1

Control Design of Dynamic Virtual Power Plants:

An Adaptive Divide-and-Conquer Approach
Verena Häberle, Michael W. Fisher, Eduardo Prieto-Araujo and Florian Dörfler

Abstract—In this paper, we present a novel control approach tracking set points. The key to success is heterogeneity: Only
for dynamic virtual power plants (DVPPs). In particular, we a sufficiently heterogeneous group of devices (complementing
consider a group of heterogeneous distributed energy resources each other in terms of energy/power availability, response
(DERs) which collectively provide desired dynamic ancillary
times, and weather dependency) can reliably provide dynamic
arXiv:2108.00925v5 [eess.SY] 11 Sep 2023

services such as fast frequency and voltage control. Our control

approach relies on an adaptive divide-and-conquer strategy: ancillary services across all power and energy levels and time
first, we disaggregate the desired frequency and voltage control scales, while none of the individual devices is able to do so.
specifications of the aggregate DVPP via adaptive dynamic par- Motivating examples of collections of heterogeneous en-
ticipation matrices (ADPMs) to obtain the desired local behavior ergy sources for dynamic ancillary services provision include
for each device. Second, we design local linear parameter-varying
(LPV) H∞ controllers to optimally match this local behaviors. hydro-power with initially inverse response dynamics com-
In the process, the control design also incorporates the physical pensated by batteries on short time scales [8], synchronous
and engineered limits of each DVPP device. Furthermore, our condensers (with rotational energy) paired with converter-
adaptive control design can properly respond to fluctuating device based generation [9], or hybrid storage pairing batteries with
capacities, and thus include weather-driven DERs into the DVPP supercapacitor providing regulation on different frequency
setup. Finally, we demonstrate the effectiveness of our control
strategy in a case study based on the IEEE nine-bus system. ranges [10]. However, the coordination of all these collections
is highly customized, and not (even conceptually) extendable
Index Terms—Dynamic virtual power plant, fast ancillary to other device aggregations. Further, none of these collections
services, matching control.
are controlled to match a desired aggregate dynamic behavior,
therefore lacking optimal performance and reliability during
I. I NTRODUCTION ancillary services provision. In contrast, other works in [11],
[12] propose more versatile DVPP approaches to achieve a
F UTURE power systems will contain an increasing pen-
etration of non-synchronous distributed energy resources
(DERs). In this regard, reliable ancillary services provision,
desired short-term frequency response on an aggregate level.
In particular, [12] relies on static participation factors and a
as currently ensured by conventional generators, has to be coordinated control signal which is communicated to each de-
shouldered by DERs. This imposes great challenges to cope vice, but therefore subject to communication delays and single
with the fluctuating nature of renewable energy sources [1], point of failure risk. As opposed to this, [11] presents a fully
as well as their device-specific limitations. decentralized control strategy based on dynamic participation
As early as 1997, the concept of virtual power plants (VPPs) factors, which can be used to take local device dynamics into
has been proposed to pave the way for future ancillary services account. However, both [12] and [11] are restricted to provide
by DERs [2]. VPPs are collections of distributed generators frequency control, do not consider device-level constraints, and
(all with individual device limitations), aggregated to have the are non-adaptive, therefore prone to failure during temporal
same visibility, controllability and market functionality as a variability of weather-driven DERs.
unique power plant [3]–[5]. Today, most commercial imple- In this work, we present a novel multivariable control
mentations as well as the scientific landscape are restricted to approach for DVPPs, capable of providing multiple desired
VPPs providing static ancillary services in the form of tracking dynamic ancillary services at once. We particularly focus
power and voltage set points, see, e.g., [6]. on fast frequency and voltage control objectives, specifying
In this work, we are interested in the vastly underexplored them as a desired dynamic multi-input multi-output (MIMO)
concept of a dynamic virtual power plant (DVPP) consisting of behavior of the aggregate DVPP, given in terms of a desired
heterogeneous DERs, which all-together can provide desired transfer matrix from frequency and voltage to active and
dynamic ancillary services beyond mere set point tracking [7]. reactive power. In addition to the desired aggregate output,
In particular, we are interested in dynamic ancillary services on our DVPP control strategy also incorporates the DVPP internal
faster time scales, such as fast frequency and voltage control, constraints of the devices (e.g. speed limitations, capacities,
which cannot be provided by existing VPP setups restricted to current constraints, etc.), to ensure they are not exceeded
during normal operating conditions. We pursue a local control
This paper is based upon work supported by the King Abdullah University strategy and design individual feedback controllers for each
of Science and Technology (KAUST) Office of Sponsored Research (award DVPP device, subject to its own limitations, but so that the
No. OSR-2019-CoE-NEOM-4178.11) and by the European Union’s Horizon aggregate behavior meets the desired MIMO specification.
2020 research and innovation program (grant agreement No. 883985).
V. Häberle, M. W. Fisher and F. Dörfler are with the Automatic Control More specifically, our control approach relies on an adaptive
Laboratory, ETH Zurich, 8092 Zurich, Switzerland. divide-and-conquer strategy composed of two steps: first, we
E. Prieto-Araujo is a Serra Húnter Lecturer with the Centre d’Innovació Tec- disaggregate the MIMO specification among the devices using
nològica en Convertidors Estàtics i Accionamients, Department d’Enginyeria
Elèctrica, Universitat Politècnica de Catalunya, 08028 Barcelona, Spain. adaptive dynamic participation matrices (ADPMs) which take
Email:{verenhae,mfisher,dorfler}@ethz.ch; eduardo.prieto-araujo@upc.edu the form of MIMO transfer matrices, and basically represent
 2

TABLE I: List of notation for the DVPP control setup.

a multidimensional and adaptable version of the dynamic
participation factors in [11]. Second, we employ local lin- Description Symbol
ear parameter-varying (LPV) H∞ methods [13] to optimally Set of non-controllable DVPP devices N
Set of controllable DVPP devices C
match the obtained local desired behavior of each device, DVPP device index i
while satisfying transient device-level constraints. Further, we Measured bus frequency deviation ∆f
propose centralized and distributed update strategies to adapt Measured bus voltage magnitude deviation ∆v
the ADPMs online towards capacity fluctuations, so that our Active power deviation output of device i ∆pi
Reactive power deviation output of device i ∆qi
control design can properly respond to temporal variability of Aggregate active power deviation output of the DVPP ∆pagg
DERs. This allows for a DVPP setup including weather-driven Aggregate reactive power deviation output of the DVPP ∆qagg
DERs, which are typically treated as non-dispatchable, and not Desired active power deviation output of the DVPP ∆pdes
employed for fast ancillary services provision [14]. Desired reactive power deviation output of the DVPP ∆qdes
Local closed-loop transfer matrix of device i Ti (s)
The remainder of the paper is structured as follows: in Sec- Desired MIMO transfer matrix of the DVPP Tdes (s)
tion II, we introduce the novel DVPP control setup for Desired DVPP transfer function for the f-p channel fp
Tdes (s)
vq
fast frequency and voltage control. We provide a simplified Desired DVPP transfer function for the v-q channel Tdes (s)
setup using the formalism of linear systems, which makes it
convenient to develop our control design. Section III presents (deviating from the respective power set point), sum up to the
the divide-and-conquer strategy involving the disaggregation aggregate active and reactive power deviation output of the
via ADPMs and the local H∞ control. In Section IV, we DVPP, namely ∆pagg and ∆qagg , respectively, i.e.,
demonstrate the performance of our control design on a test    
case of the IEEE nine-bus system using detailed and nonlinear ∆pagg P ∆pi
= i∈N ∪C . (1)
system and device models. Section V concludes the paper. ∆qagg ∆qi
We assume that all non-controllable devices i ∈ N have
II. DVPP C ONTROL S ETUP a pre-existing frequency and voltage control (e.g., installed
We consider a DVPP control setup for a group of heteroge- turbine and governor controls or automatic voltage regulators
neous DERs (Fig. 1 and Table I), including both a collection N (AVRs)). Hence, their local closed-loop transfer 2×2 matrices
of non-controllable devices (e.g., installed synchronous gener- Ti (s) (from frequency and voltage magnitude to active and
ators and condensers), as well as a collection C of controllable reactive power) are considered as fixed within the DVPP setup.
devices (e.g., converter-based generators). We assume that all The local closed-loop transfer 2 × 2 matrices Ti (s) of the
devices of a DVPP are connected at the same bus of the controllable devices i ∈ C, in turn, can be shaped freely by
transmission grid, where they receive an input signal in terms employing appropriate feedback controls (cf. Section III-C).
of the measured bus frequency ∆f and voltage magnitude Considering the local closed-loop transfer matrices Ti of
deviation ∆v (Fig. 1). The active and reactive power deviation both the non-controllable and controllable devices i ∈ N ∪ C,
output of each device i, namely ∆pi and ∆qi , respectively the aggregate DVPP behavior is given by
   
∆pagg (s) P ∆f (s)
DVPP: collection of heterogeneous DERs = i∈N ∪C Ti (s) . (2)
∆qagg (s) ∆v(s)
set of non-controllable devices N
To compensate for ancillary services conventionally pro-
T1 (s)
vided by synchronous generators in transmission networks, a
 
∆p1
∆q1
non-controllable device 1 decoupled f-p and v-q behavior is specified for the aggregate
+ +
DVPP as a desired diagonal1 MIMO transfer matrix as
Tr−1 (s)     fp
∆pr-1
  
∆qr-1 ∆pdes (s) Tdes (s) 0 ∆f (s)
= . (3)
   
∆f non-controllable device r-1 ∆pagg
∆v ∆qagg ∆qdes (s) 0 T vq (s) ∆v(s)
+
| {z des }
set of controllable devices C +
=:Tdes (s)
Tr (s)
Since the transfer matrices Ti of the non-controllable de-
plant r  
∆pr + +
∆qr vices N are fixed, the DVPP control design problem is to find
control
local controllers for the controllable devices C, such that the
following aggregation condition holds2 :
controllable device r
P !
Tn (s) i∈N ∪C Ti (s) = Tdes (s). (4)

plant n 
∆pn
 Furthermore, during the control design it is important to
∆qn ensure that physical and engineered device limitations, includ-
control ing response time constraints as well as (potentially time-
varying) limits on power availability and current capacity, are
controllable device n
not exceeded during normal operating conditions.
≈
   
