You are on page 1of 6

An Economic Optimization for BESS Sizing in a

Hybrid PV and Wind Power Plant


Yuqing Yang∗ , Stephen Bremner∗ , Chris Menictas† , Ke Meng‡ , Zhao Yang Dong§ and Merlinde Kay∗
∗ School of Photovoltaic and Renewable Energy Engineering, UNSW Australia.
† School of Mechanical and Manufacturing Engineering, UNSW Australia.
‡ School of Electrical and Information Engineering, The University of Sydney, Australia.
§ School of Electrical Engineering and Telecommunications, UNSW Australia.

Abstract—This paper presents an analytical sizing strategy for between the predicted and actual power generation is required
a Battery Energy Storage System (BESS) in a Hybrid PV and to be met by ancillary services if the actual power generation
Wind Power Plant (HPP), aiming at maximizing the operating is lower than the bid or to be dumped if it is over the bid [6].
profits of the HPP. This goal was achieved mainly through
applying the battery to compensate for forecasting errors, i.e., Therefore, when both solar and wind energy are integrated,
reducing renewable energy curtailment and ancillary service a higher accuracy level of renewable energy forecasting is
requirement under the electricity market operation rules. A novel necessary for renewable energy dispatch [7].
reformulation for the optimization problem was proposed to By improving the forecasting accuracy, the penalty from
convert the original nonlinear programming problem into a linear ancillary services can also be avoided. Besides implementing
programming problem. Two preliminary forecasting methods
were applied as comparable studies to verify the effectiveness different forecasting methods to improve the accuracy, an
of installing the BESS. The numerical simulation results have appropriately sized Battery Energy Storage System (BESS)
showed the detailed operating profits in each scenario with the can also be employed to store the energy when the predicted
corresponding BESS power and energy capacity. renewable production is less than the actual power generation
Index Terms—Battery Energy Storage System, Hybrid PV and to avoid the curtailment, and release the energy when the
Wind Power Plant, Forecasting Errors, Operating Profits.
predicted renewable production exceeds the actual power gen-
I. I NTRODUCTION eration to avoid the penalty from ancillary services. Moreover,
extra profits can also be gained from charging the battery when
As the technologies of wind and PV became mature, both the electricity price is lower and discharging the battery when
wind and PV farms have been developed all over the world the electricity price is higher. Therefore, in this paper, the
[1]. At the same time, the penetration of renewable energy BESS size will be determined by maximizing the operating
has increased significantly, in which wind and PV have made profits, including reducing the penalty from ancillary services,
a huge contribution. It can be envisioned that wind and avoiding renewable energy curtailment and energy arbitrage.
PV will still be two main renewable resources for future There are also numerous studies that have been conducted
power systems with higher penetration [2]. In this regard, the to solve battery sizing issues. For example, in [8], [9],
combination of wind and PV farms, i.e., a Hybrid PV and [10], battery sizes were determined based on wind energy
Wind Power Plant (HPP), will be a new trend to increase the forecasting errors. However, only the uncertainty from wind
renewable penetration and overcome the lower capacity factor energy was considered. Apart from that, in [11], [12], [13],
of individual solar/wind farms [3]. both PV and wind energy were considered for battery size
As previous studies show, solar and wind energy have a determination, but in the context of a microgrid. Hence, in
good complementary feature [4]. This is simply because solar this paper, an analytical BESS sizing strategy for a HPP was
energy can be produced during the daytime and wind energy is proposed to compensate for forecasting error and improve
more dominant during the night. Besides the complementary the operating profits at the same time. Moreover, this paper
characteristic in their generation profiles, the study of [3] also proposes a reformulation technique to convert the original non-
demonstrated that synergistic gains can be achieved to help linear programming problem to a linear programming problem.
reduce overall cost by co-locating wind and solar farms. In Therefore, it can be solved faster and more accurately.
this sense, constructing a hybrid PV and wind power plant The organization of this paper is as follows. Section II il-
