Microgrid paper camera copy final draft

of 8
Architecture Study of an Energy Microgrid
Ravi Patel, Walter Paleari and Daniel Selva
Systems Engineering Program
Cornell University
Ithaca, NY, USA
{rpp63,wp243,ds925}@cornell.edu
Abstract— In the last decade, there has been a push to achieve
regional energy independence by developing small, self-sufficient
microgrids that complement, and in some cases, replace the main
centralized grid. This sort of distributed energy system has
numerous advantages. One of them is the ability to disengage and
function independently from the main grid in the event of a
catastrophic failure. Additionally, they allow for a far greater
penetration of renewable energy sources, thus allowing for a much
cleaner energy system with a diverse set of energy sources, and
limited dependence on fossil fuels. Lastly, the proximity of the
energy production and end user allows for the excess energy,
generally dissipated, produced during the power generation
process to be leveraged into a parallel heating/cooling cycle, thus
increasing the energy efficiency of the entire process.
While the concept of a distributed energy system and its
merits are easy to see, industry experience shows that effectively
designing such a system is a far more complicated task. Most such
systems fail to generate at their potential due to the lack of
appropriate configuration. The architecture design of a microgrid
is complex due its dependence on a number of project-specific
parameters such as stakeholder needs, resource availability,
existing legacy infrastructure, and demand among others.
The purpose of this paper is to study the use of a System
Architecture approach to designing a microgrid for Ithaca NY.
Such an approach involves examining the needs of the
stakeholders, determining system goals, selecting a concept, and
developing an architectural model, a mathematical construct that
is used to generate alternative architectures and evaluate their
cost, performance, and risk. The space of alternative architectures
is explored by means of a multi-objective evolutionary
optimization algorithm. Data mining and sensitivity analysis
algorithms are used to determine design features that are common
in good architectures. Finally, a small set of promising
architectures is selected.
Keywords—system architecture; microgrid; evolutionary
optimization; knowledge discovery.
I.INTRODUCTION
1. General Introduction
A microgrid is a system that generates, stores and transmits
electricity to local loads, which can function in parallel to, or
independently from the main grid. A microgrid generally
consists of a power generation unit, a power management
system, an energy storage system, and a utility connection. The
power generation unit is generally diversified into multiple
sources, both renewable and conventional. The power
management system is effectively the brain of the microgrid,
drawing power from different sources to meet the demand. The
storage system handles the fluctuations of the output, both in
terms of actual output and frequency fluctuations. And finally
the connection to the utilities enables the microgrid to exchange
power with the grid when required, drawing power when there
is a generation deficit, and supplying power when there is an
excess in production.
Microgrids are growing more popular both in the developed
and developing world because they represent a shift towards a
more energy independent, sustainable future [1]–[3]. As
mentioned earlier, they protect from a larger grid failure should
there be an extreme weather event or other technological issues.
A study conducted by the US Department of Energy estimated
that “sustained power interruptions” (over 5 minutes) cost the
US $25 billion to $70 billion annually [4].
A microgrid can technically be powered by any energy
source, but the general push [5][6][7] has been towards
incorporating renewable energy sources into the microgrid’s
capacity, despite the associated challenges. The sources
generally considered are solar, hydro, wind, geothermal, and
biomass among others. The choice of sources is extremely site
and project-specific. However, most renewable sources are
inherently intermittent. These technologies, therefore, typically
work in tandem with more conventional backup sources, like
diesel, natural gas, fuel cells etc., or battery storage to balance
out any fluctuations in their output.
2. New York prize & Ithaca
After the rolling blackouts that plagued New York State in
the wake of Hurricane Sandy, there has been a movement
towards decentralizing power generation. With its aging
infrastructure and growing complexity, the main grid has to
constantly undergo upgrades, making it expensive as well as
unreliable and unpredictable, especially against increasingly
common extreme weather phenomena [4]. In an effort to relax
the load on the main grid, as well as increase energy
independence in the state of New York, the New York prize, a
$40 million competition aimed at helping communities set up
independent energy systems, was founded.
The City of Ithaca, like many communities in New York
State, is vulnerable to grid-wide power outages. In addition,
Ithaca is highlighted in the New York Prize’s Finger Lakes
"Opportunity Zone" as an area where microgrids may reduce
utility system constraints and defer expensive infrastructure
investment costs.