∆f ∆pdes
∆v ∆qdes
Tdes (s) 1 Our formalism directly extends to full MIMO specifications, potentially

desired aggregate behavior relevant for future ancillary services or in other types of networks [15].
!
2 The sign “ = ” indicates that the terms on the left hand side of the
Fig. 1: Schematic of the DVPP control setup. equality must be designed in such a way that the equality is satisfied.
 3

Of course, to meet the aggregation condition (4), the power Algorithm 1 Sort Algorithm for Channel k ∈ {fp, vq}
park comprising the DVPP has to be sufficiently diverse 1: // Fix ADPFs of non-controllable devices 1, ..., r−1 via (9)
covering all time scales and energy/power levels. Further, it 2: // Define steady-state ADPFs as LPFs
is assumed that the desired behavior Tdes is provided by the 3:
θik
mki (s) ← τ k s+1 k
, ∀i ∈ Clpf
power system operator, therefore reasonably specified so that i

it is collectively achievable by the devices, while rendering the 4: // Fix intermediate ADPFs as BPFs
k
closed-loop power system stable, and being robust to model 5: for i = r + |Clpf |:n−  1 do 
k 1 1
Pi−1
uncertainties and parameter variations in the rest of the grid. 6: mi (s) ← (τ k s+1)di τ k s+1 − l=1 mkl (s) , di ∈ N0
i i
7: end for
III. A DAPTIVE D IVIDE - AND -C ONQUER S TRATEGY // Fix fastest
 device’s
8:
Pn−1ADPF  as HPF
Our approach to solve the previous DVPP control design 9: mkn (s) ← 1 − i=1 mki (s)
problem is based on an adaptive divide-and-conquer strategy,
composed of two steps:
so that the matching condition (5) holds trivially. Given the
1) Disaggregate the desired DVPP behavior by dividing the fixed ADPFs in (9), the ADPFs of the controllable devices
MIMO transfer matrix Tdes among the DVPP devices are selected such that the participation condition in (8) is
using adaptive dynamic participation matrices (ADPMs) satisfied, while simultaneously respecting the heterogeneous
to obtain local desired behaviors. The latter are defined time scales of local device dynamics along with steady-state
by the product of each ADPM and Tdes , respectively. power capacity limits. Hence, for each ADPF, we envision
2) Design a local feedback control for each device to opti- (see case studies in Sections IV-C and IV-D for examples)
mally match the local desired behavior. We will resort to
• a low-pass filter (LPF) participation factor for devices that
a linear parameter-varying (LPV) H∞ method.
can provide regulation on longer time scales on channel
k including steady-state contributions,
A. Disaggregation via ADPMs • a high-pass filter (HPF) participation factor for devices
We disaggregate the desired MIMO transfer matrix to the able to provide regulation on very short time scales on
individual devices by imposing the local matching condition channel k, and
! • a band-pass filter (BPF) participation factor for devices
Ti (s) = Mi (s) · Tdes (s), ∀i ∈ N ∪ C, (5) able to cover the intermediate regime.
where the 2 × 2 transfer matrices Mi are adaptive dynamic To accomplish this, we specify the ADPFs by two parameters:
participation matrices (ADPMs) of the form a channel-specific time constant τik for the roll-off frequency
 fp  to account for different time scales of local device dynamics
mi (s) 0
Mi (s) = , ∀i ∈ N ∪ C, (6) on channel k ∈ {fp, vq}, and a DC gain mki (s = 0) := θik
0 mvq
i (s) to account for device power capacity limits. In particular,
with the diagonal elements mfp vq
i , mi being adaptive dynamic
the ADPFs with a BPF or HPF behavior will always have
participation factors (ADPFs) for the f-p and v-q channel, a constant zero DC gain by definition, i.e., mki (s = 0) = 0. In
k
respectively (see below). Using the matching condition (5), contrast, for all devices Clpf participating as a LPF on channel
k k
the aggregation condition (4) can be disaggregated as k, the LPF DC gains θi , i ∈ Clpf have to satisfy
P ! P k
i∈N ∪C Mi (s) · Tdes (s) = Tdes (s), (7)
P
i∈N ∪C Ti (s) = i∈N ∪C mi (s = 0) = 1, k ∈ {fp, vq} (10)
P !
where i∈N ∪C Mi (s) = I2 , and I2 is the identity matrix.
to meet the participation condition in (8).
This results in the participation condition Finally, for each channel k ∈ {fp, vq} separately, we
P fp ! P vq ! sort the devices in descending order w.r.t. their channel-
i∈N ∪C mi (s) = 1, i∈N ∪C mi (s) = 1. (8) specific time constant, and apply Algorithm 1 to compute the
Considering the solvability of the local matching condition respective ADPFs as LPFs, BPFs and HPF according to the
in (5), we require that each reference model Mi · Tdes is devices’ response time and capacity limitations.
selected carefully in such a way that it can be matched by
the associated device dynamics during normal operating con- B. Online Adaptation of LPF DC gains
ditions. On the one hand, as mentioned before, we therefore
require that Tdes is reasonably specified, and on the other hand, We specify the LPF DC gains θik , i ∈ Clpf k
in such a way
as outlined in the following, we need to carefully select the that they can be adapted online, proportionately to the time-
ADPMs Mi according to the individual device limitations. varying power capacity limits of the devices. For each channel
ADPF Selection: The ADPFs for the f-p and v-q channel k ∈ {fp, vq}, we consider the optimal quadratic allocation
are selected independently, but according to the same principle. P 1 k 2
In the following, we hence address both channels simultane-
minimize i∈C k y max,k (t) (θi (t))
θik (t),∀i∈Clpf
k lpf i

ously using the variable k ∈ {fp, vq}. A list of notation is

θik (t) mkj (s = 0) = 1 (11)
P P
subject to k
i∈Clpf + j∈N
provided inTable II.
For the non-controllable devices with fixed Tik , i ∈ N , for θik (t) ≥ 0, ∀i ∈ Clpf
k
,
each channel k, the ADPFs are obtained as
where the equality constraint assures the participation condi-
mki (s) := Tik (s)(Tdes
k
(s))−1 , ∀i ∈ N , k ∈ {fp, vq}, (9) tion (8), and yimax,fp (t) ≡ pmax (t) ∈ [pmax fp
, p̄imax ], ∀i ∈ Clpf
i i
¯
 4

TABLE II: List of notation for the disaggregation via ADPMs.

a time-scale separation and are thus rather slow [19]. A more
Description Symbol careful investigation of different update and communication
Control channel index k ∈ {fp, vq} strategies, along with the impact of communication delays
ADPM of device i Mi (s)
ADPF of device i for channel k mk
(which are not explicitly modelled in this paper) is subject
i (s)
Closed-loop transfer function of device i for channel k Tik (s) to future work.
Desired DVPP transfer function for channel k k
Tdes (s) Similar to the LPF DC gains, online adaptation could also
Time constant of device i for channel k τik be applied to all other ADPF parameters.
Time varying DC gain of device i for channel k θik (t)
Upper/lower DC gain of device i for channel k θik , θ̄ik
¯ k
Set of contrl. devices with LPF participation on channel k Clpf C. Design of Local Matching Control
Set of communication partners of device i for channel k Iik
Time-varying active power capacity limit of device i yimax,fp = pmax (t)
Next, we need to find local feedback controls for the
i
Lower/Upper active power capacity limit of device i pmax
i , p̄max
i
controllable devices C to ensure their closed-loop transfer
Time-varying reactive power capacity limit of device i ¯
yimax,vq = qimax (t) matrix Ti satisfies the matching condition (5). Inspired by
Lower/Upper reactive power capacity limit of device i qimax , q̄imax existing methods on multivariable [20]–[23] and adaptive [24],
Time-varying power capacity limit of device i for channel k ¯ y max,k (t)
i [25] H∞ control of power converters, we address these local
matching control designs with a robust and optimal LPV H∞
device j device i
control, which is well-suited to deal with the ADPMs, being
plant plant parameter-varying with respect to the time-varying LPF DC
control
gains. More specifically, the LPV H∞ control is computation-
control ···
device i ally tractable and provides a suboptimality certificate for the
yjmax,k (t) θjk (t) θjk (t)
plant
θik (t)
entire parameter space of the LPF DC gains, thereby limiting
yjmax,k (t)
yimax,k (t)
··· yimax,k (t) the performance degradation over all possible operating con-
control
θik (t) DVPP ··· ditions. Further, we ensure transient time-domain limits of the
operator plant devices are not violated during normal operating conditions.
··· Tolerating Mismatch: We first relax the restrictions on
control
the matching control and therefore modify the participation
··· device j
condition in (8) by allowing a mismatch in the high frequency
(a) Broadcast communication via (b) Distributed peer-to-peer com- range of the Bode plot. In particular, since the measurement
the DVPP operator. munication. unit (e.g. a phase-locked loop (PLL)) for the bus frequency
Fig. 2: Different options of communication structures to update the ADPFs. and voltage measurement is limited by some bandwidth 1/τc ,
it suffices if the participation condition in (8) only holds for
(or yimax,vq (t) ≡ qimax (t) ∈ [qimax , q̄imax ], ∀i ∈ Clpfvq
) rep-
¯ the frequency range below, i.e.,
resents the time-varying active (or reactive) power capacity
limit of device i. The LPF DC gains are given by the optimal P k ! 1
i∈N ∪C mi (s) = τc s+1 , k ∈ {fp, vq}. (14)
k
solution of (11), i.e., for all i ∈ Clpf and k ∈ {fp, vq}, we get
Consequently, line 9 in Algorithm 1 is adjusted as
y max,k (t)
 
