Electrical Power and Energy Systems 115 (2020) 105463

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: [Link]/locate/ijepes

Control of a population of battery energy storage systems for frequency T

response
⁎,1
Zeyad Assi Obaida, , L.M. Cipciganb, Mazin T. Muhssinc, Saif Sabah Samid
a
College of Engineering, University of Diyala, Baqouba, Diyala, Iraq
b
Institute of Energy, School of Engineering, Cardiff University, Cardiff, Wales, United Kingdom
c
Faculty of Engineering, Mustansiriyah University, Baghdad, Iraq
d
Head of Statistics Department, Mayoralty of Baghdad, Baghdad, Iraq

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

Keywords: The control of multiple battery energy storage systems (BESSs) to provide frequency response will be a challenge
Battery energy storage system in future smart grids. This paper proposes a hierarchical control of BESSs with two decision layers: the ag-
Demand side response gregator layer and the BESS control layer. The aggregator layer receives the states of charge (SoC) of BESSs and
Frequency regulation sends a command signal to enable/disable the BESS control layer. The BESS controller was developed to enable
Large-scale BESS aggregation
the BESSs to respond from the highest to lowest SoC when the frequency drops, and from lowest to highest when
Power system frequency control
it rises. Hence, the BESS’s response is prioritised to reduce the impact on the power system and end-users during
the service. The BESS controller works independently when a failure occurs in the communication with the
aggregator. The dynamic behaviour of the population of the controllable BESSs was modelled based on a Markov
chain. The model demonstrates the value of aggregation of BESSs for providing frequency response and evaluates
the effective capacity of the service. The model was demonstrated on the 14-machine South-East Australian
power system with a 14.5 GW load. 254 MW of responsive capacity of aggregated batteries was effective in
reducing the system frequency deviation below 0.2 Hz following a sequence of disturbances.

1. Introduction batteries. The dynamic representation of BESS’s SoC was represented by

state transitions, from zero to full charge and vice versa [16]. Various
With the increased integration of renewable energy generation, types of batteries and their applications were presented such as behind
system frequency control using only conventional generators becomes the meter BESSs (home-based) [14], smart charging of EVs [17], and
more expensive [1–4]. The aggregation of demand devices to regulate large-scale BESSs (grid-scale BESS) [18]. The aggregation of these
the frequency mitigates the need to increase the capacity of conven- BESSs was shown to be important in regulating power system frequency
tional power generators [1–3,5]. In Great Britain’s (GB) power system, [14,17,18].
demand side response from commercial and industrial consumers is It is anticipated that by 2025/26 in the GB power system, the
considered for frequency control [4–8]. The term ‘frequency response’ fluctuation of the power from wind and solar will lead to a sharp ramp
in the UK refers to all dynamic and non-dynamic frequency services. in the system demand during the day. In addition, the combination of
Energy Storage systems are important elements of future smart grids high wind and solar output along with low demand means that a sig-
[9–11]. BESSs have been evaluated and considered in the literature for nificant number of interventions by the system operator would need to
frequency regulation [11–13]. be taken for system balancing and operability. Therefore, there are
A Markov chain has been used to represent the batteries SoC for opportunities for demand-side services during periods of low and high
electric vehicle (EVs) batteries [15] or PV batteries [16]. The modelling demand [19].
of the batteries SoC for availability of power from PV systems was BESSs are showing improvements in their technologies as well as
presented in [16]. The model was used to improve the availability of cost reduction and it is estimated that storage in the GB power system
photovoltaic generation and to represent the charge/discharge of the by 2050 will be about 10.7GW based on ‘Consumer Power Scenario’ by

Abbreviations: BESS, battery energy storage system; SoC, state of charge; DLC, direct load control; EVs, electric vehicles; DOD, depth of discharge; ΔF, frequency
deviation; MSF, membership function; GB, great britain
⁎
Corresponding author at: Senior Lecturer at department of Electrical Power and Machines, College of Engineering, University of Diyala, Iraq.
1
HCED (Higher Committee for Educational Development in Iraq) to study PhD at Institute of Energy, Cardiff University, UK.

[Link]
Received 19 March 2019; Received in revised form 7 July 2019; Accepted 1 August 2019
0142-0615/ © 2019 Elsevier Ltd. All rights reserved.
Z.A. Obaid, et al. Electrical Power and Energy Systems 115 (2020) 105463

