Electric Power Systems Research 189 (2020) 106646

Contents lists available at ScienceDirect

Electric Power Systems Research

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

Enhanced battery controller for inertia support in residential microgrid T

based on active disturbance rejection control
⁎
Bowen Wang ,a, Gregor Verbiča, Weidong Xiaoa, Archie C. Chapmanb
a
School of Electrical and Information Engineering, The University of Sydney, Sydney, Australia
b
School of Information Technology and Electrical Engineering, The University of Queensland, Brisbane, Australia

A R T I C LE I N FO A B S T R A C T

Keywords: Lack of system inertia is becoming a big concern in modern power systems with a high penetration of inverter-
Active disturbance rejection control interfaced generation. In this paper, we argue that PV-battery systems owned by end-users can be used as a
Grid-forming control source of inertia to improve the frequency performance of future grids. To that end, we implement grid-forming
Prosumers control on battery converters in a residential microgrid powered by prosumer-owned PV-battery systems and a
PV-battery systems
back-up generator. To improve performance, we propose a novel controller based on active disturbance rejection
Residential microgrid
control (ADRC). This controller observes and rejects the “total disturbance” of the system, thereby increasing the
response speed and enhancing the stability of the system. In addition, we incorporate an adaptive parameter
tuning algorithm into the ADRC controller, which automatically calculates and updates the optimal control
parameters in every sampling period. Simulation results are provided to verify the effectiveness and feasibility of
the proposed approach.

1. Introduction response because it cannot respond “naturally” to a disturbance due to

the time delay introduced by the PLL. Furthermore, grid-following
Power systems are in the midst of a rapid transformation due to the converters can only operate in a strong grid with a stiff frequency and
increasing penetration of renewable energy sources (RESs) [1]. Most voltage[5]. In a weak grid, such as microgrids with a large penetration
unconventional RESs (e.g. wind and solar), are connected to the grid of inverter-interfaced generation, the PLL used for grid synchronisation
through power electronic interfaces, so they do not provide an inertial might fail to operate properly [6,7]. Alternative methods, such as using
response in case of system disturbance. In a conventional power system, a transient frequency drift compensator [8] have been proposed to deal
inertia comes from the kinetic energy stored in the rotating masses of with the issue of synchronisation in weak grids.
synchronous generators, which serves as an immediate source of energy In comparison, in grid-forming control, the inverter acts as a voltage
that reduces frequency excursions following power imbalance. Power source, which both improves the frequency response and also enables
systems dominated by inverter-interfaced generation, on the other synchronisation in a weak grid [2]. This is made possible by ‘extracting’
hand, would experience a high rate of change of frequency (RoCoF) energy from the DC link capacitor much like synchronous machined
immediately after a disturbance [2,3]. extract kinetic energy from rotating masses; this results in a ‘natural’
Due to the fact that a large system inertia is attributed to the large inertial response compared to a ‘synthetic’ one provided by FFR. Ac-
and instant power supplied by the kinetic energy of synchronous ma- cordingly, the energy stored in the DC-link capacitor can be regarded as
chines, a possible solution in inverter-dominated power systems is to equivalent to the rotor of a synchronous generator that serves as im-
provide fast frequency response (FFR) to the grid using resources such mediate energy storage to the grid, and thereby the genuine inertia is
as wind turbines, solar photovoltaic (PV) and batteries. Due to the re- improved. The issue is that this energy is limited, so it needs to be
sponsiveness of power electronic converters, FFR can be activated much augmented by an additional energy source such as a battery or a su-
faster than primary frequency response [2]. The problem is that “con- percapacitor.
ventional” grid-following converters require a phase locked loop (PLL) A possible solution to the low-inertia problem is emerging at the
to measure the frequency, which introduces a time delay in the order of demand side of the energy supply chain, where consumers increasingly
a few hundred milliseconds [4]. Therefore, FFR is slower than inertial invest in rooftop PV-battery systems [9,10] that can be used as a source

⁎
Corresponding author.
E-mail addresses: bowen.wang@sydney.edu.au (B. Wang), gregor.verbic@sydney.edu.au (G. Verbič), weidong.xiao@sydney.edu.au (W. Xiao),
archie.chapman@uq.edu.au (A.C. Chapman).