θik (t) = 1 − j∈N mkj (s = 0) P i max,k , (12)
P  
Pn−1
l∈C k yl (t)
lpf
mkn ← τc s+11
− i=1 mki (s) , k ∈ {fp, vq}. (15)
where the quantity in the parentheses in (12) is the contribution To simplify notation, we will drop the device index i ∈ C
to the DC gain coming from the non-controllable devices. in the following, as the remainder of this subsection refers to
Obviously, the LPF DC gains in (12) are bounded on an the local matching control of one controllable device i.
interval θik (t) ∈ [θik , θ̄ik ], where the lower and upper bounds LPV H∞ matching control: For each controllable device,
depend on the lower ¯ and upper power capacity limits.
we attempt to find a matching controller where the specifica-
During power system operation, the DC gains in (12) can be tion in (5) serves as a local reference model to be matched.
updated in a centralized (broadcast) fashion, where the DVPP Let y := [∆p, ∆q]′ and w =: [∆f, ∆v]′ . The control design
fp vq
operator collects all pmax i (t), ∀i ∈ Clpf and qimax (t), ∀i ∈ Clpf , setup is shown in Fig. 3, where
and communicates back the solution (12) for k ∈ {fp, vq} to
all devices (Fig. 2a). The updates can be communicated either ẋ = Ax + Bu + Ew
(16)
continuously or in an event-triggered fashion. If a distributed y = Cx + Du + F w
implementation is more desired, one could alternatively use is the linearized reduced-order system of the plant (e.g., rep-
the consensus-filters [16] resenting the primary source technology and/or the associated
 k
θik (t)

d k
P θl (t) grid-side converter interface, see Section IV), and
dt θi (t) = l∈Iik y max,k (t) − y max,k (t) with
i
 l P  (13) ẋr = Ar (Θ(t))xr + E r (Θ(t))w
k k
P
θ (0) = 1 − m (0) (17)
y r = C r (Θ(t))xr + F r (Θ(t))w,
i∈C k i j∈N j
lpf

where Iik is the set of communication partners for all i ∈ is the LPV system representation of the local reference model
k
Clpf , k ∈ {fp, vq} to track the allocation rule (12) via peer-to- M · Tdes , which is included into the control setup, but not
peer communication of θik (t) and yimax,k (t) (Fig. 2b). Other a physical part of the system itself. The state-space matrices
possible implementations with a reduced amount of commu- in (17) depend affinely on the time-varying vector of parame-
nication could be based on adaptive-increase/multiplicative- ters Θ(t) ∈ Rlog2 (q) , which, according to the selected ADPM
decrease (AIMD) methods [17], [18], which however, require of the device, is composed of the different DC gain parameters
 5

 (l) ′ ′ ′
A Q+QA(l) +B(l) Y (l) +Y (l) B(l)

⋆ ⋆
′
 E (l) −γI ⋆ ≺ 0, l = 1, ..., q (18a)
C (l) Q+D(l) Y (l) F (l) −γI
′ ′ ′ ′
(A(l) +A(t) )Q+Q(A(l) +A(t) )′ +B(l) Y (t) +B(t) Y (l) +Y (l) B(t) +Y (t) B(l)
 
⋆ ⋆
′ ′ l = 1, ..., q−1
 E (l) +E (t) −2γI ⋆ ≺ 0, t = l+1, ..., q (18b)
(C +C )Q+D Y +D(t) Y (l)
(l) (t) (l) (t) F (l) +F (t) −2γI

θik (t), i ∈ Clpf

k
, k ∈ {fp, vq}. Since the latter are varying on matching error ε(s) = N (s)w(s) in the H∞ -norm as
the interval [θik , θ̄ik ] for all t ≥ 0 (cf. (12)), the vector Θ(t) minimize γ (21a)
¯
ranges over a fixed polytope of vertices Θ̂(1) , ..., Θ̂(q) . K(Θ)

We combine (16) and (17), and add an integral state σ̇ = ε subject to ||N (s)||∞ < γ, (21b)
of the matching error ε := y − y r as a feedback signal where N (s) is the transfer function obtained from (20). As a
to robustify the matching design against stationary model common practice in H∞ control, the H∞ -bound in (21b) can
uncertainties and nonlinearities of the actual device. This be reformulated in terms of a linear matrix inequality (LMI)
yields the augmented LPV system used for control design to obtain a tractable convex optimization problem that can
         be solved efficiently. Namely, by virtue of the Bounded Real
ẋ A O O x B E Lemma (BRL) [26], the H∞ bound in (21b) is equivalent to
ẋr =O Ar (Θ) Oxr + O u+ E r (Θ) w the LMI (⋆ indicates symmetric blocks)
σ̇ C −C r (Θ) O σ D F −F r (Θ) 
QAcl (Θ)′ + Acl (Θ)Q

| {z } | {z }| {z } | {z } | {z } ⋆ ⋆
ż A(Θ) z B(Θ) E(Θ)  E(Θ)′ −γI ⋆  ≺ 0, (22)
(19)
Ccl (Θ)Q F(Θ) −γI
 
x   
ε = C −C r (Θ) O xr + D u+ F −F r (Θ) w,
  
ensuring (open-circuit) closed-loop stability of the system
} σ |{z} |
in (20), if and only if there exists a symmetric positive definite
| {z {z }
C(Θ) | {z } D(Θ) F (Θ)
z
solution Q=Q′ ≻0 for all admissible values of Θ(t) [13], [27].
Parametric State-Feedback Gain: The key challenge of (22)
where O is the zero matrix with appropriate dimension. The is the infinite number of constraints it imposes, as we consider
associated static control law is given by u(t) = K(Θ(t))z(t), the entire polytope of possible parameter values of Θ(t). For
where the state-feedback gain K(Θ) is parameter-dependent, polytopic LPV systems, however, (22) can be reduced to a
and the resulting (open-circuit) closed-loop system is given as finite set of constraints [13], [27]. In particular, since the state-
space matrices in (19) depend affinely on Θ(t), and Θ(t) varies
ż = (A(Θ) + B(Θ)K(Θ))z + E(Θ)w in a polytope of vertices Θ̂(1) , ..., Θ̂(q) , the state-space matrices
| {z } are constrained on the polytope with vertices
Acl (Θ)    
(20) A(l), B (l), E (l), C (l), D(l), F (l) := (A, B, E, C, D, F) Θ̂(l) (23)
ε = (C(Θ) + D(Θ)K(Θ))z + F(Θ)w.
| {z }
Ccl (Θ) for l = 1, ..., q. Given the latter, it can be shown that (22) will
hold for all Θ(t) if there exist a symmetric positive definite
matrix Q and matrices Y (l) for l = 1, ..., q, such that the vertex
Under the configuration in Fig. 4, the parameter-dependent
LMIs in (18) are satisfied [13]. In this regard, we can transform
state-feedback gain K(Θ(t)) is obtained by minimizing the
the infinite-dimensional optimization problem in (21) into a
T (s) convex finite-dimensional optimization problem
linearized plant y
w minimize γ (24a)
ẋ = Ax + Bu + Ew Q,Y (1) ,...Y (q)
u y = Cx + Du + F w x

matching control
subject to Q = Q′ ≻ 0 (24b)
feedback gain (18a) & (18b), (24c)
σ R ε +
K(Θ)
xr −
to compute the vertex controllers K (l) := Y (l) Q−1 . The
LPV reference model M (s) · Tdes (s) associated state-feedback control gain is then obtained as a
ẋr = Ar (Θ)xr + E r (Θ)w convex combination of the vertex controllers, with convex
y r = C r (Θ)xr + F r (Θ)w yr
coefficients corresponding to the instantaneous Θ(t). I.e.,
Pq Pq
Fig. 3: Setup for matching control design of device i ∈ C. K(Θ(t))= l=1 λ(l) (Θ)K (l), λ(l) ≥ 0, (l)
l=1 λ =1, (25)
where, for a given Θ(t) during power system operation, the
N (s)
w ε
coefficients λ(l) (Θ) of the convex combination of the vertices
ż = A(Θ)z + B(Θ)u + E(Θ)w are computed such that [13]
ε = C(Θ)z + D(Θ)u + F(Θ)w Pq Pq
(l) (l)
l=1 λ (Θ)Θ̂ = Θ(t), λ(l) (Θ) ≥ 0, (l)
l=1 λ (Θ) = 1.
u z
K(Θ)
For the parameter polytopes we are considering in our applica-
tions, i.e., simplices and parallelotopes, there exist analytical
Fig. 4: H∞ control design configuration for model-matching. closed form-expressions of the coefficients λ(l) (Θ) [28].
 6

TABLE III: IEEE nine-bus system parameters.