of 8
In the event of a blackout, there are a number of services
that may be required to continue operating, especially in the
case of emergencies. High priority users like fire stations,
police stations, low income & senior housing, wastewater
treatment plants, etc. need to be able to function in the event of
a gird failure.
Having passed the first stage of the prize, the city of Ithaca,
led by the Ithaca Community Energy Group (ICE), must submit
possible designs for the layout and operations of their
microgrid.
3. Problem Statement and Goals
The main goal of this paper is to design at a high-level a
system to produce, store and distribute sufficient and consistent
power to certain vital amenities, in the Northern Energy district
of Ithaca, in the event of a main grid failure. The energy system,
or microgrid, should have a high penetration of renewables and
should minimize emissions. Finally, the microgrid should have
the lowest possible operational as well as capital cost. The paper
takes a Systems Architecture approach towards the design of
such a microgrid, which can provide uninterrupted & sustainable
power to high priority users, in an economically viable way.
4. Literature Review and Specific Research Goals
Many different approaches have been taken towards
designing microgrids or similar distributed energy systems.
Most approaches aim to optimize the design and operations of
the microgrid using mathematical models. Some of the models
developed optimize one metric (e.g., cost), subject to constraints
on other metrics (e.g., reliability) [8], [9], while others use multi-
objective optimization techniques [10], [11].
Hawkes & Leach used a linear programming model to
optimize Equivalent Annual Cost (EAC) for a given energy
demand [8]. EAC is a combination of electricity production, fuel
cost and maintenance cost. Arefifar at el used single objective
optimization to optimize the reliability of microgrids [9]. These
methods are limited as they only optimize one metric, thus
providing little insights about the trade-offs between different
metrics that are common in a system as complex as a microgrid.
Zhou used a two-stage stochastic programming approach to
optimize a distributed energy system [10]. The method
employed in that paper starts by deploying an energy system
superstructure. Once all possible configurations of the system
have been laid out, they develop an objective function that
consists of a deterministic term dealing with the design
decisions, and a stochastic term dealing with operational
decisions. Finally, they employ a genetic algorithm to solve the
two-stage stochastic optimization problem. Abdollahi &
Meratizaman used Multi-objective optimization [11] to optimize
three different aspects of a distributed energy system: exergetic
efficiency, total levelized cost & cost rate of environmental
impact. The paper also employs a genetic algorithm but
optimizes all three metrics at once. In general, most papers that
use the more desirable Multi-objective formulation are based on
a hypothetical energy system and do not consider important site-
specific issues such as stakeholder relationships.
The paper being presented deals with an actual real world
problem, thus the scope is wider than just optimization. The
paper uses the principles of Systems Architecture to address the
entire process of designing a micro grid. The process starts with
the examination of the different stakeholders, their needs and
relative importance. Based on the relative importance of the
stakeholders, the goals of the system were determined. The goals
were then used to define the metrics that informed concept
selection using an Analytic Hierarchy Process (AHP). This
concept determines the means of generation, distribution and
storage upon which the micro grid is to be based. Then, a two-
stage optimization is performed, using a genetic algorithm to
optimize capacity-independent design decisions in the first
stage, and brute force to optimize production and storage
decisions in the second stage. Dominant design features that are
consistently more present in good (e.g., non-dominated)
architectures than in poor architectures are identified using
association rule mining. Finally, a small set of promising
architectures is selected for further studies.
5. Paper Structure
The rest of the paper is organized as follows. Section II
describes the Stakeholder analysis for the microgrid. It provides
a list of stakeholders, and explores the needs of said stakeholders
and how they value they provide to each other. The value flows
are visualized in the form of a Stakeholder Value Network
(SVN), which is then used to identify value loops and finally
organize the stakeholders into order of importance. Section III
discusses the concept generation and selection processes used in
this paper. The section starts with the identification of possible
combinations of energy sources that could be used to power the
micro grid using morphological analysis[12]. Then the most
desirable mixes of energy sources were selected using an
Analytic Hierarchy Process[13]. Section IV explores the
architectural space generated by the main design decisions for
the system. This section describes the main design decisions, the
enumeration and evaluation of architectures that can be
generated from those decisions and finally the process of
optimization, including the genetic algorithm. Section V
discusses the results of the optimization, sensitivity analysis and
data mining processes, including the possible recommendations
of the design of the Ithaca city microgrid. Section VI states the
final conclusions of the paper and outlines some opportunities
for future work.
