Optimization of Isolated Microgrids with Cost and

Reliability Targets
Daniella C. Fronda, Kim Tristan U. Rosendo, Michael Angelo A. Pedrasa, PhD
Smart Grid Research Center
Electrical and Electronics Engineering Institute
University of the Philippines-Diliman, Quezon City, Philippines
[email protected],[email protected],[email protected]

Abstract—Due to the economic and reliability constraints in the cost or reliability parameter is set as main objective while
designing islanded power systems, it is desired to have an efficient the other serves as the constraint. This paper also utilizes the
process of finding the optimum size of equipment to be installed. said algorithms and shares the same constraint scenarios, but
This paper presents a software tool that can choose the number
and sizes of distributed generators and energy storage systems has more DGs such as biomass, diesel and storage. A similar
in an island microgrid under three planning regimes: target cost, approach to this was performed by Xu et al [4] when they used
target reliability, and high reliability. These modes allow for the GA to specifically minimize total capital cost and setting the
planning of microgrids when capital costs are limited, or if the Loss of Power Supply Probability (LPSP), a power reliability
users allow for lower supply availability to reduce capital cost, or index, as a constraint for the isolated power system. Instead of
to design a microgrid with high supply availability. The results
show that the software tool can find feasible solutions under setting the LPSP, the tool that is presented here uses Expected
different combination of inputs and constraints. Energy Not Supplied (EENS) as either an objective function
or a constraint. Other simulation tools with optimization and
user interface capabilities also exist in the market, including
I. I NTRODUCTION HOMER (Hybrid Optimization of Multiple Energy Resources)
Rural electrification has remained to be a challenge to [5], Electrical Transient and Analysis Program (ETAP) [6]
this day. Many remote areas in countries all over the world and Distributed Energy Resources Customer Adoption Model
are still not connected to their main utility grid. Though (DERCAM) [7]. But these are limited to a number of technolo-
universal access to modern energy service is considered by gies and do not consider the number of allowable interruptions
the United Nations (UN) as a necessity for an individual’s and cost limit as input. Hence, the development of a software
development and the nation’s economic growth, still over one- tool that accepts a customized degree of reliability and budget
third of the human population is currently deprived of basic allocation to yield an optimal combination of resources for the
access to electricity [1]. Hence, options such as isolated or network is needed.
standalone hybrid microgrids are becoming a viable option This paper1 describes a simulation tool with coherent and
for supplying power specifically for these areas. These power efficient graphical user interface (GUI) for the Philippines that
systems are designed to independently cater for the demands will give the user access to information they need as shown in
of the area and be sized accordingly to the resources available. Figure 1. Figure 2 shows the output interface that displays the
Isolated microgrids have no transmission costs, and may con- list of optimal DGs and ESSs chosen by the program, overall
tain environmental-friendly technologies compared to a con- system throughput and total cost of each class of DG or ESS.
ventional power system. These resources include but are not Sensitivity analysis was also applied to the combination chosen
limited to solar PV, wind turbines, hydropower, diesel engine, to produce graphs correlating changes in DER characteristics
and biomass generators. However, the renewable distributed and were plotted afterwards.
generators (DGs) installed in these islanded systems still have
limited capabilities primarily due to their intermittent nature
and dependence on geographical or atmospheric conditions
[2]. Moreover, choosing the best set of technology to be
adapted by these microgrids to service the growing demand
II. M ATHEMATICAL F ORMULATION
poses many technical and financial challenges, especially if
only a tolerable amount of service interruption is permitted. A. Cost
As a result, optimizing the network design that can aggregate This study considered several cost components of DERs
all these circumstances and come up with an optimal set of as one of its objective function. All planning costs of each
DGs and ESSs has been gaining considerable attention in resource was incorporated and was computed as net annual
recent studies. Jahangir et al [3] proposed a microgrid design worth since the analysis for the microgrid will be considered
based on unconventional Monte Carlo (MC) simulation and over one year. The capital recovery factor in Eq. 1 will be
genetic algorithm (GA). Two scenarios have been carried out
by their optimization model where, for each scenario, either 1 978-1-5386-4950-3/17/$31.00 ©2017 IEEE

Authorized licensed use limited to: IEEE Xplore. Downloaded on November 06,2023 at 18:37:03 UTC from IEEE Xplore. Restrictions apply.
 2