Controller Tuning: To ensure accurate model matching, the
H∞ design in (24) generally selects large vertex control gains Parameter Value
K (l) . This results in an overly aggressive controller which is System base power 100 MVA
System base voltage (ph-ph, rms) 230 kV
not favorable in the presence of nonlinearities, state and input System base frequency 50 Hz
constraints (e.g. transient current limits of a converter.), or Power rating, SG1 250 MVA
during off-steady-state conditions. Power rating, SG2 96 MVA
We therefore regularize the problem in (24) by incorporating Power rating, SG3 64 MVA
Voltage rating, SG1 (ph-ph, rms) 16.5 kV
state and input constraints of the system in (19) into control Voltage rating, SG2 (ph-ph, rms) 18 kV
design. To do so, we employ an idealized method for design Voltage rating, SG3 (ph-ph, rms) 13.8 kV
purposes, i.e., we do not aim to represent actual state and
input constraints, but rather provide a tool, which, after tuning we investigate a multivariable DVPP setup replacing the fast
parameters, can lead to these constraints being satisfied. frequency and voltage control of a thermal-based generator,
The tuning method for state and input constraint satisfaction while additionally including online-adaptation of the ADPMs
is based on the reformulation of these constraints in terms of to handle temporal variability of weather-dependent DERs.
LMIs, which can immediately be included into the problem Finally, in a third case study, we conceptually demonstrate the
in (24). Namely, as derived in [29], [30], for w limited benefits of our DVPP control strategy based on ADPFs over
in energy and z(0) = 0, the input and state constraints competing approaches to DVPP control in [11], [12], relying
maxt≥0 ||u(t)|| ≤ µ and maxt≥0 |zj (t)| ≤ ζj for some states on dynamic, or even only static participation factors.
j ∈ {1, ..., v}, are enforced for all t ≥ 0, if Q and Y (l) satisfy
′
" #
Q Y (l) A. System Model
2 ⪰ 0, l = 1, ..., q, (26a)
Y (l) µα I As in the nine-bus system in [32], we consider two con-
1 2
 ventional thermal-based power plants and one hydro power
α diag(ζj ) ZQ ⪰ 0, j ∈ {1, ..., v}, (26b) plant for our initial test case (Fig. 5 and Table III). The im-
(ZQ)′ Q plementation is based on the system model in [33], where the
where Z = [Z1′ · · · Zv′ ]′ and Zj is a row vector of zeros transmission lines are modelled via nominal π sections, and
with a ’1’ at position j, || · || is the Euclidean norm, and the transformers via three-phase linear transformer models.
µ, ζj , α ∈ R are tuning parameters to handle several perfor- The loads are modelled as constant impedance loads. We adopt
mance requirements (see [29], [30] for details). In this regard, an 8th-order model for the synchronous machines equipped
we also implicitly tune the integral gain of the controller by with a ST1A excitation system with built-in automatic voltage
adjusting the bound ζj for the state σ. regulator (AVR) and a power system stabilizer (PSS) [33].
We include the LMIs in (26) into the problem in (24) and The governors are modelled as a proportional speed-droop
arrive at the final H∞ model-matching problem control with first-order delay, and the hydro and steam turbine
parameters are taken from [34].
minimize γ (27a)
Q,Y (1) ,...Y (q)
subject to Q = Q′ ≻ 0 (27b) B. Grid-Side Converter Model and Control Architecture
(18a) & (18b), (27c) All DVPP case studies include (among others) converter-
based generators which are interfaced to the grid via power
(26a) & (26b), (27d) converters, and thus considered as controllable (∈ C) within
where the resulting vertex controllers K (l), l = 1, ..., q, are used their respective DVPP control setup. The proposed grid-side
to compute the LPV feedback gain K(Θ(t)) online via (25). converter model used for dynamic simulation represents an
aggregation of multiple commercial converter modules, and
is based on a state-of-the-art converter control scheme [35],
IV. T EST C ASE
into which we have incorporated the H∞ matching control
To verify our DVPP controls, we use Simscape Electrical to (Fig. 6). Namely, we employ a grid-supporting control scheme
perform an electromagnetic transients (EMT) simulation based that is synchronized with the grid voltage and contributes to
on the IEEE nine-bus system3 . In particular, to demonstrate the regulation of the grid frequency and voltage according to
the basic idea of our DVPP control strategy in an instructive the local desired DVPP specifications Mi · Tdes , respectively.
way, we consider a deliberately simple test system assembled
with DVPPs containing only a few devices. Our proposed 2 7 8 9 3

method, however, can be easily extended to larger power

systems comprising DVPPs with a larger number of devices, SG 2 SG 3
(thermal-based) 5 6 (thermal-based)
especially since our local matching control is independent of
the size of the power system and the number of DVPP devices. 4
An investigation of such scenarios will be part of future work.
In a first case study, we start with a tutorial example of a
DVPP composed of only weather-independent DERs, specified 1
to improve the fast frequency response of the initial system SG 1
(hydro)
in [32] in a non-adaptive fashion. In a second case study,
Fig. 5: IEEE nine-bus system with two thermal-based power plants and one
3 The MATLAB/Simulink implementation is available online [31]. hydro power plant [32] (SG = synchronous generator).
 7

dc energy source model power turbine [34]. As proposed in [8], [36], this poor response
ix converter
+ Rf Lf LV/MV behavior can be compensated by a battery energy storage
idc
1
τdc s+1 Gdc Cdc vdc system (BESS) connected to the same bus. In addition, we
− vc,abc to DVPP
iabc vabc connection complement the hydro turbine and BESS by supercapacitor
point
mabc
abc (sc) for fast frequency response, as in hybrid energy storage
i⋆dc vdc modulation
dc voltage control 2 v⋆
dq systems [10]. This gives us a DVPP at bus 1 (Fig. 7a), for
mabc = vc,abc θpll
i⋆dc = kdc (vdc
⋆
− vdc ) ⋆
dc
which we specify an aggregate frequency response behavior
⋆ abc
vc,abc PLL
dq ẋpll = vq identical to a proportional f-p droop control, i.e.,
abc
θ̇pll = kppll vq + kipll xpll
dq θpll −D
⋆
vc,dq
∆p(s) = Tdes (s)∆f (s), Tdes (s) := τ s+1 , (28)
vdq ẋ
inner current control loop
⋆ idq where D is the desired droop coefficient, and the denominator
i,dq = idq − idq
⋆
vc,dq = vdq + Zf idq + with τ is included to filter out high frequency dynamics. The
+kpi (i⋆dq − idq ) + kii xi,dq i⋆dq fpll ||vdq || parameter values are provided in Table V. The voltage control
idq , xi,dq outer control loop including
at bus 1 is fully provided by the installed AVR of the hydro
vdq power calculation p H∞ matching control plant and not part of the DVPP control. Thus, (28) represents a
idq
p = vd id + vq iq
q p, q ··· matching ··· one-dimensional version of the aggregate specification in (3).
q = vq id − vd iq control
To provide each DVPP device with the DVPP input signal,
Lf ωpll given by the bus frequency deviation at bus 1, we use individ-
Fig. 6: Converter model in per unit, where Zf = ( ωb
J2 + Rf I2 ) with
J2 = [0 -1; 1 0] and I2 = [1 0; 0 1].
ual bus measurements of the devices (instead of one common
bus measurement), assuming that independent measurements
TABLE IV: Converter model and control parameters. are sufficiently similar.
Parameter Symbol Value
ADPF Selection: Since the aggregate specification of
dc link capacitor Cdc 0.096 pu DVPP 1 in (28) is a one-dimensional transfer function, the
dc link conductance Gdc 0.05 pu ADPMs of the devices reduce to the ADPFs for f-p control,
RL-filter resistance Rf 0.01 pu respectively, i.e., Mi (s) = mfp
i (s), i ∈ {hydro, bess, sc}.
RL-filter inductance Lf 0.0942 pu A small-signal model of the hydro governor and the turbine
dc voltage control gain kdc 100
PLL control gains kppll
, kipll 0.4775, 5.4113
dynamics with input being the rotor frequency deviation and
Current control gains kpi
, kii 0.73, 1.19 output the mechanical active power deviation is given by5 [34]
Reactive power control gains q
kp , kiq 0.005, 0.0005 −1/Rg τr s+1 1−τw s
p ⋆ Thydro (s) = τg s+1 (Rt /Rg )τr s+1 1+0.5τw s , (29)
Converter rated at Sr , vr = 2/3 kV (ph-n, peak), and vdc = 3vr .
Parameters in per unit, normalized w.r.t. to the dc- and ac-side ratings, respectively. | {z } | {z } | {z }
speed droop transient droop turbine

Similar to [33], we assume that the dc current idc is supplied where the associated parameter values are given in Table V.
by a controllable dc current source (Fig. 6), e.g. representing Since (29) describes a pre-installed setup of the hydro unit at
the machine-side converter of a wind power plant, a PV bus 1, it is considered as fixed (∈ N ) within the DVPP setup
system, or an energy storage. In particular, we use a coarse- for f-p control. In particular, for a choice of 1/D = Rg and
grain model of the underlying primary source technology τ = τg in (28), the hydro ADPF is given as
and model its response time by a first-order delay with time mfp -1 τr s+1 1−τw s
hydro (s) = Tdes (s) Thydro (s) = (Rt /Rg )τr s+1 1+0.5τw s (30)
constant τdc [33], e.g. representing the resource associated
dynamics, communication delays and/or actuation delays. i.e., by the design choice of (28), the hydro unit establishes
The grid-side converter control is separated into two control the full DC gain of DVPP 1, so that Thydro (0) = Tdes (0).
loops for the dc and the ac side (Fig. 6). The dc-side control The converter-based BESS and supercapacitor are control-
regulates the dc voltage through the dc current source and lable (∈ C) and used to complement the hydro response on
a proportional controller [33]. The ac-side control is used to
8 9 3 101
control the network current magnitudes. It is implemented in
a dq-coordinate frame oriented via a phase-locked loop (PLL)
100
which tracks the system frequency after the RL-filter, keeping 6
SG 3

the converter synchronized with the grid voltage [35]. The ac-
4
side control is composed of a cascaded control loop, where 10-1

the outer loop, containing the H∞ matching control, provides

the ac current reference i⋆dq to the inner current control loop4 . 1 10-2
The converter and control parameters are given in Table IV. super-
BESS capacitor DVPP 1
hydro 10-3
C. Case Study I: Supplementing Hydro in Frequency Response 10-2 100 102

The initial system (Fig. 5) is characterized by a poor (a) Case study I: IEEE nine-bus (b) Magnitude Bode plots of the
short-term frequency response, caused by the transient droop system with a DVPP at bus 1. ADPFs of the DVPP 1 devices.
compensation and the non-minimum phase zero of the hydro Fig. 7: System model and ADPFs of case study I.