Table 1 – Value flow scoring system

of 8
II. STAKEHOLDER ANALYSIS
The first step in the systems architecting process is
identifying the stakeholders that will influence and in turn be
influenced by the system that is being designed. Once the
stakeholders have been identified, their needs are ascertained, in
order to develop the goals of the system. This is accomplished
by making a list of stakeholders and brainstorming their needs,
which are then projected to goals. The list of stakeholders for
this project is shown in Table 2.
The next step in the process is mapping out the connections
between the stakeholders into a Stakeholder Value Network
(SVN) [14]. The SVN for the Microgrid is displayed in Figure
1. It shows all the stakeholders of the system, along with the
value flows between each of them. In addition to being a useful
visualization for qualitative analysis, the SVN facilitates the
List of stakeholders Description Type Most Important need
Priority users
Hospitals, schools, emergency services, first responders etc, services that
are vital during blackouts or disasters
Beneficiaries
Clean, stable power, even if uncoupled from
main grid
ICE
Group of local citizens interested in working towards energy
independence for Ithaca.
Problem
Stakeholder
Minimum possible cost
IAWWTF
Plant treats waste water from region, already has a system of gas fired
micro turbines that generate power. They will be part of the priority
users, proposed site of micro-grid.
Stakeholder
Stable power, even if uncoupled from main
grid
Regulators Government regulators
Problem
stakeholder
project up to regulations
Local community
Local community who will benefit from the services provided by priority
users
Stakeholder Usage of services provided by priority users
Local government
City and town governments that will benefit from services of priority
users in cases of emergencies, also their support will be important in
setting project
Stakeholder
Services provided by priority users in case of
Emergencies
Suppliers
Businesses that will supply the equipment and raw materials for the
project
Stakeholder Business
External
consultants +Cornell
Consultant team from NY state, who are also trying to conduct a
feasibility study of micro-grid
Stakeholder Information
State government NY State government Stakeholder Project report
NYSEG Local energy provider Stakeholder Additional power to augment grid
NGOs Environmental NGOs Stakeholder clean power
Table 2 - Stakeholder and stakeholder needs
Figure 1 - Stakeholder Value Network (SVN)
Figure 2 - Stakeholder Value Network

of 8
identification of the most important stakeholders by means of a
quantitative methodology described in [15].
Once the value flows have been identified, each of them is
given a score, depending on two criteria, Supply Ranking &
Intensity of need. The Supply ranking was based on the ease of
finding alternate suppliers for the service or function, and the
intensity ranking maps the degree to which that particular good
or service is needed. The scoring system for value flows is
shown in Table 1. Scores for value loops are then obtained by
multiplying the scores of all flows in the loop, which penalizes
long value flows involving multiple stakeholders, as described
in [14].
1. Stakeholder value network
The network topology of the SVN is meshed, as opposed to
a typical hub-and-spoke configuration. This describes the
complex relationship between the 12 Stakeholders of the
network. Since the system affects a number of different
localities and towns, the needs, relationships and value
networks are very diverse and often interconnected.
The loop with the highest value is the one that contains only
the Project and ICE. This is a sensible result as ICE is the one
who has commissioned the project. The give and take between
ICE and the project will be significant. The second highest loop
has five stakeholders in it. This suggests that the value delivery
in the system is going to be governed by a complex set of sub-
requirements [15]. The rest of the top five loops contain the
highest level stakeholders, namely the priority users, local
community and the Waste Water Treatment Plant. This shows
a large degree of interconnectedness between the stakeholders,
where many stakeholder needs can be fulfilled by other
stakeholders rather than the system itself.
Once all the value loops were identified, the weighted
stakeholder occurrence, , was calculated as the sum of the
scores of all loops in which the stakeholder is present,
normalized by the sum of the scores of all loops, using equation
1 . The hierarchy of stakeholders based on importance is shown
in Table 3.