can be more profitable than individual solar/wind farms. lustrates the formulation of problem and Section III shows the
However, with two different types of renewable energy re- algorithm and computation procedure. After that, numerical
sources in the hybrid power plant, more significant uncertainty simulations and results are demonstrated in Section IV, which
will occur during its operation. Under current electricity mar- is followed by the conclusion in Section V.
ket operation rules, for example in Australia [5], the operation
of renewable energy power plants are highly dependent on the II. P ROBLEM I LLUSTRATION
renewable energy forecasting accuracy. The mismatch partition The problem formulation can be expressed in four parts,
978-1-5386-4950-3/17/$31.00 ©2017 IEEE including the complementary nature analysis of solar and wind
TABLE I: The Proportion of the Three Scenarios It is obvious that persistence forecasting can be accurate only
Scenario WCS(%) SCW(%) WSS(%) when weather routines are stable. Hence, persistence forecast-
Proportion 20.63% 14.30% 34.93% ing is usually used as a benchmark for different advanced
forecasting methods. Here, persistence forecasting can be used
to demonstrate the results for a case where a less accurate
resources, modeling of the hybrid power plant, forecasting forecasting technique is adopted.
techniques and optimization problem formulation. By applying perfect and persistence forecasting, the effec-
A. Complementary Feature of Solar and Wind Resources tiveness of using the BESS can be predicted to be between
these two extreme forecasting techniques.
In this study, Global Horizontal Irradiance (GHI) and wind
speed (10-meter height) data in Goulburn, Australia, from D. Problem Formulation
2001 to 2011 were used to investigate the complementary According to the rules of the Australian Energy Market
characteristics of solar and wind resources. The approach from Operator (AEMO), hybrid power plants can be registered as
[4] has been adopted to evaluate the following scenarios for semi-scheduled power plants. When AEMO sets its dispatch
solar and wind resources: interval flag as “TRUE”, the HPP will be dispatched the same
• Wind complements solar (WCS) as scheduled power plants. In this study of operating the HPP,
• Solar complements wind (SCW) it is assumed that AEMO always sets the dispatch flag of this
• Wind and solar synergy (WSS) HPP as “TRUE”, which means the HPP is allowed to modify
Based on the approach from [4], the number of hours in the amount of generation up to five minutes prior to each
each scenario over 10 years can be calculated. The proportion dispatch period. To simplify the process in this study, it is
of each scenario is shown in Table I. assumed that the bids from the HPP are always accepted by
From the table above, it can be observed that there is quite a the national electricity market and in this study, the wholesale
high possibility that solar and wind resources can complement spot prices of New South Wales captured from Nemsight [17]
each other in this location. are used as the bid prices.
In this study, the BESS was optimally dispatched to achieve
B. PV, Wind and Battery Modeling
multiple goals for the HPP operation, including reducing the
The PV and wind modeling procedure is simplified by dumped renewable energy, avoiding the penalty from ancillary
using HOMER [14], a software for hybrid renewable energy services and promoting the extra profits from energy arbitrage.
systems design developed by the National Renewable Energy Therefore, maximizing the operation profits has been adopted
Laboratory (NREL). Therefore, GHI, wind speed and air as the objective function to achieve the multiple operation
temperature data are required to be fed into HOMER. Besides goals at the same time. The objective function of this problem
that, a numerical model [15] is adopted for battery modeling. can be formulated as,
c
PBESS,k ηBESS ∆t
SOCk+1 = SOCk + rated
(1) GHP P = EPT (PHPP − PBESS )∆t
EBESS
− AST max(FHPP − PHPP + PBESS , 0)∆t
PBESS,k ∆t (3)
SOCk+1 = SOCk + d rated
(2) − CO&M,P V PP V,c − CO&M,wind Pwind,c
ηBESS EBESS
− CO&M,BESS · max(|PBESS |)
Where PBESS,k denotes the charging/discharging power of
the BESS at time k. SOCk+1 indicates the State of Charge Where GHP P , PHPP , EP, AS, FHPP , PHPP , PP V,c ,
(SOC) of the BESS at time k. Note that positive values Pwind,c , PBESS indicate the net operating gain of HPP, the
of PBESS indicate that the battery is absorbing energy and actual power generation of the HPP, the electricity price, the
negative values represent the battery releasing energy. More- power of ancillary service, the ancillary service price, the
