1

Frequency and voltage regulation in hybrid AC/DC

networks
Jeremy D. Watson and Ioannis Lestas

Abstract—Hybrid AC/DC networks are a key technology for frequency and voltages are appropriately regulated. This is a
sustainable electrical power systems, due to the increasing num- challenging problem since the ILC operation simultaneously
ber of converter-based distributed energy resources such as solar affects the AC frequency and the DC voltage. Moreover,
or wind. In this paper, we consider the design of control schemes
for hybrid AC/DC networks, focusing especially on the control a prescribed allocation is desired in many cases, such that
of the interlinking converters (ILC(s)). We present two control economic optimality is achieved among generating units. Fur-
schemes: firstly for decentralized primary control, and secondly, thermore, distributed techniques for generation control are de-
a distributed controller to achieve secondary control objectives sirable due to the increasing penetration of renewable sources
as well. In the primary case, the stability of the controlled of generation which significantly increases the number of
system is proven in a general hybrid AC/DC network which
may include asynchronous AC subsystems. Furthermore, it is active elements in power grids, making traditional centralized
demonstrated that power-sharing across the AC/DC network is approaches impractical and costly.
significantly improved compared to previously proposed dual Related work: Numerous controllers for either AC or DC
droop control. The proposed scheme for secondary control networks or microgrids alone have been proposed recently, e.g.
guarantees the convergence of the AC system frequencies and from simple droop-based strategies to sliding mode control for
the average DC voltage of each DC subsystem to their nominal
values respectively. An optimal power allocation is also achieved DC networks in [6], distributed consensus for DC networks in
at steady-state. The applicability and effectiveness of the proposed [7]-[8], and model predictive control in [9]. For AC microgrids
algorithms are verified by advanced simulations on a test hybrid the literature is even more extensive, as surveyed in [10] for
AC/DC network in Simscape Power Systems. example. However, we focus on the control of hybrid AC/DC
Index Terms—AC/DC grids, frequency control, voltage control, network, which presents new challenges as control actions on
distributed control either the AC or DC sides affect the entire network. It is
also more difficult to achieve optimality of power allocation
between AC and DC sources.
I. I NTRODUCTION
Both primary and secondary control strategies are required
Motivation: In view of the increasing number of converter- for AC/DC networks. In comparison to the literature on AC
interfaced energy sources such as solar or wind generation, networks or DC networks, relatively few control approaches
hybrid AC/DC networks are a key technology for sustainable for AC/DC networks have been proposed, especially for sec-
electrical power systems. The hybrid AC/DC network allows ondary control. Such control schemes are implemented via
the easy integration of such renewable energy sources and can the AC / DC sources in conjunction with the ILC and must
combine the advantages of both DC grids and AC grids into regulate the AC frequency, DC voltage, and power allocations.
one power network which operates with high efficiency. Direct We briefly review the primary and secondary control schemes
current grids have several advantages [2] over traditional AC in the literature for the ILC.
systems: lower power losses, largely due to the absence of Droop control schemes are decentralized, intuitive, and easy
reactive power; higher power transfer capability; and DC grids to implement. AC frequency droop is ubiquitous, while DC
can also facilitate the connection of asynchronous AC grids. voltage droop controllers are prevalent in the literature as well.
However, AC technology is well established and is more The DC bus voltage dynamics are comparable to traditional
suitable for some applications. Therefore, it is advantageous to AC frequency / real power control, in that an excess of
combine AC and DC networks via interlinking voltage source active power increases the voltage and vice versa. Hence,
converters (VSC) to form a hybrid AC/DC network [3]. an obvious way to control the interlinking converter is a
Hybrid AC/DC networks present new challenges in terms dual droop scheme combining the two characteristics in one
of frequency and voltage control [4]. In particular, an open controller. The DC voltage droop stabilizes the DC grid and
problem [5] is the control of the interlinking converter (ILC), the DC system participates in the frequency regulation of the
where the aim is to guarantee stability while ensuring that the connected AC grids via the ILC [11]. However, the two droop
schemes interact with each other in a way that degrades their
This work was supported by ERC starting grant 679774. performance, as noticed by [12]. A strategy using the ILC
Jeremy D. Watson and Ioannis Lestas are with the Department of Engi-
neering, University of Cambridge, Trumpington Street, Cambridge, CB2 1PZ, power to improve performance is presented in [13], although
United Kingdom (e-mail: jdw69@[Link], icl20@[Link]). the coupling between the AC and DC grids still introduces
A preliminary version of this work appeared in [1]. This manuscript some inaccuracy.
includes an improved distributed control scheme, the analytic proofs of the
main results, more advanced simulations, as well as additional discussion that Another approach for the control of the ILCs was presented
demonstrate the applicability of the proposed analysis. in [3], [16], and [17], where the per-unit values of the AC
 2