=
∑
∑
(1)
The most important stakeholder is in fact the local
community. This suggests that the priority customers should be
determined by understanding the needs of the local community,
since the needs of the local communities are directly related to
the services that are allowed to function in the event of a main
grid failure.
Based on the needs of the main stakeholders, the system
should be able to provide adequate, stable and clean power at
the lowest possible investment for services that are required by
the community. The services that were found to be essential for
the community were the police station, fire department, two
high schools which could be used as shelter during an
emergency, the waste water plant, drinking water plant and
finally a number of buildings housing low income households
and the elderly.
III. CONCEPT EXPLORATION AND SELECTION
After the analysis of the stakeholders and the assessment of
their needs, five different concepts for the power production side
of the microgrid were generated. The concepts are the highest
level system descriptions, essentially given by the possible
generation methods without delving into quantitative details.
These concepts were generated using a morphological matrix,
not reproduced here for brevity [16].
1) Solar panels, biogas and a backup of natural gas: This is
the concept currently presented by ICE for the NY Prize
application. It consists of two renewable resources (solar
and biomass) and a more secure, though relatively clean
source (natural gas). Furthermore, the waste water plant
already owns a bio-digester, whose capacity is three times
more than the current load.
2) Coal plant: Even if this concept is polluting and not
particularly suited for a microgrid, a coal power plant was
considered among the options mainly because of the very
low price of coal.
3) Storage only: This concept considers installing a mix of
different types of batteries and possibly a pumped water
storage system that will be charged when the microgrid is
connected to the national grid and will provide power to the
users if disconnected. Therefore, no real power production
unit is considered.
4) Hydroelectric, solar panels, and wind: This concept is
entirely based on non-emitting sources. While Solar and
wind are inherently intermittent, due to the clean nature of
this solution, it is worth exploring.
5) Distributed solar panels: This concept is based on the
“community microgrid” concept, where the solar panels are
not all located at a central generation hub, but rather are
distributed all over the city of Ithaca. One of the major
advantages of this concept is that it makes the acquisition of
land for setting up a central PV hub unnecessary.
Five concepts are available and must be evaluated according
to a number of criteria. The criteria used for the concept
selection are as follows:
Rank 1 2 3 4 5 6 7 8 9 10 11
Stakeholder
Local
community
Priority
users
Local
government
waste
water
plant
ICE Suppliers
State
government
NGOs
External
consultants
NYSEG Regulators
Weighted
stakeholder
occurrence
0.4587 0.4538 0.414 0.3655 0.316 0.068 0.0554 0.0404 0.0044 0.0042 0
Table 3 - Stakeholder importance table

of 8
1. Availability: This criterion represents the percentage of time
the microgrid can supply sufficient power. In emergencies
this is particularly important, as without the national grid as
a backup there is a need for a stable output of electricity.
2. Cost/revenues: Economic feasibility of the project is an
important criterion for the stakeholders involved.
3. Emissions: Given the current attention to climate change and
pollution, the emissions of the microgrid play a significant
role. Moreover, the central hub of the grid will be very close
to where the local community lives, and they will require a
certain standard of environmental sustainability.
4. Impact on the landscape: Though possibly less important
than other criteria, the impact the microgrid will have on the
landscape may vary between the different concepts, and will
therefore be considered.
5. Limitations on maximum capacity: Not all sources can
produce the same amount of power. This depends mainly on
the region for solar, wind and hydropower.
The Analytical Hierarchy Process, also known as AHP[13],
was used to evaluate all the concepts based on the criteria listed
above. The AHP confirmed the first impression that the current
concept (solar panels, biomass and natural gas) is the most
suitable for satisfying the needs of the stakeholders. Thus, the
architecture space exploration presented in the next section
explores this concept in further detail.
IV. ARCHITECTURE SPACE EXPLORATION
1. Formulation and enumeration
The architectural decisions can be divided into three subsets
of decisions: the first subset concerns the ratio between the
energy generated by solar panels and by the gas-fired unit; the
second subset is made of decisions regarding the technologies
used; finally, the last subset is about the size and the amount of
power produced by the microgrid.