4 The proposed H 5 Since the hydro power plant is naturally grid-forming, we only consider
∞ matching control scheme can also be adapted to
other type of converter controls, e.g. more classic versions where the grid- the “grid-following” part of the hydro power plant for the DVPP participation,
side converter is regulating the dc voltage. given by the hydro governor and turbine dynamics.
 8

outer control loop including H∞ matching control (DVPP 1)

id,0
Firstly, we disregard the dc side of the converter by assum-
⋆
matching control xi,d,0 ing vdc ≈ vdc and only consider the ac-side dynamics for con-
i⋆d
i⋆d,0
++ ∆i⋆d
feedback gain ∆i , ∆x
d i,d
-+ id , xi,d trol design. In particular, since the ADPFs mfpi , i ∈ {bess, sc}
σp R p
Ki
-
+ ∆p -+
include the time constants τbess and τc > τsc of the converter’s
p0
reference model Mi (s) · Tdes (s)
primary source (cf. (31) and (32)), the regulation of active
f⋆
fpll +- ∆fpll ẋr = Ari xr + Eir ∆fpll xr
power on the ac side, imposed by the local reference model
∆pr = Cir xr + Fir ∆fpll ∆pr Mi · Tdes , will be slower or equal than the dc-side control.
Considering the ac-side dynamics, we extract the PLL from
i⋆q reactive power control
ẋq = q − q ⋆
q the model, as it provides the DVPP input signal in terms
i⋆q = kpq (q − q ⋆ ) + kiq xq of the frequency deviation ∆fpll . Therefore, the remaining
system for control design consists of the RL-filter dynamics,
Fig. 8: Outer control loop with matching control for DVPP 1 converters, where
Ki , i ∈ {bess, sc} is provided with the states of the nonlinear converter the current loop, and the power computation, given in the dq-
model in Fig. 6. The reactive power set point is regulated with a PI controller. coordinate frame of the PLL as
Lf
faster time scales, therefore participating in the specified f-p ωb i̇dq = −Zf idq + vc,dq − vdq (33a)
control in (28) as BPF and HPF, respectively. Their ADPFs ẋi,dq = i⋆dq − idq (33b)
are selected by Algorithm 1, i.e., ⋆
vc,dq = vdq + Zf idq + kp,i (i⋆dq − idq ) + ki,i xi,dq (33c)
q = vq id − vd iq .
 
mfp (s) = 1 1
− m fp
(s) (31) p = vd id + vq iq , (33d)
bess τbess s+1 τbess s+1 hydro

mfp 1
− mfp fp We proceed by approximating vd ≈ v ⋆ and vq ≈ 0 as
sc (s) = τc s+1 bess (s) − mhydro (s), (32)
constants, such that the active and reactive power expressions
where the magnitude Bode plots are shown in Fig. 7b and the become decoupled, i.e., p ≈ v ⋆ id and q ≈ −v ⋆ iq , assuming to
parameter values are provided in Table V. The time constant stay close to the nominal operating point. In particular, since
τbess corresponds to the dc time constant in the associated we include the reference model Mi · Tdes into the matching
converter model in Fig. 6, representing actuation delays of the control, we can compensate for the mismatch between the
BESS technology. In contrast to the hydro unit, the frequency desired active power injection and the latter approximation.
measurements for the BESS and the supercapacitor are given ⋆ ⋆
by their respective PLL. Therefore, the HPF ADPF of the Moreover, since vdc ≈ vdc , we can assume vc,dq ≈ vc,dq
supercapacitor is cut at the PLL-bandwidth 1/τc < 1/τsc and therefore reduce (33c) and (33a) to
according to the relaxation in (15). Lf
= kp,i (i⋆dq − idq ) + ki,i xi,dq .
ωb i̇dq (34)
Since mfphydro is non-adaptive by definition and establishes
the full DC gain of DVPP 1, all ADPFs in (30) to (32) are free By linearizing around the nominal operating point, we obtain
of any adaptive DC gain parameters, which is in accordance
with the weather-independence of all three devices. Lf
ωb ∆i̇dq = kp,i (∆i⋆dq − ∆idq ) + ki,i ∆xi,dq (35a)
Local matching control: To match the local closed-loop
dynamics of the converter-based BESS and supercapacitor ∆ẋi,dq = ∆i⋆dq
− ∆idq (35b)
⋆
with their local reference model Mi · Tdes , i ∈ {bess, sc}, re- ∆p = v ∆id , ∆q = −v ⋆ ∆iq , (35c)
spectively, we employ the previously introduced H∞ matching
control in the outer control loop of their respective grid-side where all variables represent deviations from their respective
converter (Fig. 6). As shown in Fig. 8, the matching control is equilibrium point, i.e., ∆idq = idq − idq,0 , ∆i⋆dq = i⋆dq − i⋆dq,0 ,
defined by the local reference model Mi · Tdes , i ∈ {bess, sc} ∆xi,dq = xi,dq − xi,dq,0 , ∆p = p − p0 and ∆q = q − q0 .
and the associated state-feedback gain Ki . In particular, since Since we are only interested in controlling i⋆d to regulate the
the ADPFs of the BESS and the supercapacitor are fixed trans- active power, the decoupled system in (35) can be reduced to
fer functions (and with that their reference models Mi · Tdes ), only considering the d-states and the active power output, i.e.,
the matching control design simplifies in such a way, that the Lf
= kp,i (∆i⋆d − ∆id ) + ki,i ∆xi,d
ωb ∆i̇d (36a)
respective optimization problem in (27) has to be solved for
a parameter polytope consisting of one vertex only. Conse- ∆ẋi,d = ∆i⋆d
− ∆id (36b)
⋆
quently, each matching controller Kbess (Θbess (t)) = Kbess ∆p = v ∆id . (36c)
and Ksc (Θsc (t)) = Ksc is parameter-independent, and directly TABLE V: DVPP 1 parameters.
obtained as the solution of (27), respectively.
Parameter Symbol Value
Beyond that, we consider a one-dimensional DVPP specifi- Power rating, DVPP 1 Sdvpp1,r 250 MVA
cation for the f-p control of DVPP 1 (cf. (28)), and therefore Power rating, hydro Shydro,r 250 MVA
employ the matching control only in the converter’s “active Power rating, BESS Sbess,r 50 MVA
power path” to control the d-component i⋆d (Fig. 8). Power rating, supercapacitor Ssc,r 25 MVA
Hydro droop control gains Rg , R t 0.03, 0.38
Simplified Converter Model for Control Design: According Hydro time constants τg , τ r , τ w 0.2 s, 5 s, 1 s
to Section III-C, a linearized representation of the converter Device time constants τbess , τsc , τc 0.2 s, 0.01 s, 0.081 s
model in Fig. 6 is required to design the feedback gain Ki Tdes parameters D, τ 1/0.03, 0.2 s
in Fig. 8. While linearizing the converter model, we proceed αbess = αsc 5 · 10-5
H∞ tuning parameters µbess = µsc 1
by making a reasonable model reduction, which supports the ζbessσp = ζscσp 2.5 · 10-4
optimization problem in (27) in computation and scope.
 9

the v-q control, and the denominators with τp and τq are

0 included to filter out high-frequency dynamics. The associated
20 parameter values are given in Table VI.
-0.2 Due to their volatile availability, neither PV nor wind are
10 typically solely installed for frequency and voltage control.
-0.4
However, the complementary nature of wind and solar energy
5 25 50 75 100 125 150 can be exploited to compensate each other’s fluctuations [37].
0
This motivates the combination of wind, PV and STATCOM
0.4 for DVPP 3, which, thanks to our adaptive control approach
40 0.2
0
(Section III), is able to handle temporal variability of the DERs
20 -0.2 without affecting the overall DVPP response behavior.
5s 5.04s ADPF Selection: The ADPMs of wind, PV and STATCOM
0 (st) are composed of the ADPFs for f-p and v-q control, i.e.,
5 25 50 75 100 125 150 5 25 50 75 100 125 150  fp 
mi (s) 0
Mi (s) = , i ∈ {wind, pv, st}. (38)
Fig. 9: Frequency and active power deviation of the hydro unit at bus 1 in the 0 mvq
i (s)
initial system (Fig. 5) compared with DVPP 1 in case study I (Fig. 7a). The
dashed lines indicate the desired active power injection of the DVPP devices.
Further, both the wind, PV and STATCOM are converter-
based, thus considered as controllable (∈ C) within the DVPP
Finally, the system in (36) serves as a linearized plant
setup for f-p and v-q control in (37). For the ADPFs of the
model for the design of the BESS and supercapacitor matching
f-p channel, we select wind and PV to participate as LPFs:
controllers (cf. Fig. 3), where the PLL frequency deviation
∆fpll is an input to only the device’s local reference model fp
θwind
fp
θpv
mfp
wind (s) = τwind s+1 , mfp
pv (s) = τpv s+1 , (39)
Mi · Tdes (i.e., E = 0 in (16)). In this regard, the augmented
system in (19), required to solve the optimization problem where τwind , τpv are the dc time constants employed in the
in (27), is established by (36) and the individual reference associated converter model in Fig. 6. The adaptive DC gains
model Mi · Tdes , i ∈ {bess, sc}, respectively, where Tdes is fp
θwind fp
(t) and θpv (t) are defined as in (12), i.e.,
given in (28). The tuning parameters in (26) to limit transients
of the converter currents as ||∆i⋆d || ≤ µbess and ||∆i⋆d || ≤ µsc , fp pmax
wind (t) fp fp
θwind (t) = pmax ∈ [θwind
max , θ̄wind ],
wind (t)+ppv (t)
¯
as well as to shape the integral gain of the controllers as max (40)
fp ppv (t) fp fp
|σp | ≤ ζbessσp and |σp | ≤ ζscσp , are given in Table V. θpv (t) = pmax (t)+p max (t) ∈ [θpv , θ̄pv ],
Simulation Results: We compare the frequency response of wind pv ¯
the hydro unit at bus 1 in the initial system in Fig. 5 with the where the active power capacity limits vary in the interval
response of DVPP 1 in Fig. 7a in detailed simulations based pmax max max max max max
wind (t) ∈ [pwind , p̄wind ] and ppv (t) ∈ [ppv , p̄pv ], respec-
on nonlinear system and device models. In both scenarios, the ¯
tively. The ADPF of the f-p channel for ¯the STATCOM with
hydro unit is assigned 60% of the baseload, while the thermal- battery is obtained via Algorithm 1 as
based power plants at buses 2 and 3 provide the remaining
40%. Since we focus on time scales of minutes to seconds mfp
st (s) =
1
τc s+1 − mfp fp
wind (s) − mpv (s), (41)
or faster, the impact of the state of charge of the BESS and
where 1/τc is the PLL bandwidth. Note that mfp st depends on
the supercapacitor are neglected. From the simulation results fp fp
of the nine-bus system in Fig. 9, we can observe that the the adaptive DC gains θwind and θpv of wind and PV.
frequency response of the stand-alone hydro unit to a 30 MW Since the reactive power capability of the converters is
load step at bus 6 is quite poor in terms of both settling time independent of the dc source technology in our test case, we
and frequency nadir. In contrast, the DVPP highly improves select the ADPFs of the v-q channel for all three devices as
the frequency response of the hydro plant, and the BESS θ vq θ vq θ vq
mvq wind vq pv vq
wind(s) = τc s+1 , mpv(s) = τc s+1 , mst (s) = τc s+1 . (42)
st
and the supercapacitor match their desired power injection
(dashed lines) very well. We also observed that the transient 101
actuation constraints, i.e., converter current constraints, are not
8 9 3 wind
encountered during the load step (not shown). 100
PV
STATCOM 10-1
D. Case Study II: SG Replacement by Adaptive DVPP Control 6 & battery