frequency and DC voltage are synchronized by controlling by [21] for general hybrid AC/DC networks. For this
the power transfers through the ILC with a proportional- setting, we provide stability guarantees and sufficient
integral (PI) controller. This allows the ILC to relate the AC conditions for an optimal steady state power allocation.
frequencies and DC voltages to each other; and the power 2) We propose a novel distributed approach for the con-
allocation is then determined by the droop coefficients of the trol of interlinking converters and generation sources
AC and DC sources. This approach is effective, although the for secondary frequency and voltage regulation which
slower integral term makes the use of inertia through the ILC guarantees the convergence of both the AC system fre-
difficult to achieve. quencies and the weighted average DC voltage of each
A similar strategy has also been proposed for secondary DC subsystem to their nominal values. A prescribed
control, where the aim is to restore the frequencies and volt- power sharing is also achieved. We also show that virtual
ages to their nominal value and to share power equitably be- capacitance in the controller can be used to adjust the
tween sources in the AC and DC subsystems [4]. A distributed voltage deviations and improve performance.
consensus approach was proposed in [14], where again the AC 3) We perform realistic simulations with advanced con-
frequency and DC voltage are synchronized by PI control of verter and generator models in a suitable test system,
the ILC. This approach is effective for a AC / DC networks verifying the performance of our control strategies and
with only one ILC bus connection. However, in the case of DC comparing them to the traditional dual-droop controller.
subsystems with multiple interconnections at different buses, Paper structure: The paper is organized as follows. Section
regulating the local voltage deviations to zero via the ILC is II presents the network model and formulates the control
not optimal because this affects the power transfer capability problem. Our main results are given in section III, including
through such a DC subsystem. In [18], a distributed controller the proposed decentralized primary controller in section III-A
for sharing frequency reserves of asynchronous AC systems and the distributed controller in section III-B. The performance
via HVDC was designed. However the DC voltage dynamics of the two controllers is illustrated via case studies in section
were not modelled. The authors in [19] designed distributed IV and compared to traditional controllers. Finally, conclu-
controllers for distributed frequency control of asynchronous sions are presented in section V. The proofs of all the results
AC systems connected through a MTDC grid. presented can be found in the appendix.
Contribution: In view of the current literature, there is a
need for new control schemes which: can be rigorously proven II. P ROBLEM FORMULATION
to be stable in any AC/DC network topology; can achieve an
A. Network model
appropriate steady-state power allocation without the use of
communication; and can make use of the DC system(s) and We consider a general hybrid AC/DC network with the
ILC(s) to provide inertia for the AC system(s) and vice versa set of buses denoted by N = (1, 2, ..., |N |) and the set of
[20]. transmission lines by E = (1, 2, ..., |E|). The network is
In this paper, we present new voltage source converter composed of multiple AC and DC subsystems. We denote the
control schemes for the interlinking converters in hybrid set of subsystems by K = (1, 2, ..., |K|) and we also have
AC/DC networks. Inspired by the controller proposed in [21], N = (∪i Nidc ) ∪ (∪j Njac ), where Niac and Njdc denote the
where linking the DC voltage to the AC frequency was found collection of buses belonging to the AC subsystem i ∈ K and
to provide desirable stability and performance properties in DC subsystem j ∈ K respectively. Each subsystem is assumed
AC networks, we show the successful application of similar to be connected and it is connected to the rest of the network
ideas to general hybrid AC/DC networks for both primary and only via interlinking converters1 , as illustrated in Figure 1.
secondary control. The proposed schemes have advantages Each AC subsystem i is described by the connected graph
over previous proposals, including the existence of rigorous (Niac , Eiac ) with arbitrary direction, and each DC subsystem
stability guarantees in general network topologies, improved j by the connected graph (Njdc , Ejdc ) with arbitrary direction2 .
power sharing, and the ability of the proposed controllers to For each bus j ∈ N we use i : i → j and k : j → k to denote
use the energy stored in the DC capacitance as the “inertia” the predecessors and successors of bus j respectively. For con-
for the emulation of synchronous machines. In particular, our venience we also define the set of all AC buses Nac = ∪i Niac
decentralized control approach, by making use of the energy and all DC buses Ndc = ∪j Njdc , such that Nac ∪ Ndc = N ;
stored in DC-side capacitance, achieves an improved steady- likewise we define the set of all AC edges Eac = ∪i Eiac
state power allocation and primary frequency regulation com- and all DC edges Edc = ∪j Ejdc . Connections between AC
pared to schemes that directly control the power transfer. We and DC buses are facilitated by the interlinking converters,
also propose a scheme for distributed secondary frequency and the set of which is denoted by X = (1, 2, ..., |X|). The ILC
voltage control which regulates the frequency and the weighted buses are denoted by Xjac ∈ Nac and Xjdc ∈ Ndc for the AC
average voltages of DC subgrids to prescribed nominal values and DC buses, respectively, to which the ILC j is connected.
at steady state, and also leads to an optimal power sharing. The set of AC buses to which a converter is connected is
Moreover, we show that virtual capacitance in the controller denoted by Xac = (X1ac , . . . , X|X|
ac
) ⊂ Nac . Similarly, the set
can be used to further improve performance.
1 Note that this is without loss of generality since the connection of two
For clarity, we summarize the main contributions of the
collections of AC buses (or DC buses respectively) may simply be considered
paper below: as one larger subsystem.
1) We propose a decentralized VSC controller inspired 2 The results presented in the paper are unaffected by the choice of direction.
 3

of DC buses to which a converter is connected is denoted by through the ILC. Hence for a converter bus j ∈ Xac , the power
Xdc = (X1dc , . . . , X|X|
dc
) ⊂ Ndc . transfer pX
j is the AC-to-DC transfer, whereas for a converter
bus j ∈ Xdc , the power transfer pX
j is the DC-to-AC transfer.
Table I
N OTATION IN SYSTEM MODEL
We assume that such power transfers are instantaneous and
lossless, hence for an ILC x we have pX X
X dc = −pXxac .
ωj AC frequency deviation at bus j x

ηij voltage angle between two AC buses i and j

Mj inertia at bus j
pG
j generated power at bus j
pL
j load power at bus j
pX
j interlinking converter power transfer at bus j
pij power transfer between buses i and j
Bij transmission line susceptance for (i, j) ∈ Eac
Dj damping coefficient at bus j ∈ Nac \ Xac Figure 1. ILC connection diagram
Cj capacitance at bus j ∈ Ndc Given these assumptions and definitions, the hybrid AC/DC
Vj DC voltage deviation at bus j network dynamics are:
Gij line conductance for (i, j) ∈ Edc
η̇ij = ωi − ωj , (i, j) ∈ Eac (1a)
Assumption 1. We make the following assumptions regarding X X
the network: Mj ω̇j = pG L
j − pj + pij − pjk − Dj ωj ,
• 1a: Voltage magnitudes are 1 p.u. for all buses j ∈ Nac . i:i→j k:j→k

• 1b: Lines (i, j) ∈ Eac are lossless and are characterized j ∈ Nac \ Xac (1b)
by their constant reactances Xij > 0.
X X
0 = pG L
j − pj + pij − pjk − pX
j , j ∈ Xac
• 1c: Reactive power does not affect either bus voltage
i:i→j k:j→k
angles or the frequency, and is thus ignored. (1c)
• 1d: The AC system(s) are three-phase balanced. X X
• 1e: Bus voltages are close to 1 p.u. for all j ∈ Ndc , such
Cj V̇j = pG L
j − pj + pij − pjk − pX
j , j ∈ Ndc
i:i→j k:j→k
that currents and powers are approximately equivalent in
(1d)
a per-unit system. (
• 1f: Lines (i, j) ∈ Edc are characterized by their conduc- Bij sin ηij j ∈ Nac
tance Gij = R1ij , where Rij is the line resistance. The pij = , (i, j) ∈ E (1e)
Gij (Vi − Vj ) j ∈ Ndc
line losses are small and may be neglected.
Remark 1: Assumption 1 may be explained as follows: We now write the system dynamics in matrix form. The
vector of angle differences is η = [ηij ](i,j)∈Eac , the vector
• 1a-d: These are well-known assumptions for AC trans-
of AC frequency deviations from its nominal value (50 or
mission systems which are used in much of the literature.
60 Hz) is denoted by ω = [ωj ]j∈Nac , while the vector of
These assumptions allow us to model the active power
DC voltage deviations from their nominal value is denoted by
transfer through a line (i, j) as pij = Bij sin ηij where
−1 V = [Vj ]j∈Ndc . M is the diagonal matrix of the generator
Bij = 3Xij > 0.
inertias Mj , j ∈ Nac \ Xac , while the damping coefficients
• 1e-f: These are typical assumptions in DC networks [19],
are D = diag([Dj ]j∈Nac \Xac ). The frequencies at the cor-
making use of the fact that in practice the DC grid voltage
responding AC buses are denoted by ω G = [ωj ]j∈Nac \Xac ,
will typically be regulated very close to the nominal and
and the vector of frequencies at the converter buses is ω X =
that line losses are generally small. In particular, although
[ωj ]j∈Xac . The diagonal matrix of the DC bus capacitances
the percentage change in DC voltage differences between
is C = diag([Cj ]j∈Ndc ). The vector of generator powers is
buses can be large (and thus cannot be ignored), the
denoted by pG = [pG j ]j∈N , the vector of load powers by
percentage change in individual DC voltages is small
pL = [pL ]
j j∈N . We also use the notation pX j = 0 for buses
(also demonstrated in our simulations in Section IV with
without converters, i.e. j ∈ N \ (Xac ∪ Xdc ) and denote the
the proposed controller). 3
vector of converter powers by pX = [pX j ]j∈N . Similarly, at
We also consider the modelling of the interlinking converters, the converter buses j ∈ Xac we use the notation pG j = 0. The
as illustrated in Fig. 1. The AC-side bus of the interlinking powerP transfer vector is defined by pF
= [p F
]
j j∈N where each
,
converter has no inertia of its own, however power imbalances pF = p −
P
p . The matrix A is the incidence
j i:i→j ij k:j→k jk
in the AC subsystem lead to a power transfer through the matrix of the graph (Nac , Eac ). The system equations are thus:
converter. Note also that this power transfer will affect the
DC-side voltage of the converter. For each bus j ∈ Xac ∪Xdc ,
we define the power transfer pX
 G
j as the power leaving the bus ω
η̇ = AT ω = AT (2a)
ωX
3 In networks where 1e is not applicable, the DC network flows may be
G
Dω G
   
represented by currents instead of power. The proof of stability in Theorem
M ω̇
1 would then require a passivity condition from the input voltage to output  0  = pG − pL − pX +pF −  0  (2b)
current at each DC bus, along with a similar condition on the ILC equations. C V̇ 0
 4