the UK system operator [20]. A large number of these batteries will be

in distribution systems connected behind the meter [20]. Thus, con-
trolling a large number of distributed loads such as BESSs will be a
challenge [21]. BESSs offer a fast dynamic response to compensate load
variations. Controlling a large amount of distributed load such as BESSs
has been investigated using centralized and decentralized control
methods [22–24]. A centralized control method reduces the uncertainty
in the response of controllable units. However, the centralized method FGrid
has a challenge related to the communications such as the cost and
latency [24]. In contrast, the decentralized control method removes this
challenge, but it introduces uncertainty due to the independent re-
sponse of these large distributed loads [22–24].
Many previous works have considered the control of BESSs in either
centralised or decentralised control methods [12,14,17,18]. However, Fig. 1. Block diagram of the proposed hierarchical control of BESSs.
and to the author knowledge, none of them has considered developing a
control method that compromises the advantages of both centralised throughout the power system. A demand aggregator, which is a third-
and decentralised control methods. Therefore, a developed control of party company, aggregates these BESSs to offer frequency response
large distributed residential and non-residential BESSs is considered in services when required. This is done by integrating a controller into
this paper to provide frequency response services in future power sys- each BESS to control its charging/discharging processes. The demand
tems. The hierarchical control proposed in this paper is a tradeoff be- aggregator has a central controller, which is represented by the ag-
tween the centrlised and decentrlised control methods of large dis- gregator layer (see Fig. 1). The aggregator layer collects the SoC of
tributed loads. As a result, this reduces the uncertainty associated with BESSs and sends command signals to enable/disable each BESS con-
the response of the population during a frequency event and reduces troller in the BESS control layer.
the challenge associated with the cost and latency of the communica- The battery degradation can be affected by two main factors, the
tions. number of cycles and the depth of discharge (DOD) [25]. The DOD is
The application of BESSs in direct load control (DLC) is proposed in the level of SoC in the battery designed by the manufacturer. When the
[14]. The combination of electrical load, its levels in the building, and battery goes below its DOD, this will reduce the number of cycles of the
their controllable devices were considered to investigate the DLC ap- battery. Therefore, the risk of reducing the life of the battery will be
plication. The problem of controlling many distributed small-scale increased [25,26]. Hence, considering the level of the SoC for the ag-
BESSs in a building was highlighted. The scheme presented in [14] gregated responsive BESSs is important as there may be hundreds of
reduced the frequency deviation by controlling batteries that were in- cycles each year when providing frequency response service.
stalled behind the meters [14]. However, peritonising the BESSs ac- Therefore, the BESS controller has pre-set frequency bands as shown
cording to their level of SoC and the value of frequency deviation was in Fig. 2. The response depends on the frequency deviation and the
not considered in [14]. Therefore, the proposed BESSs’ controller BESS SoC. The, BESSs will respond starting from the highest level of
prioritises the response of the BESS based on the level of SoC. Hence, it SoC to the lowest level of SoC when the frequency drops below a
reduces the risk of battery degradation, and the impact on the power nominal value. When a frequency rises above a nominal value, BESSs
system and the end users will be reduced. will respond staring from the lowest SoC to the highest SoC. As a result,
Coordination methods were presented in [12] for controlling the risk of a simultaneous power change of a large number of BESSs
neighbouring batteries to regulate frequency and voltage based on local during low-frequency is reduced. In addition, prioritising the BESSs SoC
communication system on building level. However, the work presented reduces battery degradation.
in [12] had not consider the large problem of communication failures Collecting the SoC values of all BESSs allows the aggregator to de-
within a wide area power system. Therefore, The BESSs controller cide the available response capacity from the population of BESSs. In
proposed in this paper responds, even when a failure occurs in the addition, in the case of discharging the BESSs and injecting power back
communication with the aggregator control layer. In addition, the to the grid, the proposed hierarchal control allows the aggregator to
speed of the BESS response is not affected by the continuously updated decide the response time of the BESSs according to their SoC levels. For
profile set by the central controller of the aggregator. example, the response from the BESSs with the lowest SoC can be used
For the work presented in the literature, and for the author for the provision of primary response period (i.e. up to 10 s) while the
knowledge, none of them presented a control that is able to provide
different frequency response services at the same time. The proposed
BESS controller enables the population to participate in low and high
frequency response services. In addition, the batteries’ model presented
previously was not presented to evaluate the effective population ca-
pacity. Therefore, the proposed model of a developed Markov chain
evaluates the effective responsive capacity of the aggregated BESSs
during the frequency event by considering different aggregation case
studies in a multi-machine power system.
In summary, the objectives in this paper are: (i) to develop a hier-
archal control of a population of BESSs; (ii) to develop a BESS controller
for frequency response services; (iii) to model the population of bat-
teries and their control to demonstrate the potential for a batteries’
aggregator to offer frequency response services, and (iv) to evaluate the
capacity of the aggregated batteries during the frequency event.

2. Hierarchical control of BESSs

It is assumed that BESSs with different capacities are distributed Fig. 2. The desired frequency response of the BESSs during a frequency event.