c
over, ηBESS d
and ηBESS denote the charging and discharging forecasted power of the HPP, the actual power generation of
efficiencies of the BESS, respectively. ∆t denotes the time the HPP, the power capacity of PV arrays, the power capacity
interval. of wind turbines, the instant power of the battery, where
positive values represent the battery is charging while negative
C. Forecasting Techniques values suggest the battery is discharging, respectively.
The two forecasting techniques used in this study are Note that |PBESS | denotes the absolute values of each entry,
perfect forecasting [16] and persistence forecasting. Perfect rather than the norm of the vector and the bold variables are
forecasting is not a real forecast, but it can show us the all N × 1 vectors. In addition, CO&M,P V , CO&M,wind and
highest potential this system can achieve. Therefore, perfect CO&M,BESS denote the operation and maintenance cost of
forecasting represents the ideal scenario of zero forecasting PV, wind turbines and BESS per interval. I indicates a N × 1
error. This means that the forecasted power production is the vector that is all ”1”s.
same as the actual power generation. Furthermore, PHPP , PAS and PDump can be expressed
On the other hand, persistence forecasting assumes that the as,
next day's weather conditions are the same as the day before. PHPP = PPV + Pwind (4)
PAS = max(FHPP − PHPP , 0) (5) cost with absolute value operation for PBESS is also not
included in the optimization process, but corrected after that.
PDump = max(PHPP − FHPP , 0) (6)
Therefore, the HPP operating profits without consideration of
Where PPV and Pwind are the instant power generation operation and maintenance cost is denoted by G0HP P , and
from PV and wind in the HPP. the power of the BESS without consideration of charging and
Apart from the objective function, the decision variables of discharging efficiencies is represented as P0BESS . The problem
PBESS were also limited by the power capacity of the BESS. can be reformulated as,
Thus, the constraint of the BESS's charging and discharging Maximize G0HP P
boundaries can be expressed in (7), h i
G0HP P = EPT (PHPP − P0BESS ) − AST · T ∆t (12)
−PBESS,max ≤ −PBESS,k ≤ PBESS,max (7)
subject to, (7), (8) and (10)
Where PBESS,max indicates the maximum charging bound-
ary of the BESS. Meanwhile, the same magnitude was adopted T≥0 (13)
as the discharging boundary. PBESS,k denotes the k th entry
in PBESS , and similar rules apply to other N × 1 vectors. T ≥ FHPP − PHPP + P0BESS (14)
Next, the constraint of the HPP operation boundaries was Therefore, the nonlinear programming problem has been
also considered, thereby limiting the decision variable of converted into a linear programming problem through extend-
PBESS,k . ing the dimension of decision variables.
0 ≤ PHP P,k − PBESS,k ≤ PHP P,max (8) To simplify, the results from the above optimization is
regarded as the desired BESS output, then the actual BESS
Another important constraint is the SOC of the BESS. The output can be corrected through (15). Similarly, the total net
relationship between SOCk and PBESS,k follows (1-2). To operating profits can also be corrected through (16).
avoid severe degradation when the battery is operated in high
c/d
Depth of Discharge (DOD), 20% of SOC is adopted as the PBESS = ηBESS P0BESS (15)
lower boundary.
GHP P = G0HP P − CO&M,P V PP V,c
20% ≤ SOCk ≤ 100% (9) − CO&M,wind Pwind,c
(16)
The objective function and constraints above were − CO&M,BESS max(|PBESS |)
formulated in a standard form using the matrices below,
where PBESS is the decision variable.
From the formulation above, we can see that through opti-
Maximize GHP P mally charging/discharging the battery, the difference between
subject to, (7-8) the forecasted power profile and the actual power profile can be
compensated, so that the ancillary service cost and renewable
20%EBESS,max I ≤ EBESS,0 I + A · PBESS ∆t
(10) energy curtailment can be reduced. Therefore, the operation
≤ EBESS,max I profits of the hybrid power plant can be improved. Besides
It can also be noticed that the constraint (10) reflects (9). that, the excess battery capacity can be used to store the energy
In (10), we ignored the charging and discharging efficien- when the market price is lower and then sell it back when
cies, which was corrected after the optimization procedure. the market price is higher, in order to increase the operation
EBESS,0 denotes the initial energy in the BESS. The matrix profits. Hence, by maximizing the operating profits of the HPP,
A can be expressed as, multiple goals can be achieved at the same time.
 