1) Equilibrium conditions: An equilibrium of the system Using the standard method of Lagrange multipliers as in
in (2) is defined by the following conditions: [22], and defining the vector pu = [(Dω G )T 0T 0T ]T for
 G convenience, the solution pG∗ to the optimization problem is:
ω
0 = AT ω = AT (3a) Q̃11T L
ωX pG∗ = (p + pu ) (5)
Dω G 1T Q̃1
 

0 = pG − pL − pX +pF −  0  (3b)
0 III. M AIN RESULTS
A. Decentralized primary control
We assume that there exists4 some equilibrium point of (2),
We assume power-frequency droop control for the AC gen-
and denote such an equilibrium by (η ∗ , ω G∗ , V ∗ ). Individual
erators and power-voltage droop for the DC energy resources:
equilibrium values are also denoted by the superscript  
∗
asterisk, e.g. ηij , ωiG∗ , and Vj∗ . G ω
p = −Q̃ + pGnom (6)
mV
∗ π
Assumption 2. |ηij |< 2 for all (i, j) ∈ Eac . where Q̃ ≥ 0 is the inverse cost matrix of droop coefficients,
and m > 0 is a constant, and as stated previously, ω and
This assumption is often referred as a security constraint and
V are the column vectors of the AC frequency and DC
is common in the literature for power grid stability analysis.
voltage deviations, respectively. The nominal power generation
pGnom is a constant reference of the droop control scheme that
satisfies (4) for a nominal aggregate load5 and with the fre-
B. Control objectives
quencies at their nominal value.. The second control objective
The control objectives are: (limitation of frequencies and voltages deviations) may be
1) Solutions must converge to an equilibrium point. satisfied by choosing suitably large droop coefficients in Q̃.
2) For primary control, AC frequencies and DC voltages In order to simplify the presentation here we use proportional
should not deviate too far from their nominal values, droop control schemes, nevertheless this condition could be
i.e. limt→∞ |ωj (t)| < eω for all buses j ∈ Nac and relaxed to local input strict passivity of the dynamics of each
limt→∞ |Vj (t)| < eV for all buses j ∈ Ndc for some AC generator from input −ωj to output pG j and each DC
appropriate scalars eV and eω . generator from input −Vj to output pG j around their respective
3) AC frequencies and a weighted average of the DC equilibrium values ωj∗ and Vj∗ , similarly to the analysis in
voltages should converge to their nominal values for [24]. It should be noted that the majority of DC networks are
secondary control. voltage-controlled, i.e. each DC source regulates its output
4) Power sharing between all sources should be optimal. voltage (which may be droop based on current or voltage).
Such voltage-controlled DC droop sources, along with more
The last objective may be stated more formally by consid- detailed dynamic models, can also be incorporated in our
ering the minimization of a quadratic cost function [22]: analysis if a strict passivity condition from the negative output
1 G T G power −pG j (or current) to the bus voltage Vj is satisfied.
min CG = (p ) Qp (4a) This allows a wide range of DC droop control structures to
pG 2
be integrated [26].
Dω G
 
We also introduce a VSC controller based on [21]. Let the
subject to: 1T pG = 1T pL + 1T  0  (4b)
voltage angle θi at the AC-side output of an ILC x be:
0 Z
pG
j = 0, j ∈ Xac (4c) θi = mVj , i.e. the frequency is given by, (7a)

where Q is a positive definite diagonal matrix containing the ωi = θ̇i = mVj (7b)
cost coefficients for each energy resource, and 1 is the vector
of ones with appropriate dimension, and 0 is the vector of where i ∈ Xxac and j ∈ Xxdc . This relates the AC frequency
zeros with appropriate dimension. Note that constraint (4b) is deviation proportionally to the DC voltage deviation by a
a requirement for power balance at equilibrium, i.e. that the chosen constant m > 0. Hence, (7b) necessitates that the
total generation and demand are equal, while (4c) suggests that frequency at the AC side of the ILC is set directly by the ILC,
the generation at converter AC buses is zero, which holds by rather than controlling the power transfer through the ILC as
definition (note that pG in traditional ILC control schemes. The relationship between
j appears in (1c) only for convenience
in presentation). To proceed further, we define the diagonal AC frequency and DC voltage allows to provide appropriate
matrix Q̃ such that Q̃ii = Q−1 stability and optimality properties for the network, as we will
ii , i ∈ N/Xac and Q̃ii = 0, i ∈
Xac . With slight abuse of terminology, we shall refer to Q̃ as show within the paper.
the inverse cost matrix. 5 Note that p in (1) does not need to be equal to the nominal load for
L
power balance to be achieved at equilibirum, i.e. power balance will hold
4 Existence of equilibria in AC systems is beyond the scope of this paper if an equilibrium point of (1),(6),(7b) is reached with the power generated
and have been considered in e.g. [23]. determined by the deviations in frequency/voltage.
 5

We assume that the converter is lossless and that the at any arbitrary bus under the same assumptions. For droop-
internal dynamics are sufficiently fast compared to the network controlled sources to contribute power in proportion to their
nom
dynamics. In [21] the suggestion is to set m = Vω nom where droop coefficients, a system-wide synchronizing variable is
dc
ω nom and Vdc nom
are the nominal values of the AC grid required. The proposed controller (7b) achieves this by relating
frequency and the DC grid voltage. Since in this paper ωi the AC frequency to the DC voltages. By contrast, the dual
and Vj are deviations from a nominal value, other values of droop controller (8) does not provide such a relation.
m are also possible. Large values of m result in smaller DC
voltage deviations and larger AC frequency deviations, and B. Distributed control
nom
in general m = Vω nom may be too large for this scheme
dc In this section we propose a distributed controller inspired
as frequency deviations are generally less acceptable than
by [22] and [25]. The controller proposed in this section uses
voltage deviations. Instead of directly controlling the power
communication to achieve the secondary control objectives
transfer, (7b) relates the frequency and voltage within the AC
of exact power sharing, frequency and voltage regulation. It
and DC sides, respectively. Not only does this improve the
should be noted that this scheme is an alternative to that of
accuracy of the power-sharing in the primary time-frame, but
the communication-free scheme in section III-A and should
also provides fast response to AC disturbances via capacitive
therefore be used if appropriate communication is available.
inertia as discussed in [21] and [22]. In this paper, we take this
concept further and use the capacitive inertia of the entire DC The concept of network emulation can be carried further
subsystem for frequency support, and also use the inertia of with the aid of distributed communication. Let the average
the AC system to regulate the DC voltage when appropriate. DC voltage deviation of each subsystem k be represented by
the capacitance-weighted average V̄k :
Theorem 1 (Stability). Consider a dynamical system de- X
scribed by equations (1), (2) with the control scheme in (6), V̄k = Cj Vj (9)
(7b), and an equilibrium point for which Assumption 2 holds. j∈Nkdc