2
Z.A. Obaid, et al. Electrical Power and Energy Systems 115 (2020) 105463

Fig. 5. Six logic frequency bands in the BESS controller.

levels represent six bands of logic indicators as shown in Fig. 5. The

Fig. 3. Block diagram of the BESS controller. system frequency deviations are located within one of these bands using
Eqs. (6)–(11), where ΔF = FGrid − 50. The number of these bands is set
according to the aggregators’ preferences and the preferred degree of
BESSs with higher level of SoC can be used for longer service period
the frequency response smoothness. Hence, the higher the number of
such as secondary frequency response (i.e. up to 30 min). Therefore, the
bands the smoother is the response of a population of BESSs. Therefore,
uncertainty in the response of the aggregated BESSs will be reduced
the proposed BESS controller enables the population to participate in
during the frequency service period.
both low and high frequency response services.
Fig. 3 displays the BESS controller, which has three main compo-
nents: (i) Measurement of SoC levels, (ii) Measurements of frequency
deviation and (iii) The logic gates to control the BESS charging and
ΔFL1 = {1 − a ≥ ΔF ≥ −a1
0 else (6)
discharging according to (i), (ii), and the aggregator enable/disable
command signal. The command signals of the aggregator control layer
are either logic 1 to enable the BESS controller or logic 0 to disable it.
ΔFL2 = {1 − a1 > ΔF ≥ −a2
0 else (7)
The design of each component is explained as follows:

2.1. Measurements of SoC levels

ΔFL3 = {1 − a2 > ΔF
0 else (8)

The SoC of a BESS lies within one of the following levels: 0%, 25%,
ΔFH1 = {1 a1 ≥ ΔF ≥ a
0 else (9)
50%, 75%, and 100% SoC and Fig. 4 displays the logic outputs of the
indicators of these levels which are C1, C2, C3, C4, and C5. Eqs. (1)–(5)
were used to categorise all BESS into one of these indicators. These
ΔFH2 = { 1 a2 ≥ ΔF > a1
0 else (10)
indicators are the input to the logic gates of BESS charging/discharging
control. Considering these different levels reduces the risk of the battery
degradation and the risk of a simultaneous power change at the same
ΔFH3 = {1 ΔF > a2
0 else (11)
time.
2.3. The logic circuits in the BESS controller
C1 = {10 else
SoC = 0
(1)
The SoC measurements and frequency deviations measurements are
C2 = {
1 25% ≥ SoC > 0 used as inputs to the logic gates, the output of the logic gates and logic
0 else (2) control output is either enable the charging /discharging of the battery
when logic 1 or disable it when logic 0. The logic gates and the control
C3 = {
1 50% ≥ SoC > 25%
output’s switches are controlled by the command signal received from
0 else (3)
the aggregator control layer. Therefore, when the command signal is
logic 1, the BESS will provide a frequency response by charging/dis-
C4 = {
1 75% ≥ SoC > 50%
0 else charging the BESS as shown in the logic truth table in Tables 1 and 2.
(4)
As a result, the BESS controller provides a response based on the last
command received from the aggregator control layer. Hence, the con-
C5 = {
1 100% ≥ SoC > 75%
0 else (5) troller works independently when any failure occurs in the commu-
nication with the aggregator control layer. Considering different levels
of SoC and frequency reduces the impact on the BESS and the power
2.2. Measurements of frequency deviation levels system.
In addition, the speed of the BESS response is not affected by the
There are three levels of positive frequency deviations (i.e. ΔFH1,2,3) continuously updated profile sent by the central controller of the
and three levels for negative frequency deviations (i.e. ΔFL1,2,3). These
Table 1
Truth table for the control output of the charging (NC = No Change).
Output C10% SoC C2 25% SoC C350% SoC C4 75% SoC C5 100% SoC

ΔFL1 NC NC NC NC 0
ΔFL2 NC NC NC 0 0
ΔFL3 0 0 0 0 0
ΔFH1 1 1 NC NC NC
ΔFH2 1 1 1 NC NC
ΔFH3 1 1 1 1 1
Fig. 4. Logic bands for measurements of SoC level.

3
Z.A. Obaid, et al. Electrical Power and Energy Systems 115 (2020) 105463

Table 2
Truth table for the control output of the discharging (NC = No Change).
Output C10% SoC C2 25% SoC C3 50% SoC C4 75% SoC C5 100% SoC