DVPP 3
While keeping DVPP 1 at bus 1, we replace the thermal- 4
based power plant at bus 3 by another DVPP, consisting of a
wind power plant, a PV system and a STATCOM with battery 100

(Fig. 10a). We want to substitute the services of the thermal- 1

based power plant, and specify a f-p and v-q control as super- 10-1
BESS capacitor DVPP 1
"−D −H s # hydro
p p
0
   
∆p(s) ∆f (s) 10-2
τp s+1 10-2 100 102
= Tdes (s) , Tdes (s):= −Dq , (37)
∆q(s) ∆v(s) 0 τq s+1 (a) Case study II: IEEE nine-bus (b) Magnitude Bode plots of the
where Hp and Dp are the normalized virtual inertia and droop system with DVPPs at buses 1 and 3. ADPFs of the DVPP 3 devices.
coefficients for the f-p control, Dq is a high gain droop for Fig. 10: System model and ADPFs of case study II.
 10

outer control loop including H∞ matching control (DVPP 3) Finally, during online operation, we compute the state-
idq,0
matching control xi,dq,0 feedback controllers Kwind (Θwind (t)), Kpv (Θpv (t)) and
i⋆d,0 controller ∆idq , ∆xi,dq
i⋆d ++ ∆i⋆d σq R
-+ idq , xi,dq
q
Kst (Θst (t)) as the convex combination of the respective
Ki (Θ) -
+ ∆q -+ vertex controllers (cf. (25)), where the convex combination
i⋆q ∆i⋆q σp R q0 p coefficients are obtained via closed-form expressions derived
+ + ∆p -+
i⋆q,0 + -
p0 in [28], based on the instantaneous values of the vectors
reference model Mi (s) · Tdes (s)
f⋆ Θi (t), i ∈ {wind, pv, st}.
fpll ∆fpll
 
+- ∆fpll
ẋr = Ari (Θ)xr + Eir (Θ)
||∆vdq || xr Simulation Results: While keeping DVPP 1 at bus 1, we
||vdq || ||∆vdq || ∆pr ∆fpll ∆pr
   
+- ∆q r
= Cir (Θ)xr + Fir (Θ)
||∆vdq || ∆q r
replace the synchronous generator (SG) at bus 3 by DVPP 3 as
v⋆ in Fig. 10a, and investigate its frequency and voltage response
Fig. 11: Outer control loop with matching control for DVPP 3 converters. behavior in detailed simulations using nonlinear system and
device models. We adopt the same baseload supply for DVPP 3
The adaptive DC gains are defined as in (12), i.e., as for the prior SG, provided by the wind power plant and
max
vq qwind (t) vq vq
θwind (t) = ∈ [θwind
max (t)+q max (t)+q max (t)
qwind , θ̄wind ], 0 4

max
pv ¯
st
-0.02
vq q (t) vq vq 2
θpv (t) = qmax (t)+qpv
max max (t) ∈ [θpv , θ̄pv ], (43) -0.04
wind pv (t)+qst ¯
max 0
vq q (t) vq vq -0.06
θst (t) = qmax (t)+qstmax (t)+qmax (t) ∈ [θst , θ̄st ],
wind pv st ¯ -0.08 -2 2

where the reactive power capacity limit qimax is related to the -0.1
-4
0
active power capacity limit pmax i , i ∈ {wind, pv, st} via the 5 25 50
PQ-capability curve of the power converter [38]. Namely, for -2
5s 5.1 5.2s
vdc sufficiently large, the ac voltage converter limit can be
disregarded, and, under the assumption of a constant voltage
6
magnitude, the stationary current converter limit ||idq ||max = 1 0.05

4
pu relates the active and reactive power limits in per unit as
q q 0 2

qimax (t) = (||idq ||max )2 −pmax i (t)2 = 1−pmax i (t)2 . (44) 0

-0.05 -2 2
Finally, the magnitude Bode plots of all ADPFs during 1
nominal capacity conditions are shown in Fig. 10b, and the 5 25 50 0
associated parameter values are provided in Table VI. -1
Local matching control: Similar as in case study I, we
5s 5.1 5.2s
employ the H∞ matching control in the outer control loop
of each grid-side converter to match the local closed-loop 5 25 50

dynamics of the DVPP 3 devices with their local reference Fig. 12: System response in case study II after load step at bus 3. The dashed
model Mi · Tdes , i ∈ {wind, pv, st} (Fig. 11). However, in lines indicate the desired power injection of the DVPP devices.
contrast to the DVPP 1 converters, we now aim to regulate
0 4
the entire reference current i⋆dq to participate in the desired
-0.02
f-p and v-q control of DVPP 3 in (37). In this regard, we 2
-0.04
consider the previous linearized converter model in (35) with 0
both the d- and the q-states, as well as the active and reactive -0.06

power outputs as our plant model for control design. -0.08 -2 2

Furthermore, the ADPFs of the DVPP 3 devices (and with -0.1

-4 0
that their local reference models Mi · Tdes , i ∈ {wind, pv, st}) 5 25 50
rely on the adaptive DC gains introduced in (40) and (43), so -2
5s 5.1 5.2s
that the associated state-feedback gains Ki (Θ) are parameter-
dependent now. More precisely, the parameter vector Θi (t) of
6
each device’s augmented LPV system in (19) is given by the 0.05

DC gains appearing in the individual ADPM in (38), i.e., 4

fp vq 0 2
Θwind (t) = [θwind (t),θwind (t)]′, Θpv (t) = [θpv
fp vq
(t),θpv (t)]′,
fp vq
(45) 0
Θst (t) = [θwind fp
(t),θpv (t),θst (t)]′ . -0.05 -2 2
1
Hence, for each device, the optimization problem in (27) 5 25 50
(1) (4) 0
is solved for the respective set of vertices Θ̂wind , ..., Θ̂wind , -1
(1) (4) (1) (8)
Θ̂pv , ..., Θ̂pv and Θ̂st , ..., Θ̂st , to compute the associated 5s 5.1 5.2s
vertex controllers, respectively. The tuning parameters in (26)
5 25 50
to limit transients of the converter currents as ||∆i⋆dq || ≤ µi ,
and to shape the integral gains of the controllers as |σp | ≤ ζiσp Fig. 13: System response in case study II after a PV generation loss at bus 3.
and |σq | ≤ ζiσq , ∀i ∈ {wind, pv, st}, are given in Table VI. The dashed lines indicate the desired power injection of the DVPP devices.
 11

101 101 control provided by the PV plant (see Fig. 14 in comparison

to Fig. 10b). Note that if we would stick to a non-adaptive
100 100 controller without adapting the ADPMs online, one would
observe a larger steady-state deviation in frequency due to
10-1 10-1
the absent DC gain contribution of the PV plant.