https://doi.org/10.1016/j.epsr.2020.106646
Received 4 October 2019; Received in revised form 19 April 2020; Accepted 1 August 2020
Available online 12 August 2020
0378-7796/ © 2020 Elsevier B.V. All rights reserved.
B. Wang, et al. Electric Power Systems Research 189 (2020) 106646

of inertia. In Australia, for example, the penetration of integrated pro- 2. Battery-based grid-forming converter
sumer-owned PV-battery systems has increased from 7500 installed
systems in 2016 to 22,000 in 2018[11]. It is widely accepted that the 2.1. Inertial response in a conventional power system
uptake of these technologies will lead to an increasingly decentralised
energy system, which will enable the shift in the power system oper- We assume that n synchronous generators are installed in a con-
ating paradigm from generation-following-load to load-following-genera- ventional power system. The energy stored in the rotors of synchronous
tion [2]. This shift will be made possible by the services provided by the machines during steady state can be expressed as:
behind-the-meter distributed energy resources (DERs), including in- 1 1 1 1
ertial response, FFR, and primary frequency control. E= J1 ωg2 + J2 ωg2 + ⋯+ Jn ωg2 = Jωg2
2 2 2 2 (1)
Against this backdrop, the emergence of PV-battery systems offers a
possibility to use batteries as inertia providers. In our previous work where J = J1 + J2 + ⋯+Jn is the system inertia and ωg is the grid an-
[12], we considered a typical scenario for a PV-battery system, in which gular frequency. During the first instant of a contingency when the grid
PV and battery are connected in parallel via individual DC/DC con- frequency drops from ωg to ω, the energy difference required from the
verters. On the control level, the PV-battery system is interfaced with rotors is:
the grid by a hybrid inverter controlled in a grid-following mode, ef- 1 2 1 1
ΔE = Jωg − Jω2 = J (2ωg − Δω)Δω
fectively acting as a controlled current source. This paradigm offers a 2 2 2 (2)
chance to make batteries serve as the energy sources in grid-forming
where Δω = ωg − ω is the frequency difference. As Δω ≪ 2ωg, (2) can
converters to augment the energy capacity of the DC-link capacitor.
be simplified as:
Against this backdrop, the first contribution of the paper is to design
a novel battery-based grid-forming converter (BGFC) for frequency ΔE = Jωg Δω. (3)
support in residential microgrids. A combination of prosumer-owned
Observe in (3) that Δω is small during the first instant of a transient
PV-battery systems is considered for the case study, which includes
state provided that J is sufficiently large, which highlights the im-
passive loads and a back-up diesel generator, as illustrated in Fig. 1. We
portance of inertia.
assume that battery-based inverters operate in a grid-forming mode to
With the penetration of inverter-interfaced RESs with no physical
support voltage and frequency of the grid. There exist several grid
inertia, the number of synchronous generators is reduced to m. Thereby
forming control methods [13,14]; in this paper we use the matching
the inertial response is:
control. In matching control, the DC-link voltage is regulated by
adopting a droop characteristic to the voltage regulator to generate the 1 1 1
E= J1 ωg2 + ⋯+ Jm ωg2 + ERES = J^ωg2 + ERES
desired current reference from the DC source. Hence, the voltage con- 2 2 2 (4)
troller should be implemented on the DC side. The battery charging
where J^ is the reduced inertia and ERES is the energy stored in the RES.
controller, therefore, should include the voltage droop control and be
As J^ is much smaller than J, RES should contribute to the compensation
implemented into the bidirectional DC/DC converter (BDC). This is
of the load variation. During the inertial response, the energy difference
particularly important with significant disturbances where the energy
is expressed as:
stored in the DC-link capacitor is insufficient for a meaningful inertial
response. ΔE = J^ωg Δω + ΔERES. (5)
The second contribution of this paper is a novel battery charging
controller adopting active disturbance rejection control (ADRC) and Hence, ΔERES should be sufficient to be equivalent to the required in-
adaptive parameter tuning. Building on our previous work [12], this ertia, which compensates the power difference and makes J^ωg Δω → 0 .
paper focuses on its implementation for grid-forming control. After Furthermore, fast dynamic response is required to compensate the
applying the matching control, the frequency regulation problem can power difference. The grid-forming control should be properly designed
be transformed into the DC-link voltage regulation problem. The fre- to avoid time delay introduced by the PLL.
quency deviation is revealed by the change of DC-link voltage and this
voltage change can regarded as the “total disturbance” of the converter. 2.2. Battery-based grid-forming converter
Adopting the concept of an “extended state observer”, this “total dis-
turbance” is observed and immediately rejected [15]. To reduce the Matching control, one of the approaches for grid-forming control, can
state estimation error, an adaptive tuning method is designed in the be understood by the structural similarity and the duality between the
extended state observer to tune the controller parameters. generator model and the two-level power converter model [5]. This
The remainder of the paper is organized as follows: Section 2 de- reveals a link between the DC link voltage and the rotor angular fre-
scribes the basics of the inertial response and the BGFC. Section 3 ex- quency. Frequency regulation can be realized by controlling the DC link
plains the proposed ADRC-based controller and the adaptive tuning voltage, which obviates the need to measure the AC quantities.
method. Section 4 describes the simulation results for a residential The matching control applied to a BGFC is illustrated in Fig. 2. The
microgrid as a case study. Section 5 concludes the paper. “matching part” that mimics the synchronous generator is implemented
in the inverter control, which generates modulation signals. Matching
approach relies on a measurement of the DC voltage vdc, which is then
translated into the voltage frequency ωc using the constant η = ωref / vdc
*
that encodes the ratio between the nominal grid frequency ωref and the
DC voltage reference vdc * and vbat represents the nominal
* . Voltage vmag
AC voltage reference magnitude and battery output voltage, respec-
tively.
Following the approach, the frequency regulation problem is
translated into the DC voltage control problem, which is realized in the
battery charging controller. Fig. 2 shows the droop characteristic when
a droop gain of mp is programmed within the controller to achieve
active power sharing.
The reference for the current discharged from the battery is sym-
Fig. 1. Residential Microgrid for case study. * . The value is determined by the droop output and the
bolized by ibat