ΔFL1 NC NC NC NC 1
ΔFL2 NC NC NC 1 1
ΔFL3 1 1 1 1 1
ΔFH1 0 0 NC NC NC
ΔFH2 0 0 0 NC NC
ΔFH3 0 0 0 0 0

aggregator. The BESS provide a fast response based on its SoC level.
Fig. 7. State Transition Diagram is representing the dynamic behaviour of the
When the controller is enabled by logic 1 signal, it considers two fac- BESSs population according to the levels of SoC (adopted from [16]).
tors, which are the SoC level and the value of frequency deviation.
Therefore, the controller was designed with the assumption that there is
no cycle limit for the controller response outside these factors. battery SoC for electric vehicles batteries [15] or for PV charging-based
batteries [15,16]. Hence, a Markov-based model was developed to re-
present these two steps of the dynamic behaviour.
3. Modelling a population of controllable BESSs
(A) Modelling the dynamic behaviour of the BESSs population just before a
A model of the dynamic behaviour of BESSs was used to quantify the frequency event
effective response capacity during the provision of the frequency ser-
vice. This model considers the control concept of large distributed A Markov-based state diagram was used to represent the dynamic
BESSs proposed in this paper. The aggregator layer collects the states of behaviour of BESSs states as shown in Fig. 7, this Figure was drawn
BESSs according to five different states 0%, 25%, 50%, 75%, and 100% based on [16]. Each state represents a set of BESSs in the population
SoC. Therefore, modelling the dynamic behaviour of the population is according to their SoC level. The dynamic transition from the left (un-
divided into two steps. (A) Modelling the population of BESSs at the charged) to the right (fully-charged) was represented by the ‘charging’
moment just before the frequency event based on their nominal power transition probabilities P1, P2, and vice versa is by P−1, P−2. The
(assumed initial condition of BESSs according to the SoC states). (B) probabilities of the states to remain at zero SoC and full SoC are re-
Modelling the dynamic switching of charging/discharging operation of presented by K11 and KNN.
controllable BESSs during a frequency event, see Fig. 6. The state diagram of Fig. 8 was represented by a 5x5 state transition
In (B), the probability of the aggregated power deviation from matrix as presented in Eq. (12). Each state in Fig. 8 has an initial
the population’s nominal power (of (A)) during a frequency event is condition at the moment just before a frequency event. The matrix in
calculated. For example, if the population of BESSs is procured to Eq. (15) represents the initial conditions of the five states, where i (1)
provide the secondary frequency response service to the GB power represents the BESSs population with 0% SoC, and i (5) represents the
system, its response could be sustained up to 30 min. Therefore, it is BESSs population with 100% SoC. This initial condition is assumed to
necessary to represent the dynamic behaviour of the aggregated power represent the capacity of each state of BESSs at the moment just before
deviation of the BESS population during any service period. Markov- the frequency event.
chain was previously used to represent dynamic behaviour of the

Fig. 8. The Membership Functions for the switching of the BESSs population,
Fig. 6. Steps of modelling the dynamic behaviour of BESSs based on the pro- (a) switching probabilities into ‘charging’ state, (b) switching probabilities into
posed control scheme. ‘discharging’ state.

4
Z.A. Obaid, et al. Electrical Power and Energy Systems 115 (2020) 105463

⎡ K11 P1 P2 0 0 ⎤ ⎡ C1 ⎤
⎢ K11 0 P1 P2 0 ⎥ ⎢C 2 ⎥
P (tn) = ⎢ P−2 P−1 0 P1 P2 ⎥ C = ⎢ C3 ⎥
⎢ ⎥ ⎢C 4 ⎥
⎢0 P−2 P−1 0 KNN ⎥ ⎢ C5 ⎥
⎢ KNN ⎥ ⎣ ⎦5 * 1 (20)
⎣0 0 P−2 P−1 ⎦5 (12)

where Yi (tn) = [C1xP (1) + C 2xP (2) + C 3xP (3) + C 4xP (4) + C 5xP (5)] (21)

The dynamic aggregation is implemented based on the proposed

K11 = 1 − ∑ Pi
control scheme using matrix ‘C’. This is done by assigning the value of
i ∈ {1, M } (13)
the matrix’s parameters (C1, C2 ….) to either 0 or 1 according to the
value of Δ F as explained in Section 2. For example, to stop charging of
KNN = 1 − ∑ P−i BESSs with 75% SoC and 100% SoC during the second frequency band,
i ∈ {1, M } (14)
(between ‘a1’ and ‘a2’), parameters C1, C2, and C3 are set to 0, while C4
and C5 are set to 1. The power deviation of the BESSs population is
Pi (0) = [i (1) i (2) i (3) i (4) i (5)]1 (15) calculated using Eq. (22), where the responsive capacity of the BESSs in
(21) is subtracted from the total capacity of the aggregated BESSs
(B) Modelling the dynamic switching behaviour of BESSs during a frequency (Ptot ).
event
Y (tn)1 = Ptot − Yi (tn) (22)
When a frequency deviation occurs (for example, frequency drops), For example, when ΔF = −0.08 Hz where ΔF = FGrid −50, at this
the population of BESSs should start responding immediately and the moment, the grid frequency is less than the nominal frequency and the
BESS controller prioritise the population response based on the value of Ì¿