Figure 1. Input GUI

Figure 2. Output GUI

computed for each DER since each of them have different management cost was computed as a percentage of the capital
lengths of useful life as assumed in Table 1. The interest rate cost, from collected data in IRENA [8-11].
of 10% was be used for all computations.
Type Useful life (n) AWstd = CAPstd ∗ ([A/P ] + omstd ) (2)
Solar Panels 35
Wind Turbines 20 Where,
Hydro Generators 50 AWstd is the Annual Worth for solar/wind/hydro DGs &
Biomass Generators 25 ESS
Diesel Generators 20
Storage 10 CAPstd is the Capital Cost for solar/wind/hydro DGs & ESS
Table I
omstd is the O&M Cost as percentage of Capital Cost
U SEFUL L IFE CONSIDERED FOR EACH D ISTRIBUTED G ENERATORS For biomass DGs, two forms of O&M cost were considered:
fixed and variable cost. Fixed O&M cost data was obtained
from IRENA [12] and the variable cost was computed. The
A i(i + 1)n variable O&M cost for biomass generators is comprised of
= (1) the annual fuel cost and transportation costs. Both of these
P (i + 1)n − 1
are dependent on the type of feedstock used for the generator.
Where,
A/P is the Capital Recovery Factor
OMvar = AF Cbio + OMtranspo (3)
i is the Interest Rate
n is the Useful Life
For solar, wind, hydro DGs and ESS, a standardized for- $ kg hrs
mula (Eq. 2) for annual worth was used. The operation and AF Cbio = ∗ ∗ Rating(kW ) ∗ (4)
kg kW h yr

Authorized licensed use limited to: IEEE Xplore. Downloaded on November 06,2023 at 18:37:03 UTC from IEEE Xplore. Restrictions apply.
 3

ED is the total dumped energy

Pexcess is the total excess power produced from all the DERs
AWbio = CAPbio ∗ [A/P ] + OMvar + OMf ixed (5)
in a year
Where, Pload is the total load demand in a year
OMvar is the Variable O&M Cost for biomass DGs
AFCbio is the Annual Fuel Cost for biomass DGs C. Modes of Operation
OMtranspo is the O&M transportation Cost 1) High Reliability: This setting chooses an optimal system
AWbio is the Annual Worth for biomass DGs which can meet the load demand for the given DER mix at
CAPbio is the Capital Cost for biomass DGs the cheapest price possible. That is,
OMfixed is the Fixed O&M Cost for biomass DGs
For diesel DGs, the annual worth computation is similar M in (Cnet ) (11)
to that of the aforementioned standard formula, but with the
added annual fuel cost. The AFC considers the liters per Constricted to,
hour consumed by a certain DG and its power rating. Diesel EEN S = 0 (12)
generators with smaller power ratings consume less fuel with 2) Target Cost: Microgrid planning that involves this mode
the same output as that of their larger counterparts. finds the DER combination with the least total annualized cost
$ L hr with a specified threshold for reliability assessment criterion.
AF Cdie = ∗ ∗ (6) Thus, minimizing the reliability objective function obtained in
L hr yr
Eq. 9 such that cost does not exceed the maximum budget
allocation. Mathematically, it can be written as:
AWdie = CAPdie ∗ ([A/P ] + omdie ) + AF Cdie (7)
M in (EEN S) (13)
Where,
CAPdie is the Capital Cost for diesel DGs Subject to,
AFCdie is the Annual Fuel Cost for diesel DGs
AWdie is the Annual Worth for diesel DGs Cnet ≤ Cnetmax (14)
Combining all these equations into one, the total cost can
Where,
be expressed as shown in Eq. 8
Cnetmax is the maximum allowable limit of total cost
3) Target Reliability: The last of the target scenarios is
Cnet = AWs +AWw +AWh +AWdie +AWbio +AWsto (8) concerned with creating the optimal combination of DGs and
ESSs with the capability to meet the reliability assessment
B. Reliability criterion. Therefore, the goal is to minimize the cost objective
function obtained in Eq. 8 under reliability constraint in Eq. 9.
Like costs, another important aspect to look at is the
Mathematically, it can be shown in the following relationship:
system’s reliability. The quantity called Expected Energy
Not Supplied (EENS) was used as the main benchmark to M in (Cnet ) (15)
check the demand servicing capability of the DERs chosen.
Subject to,
It provides an estimation of the expected disservice to the
consumer in kilowatts per hour (kWh) summed for a whole EEN S ≤ EEN Smax (16)
year as computed in Eq. 9.
Where,
8760
X EENSmax is the maximum allowable limit of energy not
EEN S = PDi − PSi (9) served
i=1