Ratio between solar panels and gas fired unit
There are four decisions regarding this ratio and each one
represents the percentage of power generated by solar panels in
a specific quarter of the year. It has been decided to divide the
year into four quarters as it is a good balance between a monthly
division, which has a higher resolution but is more demanding
in terms of computational power, and a very low resolution of
just a single term for the whole year.
For each decision, 10 options are available (from 0% PV to
90% PV) as, given the high instability of the supply of energy
from solar panels, at least 10% of gas production is needed to
meet the variations of the demand.
Technologies
Four decisions have to be made regarding technologies used.
Fuel: Two options are available for the fuel: biogas, deriving
from an aerobic bio-digester, and Syngas, produced with the
gasification of the biomass. They both derive from biomass, but
the process the biomass undergoes is different, therefore a
different infrastructure to treat the biomass would be needed.
Power production unit: Five options are available for the power
production unit: Standard gas turbine, Steam injection gas
turbine (STIG), Micro gas turbines, Standard reciprocating
engine and Stirling engines. They are very different from each
other in terms of cost, emissions and size.
Size of the storage: The different options for this decision have
been narrowed down to three options: 2MWh, 6MWh, and
10MWh. A larger storage system would allow more
independence of the system in case of blackout (both of the
microgrid and of the national grid), but will be significantly
more expensive. The storage sizes were determined based on
modularity of available battery types:
2 MWh: this battery can meet the average load of the priority
users (2MW) for one hour. Therefore, it is not intended to
be a backup, but just a way to stabilize the output to meet
the demand variations.
6 MWh: this battery can meet the average load for three
hours. It can stabilize the output and be a backup for a short
amount of time.
10 MWh: this battery can meet the average load for 5 hours.
It can be used to stabilize the output and be a backup for a
medium amount of time.
Storage system: The number of options for this decision has
been narrowed to two systems: flywheel, a mechanical storage
system that consists of a spinning object with high momentum;
and Aqueous hybrid ion storage (AHI™), a chemical storage
technology. The size of the storage can also be divided equally
between more than one storage system. The two methods of
storage considered are flywheels and AHI™. Even if hard to
scale up, the flywheel is very efficient in meeting the high
frequency variations in the demand. The AHI™ is very mature
and reliable albeit expensive.
Size of the microgrid
As previously done with the decisions regarding the mix of
the energy sources, the year has been divided in four quarters
and each decision represents the percentage of power produced
out of the maximum demand of that quarter.
For each decision, 11 options are available (from 0% to
100%). The assumption made here is that 0% represents the
minimum generation requirement to meet the needs of the
priority users (2MW), while 100% is the generation required to
meet the total needs of the area, including non-priority users
such as housing complexes, shops etc. (6MW). Each of the
options is 0.4MW greater than the previous one.
In total, there are 12 independent decisions, therefore the
architectures can be represented by an integer vector of length
12. The size of the architectural space can be computed by
multiplying the number of options for each decision; 10 ⋅ 10 ⋅
10 ⋅ 10 ⋅ 2 ⋅ 5 ⋅ 3 ⋅ 3 ⋅ 11 ⋅ 11 ⋅ 11 ⋅ 11 = 13,176,900,000
2. Evaluation
The large size of the architecture space makes solving the
global optimization problem by brute force (i.e., full factorial
evaluation) impossible. However, we can exploit the structure
of the global problem and decompose it into two decoupled

of 8
optimization problems. The first optimization problem will
focus on optimizing metrics per unit energy/power. Then, the
optimal architecture fragments (i.e., those on the Pareto front)
will be combined with the full-factorial enumeration of the
options for the last five decisions (which represent the total
energy production and storage) and optimized. By doing this,
the first evaluation will have to deal with only 10 ⋅ 2 ⋅ 5 ⋅ 3 =
300,000 architectures, and the results still reflect the true global
optima thanks to the decoupled structure of the problem. The
size of the second architectural space will depend on the size of
the Pareto front from the first optimization. The flow of inputs
and outputs of the optimizations is shown in Figure 4.
3. Optimization of capacity-independent metrics.
In the first optimization problem, all the metrics are
completely unrelated to the size of the microgrid. Therefore, this
optimization represents architectures as integer vectors of length
7.