2
B. Wang, et al. Electric Power Systems Research 189 (2020) 106646

Fig. 3. Bidirectional DC-DC Converter.

3.1. ADRC-based current regulator

Fig. 3 shows the DC/DC converter circuit used for battery interface.
The small-signal model transfer function of BDC can be expressed as
[17]:
i˜bat (s ) K (βs + 1)
Fig. 2. Battery-based grid-forming converter (BGFC). Gdi (s ) = = 2 di
d˜ (s ) s + a1 s + a2 (9)
Vdc 1 (1 − D)2
initial power setpoint P*, which is motivated by the analogy between where K di = , β = R dc Cdc, a1 = , a2 = .
generator torque and the converter DC current. The role of the current R dc LCdc R dc Cdc LCdc
In (9), Vdc and D represent the dc component of the DC voltage and
regulator is to regulate the current error, which determines the beha-
the dc component of the duty cycle, respectively. Rdc represents the
viour of the BGFC. equivalent line resistance. The small signals i˜bat and d̃ are the state
variables.
3. Current regulator based on active disturbance rejection control Based on the model of the converter and the ADRC theory, the
and adaptive parameter tuning ADRC-based current regulator can be designed. As ibat − d system is a
second-order system. Taking the form of (6) and n = 2, the state
The extended state observer is the core part of ADRC, which can be equation of the ibat − d system can be expressed as:
explained as follows [16]. Consider the following state-space model of a
(1)
continuous dynamical system uncertainty: ibat ′ = ibat (10a)
x ’ (t ) = Ax (t ) + BF (x (t ), t ) + bBu (t ) (6a) (1)
ibat ″ = F (ibat , ibat , d (1), d ) + bi d. (10b)
y (t ) = x1 (t ) (6b) (1)
− a2 ibat + K di βd (1) + (K di − bi ) d as the “total dis-
Regard F = −a1 ibat
where x ∈ Rn is the vector of continuous state variables, y(t) is the turbance”, the observer for (10) can be designed as:
output to be controlled, u(t) is the control variable, b is its gain,
z i1’ = zi2 + l i1 (ibat − zi1) (11a)
⎡0 1 ⋯ 0⎤ ⎡0⎤
⋮ ⋱ ⋮⎥ ⋮ z i2’ = zi3 + l i2 (ibat − zi1) + bi d (11b)
A=⎢ and B = ⎢ ⎥ .
⎢0 0 ⋯ 1⎥ ⎢0⎥
⎣0 0 ⋯ 0⎥
⎢ ⎦n × n ⎢
⎣1⎥ ⎦n × 1 z i3’ = l i3 (ibat − zi1). (11c)