Where, III. P ROGRAM I MPLEMENTATION

PDi is the total hourly demand in the ith hour of the year A. Microgrid Simulation
PSi is the total hourly energy supplied by the chosen
combination in the i hour of the year The simulation tool created by Gan and Luya [13] which
The total instances of interruptions or SAIFI (System Aver- observes patterns and determines the effects of a DG mix given
age Interruption Frequency Index) and its average duration in load demand for a supply deficient microgrid was run on each
hours or SAIDI (System Average Interruption Duration Index) and every DG combination generated by the GA through an
was also counted everyday. In cases that there are oversupply 8760 hour (one year) simulation. As an integral part of the
of energy produced, it was also counted and was expressed as whole process, it will serve as the main generator of DER
an energy dumped as shown in Eq. 10. information such as number of expected interruptions, total
number of energy deficit or excess and total number of power
8760 contributed per resource. All of these results will then be
X Pexcess
ED = (10) passed back and evaluated by decision making process (GA)
i=1
Pload
of the proposed optimization tool. Its full implementation was
Where, discussed in detail in the paper presented in [13].

Authorized licensed use limited to: IEEE Xplore. Downloaded on November 06,2023 at 18:37:03 UTC from IEEE Xplore. Restrictions apply.
 4

Simulation SD SWD SWHD SBSto SDSto

Cost ($) 100976.8 101156 35953.7 168319.8 122368
EENS (%) 0 0 0 0 0
Demand (kWh) 710009.4 710428.1 710189.8 7109004 709647
Oversupply (%) 2.68E-6 6.76E-6 4.64E-1 1.97E-7 3.16E-5
Diesel Used 147367.4 145212 3504.5 0 145661
Feedstock Used 0 0 0 98770.9 0
Storage Cycles # 0 0 0 0 85.978
Table II
C OST & E NERGY C HARACTERISTICS OF 5 T EST C ASES IN H IGH
R ELIABILITY O PTIMIZATION

Target Cost (TC)

On this optimization mode, the solutions with the highest
cost are truncated, keeping 80% of the current population. The
population keeps increasing with each iteration until there are
enough solution sets in the population that meet the threshold.
The population is then sorted by EENS and those with the
highest values are truncated while the population number is
maintained.

Target Reliability (TR)

Summarily, it is the inverse of the Target Cost mode, and
the two phases are also used in this mode. Instead of sorting
by cost in the first phase with increasing population, the
population is sorted by EENS. The elements with higher values
Figure 3. Overall Genetic Algorithm Flow of EENS are truncated and 80% of the population is kept. On
the second phase, the solutions are then sorted based on cost.
B. Genetic Algorithm
IV. C ASE S TUDIES
The backbone of the program is the genetic algorithm. It
makes use of the cost evaluation and microgrid simulation The following case studies were configured to demonstrate
of possible solution sets for minimizing objective functions the usefulness of the optimization algorithm. These cases
or constraints. Figure 3 shows the detailed implementation exhibit the output characteristics of the network under different
of the genetic algorithm that was form fitted to achieve the inputs and constraints.
optimization characteristics desired; form fitted in terms of
parameters such as crossover and mutation rate, number of A. Case 1 - High Reliability Optimization: 5 Test Cases
iterations, population, chromosome generation and selection The five main DER combinations that were used for the
methods. test cases are: solar-diesel (SD), solar-wind-diesel (SWD),
1) Truncation: The population offers a diverse set of solu- solar-wind-hydro-diesel (SWHD), solar-biomass-storage (SB-
tions for the optimization problem. Since there are different Sto) and solar-diesel-storage (SDSto). These were selected
optimization modes or problems, the GA uses different forms based on practicality and availability in the market. Table 2
of truncation. Sorting and truncating in the population based is a summary of each unique DER combination’s quantitative
on EENS and cost is necessary to make sure that the solution characteristics after high reliability optimization.
converges to an optimal solution because only the fittest It can be observed that increasing the sizes and kinds of
solutions should be carried on to the next iteration, a process renewable sources allows the diesel generator to reduce the
known as Elitism. amount of fuel it consumes hourly. The only system with a
hydro generator was the most cost efficient among the five test
High Reliability (HR) combinations evident in Figures 4 and 5.
This is the default mode of optimization which selects a so-
lution set with the least cost that meets the given load demand B. Case 2 : Three Optimization Modes using Solar-Diesel-
based on a combined fitness function for each solution. The Storage
sum of cost and EENS are used as the fitness function. The two The three optimization modes were observed for the solar-
factors are given weights which can be calibrated to improve diesel-storage set. Since the cost of building and operating
the truncation. Elitism was used by sorting the population’s these systems increases with the amount of energy serviced,
fitness and truncating those that exceed the number of elements the balance had to be found for the three different priority
in the original population. settings. Once the threshold was met for target cost or target