ΔF and the SoC level. Therefore, the charging is stopped, starting from controlled demand must be responded. TheP (i) = N andP (−i) = N ac-
Ì¿
the BESSs with the highest to the lowest SoC. cording to the Fig. 9 and the response is under the control of NandN .
The transition probabilities (P1, P2), and (P-1, P-2) shown in Fig. 7 The control is also linked with the battery SoC according to Fig. 4, the
have a value from 0 to 1 according to the basic concept of Markov chain controlled demand will be responded for all batteries above 50% SoC.
[16]. Hence, to represent stop charging of the BESSs, the switching Eqs. (18) and (19) will dynamically updated to calculate the total re-
probability (P1, P2) will be set towards 0 and vice versa for (P-1, P-2). In sponded capacity.
addition, not all BESSs respond in the same time, the behaviour is a The flowchart shown in Fig. 9 represents the complete modelling of
gradual transition and is dynamically linked with the level of frequency the proposed control of BESSs for the negative ΔF bands (when the
drop. The transition starts from 0 when no BESSs response and ends at 1 frequency drops). When the frequency goes below a nominal value
when all BESSs should respond. ‘−a1’, the prosed dynamic model of BESSs population is activated. The
Fuzzy membership functions (MSFs) has a gradual transition from 0 proposed control which associated with the dynamic model selects the
to 1[4,27–30]. Therefore, MSFs were used to model this gradual tran- response of controllable batteries according to the proposed criterion,
sition behaviour of the (P1, P2) and (P-1, P-2) as shown in Fig. 8. which is based on the value of ΔF and the level of SoC. Finally, the
The population response starts with ‘a’ value which represents the dynamic aggregation will be updated to calculated the responded ca-
beginning of the first frequency band and ends at ‘a2’ value, which pacity. The complete flowchart for the positive ΔF is the same as in
represents the beginning of the last frequency band, where all BESSs
should respond. The MSF dynamically updates the value of the Start
switching probabilities (P1, P2) and (P-1, P-2) of the population. Th
Calculations of Population
updated value of P1, P2) and (P-1, P-2) is according to the switching Initial conditions
rules; Pi (for P1, P2) and P−i (for P-1, P-2) as shown in Eqs. (16) and (17).

P if ΔF ≥ a No
P (i)=⎧ ¨)” a +] End:
⎨ ¯
⎩ N if ΔF ≤ −a (16) No Change
Yes
P¯ if ΔF ≥ a Run dynamic Representation
P (−i)=⎧ of the BESSs Population
⎨
⎩ N if ΔF ≤ −a (17)
Updating the BESSs representation
The updated values of (P1, P2) and (P-1, P-2) are used to dynamically according to the value of ¨)
Representing the BESS Controller
re-update the state transition matrix of Eq. (12). Therefore, the initial
condition of the five states, which represented the capacity of the BESSs a•¨)•a +]
No
a!¨)•a2 +]
No
a2 ! ¨)
No
states before the frequency deviation, are then dynamically updated
during the frequency drop/rise period. This is done in a second-by- Yes Yes Yes
second timeframe using Eq. (18). Where, P (tn) is the new state tran-
C2=C3=C4=0; C2=C3=0;
sition matrix with the updated values of (P1, P2) and (P-1, P-2). C2=C3=C4=C=
C= C4=C=

P (tn + 1)1 = Pi (0) P (tn) (18)

The dynamic aggregation of the total responsive capacity of the Dynamic aggregation of the
BESSs population capacity
population of BESSs is done by using Eq. (19). Combining Eqs. (19) and
(20) results in Eq. (21), where P(1) to P(5) represent the dynamic up-
dated initial condition (capacity) of each state after a frequency drop/ Reserve service
time occurred?
End
rise. No Yes

Yi (tn) = [P (tn + 1)] C (19) Fig. 9. Flow chart of the modelling of the proposed control of BESSs.

5
Z.A. Obaid, et al. Electrical Power and Energy Systems 115 (2020) 105463

Table 3
Aggregators and their BESSs population assumption types.
Non-residential BESSs Residential BESSs Total
(MW) (MW) (MW)
Aggregator 3
1-Bus 206 (area 2) 30 50 80
2-Bus 312 (area 3) 42 50 92
3-Bus 408 (area 4) 50 60 110

Table 4
Study cases for the simulation results with the South East Australian power
system.

Aggregator 1 Study cases

Aggregator at: A1 (MW) A2 (MW) A3 (MW)

1-Bus 206 (area 2) 80 80 80

2-Bus 312 (area 3) 0 0 110
3-Bus 408 (area 4) 0 92 92

Total (MW) 80 172 282

Aggregator 2
when 123 MW of wind generation were lost followed by another loss of
192 MW wind generation after 6 s. This loss of approximately 311 MW
generation led to 560 MW interconnector tripping. This event of gen-
eration and interconnectors loss sequence was modelled and applied to
the test system at t = 5 s, t = 11 s, and t = 13 s, and the simulation
results were captured. This disturbance was simulated as a sudden in-
crease in the load at busbar 405 near generator GPS_4 in area 4 (see
Fig. 10), the impact of the location of the disturbance was out of the
scope of this paper.
Three case studies are considered to represent three realistic pos-
sibilities to integrate different aggregators with different capacities (see
Fig. 10. IEEE 14-generator 59-bus, 5-areas, The South East Australian Power Table 4). Also, the initial condition of BESSs according to the level of
System [32,33]. SoC is assumed as shown in Table 5. In addition, the frequency bands
parameters, a, a1, and a2 also have an impact on the population re-
Fig. 9, except the value of ΔF for the value of a, a1, and a2 will be for sponse. Therefore, three different values were considered for the si-
the right side of Fig. 5. mulation comparison as shown in Table 6.