Then there exists an open neighbourhood of this equilibrium As in the second distributed MTDC controller proposed
point such that all solutions of (1),(2),(6),(7b) starting in this in [25], the DC voltages within each subsystem are either
region converge to the set of equilibrium points as defined in communicated within the network so as to obtain V̄k (for
(3). small subsystems) or V̄k is obtained via an appropriately fast
Theorem 1 demonstrates the local convergence of solutions distributed approach, such that its dynamics can be decoupled
to (1),(2),(6),(7b) to the set of its equilibria. Note that the result from the stability analysis in this paper. From (9) we have:
is local due to the sinusoidal power transfers in (1e) and that
V̄˙ k =
X X
Cj V̇j = (pG L X F
j − pj − pj + pj )
it becomes global if those are linearized. (10)
j∈Nkdc j∈Nkdc
The following theorem demonstrates that when line resis-
tances become arbitrarily small, then the equilibria of the The DC branch-flows pF j cancel out within the subsystem, and
considered system tend towards the global minimum of (4). we therefore have the following expression which resembles
Theorem 2 (Power sharing). As the DC line resistances the swing equation:
become arbitrarily small, the power sharing of the system (1), V̄˙ k =
X
(pG L X
j − pj − pj ) (11)
(2) with the control scheme (6), (7b) becomes arbitrarily close j∈Nkdc
to the solution of the optimization problem (4).
We now introduce the concept of virtual frequency deviation
Remark 2: In a practical network there will always be ω̂ which is defined for the entire network as follows:
some small DC line resistances which affect power sharing. (
A fundamental trade-off exists between voltage regulation and ωj if j ∈ Nac
ω̂j = (12)
power sharing accuracy for linear droop-controlled DC grids mV̄k if j ∈ Nkdc
[27]-[28] which can be adjusted by changing the magnitude of
droop gains. Nevertheless, if these line resistances are small, where k is the DC subsystem to which all nodes in the
the voltage deviations will also be small and the power sharing associated set Nkdc belong. We will denote the vector of
will be close to optimal. average DC subsystem voltages by V̄ . The converter which
Remark 3 (Power sharing in a dual-droop ILC controller interlinks AC bus i and the DC subsystem k is governed by
scheme): Consider the dual-droop scheme (8) often used in ωi = mV̄k (13)
the literature for primary control of the ILC,
where m > 0 is a positive coefficient. Similarly to the primary
pX ω V
i = Ki ωi − Kj Vj (8) controller (7), we control the AC frequency of the interlinking
converter instead of directly controlling the power transfer. A
where Kiω and KjV are the respective droop coefficients, and
common approach to achieve optimal power sharing in sec-
the power transfer is directly controlled6 . It is clear that (8)
ondary control is to introduce a synchronizing communicating
is unable to guarantee correct power-sharing for a disturbance
variable ξ, e.g. [22], and update these values via distributed
6 In practice, the ILC controls the power transfer by varying its output averaging through an undirected connected communication
voltage angle until (8) is satisfied. graph. In particular, we denote this graph by (N, Ẽ), where
 6

Ẽ denotes the set of communication links, and also denote by Theorem 4 (Convergence to optimality). Consider the dy-
L the Laplacian of (N, Ẽ), defined as namical system described in (1),(2) with the control scheme
 (12),(13),(15) and an equilibrium point for which Assump-
deg(i), if i = j,
 tion 2 is satisfied. Then, there exists an open neighbour-
Lij = −1, if (i, j) ∈ Ẽ (14) hood of this equilibrium point such that all solutions of


0, otherwise (1),(2),(12),(13),(15) starting in this region converge to a set of
equilibria that solve the optimization problem (4), with ω ∗ = 0
where deg(i) denotes the degree of node i. Then the distributed and V̄ ∗ = 0.
controllers for the hybrid AC/DC system are:
Theorem 4 demonstrates that all solutions of the considered
Tξ ξ˙ = −Lξ − Q̃ω̂ (15a) system locally converge to an optimal solution of (4).
pG = Q̃ξ − Kω̂ (15b) Remark 4: Our proposed controller is distributed in the
sense that its implementation in a DC subgrid makes use
where Tξ denotes the diagonal matrix with positive time con- of voltage measurements only within that subgrid, and is
stants, ξ is the column vector of the synchronizing variables also fully distributed in the AC subgrids of the network.
ξj , and Q̃ is the inverse cost coefficient matrix as before It should be noted that relaxing (9) to a fully distributed
and K is a diagonal matrix of positive coefficients used to controller that makes use of only local voltage measurements
improve performance and determine power contributions from without additional information transfer, while retaining the
each generator in the primary time-frame. The proportional stability and optimality properties presented, is a highly non-
term Kω̂ is effectively a primary (droop) controller while trivial problem as this would distort the synchronization of the
the slower secondary term Q̃ξ integrates the frequency and communicating variable ξ needed for optimal power sharing.
average voltage deviations to a steady-state value of zero, and
leads also to an optimal power sharing (Theorem 3 below).
One interesting feature of this controller is that the DC bus IV. C ASE STUDIES
voltages are weighted by the associated capacitances. This is In order to demonstrate our results, we study the perfor-
in order to capture the dynamics of the physical energy of the mance of our controllers in two hybrid AC/DC systems, the
subsystem as follows from (10). A potential objection could be first as shown in Fig. 2, and the second being the ring network
that buses with low capacitance could have voltages far from in Fig. 3. All studies are carried out in MATLAB / Simscape
the nominal while still satisfying V̄k = 0, due to their faster Power Systems.
response. Nevertheless, small voltage deviations at such buses
may still be maintained in two ways. Firstly, the steady-state
DC bus voltages must still satisfy the power-flow equations.
This is a constraint that does not depend on capacitances
and can potentially restrict large voltage deviations (will be
dependent on the power flows and line resistances). Secondly,
virtual capacitance CjV may easily be added at any DC source Figure 2. Example AC/DC network 1 (mixed topology).
bus j via a derivative term in the DC source dynamics, e.g.:
V ˙
pG
j = Q̃jj ξj − Kjj ω̂j − Cj Vj

The addition of the derivative term will not affect the steady-
state value of pG j , allowing its optimality properties to be
retained.
Our first result, proven in the appendix, demonstrates that
the introduction of the controller (12),(13),(15) ensures that
the equilibria of the system (2),(12),(13),(15) coincide with
the global minimum of the optimization problem (4). Figure 3. Example AC/DC network 2 (ring topology).
Theorem 3 (Power sharing). An equilibrium of the system The parameters of the networks are given in Table II. For
(1),(2) with the control scheme (12),(13),(15) solves the opti- simplicity, we have chosen the same parameters for both
mization problem (4). networks and have numbered the buses such that the param-
eters and description of loads and generation applies to both
The following theorem, proven in the appendix, demon-
networks. Varying line parameters within reasonable bounds
strates the local convergence of solutions of the dynamical
does not significantly affect the performance of the controllers,
system (1),(2), when the controller (12),(13),(15) is applied,
except the effect on power sharing within the DC subsystem(s)
to the global minimum of the optimization problem (4).
in the primary time-frame, as expected [28]. The synchronous
Furthermore, it guarantees that frequency returns to its nominal
machine is 4 MVA, 13.8kV and is connected to the AC
value at equilibrium, i.e. that ω ∗ = 0, and that the average
subsystem at bus 5 via a 4 MVA step-down transformer, and
voltage deviation in every DC subsystem is zero, i.e. that
there are four distributed DC generators across the two sub-
V̄ ∗ = 0.
systems at buses 1 and 9 (1 MW) and buses 3 and 7 (3 MW).
 7

The simulation model is more detailed and realistic than our Network 1 (mixed topology)
80
analytical model, and it includes the inverter dynamics with DC Source - Bus 1

Active power (p.u.)