Authorized licensed use limited to: IEEE Xplore. Downloaded on November 06,2023 at 18:37:03 UTC from IEEE Xplore. Restrictions apply.
 5

Energy HR TR TC
Target EENS (GWh) n/a 0.5 n/a
EENS (GWh) 0 0.499222 0.5347338
Computed EENS (%) 0 70.348 75.239
Demand (GWh) 0.709647 0.7096471 0.7107112
Total oversupply (Wh) 224 0 0
Table IV
E NERGY CHARACTERISTICS OF SOLAR - DIESEL - STORAGE IN 3
OPTIMIZATION MODES

Target Cost: $20,000

Diesel Output Rating 100 75 60 40 20
PW Capital Cost 19010 19949 19988 19937 19935
Table V
Figure 4. Cost Distribution for 5 DER systems with high reliability C APITAL C OST FOR VARYING D IESEL G ENERATOR O UTPUT R ATINGS

C. Case 3 : Diesel Sensitivity Analysis: Solar-Wind-Diesel

with Varying Diesel Rated Power Output
The solar-wind-diesel combination was observed for cost
and energy patterns. The diesel ratings were chosen such that
the cost threshold could be met for supply-deficient systems
under target cost optimization. All of the present worth capital
costs for varying diesel generator output ratings satisfy the cost
constraint as shown in Table 5. It is desirable for the computed
cost to attempt to reach the highest cost acceptable, meaning
its upper threshold of 20,000 USD, since this indicates that it
the planner can purchase more generators and service more of
Figure 5. Energy output distribution for 5 DER systems with high reliability
the load.
The size and number of solar PVs were observed for effects
due to the increasing fuel consumption in increasing diesel
reliability, the GA attempts to minimize the other parameter,
power ratings. Overall, the number of solar PVs purchased
cost or EENS, without violating the constraint. For a target
decreased for sets with higher diesel ratings (75kW and higher)
cost setting, as the set reaches the threshold it begins to
as shown in Figure 6. This can be attributed to the cost
minimize EENS. Similar observation can also be recognized
efficiency of just buying diesel fuel instead of buying large
for target reliability. Once the EENS settles between 0 and the
numbers of solar panels to achieve the same power output.
threshold, the GA begins cost minimization.
Also, increasing the rating of the diesel shows a decrease in
Finally, the high reliability mode must find a balance
EENS (Figure 7) because it can supply the demand better.
between cost and energy not served. As we had observed
previously in the precision test, the GA does attempt to mini-
mize both parameters and finds a consistent enough solution.
Compared to target cost and reliability, the cost for a high
reliability system was higher since the required EENS was
zero.
Tables 3 and 4 summarizes the different behaviors of the
same set under the three optimization modes. It can be shown
that the software tool was able to find an optimum set with a
cost or EENS approximately equal to the input threshold for
target cost and reliability modes respectively.

Cost HR TR TC
Target Cost ($) n/a n/a 80000 Figure 6. Energy Output Distribution for varying Diesel Rated Output
PW Capital n/a n/a 79994
Computed Cost ($) 122368 56479.39 54967.99 The behavior of net annual worth and annual operation
AW Capital Cost 29752.6 15026.39 12814.28
O&M Cost 92615.6 41453 42153.71
& maintenance cost linearly increase with increasing diesel
output rating. Capital cost does not follow this pattern. At
Table III
C OST CHARACTERISTICS OF SOLAR - DIESEL - STORAGE IN 3 OPTIMIZATION
higher diesel ratings, the capital cost slope decreased while
MODES operation & maintenance cost slope increased in Figure 8
because the planning algorithm was required to allocate more

Authorized licensed use limited to: IEEE Xplore. Downloaded on November 06,2023 at 18:37:03 UTC from IEEE Xplore. Restrictions apply.
 6