4.1. Scenario 1: Different cases of the aggregation capacity

4. Demonstration of the proposed control of BESSs.
In this section, the aggregators’ cases shown in Table 4 and value 1
The modelling and simulation results of the controllable BESSs are in Table 6 are considered. Increasing the number of aggregators and the
carried out using MATLAB® and MATLAB®/SimPowerSystems™. amount of controllable BESSs leads to a significant reduction in the
Comparing to previous work which was done on a small representation frequency deviation and frequency error (see Fig. 11). In case of A3,
of power system, the 14-machines South-East Australian power system there are three aggregators with 282 MW of controllable BESSs and
was used to evaluate the proposed control scheme (see Fig. 10). This initial SoC levels as shown in Table 5, which reduces the highest fre-
large dynamic benchmark power system with load case 4 [31,32] was quency deviation from 0.23 Hz to 0.15 Hz for the biggest disturbance
used as the base case of the system load, which is approximately sequence at t = 13 s.
14.5GW. this an opportunity to increase the realistic of the response Other bands comparison is shown in Table 7 with the responsive
behaviour than work presented in the literature. This system contains a capacity of each aggregation case study. Case A1 and Case A2 provided
primary droop-based frequency response. No other frequency control less impact on the response, which is reasonable due to their capacity
loop is considered in this work. against the size of the disturbance.
The results are stored as vectors to visualise and compare the re-
sults. This model was used for testing new control techniques in a
4.2. Scenario 2: Different value of frequency bands
power system; further details were presented in [31,32]. The control of
a population of BESSs was implemented using a dynamic controllable
In this section, different values of ‘a, a1, a2’ (Table 6) and case A3 in
load. Each controllable load represents an Aggregator for different size
Table 4 were considered, these values are set by an aggregator and can
of BESSs. Three different Aggregators in different areas are considered
be updated if necessary. Through these values, the BESS response can
as shown in Table 3 and Fig. 10.
The proposed control is applicable for any battery size in a re-
Table 5
sedintional or non-residentional uses. Therefore, the requirement of the
SOC initial conditions study cases.
battery size was left to the aggregator and the market preferences and
was not covered in this paper. 0% SoC 25% SoC 50% SoC 75% SoC 100% SoC (C5)
(C1) (C2) (C3) (C4)
A large frequency disturbance was considered which recently took
place on 28th of September 2016 in the South Australian power system Pi (0) 0.1 0.1 0.1 0.3 0.4
to evaluate the proposed population control. The disturbance started

6
Z.A. Obaid, et al. Electrical Power and Energy Systems 115 (2020) 105463

Table 6 Table 7
Different values of frequency bands parameters. Comparison of the results shown in Fig. 12 using SOC initials of Table 5 ac-
cording to the control scheme proposed in Fig. 2.
a a1 a2
at Frequency limit Total responsive capacity of Max
Value 1 (Hz) 0.015 0.05 0.1 BESSs value of
Value 2 (Hz) 0.015 0.03 0.05 Δ F (Hz)
Value 3 (Hz) 0.015 0.02 0.04
Band 1 (Green) Below 49.985 A1 0.4 * 80 = 32 MW 0.06
(Δ F ≤ −0.015) A2 0.4 * 172 = 68.8 MW 0.06
be controlled according to frequency bands and SoC levels. These va- A3 0.4 * 282 = 112.8 MW 0.03

lues have an impact on the frequency response of the population of Band 2 (Orang) Below 49.95 A1 0.7 * 80 = 56 MW 0.12
BESSs. The reduction of the values of the bands improved the frequency (Δ F ≤ −0.05) A2 0.7 * 172 = 120.4 MW 0.12
A3 0.7 * 282 = 197.4 MW 0.09
response and vice versa (see Fig. 12). The impact of the reduction is in
relation to the aggregation capacity and the service type (i.e. frequency Band 3 (Red) Below 49.9 A1 0.9 * 80 = 72 MW 0.22
(Δ F ≤ −0.1) A2 0.9 * 172 = 154.8 MW 0.20
regulation, primary or secondary… etc.).
A3 0.9 * 282 = 253.8 MW 0.15
There was an oscillatory behaviour in the response in Fig. 11 and
Fig. 12. The largest responsive capacity was about 250 MW in A3 si-
mulation cases, while the total disturbance was more than 800 MW.
Therefore, this oscillatory in this situation is a normal behaviour due to
this huge disturbance comparing to real cases occurred in the literature.
The simulation results in this paper proposed more realistice frequency
response behaviour than work presented in the literature such the work
presented in [12,14,17,18,12,24].