1 0 0 ... 0 III. A LGORITHM AND C OMPUTATION
1 1 0 . . . 0
  In this study, the optimal power and energy capacities of the
. . . . . . . . . . . . . . . .
BESS are determined in an analytical process shown in Fig.1.
1 1 1 ... 1 In each scenario, the power and energy capacity of the battery
Therefore, the optimization was formulated above as a will serve as the constraints during the optimization process.
nonlinear programming problem due to the existence of the The main target of dispatching the battery in this paper is to
max operator. Here we reformulated the problem into a linear maximize the operation profits of the HPP. The optimization
programming problem by extending the dimension of the problem is a linear programming problem that can be solved
decision variables. The added decision variable T is also a directly by the MATLAB built-in solver.
N × 1 vector, which can be expressed as, Here are the optimization procedures explained in steps
below.
T = max(FHPP − PHPP + PBESS , 0) (11)
Step 1: Initialization. Input the power capacities of PV and
Besides the simplification of the charging and discharging wind turbines; input the actual PV and wind power generation;
efficiencies, the term showing the operation and maintenance input the forecasted PV and wind power generation; input the
TABLE II: Numerical Simulation Results of Cases 1 & 2
Scenario Net Operating Demanded An- Dumped
Profits (Australian cillary Service Energy (MWh)
$) (MWh)
Case 1 1.3812×106 0 0
Case 2 8.5142×105 1.5460×104 1.5422×104