DC Source - Bus 3
switching (two-level pulse width modulation), line dynamics, 70 DC Source - Bus 7
DC Source - Bus 9
detailed generator, turbine-governor, and exciter dynamics, and 60 Synchronous generator
realistic communication delays. The switching of the two-level
50
PWM converters causes the DC ripple seen in some of the
figures. More detailed parameters of the test system are given 40
0 5 10 15 20 25 30 35 40 45
in Table IV. Time (s)
Network 2 (ring topology)
Table II 80
DC Source - Bus 1
H YBRID AC/DC NETWORK PARAMETERS

Active power (p.u.)

DC Source - Bus 3
Description Parameter Value 70 DC Source - Bus 7
DC Source - Bus 9
Bus capacitances C1 , C3 , C7 ,C9 6 mF Synchronous generator
60
DC load resistances Rj , j ∈ Ndc 600 Ω
DC line resistances R12 , R19 , R23 , R78 ,R89 0.1 Ω 50
DC line inductances L12 , L19 , L23 , L78 ,L89 1 mH
DC switched loads P3 , P7 1.2 MW 40
Rated DC voltage ∗
Vdc 10 kV 0 5 10 15 20 25 30 35 40 45
Time (s)
Converter DC capacitances C3 , C7 2 mF
Converter AC filter parameters R(Ω), L(mH), C(µF ) 0.1, 1, 10
Converter ω /V ratio m 0.002 Figure 5. Power-sharing response with the decentralized primary controller
AC voltage Vac 3.3 kV (6),(7)
AC line resistances Rij 0.1 Ω
AC line inductances Lij 1 mH
AC load resistances (per phase) R4 , R6 1000 Ω
AC load active power R5 1 MW 1500 W / V) and the dual-droop scheme (50 W / V, 250 kW
Transformer reactance X5 4% / (rad/s)) are chosen to achieve a reasonable power-sharing
Generator inertia constant M 3.2 s−1 (hence the smaller source droop gains and the large dual-
ω
AC frequency f = 2π 50 Hz
droop gains). Note it is impossible to achieve comparable
steady-state power-sharing to our proposed scheme. The other
A. Decentralized primary control control parameters are then tuned to achieve the best possible
performance. It is possible to improve either the transient
Using the controllers (6), (7b) we show that the voltages performance or the power-sharing optimality, however both
and frequencies converge to equilibrium values and that the cannot be improved simultaneously. Figs. 6-7 show that the
power-sharing is close to the optimal values irrespective of the voltage / frequency deviations at equilibrium as well as the
location of the disturbance. The droop coefficients in Q̃ are power sharing are inferior to the proposed method, with
set proportionally to the source ratings with gains of (500 kW larger deviations and sub-optimal power-sharing. Increasing
/ (rad/s), 1 kW / V, 3 kW / V) for the synchronous generator, the droop gains decreases the voltage and frequency deviations
DC sources at buses 1 and 9, DC sources at buses 3 and 7 at steady-state, however unacceptable oscillations were noted
respectively. In Fig. 4 we show the AC frequency and DC in simulations. Unlike with the proposed primary controller,
voltage response to the same step disturbances at time t = 3s the topology has a strong effect on the power sharing perfor-
and t = 23s. The magnitude of the disturbance at t = 3s is mance. The dual-droop control performs considerably worse
1.2 MW (nominal added demand) located at bus 3 within DC in the first test network with three subsystems compared to
subsystem 1, while the disturbance t = 23s is 1.2 MW reduced the two subsystem ring network. In general, power sharing
demand at bus 7 within DC subsystem 2. Fig. 5 shows that performance with dual-droop control is poorer in larger and
the power allocation converges to values close to proportional more spread out networks with more subsystems, since no
as required. prescribed relation between the AC frequency and DC voltages
10.1
Network 1 (mixed topology)
50.05
is maintained among subsystems.
DC Voltage - Bus 1
DC Voltage - Bus 3
10.05
Frequency (Hz)

DC Voltage - Bus 7
Voltage (kV)

DC Voltage - Bus 9 50 Network 1 (mixed topology)

10
Synchronous generator 10.2 50
DC Voltage - Bus 1
9.95 DC Voltage - Bus 3 49.99
Frequency (Hz)

49.95 10.1 DC Voltage - Bus 7

Voltage (kV)

9.9 DC Voltage - Bus 9 49.98

10 Synchronous generator
9.85 49.9 49.97
0 5 10 15 20 25 30 35 40 45 9.9
Time (s)
49.96
Network 2 (ring topology) 9.8
10.1 50.05
DC Voltage - Bus 1 49.95
DC Voltage - Bus 3 0 5 10 15 20 25 30 35 40 45
10.05
Frequency (Hz)

DC Voltage - Bus 7 Time (s)

Voltage (kV)

DC Voltage - Bus 9 50
10 Network 2 (ring topology)
Synchronous generator 10 49.99
9.95
49.95
Frequency (Hz)

49.98
Voltage (kV)

9.9 9.95

9.85 49.9 DC Voltage - Bus 1 49.97

0 5 10 15 20 25 30 35 40 45 DC Voltage - Bus 3
9.9
Time (s) DC Voltage - Bus 7
DC Voltage - Bus 9 49.96
Synchronous generator
9.85
49.95
Figure 4. Frequency and voltage response with the decentralized primary 0 5 10 15 20 25 30 35 40 45
Time (s)
controller (6),(7)
For comparison we also study the same hybrid AC/DC Figure 6. Frequency and voltage response with traditional dual-droop control
(8)
networks with traditional dual-droop controlled ILCs. The
droop gains for the sources (250 kW / (rad/s), 500 W / V, The simulations show several advantages of our proposed
 8

Per unit voltages and frequency

Network 1 (mixed topology) Network 1 (mixed topology)
1.002
60 Average voltage - subsystem 1
Active power (p.u.)

1.001 Average voltage - subsystem 2

55 Frequency

50 DC Source - Bus 1 1
DC Source - Bus 3
45 DC Source - Bus 7
0.999
DC Source - Bus 9
Synchronous generator
40
0.998
0 5 10 15 20 25 30 35 40 45 0 5 10 15 20 25 30 35 40 45
Time (s) Time (s)

Per unit voltages and frequency

Network 2 (ring topology) Network 2 (ring topology)
1.002
65
Average DC voltage
Active power (p.u.)