Note that F (x (t ), t ) is the “total disturbance” of the system, in- According to (8), the control input is:
cluding both uncertain internal dynamics and external disturbances of
zi3 (t )
the system. The objective here is to make y(t) as desired by using u(t) as * − ibat ) + Ki2 ((ibat
d = Ki1 (ibat * )′ − ibat
′ )− .
bi (12)
the controlled variable.
Many approaches require the knowledge of F (x (t ), t ) to design the
controller. Using the extended state observer, on the other hand, the 3.2. Adaptive parameter tuning
exact knowledge of F (x (t ), t ) is not necessary. Instead, F (x (t ), t ) can be
observed from the output. As observed in (12), the ADRC-based current regulator requires the
Regarding F (x (t ), t ) as the extended state variable, the linear ex- tuning of the gains of the extended state observer li1, li2, li3, which
tended state observer is introduced as determine the convergence rate and the observation accuracy of the
controller. The authors in [18] proposed the parameter setting on the
basis of the system bandwidth, where these gains can be selected as
(7) l i1 = 3p , l i2 = 3p2 , l i3 = p3 , respectively, where p determines the band-
where Li ∈ Rn + 1 is the gain of the extended state observer in which width of the observer. The authors in [12,19] have analysed how p
i = 1…n + 1. By tuning Li, z = [z1, z2, …, z n]⊤ and z n + 1 (t ) can serve as the influences the system stability and verified the feasibility of this para-
estimation of x and F (x (t ), t ) . Then taking u(t) as: meter setting process. In this case, adjusting li1, li2, li3 is equivalent to
z n + 1 (t ) setting p.
u (t ) = K (R (t ) − x (t )) − Rewrite (11) in discrete form as follows:
b (8)
where K = [K1, …, Kn]⊤ , R = [R1, …, Rn]⊤ , Ri = R1(i − 1) ,
system (6) is trans- zi1 (k + 1) = zi1 (k )
formed to a cascade n-order integral system with a superior tracking + Ts zi2 (k ) + Ts (3p (k )(ibat (k ) − zi1 (k ))) (13a)
characteristic.
zi2 (k + 1) = zi2 (k ) + Ts zi3 (k )
+ Ts (3p2 (k )(ibat (k ) − zi1 (k )) + bi d (k )) (13b)

3
B. Wang, et al. Electric Power Systems Research 189 (2020) 106646

zi3 (k + 1) = zi3 (k ) + Ts p3 (k )(ibat (k ) − zi1 (k )) (13c)

Define:
* (k − 1)
e0 (k ) = zi1 (k ) − ibat (14)
as the tracking error. Then the adaptive tuning law for p(k) can be
designed as:
p (k ) = e ξ (k ). (15)
Parameter ξ(k) is updated in every sampling period with the following
algorithm:

Δξ (k ) = ξ (k + 1) − ξ (k )
1 e0 (k )
= M·
3Ts p (k ) − 1 zi1 (k ) − ibat (k ) (16)