IV. N UMERICAL S IMULATIONS


A. Case Setting
In this paper, four cases have been set up to simulate the
operation of the HPP under different forecasting methods and
with or without BESS, in order to evaluate the impacts of using
different forecasting techniques and the impacts of installing
the BESS in the HPP.
• Case 1: HPP without BESS under perfect forecasting
• Case 2: HPP without BESS under persistence forecasting
• Case 3: HPP with BESS under perfect forecasting
• Case 4: HPP with BESS under persistence forecasting
The power capacities of PV and wind turbines in this study
were assumed as 15MW. Therefore, the power capacity of
the HPP is 30MW. With regards to the scale of the HPP,
the maximum power and energy capacities of the BESS
were assumed as 15MW and 75MWh (5 hours storage). The
simulation interval is 1 hour and the simulation duration is
1 year. The operation and maintenance costs of PV, wind
turbines and battery used here are $16, $33 [19] per year
Figure 1. Flow Chart of Optimization per kW and $17 (Lead-acid battery) [20] per year per kWh.
In addition, both charging and discharging efficiencies in this
paper are adopted from [15] as 0.87 (Lead-acid battery).

electricity price; set time duration (1 year, i.e., 8760 hours) B. Results
and simulation interval (1 hour). Then, go to Step 2. The operation results of Cases 1 and 2 without battery are
demonstrated in Table II.
Step 2: Start with the initial power capacity and energy
From the table above, both cases were operated without
capacity of the BESS. Then, go to Step 3.
battery, it can be observed that there is a significant profit
Step 3: In the scenario where there is no battery, when reduction from perfect forecasting to persistence forecasting.
the actual power production exceeds the forecasted power The difference of the profits between the two cases mainly
generation, the extra energy has to be curtailed. On the other come from the ancillary service penalty and the dumped
hand, when the actual power production is less than the energy, the profits from which is supposed to be earned rather
forecasted power generation, the hybrid power plant needs than to be dumped.
to pay the penalties from ancillary services. In the scenarios The results also indicate that the operating profits using
where there is a battery, the optimization procedure will other advanced forecasting techniques are supposed to achieve
be performed with the given power and energy capacity of the profits between perfect and persistence forecasting. Also,
the BESS as optimization constraints. The results from the the energy deficiency that needs ancillary services and the
optimization process is the optimal charge/discharge schedule extra generation that needs to be dumped are apparently
of the battery. This optimal schedule can guarantee the HPP not identical, which indicates a slight bias from persistence
achieves the highest operation profits on a daily basis. Then, forecasting. Moreover, the former energy amount is larger than
go to Step 4. the latter. In other words, the persistence forecasting in this
case tends to overestimate the power production of PV and
Step 4: Increase the power capacity of the BESS, and
wind energy.
go back to Step 3. However, if the preset maximum power
Fig.2 and 3 show the net operating profits in Cases 3 and 4,
capacity is reached, go to Step 5.
respectively. A straight-forward outcome from Fig.2 and 3 is
Step 5: Increase the energy capacity of the BESS, and that the profits of the HPP grow significantly with the increase
go back to Step 3. However, if the preset maximum energy of the BESS power and energy capacities, but at different rates.
capacity is reached, terminate the simulation procedure. In both cases, the increase of the HPP profits are more sensitive
Figure 2. The Net HPP Operating Profits in Case 3 Figure 5. The Dumped Energy in Case 4

of 5 hours storage, resulting in a relatively large BESS energy


capacity.
The optimized HPP profits for Case 4 in Fig.3 are gener-
ally lower than the profits in Fig.2. This is because perfect
forecasting was applied in Case 3, which indicates that there
was no ancillary service needed and no dumping of renewable
energy occurred. Thus, the power and energy capacities of the
BESS are predominantly used to achieve energy arbitrage by
storing energy when the electricity price is relatively low and
then selling it back to the market when the electricity price is
high. However, in Case 3, a part of the battery's capacity has to
contribute to reducing the ancillary services and absorbing the
curtailed energy, thereby contributing less in energy arbitrage.
Fig.4 and 5 show the required energy for ancillary services
Figure 3. The Net HPP Operating Profits in Case 4 and the dumped energy in Case 4, respectively. The results
from Fig.4 and 5 show a consistent outcome of implementing
the BESS for reducing ancillary service and dumped energy.
It can be noticed that around 5 MW and 30 MWh power
and energy capacity is a good choice as a battery size for
reducing ancillary service and dumped energy. Then, any extra
battery capacity can be used to achieve higher operating profits
through energy arbitrage. For clarity, the reason that the energy
required for ancillary services and the dumped energy did not
reach zero is due to the limitation of the BESS power and
energy capacities.

V. C ONCLUSION
In this paper, an analytical process was performed to
determine the BESS size for a HPP with consideration of
two different forecasting techniques. In contrast to the sce-
narios where no BESS participates in the HPP operation,
Figure 4. The Required Energy for Ancillary Services in Case the results have shown the effectiveness of using the battery
4 in a HPP. Furthermore, a reformulation was proposed to
convert the original nonlinear programming problem into a
linear programming problem by extending the dimension of
to the BESS power capacity when its energy capacity is large decision variables. The simulation results have shown the
enough, in contrast to the increase of its energy capacity when differences and trends in the operating profits of the HPP for
its power capacity is large. This is probably due to the setting each combination of battery power and energy capacities. The
results can be used as critical information for decision makers [19] NREL, “Distributed generation renewable energy estimate of costs,”
to determine the size of the BESS in a HPP. Future work will accessed on 12 May, 2017. [Online]. Available: http://www.nrel.gov/
analysis/tech lcoe re cost est.html
explore more on the impacts of battery aging on the process [20] M. Ross, R. Hidalgo, C. Abbey, G. Jo, and x00F, “Analysis of energy
of its size determination. storage sizing and technologies,” in Electric Power and Energy Confer-
ence (EPEC), 2010 IEEE, 2010, pp. 1–6.
ACKNOWLEDGMENT
This research is supported by the University International
Postgraduate Award (UIPA) scholarship funded by UNSW.