1.001 Frequency
60
DC Source - Bus 1 1
DC Source - Bus 3
55 DC Source - Bus 7
0.999
DC Source - Bus 9
Synchronous generator
50 0.998
0 5 10 15 20 25 30 35 40 45 0 5 10 15 20 25 30 35 40 45
Time (s) Time (s)

Figure 7. Power-sharing response with traditional dual-droop control (8) Figure 8. Frequency and voltage response with the distributed secondary
controller (13),(15)

primary controller. Firstly, the power requirement of any load Network 1 (mixed topology)
change or disturbance can be effectively shared by sources 0.65
DC sources - Subsystem 1

Active power (p.u.)

across the entire hybrid AC/DC network. In contrast, the dual- 0.6 DC sources - Subsystem 2
Synchronous generator
droop scheme requires nearby sources to support most of the
0.55
power requirement, which is not only less optimal but may also
cause device ratings to be exceeded. Secondly, our proposed 0.5

scheme is able to use the inertia of the synchronous machine 0.45

0 5 10 15 20 25 30 35 40 45
and the capacitance of both DC subgrids to quickly contribute Time (s)
to the power balancing. As seen in Fig. 5, the inertia of the Network 2 (ring topology)
0.65
synchronous machine is used to supply the load demand in DC sources
Active power (p.u.)

0.6 Synchronous generator

the first few seconds. Since synchronous machines are able
to handle short-term overloads much more successfully than 0.55
power-electronic converters, this is preferable to the local DC-
0.5
side regulation of the dual-droop scheme.
0.45
0 5 10 15 20 25 30 35 40 45
Time (s)
B. Distributed control
We also consider the performance of the controller (13),(15) Figure 9. Power-sharing response with the distributed secondary controller
on the test networks under identical conditions, using the same (13),(15)
droop coefficients Q̃ = K for simplicity. In the simulation, we
obtain V̄k in (12) via propagation through the network with a proposed that guarantees stability while achieving exact opti-
communication delay of 10 ms between neighbouring buses. mal power sharing and that bus frequencies and the weighted
Simulations where V̄k is evaluated via consensus schemes average of DC voltages return to their nominal values at
were also carried out and a similar performance was achieved steady state. Finally, the proposed algorithms were verified
for small communication delays < 10 ms, however the per- by simulation and compared to traditional dual-droop control.
formance deteriorated significantly for larger communication
delays. Figs. 8-9 show that the distributed controller regulates A PPENDIX
the voltages and frequencies of the hybrid AC/DC network to The appendix includes the proofs of the results presented
their nominal values while guaranteeing optimal power-sharing in the main text. We provide first some notation that will be
at steady-state. used within the proofs. Given some column vector z with
length |N |, we use the subscripted vector zac to denote the
V. C ONCLUSION vector that includes the elements of z with indices in Nac , i.e.
We have proposed a new method for the control of inter- zac = [zj,j∈Nac ]. Likewise the subscripts zGac , zXac , zXdc , zdc
linking converter(s), used in conjunction with traditional droop denote the vectors that include those entries for which j ∈
control to guarantee stability and accurate power sharing in a Nac \ Xac , j ∈ Xac , j ∈ Xdc and j ∈ Ndc respectively.
general hybrid AC/DC network. The stability of the controlled The following relations therefore hold: z T = [zac T T
, zdc ],
T T T
system was proven and it was shown that power-sharing across zac = [zGac , zXac ]. For convenience we define Γ > 0 as the
the AC/DC network is significantly improved compared to diagonal matrix of all Bij , (i, j) ∈ Eac , and G is defined as the
dual-droop control. A secondary control scheme has also been conductance matrix of the DC subsystem(s); i.e. the Laplacian
 9

weighted by the conductances Gij . Finally, summation

P over Using Assumption 2, WE has a strict local minimum at
all the DC subsystems is denoted by the shorthand k,dc . η = η ∗ . Likewise WG and WV have strict global minima
Proof of Theorem 1: We prove our claim in Theorem 1 by at ω G∗ and V ∗ respectively. Thus W has a strict minimum
finding a suitable Lyapunov function for the system (2), (6), at Z ∗ = (η ∗ , ω G∗ , V ∗ ). Since ω X is uniquely determined
(7b). Consider the following Lyapunov candidate: by V , we can then choose a neighbourhood of Z ∗ in the
coordinates (η, ω G , V ). (19) further shows that W is a non-
W (η, ω G , V ) =WG + WE + WV increasing function of time. Hence the connected set T =
1 {(η, ω G , V ) : W ≤ } for some sufficiently small  > 0 is
= (ω G − ω G∗ )T M (ω G − ω G∗ )
Z η 2 compact, forward-invariant and contains Z ∗ . We then apply
1 LaSalle’s Theorem, with W as the Lyapunov-like function,
T
+1 Γ sin(φ) − sin(η )dφ + m(V − V ∗ )T C(V − V ∗ ).
∗
η∗ 2 which states that all trajectories of the system starting from
(16) within T converge to the largest invariant set within T that
satisfies Ẇ = 0. Since both Q̃G and D are positive definite
The time derivative of the first term is given by
matrices, clearly Ẇ = 0 implies (ω G , V ) = (ω G∗ , V ∗ ) and
W˙G = (ω G − ω G∗ )T (pG L X F G
Gac − pGac − pGac + pGac − Dω )
therefore ω̇ G = V̇ = 0. This in turn implies by (3) that the
converter AC-side frequencies ω X = ω X∗ . Furthermore, from
= (ω G − ω G∗ )T (pG L X F G
Gac − pGac − pGac + pGac − Dω )
the equilibrium conditions we deduce that the frequency in
+ (ω X − ω X∗ )T (pG L X F
Xac − pXac − pXac + pXac ) each AC subsystem synchronizes to a common value, hence
noting that the second expression follows by adding the term the angle differences η converge also to some constant value.
for the converter buses which is equal to zero by (1c). Using Therefore, by LaSalle’s Theorem we have convergence to the
the equilibrium conditions (3), noting that pX set of equilibrium points as defined by (3). Finally, choosing
Gac is a zero
vector, and rearranging results to: S such that it is open, includes Z ∗ , and S ⊂ T completes the
proof.
W˙G = (ω−ω ∗ )T (pG G∗ X
ac −pac )−(ω − ω
X∗ T X
) (pXac − pX∗
Xac )
Proof of Theorem 2: From the equilibrium conditions (3),
[ω T mV T ]T is arbitrarily close to the vector of equilibrium
− (ω − ω ∗ )T (pF F∗ G
ac − pac ) − (ω − ω
G∗ T
) D(ω G − ω G∗ ).
frequencies 1ω ∗ as the line resistances become arbitrarily
(17)
small. This follows from the equilibrium conditions (3) and
The time derivative of the second term in (16) is, again (1e), where if the conductances Gij are arbitrarily large, the
using the equilibrium conditions (3): voltage differences Vi − Vj are arbitrarily small. Therefore,
to find the power allocation when the line resistances are
W˙E = (Γ(sin(η) − sin(η ∗ )))T AT (ω − ω ∗ ) arbitrarily small, we solve the equilibrium conditions:
= (ω − ω ∗ )T (pF F∗
ac − pac ),
−Q̃1ω ∗ + pG L X F u
nom − p − p + p − p = 0
thus canceling the power transfer term in (17). The time
derivative of WV is given by: −1T Q̃1ω ∗ + 1T pG T L u T X T F
nom − 1 (p + p ) − 1 p + 1 p = 0