Theorem 1. Let (16) be the algorithm to update ξ(k). Then the tracking
error e0(k) can converge to zero and the system stability can be achieved if
the following condition is satisfied: Fig. 4. Adaptive ADRC controller for the BGFC.
0 < M < 2. (17)
Proof. On the basis of Lyapunov theory, select the Lyapunov function overshoot and this rate is expected to be relatively large during the
as: period of small tracking error converging to zero. Hence a power
function can be adopted to determine the value of M.
Vp (k ) = e02 (k ). (18) In this paper, we use the following expression to determine M:
In this case,
⎧|e0 | < δ: e0
▵Vp (k + 1) = Vp (k + 1) − Vp (k ) M= δ 1 − α 0 < M < 2,
⎨
|e
⎩ 0 | ≥ δ : |e0 |α (27)
= 2e0 (k )Δe0 (k + 1) + Δe02 (k + 1). (19)
where 0 < α < 1 and 0 < δ < 1.
The expression of Δe0 (k + 1) can be deduced as follows:
The structure of the controller for the BGFC is illustrated in Fig. 4.
Δe0 (k + 1) ∂e (k + 1) ∂z i1 (k + 1) ∂p (k ) ∂ξ (k )
Δe0 (k + 1) = = 0 · · · .
k+1−k ∂z i1 (k + 1) ∂p (k ) ∂ξ (k ) ∂k 4. Simulation verification
(20)
According to (13): In this section, we outline the results of a system-level simulation
case of a residential microgrid supplied by prosumer-owned PV-battery
∂e0 (k + 1) * (k ))
∂ (zi1 (k + 1) − ibat 1 systems and a diesel generator, illustrated in Fig. 5.
= =1−
∂zi1 (k + 1) ∂zi1 (k + 1) 3Ts p (k ) (21)

∂zi1 (k + 1)
= −3Ts (zi1 (k ) − ibat (k )).
∂p (k ) (22)
According to (15):
∂p (k )
= p (k ).
∂ξ (k ) (23)
According to (16):
∂ξ (k ) 1 e0 (k )
= Δξ (k ) = M· · .
∂k 3Ts p (k ) − 1 zi1 (k ) − ibat (k ) (24)

Combining Eq. (20), (21), (22), (23) and (24):

Δe0 (k + 1) = −Me0 (k ), (25)
which renders:

ΔVp (k + 1) = 2e0 (k )Δe0 (k + 1) + Δe02 (k + 1)

= −M (2 − M ) e02 (k ). (26)
Based on the Lyapunov theory, e0(k) will converge to zero if
ΔVp (k + 1) < 0 . Hence we conclude that 0 < M < 2 [20].
Parameter M is defined as the self-adaptive rate of the updating
algorithm, which determines the convergence rate and accuracy of the
extended state observer. A large value of M helps achieving fast con-
vergence but with a possibly large overshoot. This value can be adjusted
according to the tracking error. When the tracking error is relatively
large, the rate of change of M is expected to be small to avoid a large Fig. 5. Residential microgrid test system.

4
B. Wang, et al. Electric Power Systems Research 189 (2020) 106646

4.1. Case study

The Matlab/Simulink microgrid model consists of the following

components:

1. PV generator. The PV-interfaced generation module consists of a

PV array, a boost converter and an inverter. The DC-DC boost con-
verter that connects the PV array to the DC bus is implemented with
maximum power point tracking; the DC/AC three-phase grid in-
verter is controlled in a grid-following mode.
2. Battery grid-forming controller (BGFC). Two identical BGFCs are
controlled by the proposed control scheme. We use two BGFCs to
study the power sharing between them. All parameters in both
BGFCs are the same except for the initial power set-point. Both
BGFCs are contributing to the inertial response and primary fre-
quency control.
3. Diesel generator. A diesel-driven three-phase synchronous gen-
erator is implemented to provide the reference grid voltage. The
primary frequency control of the generator is disabled to see how
the BGFCs help regulate the frequency. The generator, however,
provides inertial response, which helps in frequency control in the
first few seconds after a disturbance.
4. Home load. Residential home loads are connected to the microgrid Fig. 6. Power flows (PI). (a) load power; (b) diesel generator power; (c) PV
as power consumers. The load is modelled as a constant impedance. power; (d) BGFC1 power; (e) BGFC2 power.