Figure 7. EENS for varying Diesel Rated Output

Figure 9. Fuel Consumption Characteristics for varying Diesel Rated Output

R EFERENCES
[1] R. Shrestha J. Archarya. Sustainable energy access planning: A
framework. Technical report, Asian Development Bank, 2015.
[2] S. T. Bahta. Design and analyzing of an off-grid hybrid renewable
energy system to supply electricity for rural areas: Case study: Atsbi
district, north ethiopia. Master’s thesis, KTH Industrial Engineering and
Management, 2013.
[3] A. Ahmadian H. Jahangir and M. A. Golkar. Optimal design of stand-
alone microgrid resources based on proposed monte-carlo simulation.
IEEE Innovative Smart Grid Technologies - Asia (ISGT ASIA), 2015.
[4] Liuchen Chang Daming Xu, Longyun Kang and Binggang Cao. Optimal
sizing of standalone hybrid wind/pv power systems using genetic algo-
rithms. Canadian Conference on Electrical and Computer Engineering,
2005.
Figure 8. Cost Characteristics for varying Diesel Rated Output [5] "HOMER Energy." HOMER Energy LLC, 2015. [Online]. Available:
http://www.homerenergy.com/software.html.
[6] "ETAP Microgrid: Integrated Power System Simulation, Planning,
Protection and Real-Time Microgrid Master Controller," Electri-
of the budget to diesel generators. Purchasing diesel generators cal Transient and Analysis Program. 2016. [Online]. Available:
instead of solar PV or wind turbines reduces capital cost but http://etap.com/microgrid/etap-microgrid.htm.
[7] "Distributed Energy Resources Customer Adoption Model (DER-
this means higher fuel costs. And in Figure 9, diesel fuel and CAM)," Microgrids at Berkeley Lab. 2015. [Online]. Available:
operation & maintenance cost are directly proportional. http://www.homerenergy.com/software.html.
[8] International Renewable Energy Agency (IRENA). Wind power. Re-
newable Energy Technologies: Cost Analysis Series, Volume 1: Power
V. C ONCLUSION Sector:Issue 5/5, 2012.
[9] International Renewable Energy Agency (IRENA). Solar photovoltaics.
This paper describes a software tool that can design isolated Renewable Energy Technologies: Cost Analysis Series, Volume 1: Power
microgrids under different design conditions. Upon doing Sector:Issue 4/5, 2012.
various test cases and performing sensitivity analysis, this [10] International Renewable Energy Agency (IRENA). Market status and
technology outlook. Battery Storage For Renewables, 2015.
program has proved its capabilities in planning viable DER [11] International Renewable Energy Agency (IRENA). Hydropower. Re-
systems with financial considerations. It is also flexible to newable Energy Technologies: Cost Analysis Series, Volume 1: Power
varying conditions and constraints a such as hidden costs, Sector:Issue 3/5, 2012.
[12] International Renewable Energy Agency (IRENA). Biomass for power
supply deficiencies and availability of resources. However, due generation. Renewable Enery Technologies: Cost Analysis Series,
to the fact that there is a large discrepancy in prices and power Volume 1: Power Sector:Issue 1/5, 2012.
rating of the DER models available, the optimization algorithm [13] E. Luya P. Gan. Standalone microgrid simulation platform. IEEE
Innovative Smart Grid Technologies - Asia (ISGT ASIA), 2016.
tends to favor some DGs like hydro and diesel generator [14] International Renewable Energy Agency (IRENA). Statistical issues:
over other renewable energy resources, most noticeably wind bioenergy and distributed renewable energy. Statistical issues: bioenergy
turbines. Although this is the case, the results were still and distributed renewable energy, 2013.
[15] K. M. Iromi Udumbara Ranaweera M. Kolhe and A. G. B. Sisara
intuitive and consistent as there are observable trends in other Gunawardana. Techno-economic optimum sizing of hybrid renewable
parameters such as total number of blackouts, power ratings energy system. Industrial Electronics Society, IECON 2013 - 39th
of the DERs and fuel consumption for biomass and diesel Annual Conference of the IEEE, 2013.
[16] Marek Obitko. Introduction to genetic algorithm, 1998. Online.
generators. [17] Neves J. Rocha M. Preventing premature convergence to local optima
in genetic algorithms via random offspring generation. IEA/AIE, 1999.
ACKNOWLEDGEMENT [18] H. Zhou Y. Lin L. Guo, C. Li and W. Liu. Optimization of capacity con-
figuration in island mode micro-grid. 2015 IEEE Advanced Information
The authors acknowledge the support from the Commission Technology, Electronic and Automation Control Conference (IAEAC),
pages 137–141, 2015.
on Higher Education -- Philippines California Advanced Re-
search Institutes (PCARI) Project IIID-54: Resilient Societal
Cyber-Physical Systems.

Authorized licensed use limited to: IEEE Xplore. Downloaded on November 06,2023 at 18:37:03 UTC from IEEE Xplore. Restrictions apply.