1. Operating cost/kWh produced: This metric is the cost of
producing one kWh. It is the sum of the cost of the fuel, the
operating costs of the power production unit and the
operating costs of the PV. It depends not only on the
technologies of choice, but also on the mix of gas/PV
chosen for each quarter of year. Even if the main purpose
of this microgrid is to provide sustainable power, especially
during emergencies, the cost of this power is very important
to the stakeholders. The Operation cost/kWh,
was calculated by using equation 2, where %Gas, %PV are
the percentages of Gas and PV, is the operation
cost/kWh of gas engines, and is the operation
cost/kWh for solar. The operational cost was then averaged
across all four quarters to get the final Operational cost for
each architecture .
= % . ( + )
= % . (2)
2. Capital cost/kW installed: This metric is the cost of
installing one kW. It is the sum of the costs of the fuel
processing unit, , the cost of the power production
unit, and of the solar panels arrays, .
Just like the operating costs, it depends on both the
technologies chosen and the mix of sources and it represents
an important objective for the stakeholders. The capital cost
of each architecture was calculated using equation 3.
=
%
∗ 8760
∗
=
%
∗ 8760
∗ ( + )
= + (3)
3. Emissions/kWh produced: This metric represents the grams
of pollutants emitted by the microgrid per kWh produced.
As one of the main needs of the stakeholders is to have an
environmentally sustainable power source, this metric is
extremely important. Emission/kWh was calculated by
equation 4, using the quarterly percentage of gas
production, , and Emissions/kWh of Gas
engine, .
=
∑ %
∗ (4)
4. Risk metrics: The risk metrics for this system were divided
into two separate metrics; Risk metric 1, that measures
reliability of the system providing some power vs no power,
and Risk Metric 2, a metric that measures how reliably the
system provides the amount of power needed.
The mean time to failure was used as the primary data
for the calculation of reliability. Many of the components in
the system should typically have a Weibull function for the
failure rate, however the assumption was made that the time
horizon for the reliability calculations, 20 years, falls within
the “useful” life of the component, and hence the failure rate
can be assumed to be constant.
Figure 4 shows the layout of the microgrid. The system consists
of 2 sources of generation (gas powered & Photovoltaic) with
their respective control system C1 (e.g. ABBs gas generator
controller, MGC600G) and C2 (e.g. ABBs PV controller,
MGC600P), the main control systems (e.g. ABBs central
control system, MGC600N with a complementary storage
system controller) and finally the power storage B (e.g. either
ABBs flywheel based PowerStore™ and/or Aqueous Hybrid
Ion(AHI) Energy storage). After obtaining the mean time to
failure of the components of the system, the figures were
converted into failure rate by taking the reciprocals and then
using the power law, calculated the reliability (R = e-λt
, where
λ = failure rate). The risk metrics are then calculated by
Figure 3 - System network chart
Figure 4 – Flow chart for optimization process

of 8
identifying the system minimal cut sets, computing the
switching function and replacing the indicator variables with
the unreliabilities [18]. Based on the system layout shown in
Figure 3, the following are the minimal cut sets for each of the
metrics.
Risk Metric 1 – represents the reliability of the system when
the failure is represented by 0 power output to users.
• Min cut sets:
{ 1, 2}, { 3}, { , }, { }, { , 2}, { , 2}, { 1, }
Risk Metric 2 – represents the reliability of system when the
failure is represented by insufficient power output to users (this
includes zero output)
• Min cut sets:
{ 1}. { 2}, { 3}, { }, { }, { }, { }
Given the following minimal cut sets the reliability of the
system, was calculated by first calculating the probability
of failure of each minimal cut set, using equation 5, and
then the reliability of the system using equation 6.
= ∏(1 − ) (5)
= ∏(1 − ) (6)
1) Optimization of capacity-independent decisions
The first optimization was performed using NSGA-II, a
multi-objective genetic algorithm [19]. For the first optimization
the first 7 decisions were considered(% of PV for the 4 quarters,
fuel source, gas engine type and types of battery). An initial
population of 150 architectures selected by means of an
orthogonal array was used. The fitness function was developed
so that it would favor low values of emissions, capex, opex, and
high values of the two risk metrics. The genetic algorithm was
run until the change in average spread in the Pareto front with
every successive iteration fell below 1
The final results of the genetic algorithm for the first
optimization yielded 80 non-dominated architectures. Due to the
varied nature of the architectures, the data mining and sensitivity
analyses conducted at this stage did not reveal much insight and
are therefore not discussed in this paper. The set of 80
architectures were used as an input into the second optimization.