Model parameters are provided in Table 1: the microgrid is supplied

4.2. Simulation I: PI-based current regulator
by a 400V, 50Hz, 60kW three-phase diesel generator, a PV array, and
one BGFC, providing, respectively 1kW, 9.4kW (maximum) and 5kW to
To effectively compare the proposed scheme with the PI control, the
the 15kW load. The other BGFC serves as the backup power supply, so it
parameters of the PI current controller have been tuned to ensure a
generates no power during the steady state. As load disturbance, we
satisfactory performance (Kpi = 2, Kii = 5). The droop coefficient was
increase the active power consumed by the home load.
set to m p = 5. The tuning of the PI gains is explained in the Appendix.
The simulation operation is described as follows:
Figs. 6–8 show the results of the simulation with PI control. Fig. 6
shows the power flow of each power unit. PV array generates a constant
(i) During 0-8s, the system is in the initial steady state and operating
power which reaches its maximum power point. As the sum of set-point
continuously into the steady state.
powers of all units exceeds the amount of the load power, excessive
(ii) At 8s, the load in the system increases by 0.6kW. The diesel gen-
power would flow to the batteries, which can be seen in Fig. 6 (d) and
erator, and the two BGFCs contribute to the inertial response.
(e). That is why the BGFC1 generates less than 5kW and why the power
of BGFC2 is negative. At 8s, the load changed slightly which resulted in
To control the current of the BGFC, we consider both the conven-
a power imbalance. From Fig. 6 (b), (c) and (d), it can be seen that the
tional PI control and the proposed adaptive ADRC control, using the
BGFC powers increased rapidly at 8s, reaching the peak at 8.02s, and
same system conditions in both cases.
then dropped back until 8.04s. After that, the BGFC primary frequency
control is activated, resulting in a steady increase of the power to
compensate for the frequency deviation.
By comparison, the diesel generator power increased smoothly at 8s
Table 1
residential microgrid system parameters.
Diesel generator

P*
diesel Diesel nominal power 60kW
Pdiesel Diesel output power 1kW
Vbase Grid nominal line-line voltage 400V
f Grid frequency 50Hz
Battery grid-forming converter (BGFC)
PB1 BGFC1 desired output power 5kW
PB2 BGFC2 desired output power 0
v bat
* Nominal battery output voltage 400V
C*
bat Nominal battery capacity 400A h
SOCini Initial state-of-charge 80%
v*
dc DC-link voltage reference 800V
Cdc DC-link capacitor 2000μF
Rdc Series resistance 10mΩ
L Series inductance 1.25mh
fs Switching frequency 5000Hz
PV Array
max PV maximum power 9.4kW
Ppv
Load
Pload Nominal load power 15kW
ΔPload Load power change 0.6kW
Fig. 7. DC-link voltages. (a) BGFC1; (b) BGFC2; (c) PV generation.

5
B. Wang, et al. Electric Power Systems Research 189 (2020) 106646

Fig. 8. Grid voltage (PI): (a) and (c); grid frequency: (b) and (d).

with a slower speed and reached its peak at 8.04s. Then it dropped back
slowly to 1kW. The sudden increase in power from 8s to 8.04s is the
inertial response provided by the generator and BGFCs. Observe that
the inertial response of the batteries is faster than the diesel generator,
but also shorter in duration due to the dominant inertial response of the Fig. 9. Power flows (ADRC). (a) load power; (b) diesel generator power; (c) PV
diesel. As both BGFCs have identical controllers, they provide the same power; (d) BGFC1 power; (e) BGFC2 power.
frequency support as illustrated in Fig. 6 (d) and (e).
Fig. 7 shows the difference between the grid-forming control (bat-
teries) and grid-following control (PV). Observe that the DC link voltage
of the PV array is constant during the load change, while BGFCs ex-
perienced a slight voltage drop due to the energy being released for
frequency support.
Fig. 8 shows the grid voltage and the grid frequency. It can be seen
that the frequency stays stable at about 50.014Hz and experiences a
sudden drop at the first instant of the load change. The frequency then
quickly recovers due to the combined inertial response from the diesel
generator and the BGFCs, followed by a steady transition to the post-
disturbance level of about 49.97Hz due to the action of droop control.
Observe that the RoCoF is comparable to the one experiences in con-
ventional power systems, which is analysed in Simulation II.

