’ way without any
These magnitudes are portioned to the appropriate d- and hardware- (charging resistors) or software features (artificial
q-component with respect to the ratios of their components decreasing the slope of the reference variable) to 3 kA. The
according to [8]. waveforms of the grid voltages and the converter currents are
+
shown in Fig. 4 and Fig. 5.
= |i+s |max · cos(γ + )
!
isd,max (23)
+
= |i+s |max · | sin(γ + )|
!
isq,max (24) 600
vgab
! vgbc
−
isd,max = |i−s |max · cos(γ − ) (25) 400
vgca
!
−
isq,max = |i−s |max · | sin(γ − )| (26) 200
with Voltage/kV 0
+
!
isq
γ + = arctan + (27) −200
isd
−
! −400
isq
γ = arctan
−
− (28)
isd −600
0.00 0.01 0.02 0.03 0.04 0.05
The flow diagram in Fig. 3 summarizes the procedure of Time/s
DPI-MPC.
Fig. 4. Phase-to-phase voltages during ramp-up process

x
3
isa
2 isb
Reconfiguration
isc
1
Current/kA

0
Prediction of future Calculate limits on
trajectories DDSRF-components −1

−2

Solve optimization
−3
0.00 0.01 0.02 0.03 0.04 0.05
problem under constraints
Time/s

Fig. 5. Converter currents during ramp-up process

Actuate input variables This scenario demonstrates the stable operation and optimal
utilization of hardware limits of the HVDC system during
voltage ramp-up procedure.

u B. Balanced grid-forming under unbalanced conditions

The restoration of a balanced phase-to-phase voltage of
Fig. 3. Flow diagram of DPI-MPC 380 kV under unbalanced load currents is tested in the second

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE MINAS GERAIS. Downloaded on April 25,2022 at 23:02:35 UTC from IEEE Xplore. Restrictions apply.
 TABLE I 8
S IMULATION PARAMETERS isa
6
isb
f 50 Hz 4 isc

Current/kA
Rs 5Ω 2
Ls 150 mH 0
Cg 15 µF −2
Ru = Rl 1.5 Ω −4
Lu = Ll 50 mH −6
−8
0.05 0.10 0.15 0.20 0.25
scenario. After 50 ms a unbalanced load of 2000 MW is Time/s
ramped-up within 100 ms. During this time the drop of the
Fig. 8. Converter currents during balancing process
amplitude of the grid-voltages and the unbalance factor are
limited by the controller. In steady state, the control error is
reduced to zero by feeding the negative sequence load currents V. C ONCLUSION
through the MMC without affecting the grid voltages. Fig. 6-
This paper deals with the grid restoration process using
8 show the disturbance currents, the grid voltages, and the
a VSC-HVDC system. To avoid unsafe operation and grid
converter currents.
disconnection during this process, a novel control algorithm
called dynamic protection integrated model predictive control
8
(DPI-MPC) is proposed. By incorporating voltage controllers
6 iga
for both positive and negative sequence components and
igb
4 a current limiting scheme, this control concept enables to
igc
Current/kA

2 reconstruct a balanced grid voltage, even under the presence

of unbalanced loads or producers, while limiting the converter
0 currents in an inherent way. HVDC stations, equipped with this
−2 control algorithm, open new reliable restoration paths from a
−4 far distance. New strategies to restore the utility grid after a
−6 blackout can be developed.
−8 R EFERENCES
0.05 0.10 0.15 0.20 0.25
[1] Y. Jiang-Hafner, H. Duchen, M. Karlsson, L. Ronstrom, and B. Abra-
Time/s hamsson, “Hvdc with voltage source converters - a powerful standby
black start facility,” in 2008 IEEE/PES Transmission and Distribution
Fig. 6. Disturbance currents during balancing process Conference and Exposition. IEEE, 2008, pp. 1–9.
[2] G. Xu, L. Tu, H. Lan, and C. Xia, “Analysis on hvdc start-up in system
black-start,” in 2015 5th International Conference on Electric Utility
Deregulation and Restructuring and Power Technologies (DRPT). IEEE,
600 2015, pp. 1705–1709.
vgab [3] R. Shah, M. Barnes, and R. Preece, “Offshore ac grid management for
an ac integrated vsc-hvdc scheme with large wpps,” in 2016 IEEE Power
400 vgbc and Energy Society General Meeting (PESGM). IEEE, 2016, pp. 1–5.
vgca [4] H. Becker, A. Naranovich, T. Hennig, A. Akbulut, D. Mende, S. Stock,
200
Voltage/kV

and L. Hofmann, “System restoration using vsc-hvdc connected offshore

wind power plant as black-start unit,” in 2017 19th European Conference
0 on Power Electronics and Applications (EPE’17 ECCE-Europe). IEEE,
2017.
−200 [5] K. Sharifabadi, H. P. Nee, L. Harnefors, R. Teodorescu, and S. Norrga,
Design, Control and Application of Modular Multilevel Converters for
HVDC Transmission Systems. Chichester, West Sussex, United Kingdom:
−400 John Wiley & Sons Inc, 2016.
[6] R. Teodorescu, M. Liserre, and P. Rodrı́guez, Grid Converters for
−600 Photovoltaic and Wind Power Systems. Chichester, West Sussey: Wiley
0.05 0.10 0.15 0.20 0.25 IEEE, 2011.
Time/s [7] T. Geyer, Model predictive control of high power converters and industrial
drives, first edition ed. Chichester, West Sussex, United Kingdom: Wiley,
2017.
Fig. 7. Phase-to-phase voltages during balancing process [8] A. Gkountaras, “Modeling techniques and control strategies for inverter
dominated microgrids,” Ph.D. dissertation, Technical University of Berlin,
This scenario shows the ability to reconstruct a balanced 2016.
voltage system under unbalanced conditions.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE MINAS GERAIS. Downloaded on April 25,2022 at 23:02:35 UTC from IEEE Xplore. Restrictions apply.