W˙V = m(V −V ∗ )T (pG G∗ ∗ T X X∗

dc −pdc )−m(V − V ) (pdc − pdc ) Clearly 1T pF = 0 in a lossless network and 1T pX = 0
− m(V − V ∗ )T G(V − V ∗ ). (18) for lossless converters. Note also that the nominal power
generation may be expressed as pG nom = −Q̃1ζ, where ζ is a
Since G is the conductance matrix of the DC graph by constant. Hence, solving for 1ω ∗ and substituting into (6):
definition we have −m(V −V ∗ )T G(V −V ∗ ) ≤ 0 since m > 0.
Furthermore, using (7b) and noting that pX X
X dc = −pXxac : 11T
x 1ω ∗ = − (pL + pu ) − 1ζ
1T Q̃1
m(V − V ∗ )T (pX X∗ X
dc − pdc ) = (ω − ω
X∗ T X
) (pXdc − pX∗
Xdc )
Q̃11T L
= −(ω X − ω X∗ )T (pX X∗
Xac − pXac )
pG = −Q̃1ω ∗ − Q̃1ζ = (p + pu )
1T Q̃1
Hence the ILC terms in (17), (18) can be canceled out.
We also note that converter buses Xac have no frequency- yields the solution (5) to the optimization problem (4).
dependent generation nor any damping, and that the respective Proof of Theorem 3: This is analogous to that of Theorem
entries of the diagonal matrix Q̃ are zeros, while all other 2. By premultiplying (15a) by 1T and noting (13) and the
entries of Q̃ are positive. Therefore, we define Q̃G > 0 as the synchronization of frequencies at steady state, it follows that
diagonal matrix with dimension |N |−|X| which includes only at equilibrium ω̂ = 0. The latter shows from (15a) at steady
the non-zero terms in Q̃. Putting it all together and substituting state that ξ ∗ = 1ξ¯ where ξ¯ is the identical equilibrium value of
(6): the individual values ξj at node j. Then from the equilibrium
conditions (3) we have:
 G T
(ω − ω G∗ )
 G
(ω − ω G∗ )

Ẇ ≤ − Q̃G
m(V − V ∗ ) m(V − V ∗ ) Q̃ξ ∗ − pL − pX + pF = 0 (20a)
G G∗ T G G∗ T ∗ T L T X T F
− (ω − ω ) D(ω − ω ) ≤ 0. (19) 1 Q̃ξ − 1 p − 1 p + 1 p = 0 (20b)
 10

Clearly 1T pF = 0 in a lossless network and 1T pX = 0 for using (10) and (11) to simplify ẆV . Clearly −(ξ − ξ ∗ )T L(ξ −
lossless converters. Hence: ξ ∗ ) ≤ 0 from the definition of the Laplacian matrix L. We also
simplify further by cancelling like terms and thus obtain:
1T Q̃1ξ¯ − 1T pL = 0 (21a)
T L Ẇ ≤ −(ω G )T (D + Kac )ω G − (ω X )T (pX X∗
Xac − pXac )
11 p
ξ = 1ξ¯ =
∗
(21b)
X X
1T Q̃1 −m (V̄k Kjj V̄k )−mV̄ T (pX X∗
Xdc − pXdc )
k,dc j∈Nkdc

Finally, substituting pG = Q̃ξ ∗ from (15b) into (21b) yields

P P
Noting that −m k,dc (V̄k j∈N dc Kjj V̄k ) ≤ 0 since Kjj ≥ 0
k
the solution to the optimization problem (4). for all j ∈ N and using (13) and pX X
Xxdc = −pXxac to cancel the
Proof of Theorem 4: We use L = (1, 2, ...|L|) to represent second and fourth terms, we finally have:
the set of nodes of the communication graph. For convenience
we also write the decomposition of Q̃ into its corresponding Ẇ ≤ −(ω G )T (D + Kac )ω G ≤ 0 (23)
AC and DC terms as Q̃ = diag(Q̃ac , Q̃dc ) where Q̃ac and Q̃dc where the damping matrix D is positive definite and can be
are diagonal matrices containing the inverse cost coefficients increased by proportional control of the AC sources via Kac .
for the AC and DC generators respectively. Likewise, we We then apply LaSalle’s Theorem. Clearly, W is minimized
decompose K into corresponding AC and DC matrices as at η = η ∗ , ω G = 0, V̄ = 0 and ξ = ξ ∗ . Therefore we
K = diag(Kac , Kdc ), where Kac = {Kjj : j ∈ Nac } consider the set T which includes (η ∗ , 0, 0, ξ ∗ ) and is defined
and Kdc = {Kjj : j ∈ Ndc } are diagonal matrices of the by {(η, ω G , V̄ , ξ) : W ≤ } for some positive constant
corresponding gains in K, appropriately ordered without loss . Since W is non-increasing with time, T is a compact,
of generality. We consider the following candidate Lyapunov positively invariant set for  sufficiently small. LaSalle’s The-
function: orem states that trajectories beginning in T converge to the
largest invariant set within T for which Ẇ = 0. We therefore
W = WG + W E + W V + Wξ (22a) examine the equality condition of (23). Ẇ = 0 implies that
Z η
1 ω G = 0, which from the system dynamics (2) implies that
= (ω G )T M ω G + 1T Γ (sin(φ) − sin(η ∗ ))dφ
2 η∗ ω X = mV̄ = 0, V̄ = 0, and η is constant. Finally, from
1 1 (15) if ω̂ = 0 then Lξ = 0 which from the definition of
+ mV̄ T V̄ + (ξ − ξ ∗ )T Tξ (ξ − ξ ∗ ) (22b) the Laplacian communication graph implies that all values
2 2
ξj , j ∈ L converge to some network-wide constant value ξ¯
The time derivatives of WG are, again adding the term from and thus ξ ∗ = 1ξ. ¯ Hence the largest invariant set Ξ within
X ¯ for
(1c) and noting that pj = 0 for all buses j ∈ Nac \ Xac : T for which Ẇ = 0 satisfies (η, ω G , V̄ , ξ) = (η̄, 0, 0, ξ)
¯ X
constant η̄ and ξ. Furthermore, p trivially converges from
ẆG = (ω G )T (pG Gac − p L
Gac + pF
Gac − Dω G
) (1c). To show that within Ξ, V takes some constant value V̂
G T G L F
= (ω ) (pGac − pGac + pGac − Dω ) G consider (1d) and (1e) and note that variables pG , pL and pX
are constant. Then defining V j = Vj − V̂j , it follows that the
+ (ω X )T (pG L X
Xac − pXac − pXac + pXac )
F
dynamics of V within Ξ satisfy C V̇ = −LDC V where LDC
∗
= (ω)T Q̃ac (ξac − ξac ) − (ω X )T (pX X∗
Xac − pXac ) is the Laplacian of the graph (Ndc , Edc ), defined in analogy
T F F∗ G T G to (14). It is easy to see that the only invariant set of this
+ ω (pac − pac ) − (ω ) (D + Kac )ω
linear ODE is V ∈ Im(1), where Im(1) denotes the image of
The time derivatives of the other functions comprising W are: 1, which together with V̄ = 0 results to V = 0. Therefore
by LaSalle’s theorem the trajectories of the system starting
ẆE = − (Γ(sin(η) − sin(η ))) A ω = − ω (pac − pac ) within T converge to the set of equilibrium points. This, in
∗ T T T F F∗
conjunction with Theorem 3 which suggests that equilibria
of (1),(2),(12),(13),(15) are solutions to (4) completes the
˙ T ˙ proof.
WV = mV̄ V̄
X X
= m V̄k [(pG G∗ X X∗
j − pj ) − (pj − pj )] R EFERENCES
k,dc j∈Nkdc [1] J. Watson and I. Lestas, “Frequency and voltage control of hybrid AC/DC networks,”
X X in 57th IEEE Conference on Decision and Control, Dec. 2018.
= m V̄k (Q̃jj (ξj − ξj∗ ) − Kjj V̄k ) [2] D. Jovcic et al, “Feasibility of DC Transmission Networks,” 2011 IEEE Conference
on Innovative Smart Grid Technologies Europe (ISGT Europe), 5-7 Dec. 2011.
k,dc j∈Nkdc [3] P. Wang et al, “Hybrid AC/DC Micro-Grids: Solution for High Efficiency Future
Power Systems,” in N.R. Karki et al. (eds.), Sustainable Power Systems, Reliable
− mV̄ T (pX X∗
Xdc − pXdc ) and Sustainable Electric Power and Energy Systems Management, Springer, 2017.
[4] S.M. Malik et al, “Voltage and frequency control strategies of hybrid AC/DC
Ẇξ = (ξ − ξ ∗ )T (−L(ξ − ξ ∗ ) − Q̃(ω̂)) microgrid: a review,” IET Generation, Transmission and Distribution, vol. 11, no.
2, 2017, pp. 303-313.
= − (ξ − ξ ∗ )T L(ξ − ξ ∗ ) − (ξ − ξ ∗ )T Q̃ω̂ [5] E. Unamuno, J.A. Barrena, “Hybrid ac/dc microgrids - Part II: Review and
= − (ξ − ξ ∗ )T L(ξ − ξ ∗ ) − ω T Q̃ac (ξac − ξac
∗
) classification of control strategies,” Renewable and Sustainable Energy Reviews,
vol. 52, December 2015, pp. 1123-1134.
[6] M. Cucuzzella, et al, “A Robust Consensus Algorithm for Current Sharing and
X X
∗
−m V̄k Q̃jj (ξj − ξj ) Voltage Regulation in DC Microgrids,” in IEEE Transactions on Control Systems
k,dc j∈Nkdc Technology, vol. 27, No. 4, July 2019, pp. 1583-1595.
 11