5. Conclusions

A hierarchical control was proposed to aggregate different size of

BESSs to provide frequency response services. The BESS controller can
respond to either negative or positive frequency deviations. Hence, can
participate in either high or low-frequency response services. The BESS
controller also enables BESS to work independently when any failure
occurs in the communications with the aggregator.
A model of a population of BESSs was developed for the proposed
hierarchical control to demonstrate the potential to provide frequency
response service and to evaluate the effective capacity during a fre- Fig. 12. Frequency response at power station of busbar 404 in Scenario 2 using
quency event. The model divides the population of BESSs into five different values of frequency bands in Table 6.
states based on their SoC, and dynamically control the BESSs according
to their SoC levels during the provision of frequency response services. Considering the battery SoC to reduce the risk of battery degrada-
The control scheme was demonstrated using various case studies on tion and not to affect the user comfort, and eliminating the cost of real-
the 14-machines South-East Australian power system. Comparing to time communications, enables the proposed control of BESSs to be
previous work, a large dynamic benchmark system with a system load applied in (i) Residential and non-residential BESSs, (ii) Large-scale
base equal to 14.5GW was used in this work to demonstrate the pro- BESSs, and (iii) Vehicle-to-Grid (V2G), and (iv) Virtual Power Plant
posed design. A 253.8 MW of responsive capacity of controllable BESSs (VPP). In addition, the proposed design is applicable to provide high
was effective in reducing frequency deviations following a large simu- frequency response services, hence, it is recommended for future work
lated real disturbance sequence.

50.05

50
Band 1 Responsive BESSs with SoC >75%
49.95
Band 2
Frequency (Hz)

Responsive BESSs
with SoC >50%
49.9
Band 3
Responsive BESSs with SoC >0%
49.85

49.8

49.75
0 5 10 15 20 25 30
Time (s)
Fig. 11. Frequency response at Power Station of busbar 404 in Scenario 1 with the aggregators’ cases in Table 4.