4.3. Simulation II: ADRC-based current regulator

The ADRC-based controller is used to form the BGFCs and compare Fig. 10. Grid voltage (ADRC): (a) and (c); grid frequency: (b) and (d).
the performance with the PI regulator. The control parameters of the
enhanced ADRC-based controllers for both BGFC1 and BGFC2 were set
V 5. Conclusion
to Ki1 = 1, Ki2 = 0.2, bi = Ldc = 6.4 × 105, α = 0.5, δ = 0.5. The other
operating conditions are the same as in Simulation I. Fig. 9 shows the
This paper argues that inertia provided by prosumer-owned bat-
power flow of the system. It is clear that during the first instants after
teries can constitute an important part of the solution to address the
the load change, the BGFC powers experienced a larger instantaneous
issue of frequency performance in future grids. This will require the use
increase (from 4860W to 5040W and -140W to 40W, respectively) than
of grid-forming control for grid-interfaced inverters, which obviates the
that in Fig. 6 (from 4860W to 5000W and -140W to 0W, respectively).
need for the PLL and consequently the time delay in the provision of
The trend implies that using the proposed ADRC-based controller, the
”synthetic inertia”, which is an issue with conventional grid-following
BGFCs’ inertial contribution is noticeably higher than using PI control;
inverters.
this is also confirmed in Fig. 10 showing the grid voltage and the fre-
Hence we propose to use a residential battery interfaced to the grid
quency change.
by a grid-forming inverter using matching control to ensure inertial
Observe that the inertial response in Simulation II is faster, resulting
response after a contingency. Simulation results verify that BGFCs
in a lower RoCoF. In more detail, we can calculate the RoCoF from
contribute to the frequency regulation by genuine inertial response.
Figs. 8 (d) and 10 (d) for each simulation, resulting in:
Unlike in conventional synchronous machines, the energy for the in-
RoCoF 1 = ▵f / ▵t = (50.014Hz − 49.996Hz)/0.02s = 0.9Hz/s and RoCoF
ertial response comes from the DC-link capacitor and the battery. To
2 = ▵f / ▵t = (50.014Hz − 49.999Hz)/0.02s = 0.75Hz/s . The RoCoF in
ensure a fast battery response, we implement an ADRC-based regulator
Simulation II is smaller, which confirms the superior performance of the
with adaptive parameter tuning. ADRC is known for the rejection of
adaptive ADRC regulator for inertial support.
“total disturbance”, which means all uncertain factors can be regarded
as a simple disturbance and rejected immediately. The adaptive

6
B. Wang, et al. Electric Power Systems Research 189 (2020) 106646

Fig. 11. Current PI Regulator.

parameter tuning algorithm helps track the satisfactory control para- Declaration of Competing Interest
meters for ADRC, which simplifies the parameter selecting process.
Simulation results confirm that the use of the proposed ADRC-based The authors declare that they have no known competing financial
regulator results in a smaller RoCoF compared to the PI control. interests or personal relationships that could have appeared to influ-
In future work we will consider a more realistic battery model, in ence the work reported in this paper.
particular focusing on the response time, which is crucial when the
battery is used to augment the capacity of the DC-link capacitor for
inertial response.

Appendix

The computation of PI gains in the PI current regulator is based on the affine parameterisation proposed in [21]. The closed-loop control structure
* is added before the control variable going into the PWM generation.
is illustrated in Fig. 11. A normalisation by vdc
The transfer function for this closed-loop system is:
1
C (s ) v * G PWM (s ) Gdi (s )
i˜bat (s ) dc
FQ (s ) = = 1
˜*
i bat (s ) 1 + C (s ) v * G PWM (s ) Gdi (s )
dc (A1)
Considering the expected dynamic response, the desired closed loop transfer function is defined as:
1
FQ (s ) =
τcl s + 1 (A2)
1
where τcl = 2πfcl
, and fcl is the cut-off frequency of the current loop. The cut-off frequency shall be no greater than one-tenth of the switching
frequency [17]. According to the parameters of the BDC shown in Table I, fcl is selected as 300Hz. Then we get:
1.6s + 8 × 10 4
Gdi (s ) = .
2.5 × 10−6s 2 + 0.0125s + 0.25 (A3)
Considering G PWM (s ) = 1 gives:
FQ (s )
1
Gdi (s ) v *
dc
C (s ) =
1 − FQ (s ) (A4)
Then the controller can be parametrised as:
4.710 9.42 × 10−5 K
C (s ) = 2.355 + − ≈ Kpi + ii
s 2 × 10−5s + 1 s (A5)
Hence the gains are selected as Kpi = 2.355 and Kii = 4.710 . In the simulation case, we selected parameter values of Kpi = 2 and Kii = 5.