10-2 10-2 E. Case Study III: Comparison with Existing DVPP Concepts
10-2 100 102 10-2 100 102
After having studied different use cases of our DVPP control
Fig. 14: Magnitude Bode plots of the ADPFs of the DVPP 3 devices during strategy in case studies I and II, we now aim to demonstrate
the PV generation loss. the benefits of our method over competing approaches to
DVPP control that already exist in literature, namely [11] and
TABLE VI: DVPP 3 parameters. [12]. In particular, the works in [11], [12] can conceptually
Parameter Symbol Value be considered as special cases of our method, both with
Power rating, DVPP 3 Sdvpp3,r 64 MVA respect to the aggregate DVPP specification Tdes , as well as
Power rating, wind Swind,r 70.5 MVA the disaggregation strategy by means of participation factors.
Power rating, PV Spv,r 53 MVA On the one hand, both [11] and [12] consider a desired
Power rating, STATCOM Sst,r 80 MVA
pmax max short-term frequency response on an aggregate level, therefore
Upper/lower capacity limit, wind wind , p̄wind 0 MW, 37 MW
Upper/lower capacity limit, PV ¯pmax , p̄max
pv pv 0 MW, 28 MW resembling a one-dimensional version of our aggregate MIMO
¯ specification in (3). On the other hand, the disaggregation
Device time constants τwind , τpv 1.5 s, 0.6 s
Device time constants τst , τc 0.2 s, 0.081 s strategy in [12] is based on static participation factors (SPF)
fp
Tdes parameters Dp , Hp , τp 1/0.03, 13 s, 0.2 s
vq similar to conventional secondary frequency control, while the
Tdes parameters Dq , τq 100, 0.2 s
αwind = αpv = αst 5 · 10-4
one in [11] relies on dynamic participation factors (DPF). In
µwind = µpv = µst 0.25 this regard, they represent a (static and) non-adaptive version
H∞ tuning parameters ζwindσp , ζwindσq 0.011, 0.0002 of the ADPFs in our approach (cf. Section III-A), respectively.
ζpvσp , ζpvσq 0.0047, 0.0002 To demonstrate the conceptual differences of our proposed
ζstσp , ζstσq 0.00175, 0.000175 DVPP control strategy and the existing methods in [11], [12]
with respect to their disaggregation strategies, we compare the
the PV system. The latter (wind and PV) are operated under three different types of participation factors, i.e., SPFs, DPFs
deloaded conditions with respect to their maximum power and ADPFs, for a DVPP with a one-dimensional frequency
point tracking, allowing them to participate in frequency and control specification Tdes , given by
voltage regulation [39]. The DC gains of the ADPFs are −D
adapted by the DVPP operator and communicated to all DVPP ∆p(s) = Tdes (s)∆f (s), Tdes (s) := τ s+1 , (46)
devices (the impact of communication delays is neglected). where D is the desired droop coefficient, and the denominator
We first consider nominal generation capacities of the DVPP with τ is included to filter out high frequency dynamics.
devices, i.e., the ADPFs are given as in Fig. 10b. During the The parameter values are provided in Table VIII. To get an
simulation at t = 5 s, we impose a 18.7 MW load step at illustrative comparison of the different characteristics of the
bus 3. From the results in Fig. 12, it becomes apparent how participation factors, we combine those types of DVPP devices
all DVPP devices accomplish an accurate matching of their in case studies I and II with the most distinct heterogeneities
desired active and reactive power injection (dashed lines), such into one exemplary “showcase DVPP” at bus 1 (Fig. 15).
that their specified f-p and v-q controls establish an adequate In particular, we complement the existing hydro generator,
replacement of the prior SG services. In particular, due to the characterized by a very slow short-term frequency response
design choice of Tdes in (37), the aggregate DVPP behavior behavior, by a weather-driven wind and PV power plant, and
even outperforms the frequency and voltage responses of additionally add a BESS to obtain a reliable fast frequency
SG 3 (cf. red curves in Fig. 12 from a separate simulation). response behavior. The relevant parameter values of the de-
Moreover, we also observed that the transient actuation con- vices are given in Table VIII. As in case study I, the voltage
straints, i.e., converter reference current constraints, are not control at bus 1 is fully provided by the AVR of the hydro
encountered during the load step (not shown). plant, therefore not part of the aggregated DVPP control.
To examine the DVPP control during time-varying limits
2 7 8 9 3
on generation capacity, we induce a step decrease of the PV
active power capacity limit pmaxpv (t) at t = 5 s (e.g. caused by
a cloud), resulting in an overall generation loss of 18.7 MW at SG 2
5 6
SG 3
bus 3 (Fig. 13). This causes the ADPMs of all DVPP devices
to automatically adapt in response to the PV loss, such that 4
the respective LPV controllers Kwind (Θwind (t)), Kpv (Θpv (t))
and Kst (Θst (t)) are recomputed online. In particular, as we
can see by comparing the aggregate DVPP response during 1

the PV generation loss (Fig. 13) and during the load step wind PV BESS DVPP
of equal size (Fig. 12), the overall DVPP behavior in (37) hydro
remains nearly unaffected, since the wind and STATCOM
ADPMs are adapted to compensate for the missing DVPP Fig. 15: Case study III: IEEE nine-bus system with a DVPP at bus 1.
 12

SPFs DPFs ADPFs

1/Rg 1/Rg 1/Rg
Hydro mfp
hydro = D mfp
hydro (s) = D
τr s+1 1−τw s
(Rt /Rg )τr s+1 1+0.5τw s
mfp
hydro (s) = D
τr s+1 1−τw s
(Rt /Rg )τr s+1 1+0.5τw s
p̄max p̄max pmax
wind (t)
Wind mfp fp wind
wind = (1−mhydro ) p̄max +p̄max mfp fp wind
wind (s) = (1−mhydro (s = 0)) p̄max +p̄max
1
τwind s+1 mfp fp
wind (s) = (1−mhydro (s = 0)) pmax (t)+pmax
1
wind pv wind pv wind pv (t) τwind s+1
fp p̄max
pv fp p̄ max
pv fp p max (t)
pv
PV mfp
pv = (1−mhydro ) p̄max +p̄max mfp
pv (s) = (1−mhydro (s = 0)) p̄max +p̄max τpv s+1
1
mfp
pv (s) = (1−mhydro (s = 0)) pmax (t)+pmax (t) τpv s+1
1
wind pv wind pv wind pv
BESS mfp
bess = 0
fp fp fp
mbess (s) = 1−mpv (s)−mwind (s)−mhydro (s) fp fp fp fp
mbess (s) = 1−mpv (s)−mwind (s)−mhydro (s)fp

TABLE VII: Overview of the different types of participation factors for each DVPP device in case study III. To relax the restrictions on the matching control,
we additionally apply the relaxation in the high frequency range according to (14) for the DPFs and the ADPFs (not shown in this table).

TABLE VIII: Showcase DVPP parameters.

since the DPFs are chosen in such a way that different time-
Parameter Symbol Value scales of local device dynamics are taken into account (cf.
Power rating, DVPP Sdvpp,r 250 MVA Section III-A). However, although the DPFs incorporate the
Power rating, hydro Shydro,r 250 MVA
Power rating, wind Swind,r 38 MVA response time limitations of the DVPP devices, they are not
Power rating, PV Spv,r 70 MVA able to adapt to the changing PV conditions, and hence suffer
Power rating, BESS Sbess,r 30 MVA compensating for the loss of the DVPP control provided by
Upper/lower capacity limit, wind pmax max
wind , p̄wind 0 MW, 38 MW the PV plant, which is in contrast to our superior ADPF-based
Upper/lower capacity limit, PV ¯pmax , p̄max 0 MW, 70 MW
pv pv
Hydro droop control gains
¯
Rg , R t 0.04, 0.38
strategy.
Hydro time constants τg , τ r , τ w 0.2 s, 5 s, 1 s
Device time constants τwind , τpv 1.5 s, 0.6 s V. C ONCLUSION
Device time constants τbess , τc 0.2 s, 0.081 s
We have proposed a novel multivariable control approach
Tdes parameters D, τ 1/0.025, 0.2 s
to DVPPs, with the objective to provide desired dynamic
ancillary services in the form of fast frequency and voltage
control. We employ an optimal adaptive control strategy that
60 takes into account the DVPP internal constraints of the devices,
0
and can additionally handle temporal variability of weather-
-0.1 driven DER in a robust way. Our numerical case study in the
40
-0.2 IEEE nine-bus system shows the successful performance of
20 our controls, and, in particular demonstrates how our DVPP
-0.3
control strategy can be used to improve the fast frequency
-0.4 response of the initial system, and to facilitate the dynamic
0
ancillary services provision by weather-driven DERs in future
5 25 50 75 100 125 5 25 50 75 100 125
power systems.
Ongoing research includes the incorporation of grid-forming
Fig. 16: Frequency and active power response of the DVPP in Fig. 15 for
different participation factors during a loss of PV generation. The dashed line converter controls into our DVPP setup, as well as the ex-
indicates the sum of desired active power injections of the DVPP devices. tension of the DVPP setup to a geographically distributed
scenario, where the DVPP devices are located at different
We compare the DVPP behavior for the three different types geographical regions within the power system. Moreover,
of participation factors during a step decrease of the PV active future work should address the design of multivariable and
power capacity limit pmax
pv (t) at t = 5 s, resulting in an overall robust specifications for the desired dynamic DVPP behavior.
generation loss of 45.7 MW at bus 1 (Fig. 16). To do so, we
consider three simulation runs, i.e., one for each disaggrega- R EFERENCES
tion strategy of Tdes . In order to get a fair comparison, we [1] F. Milano, F. Dörfler, G. Hug, D. J. Hill, and G. Verbič, “Foundations
use our previously introduced nonlinear system and device and challenges of low-inertia systems,” in Power Syst. Comput. Conf.
models, and employ the H∞ matching control framework for (PSCC). IEEE, 2018, pp. 1–25.
[2] S. Awerbuch and A. Preston, The virtual utility: Accounting, technology
the converter-interfaced generation units in all simulation runs. & competitive aspects of the emerging industry. Boston: Kluwer
An overview of the different participation factors for each Academic Publisher, 1997.
DVPP device is provided in Table VII. [3] H. Saboori, M. Mohammadi, and R. Taghe, “Virtual power plant (vpp),
definition, concept, components and types,” in APAC Power and Energy
Simulation results: From the simulation results in Fig. 16, Engr. Conf. IEEE, 2011, pp. 1–4.
we can clearly see how the different characteristics of the [4] R. Caldon, A. R. Patria, and R. Turri, “Optimal control of a distribution
system with a virtual power plant,” Bulk Power Syst. Dyn. and Control,
participation factors influence the frequency response behavior Cortina. d’Ampezzo, Italy, p. 18, 2004.
of the DVPP. In particular, due to the online adaptation of the [5] D. Pudjianto, C. Ramsay, and G. Strbac, “Virtual power plant and system
local device controls, the response obtained by our proposed integration of distributed energy resources,” IET Renew. Power Gen.,
vol. 1, no. 1, pp. 10–16, 2007.
ADPF-based DVPP control strategy (light red lines) correlates [6] E. Dall’Anese, S. S. Guggilam, A. Simonetto, Y. C. Chen, and S. V.
with the actual desired behavior of the DVPP (we have already Dhople, “Optimal regulation of virtual power plants,” IEEE Trans. Power
observed this in case study II when comparing Figures 12 Syst., vol. 33, no. 2, pp. 1868–1881, 2017.
[7] B. Marinescu. (2020) POSYTYF concept and objectives. [Online].
and 13). In contrast, the SPF-based DVPP control results in a Available: https://posytyf-h2020.eu/project-overview/project-structure
rather poor behavior, most of all because the slow frequency [8] H. Ghasemi and J. Melki, “Investigation of frequency containment
response behavior of the hydro power plant is not addressed reserves with inertial response and batteries,” 2019.
[9] R. W. Kenyon, A. F. Hoke, J. Tan, B. D. Kroposki, and B. S. Hodge,
by the SPFs of the faster DVPP devices. The use of DPFs, “Grid-following inverters and synchr. condensers: A grid-forming pair?”
in turn, drastically improves the transient DVPP behavior, NREL, Golden, CO (United States), Tech. Rep., 2020.
 13