7
Z.A. Obaid, et al. Electrical Power and Energy Systems 115 (2020) 105463

to demonstrate the effective responsive capacity for this service. 2016. p. 269–273.
[14] Kim Y-J, Del-Rosario-Calaf G, Norford LK. Analysis and Experimental
Implementation of Grid Frequency Regulation Using Behind-the-Meter Batteries
Declaration of Competing Interest Compensating for Fast Load Demand Variations. IEEE Trans Power Syst
2017;32(1):484–98.
The authors declared that there is no conflict of interest. [15] Iversen EB, Morales JM, Madsen H. Optimal charging of an electric vehicle using a
Markov decision process. Appl Energy 2014;123:1–12. Jun 15.
[16] Song J, Krishnamurthy V, Kwasinski A, Sharma R. Development of a Markov-chain-
Appendix A. Supplementary material based energy storage model for power supply availability assessment of photo-
voltaic generation plants. IEEE Trans Sustain Energy Apr 2013;4(2):491–500.
[17] Tang Y, Zhong J, Bollen M. Aggregated optimal charging and vehicle-to-grid control
Supplementary data to this article can be found online at https:// for electric vehicles under large electric vehicle population. IET Gener, Trans Distrib
[Link]/10.1016/[Link].2019.105463. 2016;10(8):2012–8.
[18] Alhejaj SM, Gonzalez-Longatt FM. Impact of inertia emulation control of grid-scale
BESS on power system frequency response. In: International Conference for
References
Students on Applied Engineering (ICSAE); 20-21 Oct. 2016. pp. 254–258.
[19] Grid N. System Operability Framework 2016, National Grid; 2016.
[1] Marcelo KK, Elizondo A, Calderon Christian Moya, Zhang Wei. Frequency re- [20] Grid N. Future Energy Scenario; 2017.
sponsive demand in U.S. western power system model. In: IEEE Power and Energy [21] Chapman AC, Verbic G. Dynamic distributed energy resource allocation for load-
Society General Meeting; 26–30 July 2015, pp. 1–5. side emergency reserve provision. In: 2016 IEEE Innovative Smart Grid
[2] Kalsi JHK, Fuller J, Marinovici LD, Elizondo M, Williams T, Lian J, et al. Loads as a Technologies - Asia (ISGT-Asia); Nov. 28 2016-Dec. 1 2016. pp. 1189–1194.
resource frequency responsive demand. In: Prepared for the U.S. Department of [22] Lu N. An evaluation of the HVAC load potential for providing load balancing ser-
Energy under Contract DE-AC05-76RL01830, Pacific Northwest National vice. IEEE Trans Smart Grid 2012;3(3):1263–70.
Laboratory Richland, Washington 99352, December 2015. [23] Short JA, Infield DG, Freris LL. Stabilization of grid frequency through dynamic
[3] Kalsi JLK, Marinovici LD, Elizondo M, Zhang W, Moya C. Loads as a Resource demand control. IEEE Trans Power Syst 2007;22(3):1284–93.
Frequency Responsive Demand. IN: Prepared for the U.S. Department of Energy [24] Sami SS. Virtual energy storage for frequency and voltage control. Institute of
under Contract DE-AC05-76RL01830, Pacific Northwest National Laboratory Energy, Cardiff University; 2017.
Richland, Washington 99352, September 2014. [25] Xu B, Oudalov A, Ulbig A, Andersson G, Kirschen DS. Modeling of lithium-ion
[4] Muhssin MT, Cipcigan LM, Jenkins N, Meng C, Obaid ZA. Modelling of a population battery degradation for cell life assessment. IEEE Trans Smart Grid
of Heat Pumps as a Source of load in the Great Britain power system. International 2018;9(2):1131–40.
Conference on Smart Systems and Technologies (SST). 2016. p. 109–13. [26] Lian B, Sims A, Yu D, Wang C, Dunn RW. Optimizing LiFePO4 battery energy
[5] Obaid ZA, Cipcigan LM, Muhssin MT. Design of a hybrid fuzzy/markov chain-based storage systems for frequency response in the UK system. IEEE Trans Sustain Energy
hierarchal demand-side frequency control. In: IEEE PES GM, 2017, Chicago. USA; 2017;8(1):385–94.
2017. p. 1–5. [27] Mazin LMC, Muhssin T, Obaid Zeyad A, Wissam F, AL-Ansari. A novel adaptive
[6] Trovato V, Teng F, Strbac G. Value of thermostatic loads in future low-carbon Great deadbeat- based control for load frequency control of low inertia system in inter-
Britain system. pp. 1–7. connected zones north and south of Scotland. Int J Electr Power Energy Syst
[7] Trovato V, Tindemans SH, Strbac G. Demand response contribution to effective 2017;89:52–61.
inertia for system security in the GB 2020 gone green scenario. In: IEEE PES ISGT [28] Muhssin MT, Cipcigan LM, Obaid ZA. Small Microgrid stability and performance
Europe; 6–9 Oct. 2013. p. 1–5. analysis in isolated island. In: 2015 50th International Universities Power
[8] Trovato V, Sanz IM, Chaudhuri B, Strbac G. Advanced control of thermostatic loads Engineering Conference (UPEC); 2015. p. 1–6.
for rapid frequency response in great Britain. IEEE Trans Power Syst 2016;vol PP [29] Obaid ZA, Sulaiman N, Hamidon MN. Design of fuzzy logic controller for AC motor
(99). 1-1. based on field programmable gate array. In: 2009 IEEE Student Conference on
[9] Cheng M, Sami SS, Wu J. Benefits of using virtual energy storage system for power Research and Development (SCOReD), Malaysia; 16-18 Nov. 2009. p. 487-490.
system frequency response. Appl Energy 2016. [30] Obaid ZA, Cipcigan LM, Muhssin MT. Fuzzy hierarchal approach-based optimal
[10] Li J, Xiong R, Yang Q, Liang F, Zhang M, Yuan W. Design/test of a hybrid energy frequency control in the Great Britain power system. Electr Power Syst Res Dec
storage system for primary frequency control using a dynamic droop method in an 2016;141:529–37.
isolated microgrid power system. Appl Energy 2016. [31] Moeini A, Kamwa I, Brunelle P, Sybille G. Open data IEEE test systems implemented
[11] Obaid ZA, Cipcigan L, Muhsin MT. Analysis of the Great Britain's power system with in SimPowerSystems for education and research in power grid dynamics and con-
Electric Vehicles and Storage Systems. 2015 18th International Conference on trol. In: 50th International Universities Power Engineering Conference (UPEC), UK;
Intelligent System Application to Power Systems (ISAP). 2015. p. 1–6. 1-4 Sept. 2015. pp. 1–6.
[12] Lee SJ, Kim JH, Kim CH, Kim SK, Kim ES, Kim DU, et al. Coordinated control [32] Hiskens RRAI. IEEE PES Task Force on Benchmark Systems for Stability Controls-
algorithm for distributed battery energy storage systems for mitigating voltage and Technical Report. IEEE; 2015.
frequency deviations. IEEE Trans Smart Grid May 2016;7(3):1713–22. [33] Vowles MGAD. Simplified 14-Generator Model of the South East Australian Power
[13] Sami SS, Meng C, Jianzhong W. Modelling and control of multi-type grid-scale System. School of Electrical and Electronic Engineering, The University of Adelaide,
energy storage for power system frequency response. In: IEEE 8th International South Australia; 2014.
Power Electronics and Motion Control Conference (IPEMC-ECCE Asia); 22–26 May