References PV-battery system based on active disturbance rejection control, 2018 Australasian
Universities Power Engineering Conference (AUPEC), IEEE, 2018.
[13] T. Jouini, C. Arghir, F. Dörfler, Grid-friendly matching of synchronous machines by
[1] I.R.E.A. (IRENA), The Power to Change: Solar and Wind Cost Reduction Potential to tapping into the DC storage, IFAC-PapersOnLine 49 (22) (2016) 192–197.
2025, Technical Report, (2016). [14] Arghir, C., Jouini, T., & Dorfler, F. (2017). Grid-forming control for power converters
[2] F. Milano, F. Dörfler, G. Hug, D.J. Hill, G. Verbič, Foundations and challenges of low- based on matching of synchronous machines. arXiv:1706.09495.
inertia systems, 2018 Power Systems Computation Conference (PSCC), IEEE, 2018. [15] J. Han, From PID to active disturbance rejection control, IEEE Trans. Ind. Electron. 56 (3)
[3] AEMO, E.N. Australia, Open Energy Networks, Technical Report, (2018). (2009) 900–906, https://doi.org/10.1109/TIE.2008.2011621.
[4] N. Miller, D. Lew, R. Piwko, Technology Capabilities for Fast Frequency Response, GE [16] W. Xue, W. Bai, S. Yang, K. Song, Y. Huang, H. Xie, ADRC with adaptive extended state
Energy Consulting, Tech. Rep. (2017). observer and its application to air-fuel ratio control in gasoline engines, IEEE Trans. Ind.
[5] A. Tayyebi, Z. Miletic, F. Dörfler, F. Kupzog, W. Hribernik, Gridforming con- Electron. 62 (9) (2015) 5847–5857, https://doi.org/10.1109/TIE.2015.2435004.
verters–inevitability, control strategies and challenges in future grid applications, [17] Y. Zhang, Y.W. Li, Energy management strategy for supercapacitor virtual impedance,
International Conference on Electricity Distribution (CIRED), (2018). IEEE Trans. Power Electron. 32 (4) (2017) 2704–2716.
[6] C.K. Sao, P.W. Lehn, Control and power management of converter fed microgrids, IEEE [18] Z. Gao, Scaling and bandwidth-parameterization based controller tuning, Proceedings of
Trans. Power Syst. 23 (3) (2008) 1088–1098. the 2003 American Control Conference, 2003, 6 (2003), pp. 4989–4996, https://doi.org/
[7] A. Mehrizi-Sani, R. Iravani, Potential-function based control of a microgrid in islanded 10.1109/ACC.2003.1242516.
and grid-connected modes, IEEE Trans. Power Syst. 25 (4) (2010) 1883–1891. [19] G. Wang, B. Wang, C. Li, D. Xu, Weight-transducerless control strategy based on active
[8] R. Georgious, A. Navarro-Rodríguez, J. García, P. García, S. Saeed, et al., Observer-based disturbance rejection theory for gearless elevator drives, IET Electr. Power Appl. 11 (2)
transient frequency drift compensation in AC microgrids, IEEE Trans Smart Grid 10 (2) (2017) 289–299, https://doi.org/10.1049/iet-epa.2016.0474.
(2019) 2015–2025, https://doi.org/10.1109/TSG.2017.2786085. [20] J.-s. Xi, P.-k. Liu, H. Ding, Design of an adaptive linear active disturbance rejection
[9] I.E.A. (IEA), Digitalization & Energy, Technical Report, (2017). controller, Opt. Precis. Eng. 26 (7) (2018) 1749–1757.
[10] P. Graham, L. Havas, T. Brinsmead, L. Reedman, Projections for Small Scale Embedded [21] c. Xiao, H. Wen, H.H. Zeineldin, Affine parameterization and anti-windup approaches for
Energy Technologies - Report to AEMO, Technical Report, (2019). controlling DC-DC converters, 2012 IEEE International Symposium on Industrial
[11] C.E. Council, Clean Energy Australia Report 2018, Technical Report, (2018). Electronics, (2012), pp. 154–159.
[12] B. Wang, G. Verbič, A.C. Chapman, Control scheme for bidirectional DC/DC converter in