[10] W. Li and G. Joos, “A power electronic interface for a battery super- nordic power system,” in 2016 IEEE Power and Energy Society General
capacitor hybrid energy storage system for wind applications,” in IEEE Meeting (PESGM). IEEE, 2016, pp. 1–5.
Power Electr. Specialists Conf. IEEE, 2008, pp. 1762–1768. [37] M. Nehrir, C. Wang, K. Strunz, H. Aki, R. Ramakumar, J. Bing,
[11] J. Björk, K. Johansson, and F. Dörfler, “Dynamic virtual power plant Z. Miao, and Z. Salameh, “A review of hybrid renewable/alternative
design for fast frequency reserves: Coordinating hydro and wind,” energy systems for electric power generation: Configurations, control,
2021, Submitted. [Online]. Available: https://arxiv.org/abs/2107.03087 and applications,” IEEE Trans. Sustain. Energy, vol. 2, no. 4, pp. 392–
[12] W. Zhong, J. Chen, M. Liu, M. A. A. Murad, and F. Milano, “Coordi- 403, 2011.
nated control of virtual power plants to improve power system short-term [38] S. G. Johansson, G. Asplund, E. Jansson, and R. Rudervall, “Power
dynamics,” Energies, vol. 14, no. 4, p. 1182, 2021. system stability benefits with vsc dc-transmission systems,” CIGRE
[13] V. Montagner, R. Oliveira, V. J. Leite, and P. L. D. Peres, “Lmi approach session B4-204, Paris, France, 2004.
for H∞ linear parameter-varying state feedback control,” IEE Proc. [39] M. Dreidy, H. Mokhlis, and S. Mekhilef, “Inertia response and frequency
Control Theory and Applications, vol. 152, no. 2, pp. 195–201, 2005. control techniques for renewable energy sources: A review,” Renew. and
[14] J. M. Morales, A. J. Conejo, H. Madsen, P. Pinson, and M. Zugno, Sustain. Energy Reviews, vol. 69, pp. 144–155, 2017.
Integrating renewables in electricity markets: operational problems.
Springer Science & Business Media, 2013, vol. 205. Verena Häberle is a Ph.D. student with the Auto-
[15] K. De Brabandere, B. Bolsens, J. Van den Keybus, A. Woyte, J. Driesen, matic Control Laboratory at ETH Zurich, Switzer-
and R. Belmans, “A voltage and frequency droop control method for land, since 2020. She received the B.Sc. and M.Sc.
parallel inverters,” IEEE Trans. Power Electron., vol. 22, no. 4, pp. degree in electrical engineering and information
1107–1115, 2007. technology from ETH Zurich, in 2018 and 2020,
[16] J. W. Simpson-Porco, Q. Shafiee, F. Dörfler, J. C. Vasquez, J. M. respectively. For her outstanding academic achieve-
Guerrero, and F. Bullo, “Secondary frequency and voltage control ments during her Master’s thesis at the Automatic
of islanded microgrids via distributed averaging,” IEEE Trans. Ind. Control Laboratory, ETH Zurich, under Professor
Electron., vol. 62, no. 11, pp. 7025–7038, 2015. Florian Dörfler, she was honored with the ETH
[17] X. Fan, E. Crisostomi, D. Thomopulos, B. Zhang, R. Shorten, and Medal and the SGA Award from the Swiss Society
S. Yang, “An optimized decentralized power sharing strategy for wind of Automatic Control (SSAC). Her research focuses
farm de-loading,” IEEE Trans. Power Syst., vol. 36, no. 1, pp. 136–146, on the control design of dynamic virtual power plants for future power
2020. systems.
[18] P. Ferraro, E. Crisostomi, R. Shorten, and F. Milano, “Stochastic
frequency control of grid-connected microgrids,” IEEE Trans. Power Michael W. Fisher is a postdoctoral researcher
Syst., vol. 33, no. 5, pp. 5704–5713, 2018. at ETH Zurich, Switzerland, since 2020, affiliated
[19] K. J. Åström and B. Wittenmark, Adaptive control. Courier Corporation, with both the Automatic Control Laboratory and the
2013. Power Systems Laboratory. He received his Ph.D.
[20] L. Huang, H. Xin, and F. Dörfler, “H∞ -control of grid-connected in Electrical Engineering: Systems at the University
converters: Design, objectives and decentralized stability certificates,” of Michigan, Ann Arbor in 2020, and a M.Sc. in
IEEE Trans. Smart Grid, vol. 11, no. 5, pp. 3805–3816, 2020. Mathematics from the same institution in 2017. He
[21] C. Kammer, S. D’Arco, A. G. Endegnanew, and A. Karimi, “Convex received his B.Sc. in Mathematics and Physics from
optimization-based control design for parallel grid-connected inverters,” Swarthmore College in 2014. His research interests
IEEE Trans. Power Electron., vol. 34, no. 7, pp. 6048–6061, 2018. combine power systems analysis with dynamics,
[22] S. Yang, Q. Lei, F. Z. Peng, and Z. Qian, “A robust control scheme control, and optimization of complex, networked
for grid-connected voltage-source inverters,” IEEE Trans. Ind. Electron., systems. The current focus has been on stability and control of power systems
vol. 58, no. 1, pp. 202–212, 2010. with high penetration of renewable energy.
[23] M. Chen, D. Zhou, A. Tayyebi, E. Prieto-Araujo, F. Dörfler, and
F. Blaabjerg, “Generalized multivariable grid-forming control design for Eduardo Prieto-Araujo is a Serra Húnter Lecturer
power converters,” arXiv preprint arXiv:2109.06982, 2021. with the Electrical Engineering Department at the
[24] T. Erfanmanesh and M. Dehghani, “Performance improvement in grid- Technical University of Catalonia (UPC), Barcelona,
connected fuel cell power plant: an lpv robust control approach,” Spain, where he is part of the CITCEA-UPC re-
International Journal of Electrical Power & Energy Systems, vol. 67, search group since 2010. He received the degree in
pp. 306–314, 2015. industrial engineering from the School of Industrial
[25] E. B. Muhando, T. Senjyu, A. Uehara, and T. Funabashi, “Gain- Engineering of Barcelona (ETSEIB), UPC, in 2011
scheduled H∞ control for wecs via lmi techniques and parametrically and the Ph.D. degree in electrical engineering from
dependent feedback part ii: Controller design and implementation,” IEEE UPC in 2016. During 2021, he is a visiting professor
Trans. Ind. Electron., vol. 58, no. 1, pp. 57–65, 2011. at the Automatic Control Laboratory, ETH Zurich,
[26] C. W. Scherer, “The riccati inequality and state-space H∞ -optimal Switzerland. His main interests are renewable gener-
control.” Ph.D. dissertation, Citeseer, 1990. ation systems, control of power converters for HVDC applications, interaction
analysis between converters, and power electronics dominated power systems.
[27] P. Apkarian, P. Gahinet, and G. Becker, “Self-scheduled H∞ control
of linear parameter-varying systems: a design example,” Automatica, Florian Dörfler is an Associate Professor at the Au-
vol. 31, no. 9, pp. 1251–1261, 1995. tomatic Control Laboratory at ETH Zurich, Switzer-
[28] B. Schürmann, A. El-Guindy, and M. Althoff, “Closed-form expressions land, and the Associate Head of the Department of
of convex combinations,” in Amer. Control Conf. (ACC). IEEE, 2016, Information Technology and Electrical Engineering.
pp. 2795–2801. He received his Ph.D. degree in Mechanical Engi-
[29] S. Boyd, L. El Ghaoui, E. Feron, and V. Balakrishnan, Linear matrix neering from the University of California at Santa
inequalities in system and control theory. SIAM, 1994. Barbara in 2013, and a Diplom degree in Engineer-
[30] H. Chen, P.-Y. Sun, and K.-H. Guo, “Constrained H∞ control of active ing Cybernetics from the University of Stuttgart,
suspensions: an lmi approach,” in Int. Conf. on Control and Autom. Germany, in 2008. From 2013 to 2014 he was an
(ICCA). Final Progr. and Book of Abstr. IEEE, 2002, pp. 157–157. Assistant Professor at the University of California
[31] V. Häberle, “Dynamic Virtual Power Plant case studies,” Github repos- Los Angeles. His primary research interests are
itory, 2021, https://github.com/VerenaHaeberle/DVPP-Case-Studies. centered around control, optimization, and system theory with applications
[32] P. M. Anderson and A. A. Fouad, The Elementary Mathematical Model. in network systems, especially electric power grids. He is a recipient of the
Wiley-IEEE Press, 2003, pp. 13–52. distinguished young research awards by IFAC (Manfred Thoma Medal 2020)
[33] A. Tayyebi, D. Groß, A. Anta, F. Kupzog, and F. Dörfler, “Frequency and EUCA (European Control Award 2020). His students were winners or
stability of synchronous machines and grid-forming power converters,” finalists for Best Student Paper awards at the European Control Conference
IEEE Trans. Emerg. Sel. Topics Power Electron., vol. 8, no. 2, pp. 1004– (2013, 2019), the American Control Conference (2016), the Conference on
1018, 2020. Decision and Control (2020), the PES General Meeting (2020), the PES
[34] P. Kundur, “Power system stability,” Power System Stability and Control, PowerTech Conference (2017), and the International Conference on Intelligent
pp. 7–1, 2007. Transportation Systems (2021). He is furthermore a recipient of the 2010 ACC
[35] A. Yazdani and R. Iravani, Voltage-sourced converters in power systems. Student Best Paper Award, the 2011 O. Hugo Schuck Best Paper Award,
Wiley Online Library, 2010, vol. 39. the 2012-2014 Automatica Best Paper Award, the 2016 IEEE Circuits and
[36] L. Saarinen, P. Norrlund, U. Lundin, E. Agneholm, and A. Westberg, Systems Guillemin-Cauer Best Paper Award, and the 2015 UCSB ME Best
“Full-scale test and modelling of the frequency control dynamics of the PhD award.