R EFERENCES
[1] T. I. R. E. A. (IRENA), “Remap 2030,” Tech. Rep., 2015 (accessed on
28 April, 2017). [Online]. Available: http://www.irena.org/remap/irena
remap usa report 2015.pdf
[2] IRENA, “The power to change: Solar and wind cost reduction
potential to 2025,” Tech. Rep., 2016 (accessed on April 28,
2017). [Online]. Available: http://www.irena.org/DocumentDownloads/
Publications/IRENA Power to Change 2016.pdf
[3] AECOM, “Wind-solar co-location study,” Tech. Rep., 2016 (accessed
on April 28, 2017). [Online]. Available: https://arena.gov.au/resources/
wind-solar-co-location-study/
[4] A. A. Prasad, R. A. Taylor, and M. Kay, “Assessment of solar and wind
resource synergy in australia,” Applied Energy, vol. 190, pp. 354–367,
2017.
[5] AEMC, “National energy rules,” (accessed April 28,
2017). [Online]. Available: http://www.aemc.gov.au/Energy-Rules/
National-electricity-rules/Current-Rules
[6] E. W. Law, M. Kay, and R. A. Taylor, “Calculating the financial value
of a concentrated solar thermal plant operated using direct normal
irradiance forecasts,” Solar Energy, vol. 125, pp. 267–281, 2016.
[7] B. Elliston and I. MacGill, “The potential role of forecasting for inte-
grating solar generation into the australian national electricity market,”
in Solar 2010: proceedings of the annual conference of the Australian
solar energy society, 2010.
[8] H. Bludszuweit and J. A. Dominguez-Navarro, “A probabilistic method
for energy storage sizing based on wind power forecast uncertainty,”
IEEE Transactions on Power Systems, vol. 26, no. 3, pp. 1651–1658,
2011.
[9] J. Wu, B. Zhang, H. Li, Z. Li, Y. Chen, and X. Miao, “Statistical
distribution for wind power forecast error and its application to deter-
mine optimal size of energy storage system,” International Journal of
Electrical Power & Energy Systems, vol. 55, pp. 100–107, 2014.
[10] H. Bitaraf, S. Rahman, and M. Pipattanasomporn, “Sizing energy storage
to mitigate wind power forecast error impacts by signal processing
techniques,” IEEE Transactions on Sustainable Energy, vol. 6, no. 4,
pp. 1457–1465, 2015.
[11] B. Bahmani-Firouzi and R. Azizipanah-Abarghooee, “Optimal sizing of
battery energy storage for micro-grid operation management using a new
improved bat algorithm,” International Journal of Electrical Power &
Energy Systems, vol. 56, pp. 42–54, 2014.
[12] D. W. Gao, “Sizing of energy storage systems for microgrids,” in Energy
Storage for Sustainable Microgrid. Academic Press, 2015, pp. 125–142.
[13] H. Khorramdel, J. Aghaei, B. Khorramdel, and P. Siano, “Optimal
battery sizing in microgrids using probabilistic unit commitment,” IEEE
Transactions on Industrial Informatics, vol. 12, no. 2, pp. 834–843,
2016.
[14] “HOMER Energy,” accessed April 28, 2017. [Online]. Available:
http://www.homerenergy.com/
[15] A. Cervone, G. Carbone, E. Santini, and S. Teodori, “Optimization of the
battery size for pv systems under regulatory rules using a markov-chains
approach,” Renewable Energy, vol. 85, pp. 657–665, 2016.
[16] E. W. Law, M. Kay, and R. Taylor, “Assessing the economic benefit
of forecasting concentrated solar thermal energy output,” in Solar2014:
The 52nd Annual Conference of the Australian Solar Council, Australian
Solar Council, Melbourne, Australia, 2014, pp. 131–140.
[17] “Nemsight,” accessed on 28 April, 2017. [Online]. Available:
http://analytics.com.au/energy-analysis/nemsight-trading-tool/
[18] G. Wang, M. Ciobotaru, and V. G. Agelidis, “Optimal capacity design
for hybrid energy storage system supporting dispatch of large-scale
photovoltaic power plant,” Journal of Energy Storage, vol. 3, pp. 25–35,
2015.