[7] T. Morstyn, B. Hredzak, G. D. Demetriades and V. G. Agelidis, “Unified Distributed

Control for DC Microgrid Operating Modes,” in IEEE Transactions on Power
Systems, vol. 31, no. 1, Jan. 2016, pp. 802-812,.
[8] V. Nasirian, S. Moayedi, A. Davoudi and F. L. Lewis, “Distributed Cooperative
Control of DC Microgrids,” in IEEE Transactions on Power Electronics, vol. 30,
no. 4, April 2015, pp. 2288-2303.
[9] L. Papangelis et al, “Coordinated Supervisory Control of Multi-Terminal HVDC
Grids: a Model Predictive Control Approach,” IEEE Transactions on Power Deliv-
ery, vol. 32, no. 6, Nov. 2017, pp. 4673-4683.
[10] J. Rocabert, et al, “Control of Power Converters in AC Microgrids,” in IEEE
Transactions on Power Electronics, vol. 27, no. 11, Nov. 2012, pp. 4734-4749.
[11] N. Chaudhuri, R. Majumder, and B. Chaudhuri, “System frequency support through
multi-terminal DC (MTDC) grids,” IEEE Transactions on Power Systems, vol. 28,
Feb 2013, pp. 347–356.
[12] S. Akkari, J. Dai, M. Petit and X. Guillaud, “Interaction between the voltage-
droop and the frequency-droop control for multi-terminal HVDC systems,” in IET
Generation, Transmission & Distribution, vol. 10, no. 6, 2016, pp. 1345-1352.
[13] S. M. Malik et al, “A Generalized Droop Strategy for Interlinking Converter in a
Standalone Hybrid Microgrid,” Applied Energy, vol. 226, 2018, pp. 1056-1063.
[14] H. Yoo, T. Nguyen and H. Kim, “Consensus-based Distributed Coordination
Control of Hybrid AC/DC Microgrids,” in IEEE Transactions on Sustainable
Energy, in press, 2019.
[15] L. Papangelis et al, “A receding horizon approach to incorporate frequency support
into the AC/DC converters of a multi-terminal DC grid,” Electric Power Systems
Research, vol. 148, 2017, pp. 1-9.
[16] P. C. Loh et al, “Autonomous Control of Interlinking Converter With Energy Stor-
age in Hybrid AC–DC Microgrid,” in IEEE Transactions on Industry Applications,
vol. 49, no. 3, May-June 2013, pp. 1374-1382.
[17] F. Luo et al, “A Hybrid AC/DC Microgrid Control Scheme With Voltage-Source
Inverter-Controlled Interlinking Converters”, Power Electronics and Applications
(EPE’16 ECCE Europe), 18th European Conference on, 2016.
[18] J. Dai and G. Damm, “An improved control law using HVDC systems for frequency
control,” in Power Syst. Comput. Conf., 2011.
[19] M. Andreasson et al, “Distributed Frequency Control Through MTDC Transmission
Systems,” IEEE Transactions on Power Systems, vol. 32, no. 1, Jan. 2017, pp. 250-
260.
[20] A. Ordoño, et al, “Interlinking converters and their contribution to primary
regulation: a review,” International Journal of Electrical Power and Energy Systems,
vol. 111, 2019, pp. 44-57.
[21] T. Jouini et al, “Grid-Friendly Matching of Synchronous Machines by Tapping into
the DC Storage,” IFAC-PapersOnLine, vol. 49, issue 22, 2016, pp. 192-197.
[22] P. Monshizadeh, et al, “Stability and Frequency Regulation of Inverters with
Capacitive Inertia,” in IEEE 56th Annual Conference on Decision and Control
(CDC), 2017.
[23] F. Dörfler, M. Chertkov, and F. Bullo, “Synchronization in complex oscillator
networks and smart grids”, in Proceedings of the National Academy of Sciences
of the United States of America, 2013.
[24] A. Kasis, E. Devane, C. Spanias, and I. Lestas, “Primary frequency regulation
with load-side participation - Part I: stability and optimality,” IEEE Trans. Power
Systems, vol. 32, no. 5, Sept. 2017, pp. 3505-3518.
[25] M. Andreasson et al, “Distributed Controllers for Multi-Terminal HVDC Trans-
mission Systems,” IEEE Transactions on Control of Network Systems, vol. 4, no.
3, Sept. 2017, pp. 564-574.
[26] F. Thams, R. Eriksson and M. Molinas, “Interaction of Droop Control Structures
and Its Inherent Effect on the Power Transfer Limits in Multiterminal VSC-HVDC,”
in IEEE Transactions on Power Delivery, vol. 32, no. 1, pp. 182-192, Feb. 2017.
[27] T.M. Haileselassie and K. Uhlen, “Impact of DC Line Voltage Drops on Power
Flow of MTDC Using Droop Control,” IEEE Transactions on Power Systems, vol.
27, no. 2, August 2012, pp. 1441-1449.
[28] F. Gao et al, “Primary and secondary control in DC microgrids: a review”, Journal
of Modern Power Systems and Clean Energy, vol. 7, 2019, pp. 227–242.