2) Optimization of system capacity.
The architectural array that was the output of the first
optimization contained only 7 of the 12 decisions in the final
architectural array. The first step in making it compatible for the
second optimization was the addition of the last 5 decisions
(battery storage size, and % of max production per quarter) that
had to do with the size of the microgrid. Using the first 80
architectures from the Pareto front of the first optimization, the
total set of 3,513,840 architectures that formed the trade-space
for the second optimization were produced using a full factorial
enumeration including the last 4 decisions, each with 11 possible
values.
The second optimization deals mainly with the size of the
microgrid, the generation and the size of the storage. The main
set of metrics that were used to evaluate the architectures in the
second optimization was as follows:
1. Net present Value: As per the stakeholder needs, one of the
main goals was maximizing return on investment. The
method used in this optimization to quantify this return on
investment is Net present value. Given that ICE, the
organization that is going to be installing and running the
microgrid, is not a profit driven organization, a modest
discount rate of 5% was selected. The lifecycle of the
project was assumed to be 20 years.
The assumption was made that, when the microgrid
operated in parallel, all the power produced can either be
sold to the main grid, or can be directed to the users directly,
in lieu of main grid power. Based on this assumption the
revenues were calculated using equation 7.
= ∗ (8)
Given these revenues and the discount rate (D), the NPV
was can be calculated by considering the net cash flow,
using equation 8.
ℎ = − −
= − + ∑
( )
(7)
2. Risk metrics: The risk metrics used in the second
optimization are the same used in the first optimization.
3. Total yearly emissions: One of the stakeholder needs was
to produce a microgrid that was clean. This metric is used
to evaluate the total annual emission for each of the
different architectures. The calculation of this metric is a
fairly simple process. Given the total production annually,
the total yearly emission figure can be
determined by equation 9.
= ∗ ∗ 8760 (8)
4. Robustness: This metric is a subjective metric that is used
to evaluate the storage sub-system. The metric shows the
system’s ability to provide power in the event of failure of
the power generation sub-system. This metric combines the
reliability of the energy storage sub-system with the amount
of power the storage system can provide. Reliability
considers the number of batteries used, while the amount of
power the storage system can provide depends on its size.
Thus this metric is determined by a function that takes into
account the size of the battery storage as well as the number
of different types of batteries. A system with more batteries
and more storage capacity is given a higher score, while a
system with lower capacity and a single battery is given a
lower score. This metric is important as the system should
be robust enough to provide power in the event of
temporary shutdown of the generation infrastructure in the
event of an emergency.

of 8
The second optimization was performed by brute force
applying non-dominated sorting on the complete tradespace of
3,513,840 architectures. The resulting Pareto front contained
684 architectures. The given architectures were examined using
data mining techniques.
V. ARCHITECTURE SELECTION AND DATA MINING
By observing the 684 architectures, certain common features
were determined and studied using data mining. The most
dominant feature is the maximization of PV as well as
production in the 2nd
and 3rd
quarters (spring and summer). This
is a sensible result as with the increased levels of sunlight, the
production of PV increases in efficiency. As a result, the overall
production can be increased without increasing operating costs,
thus increasing cash flow, and hence the NVP. The use of a bio-
digester, a Flywheel and 6MW of storage were also seen to be
common features among the architectures on the Pareto front.
Additionally, when the architectures maximizing each of the
metrics were studied, the following trends emerged. The
maximization of NPV was achieved by maximized production
in all 4 quarters, maximized use of PV in the 2nd
and 3rd
quarters,
and use of a bio-digester. Architectures minimizing emissions
generally maximized PV, while minimizing overall production.
These architectures also had a higher battery storage. Maximum
battery storage with both types of storage featured in those
architectures that maximized robustness. For the most reliable
architectures, both in terms of resilience to blackouts (risk metric
1) and brownouts (risk metric 2), microturbines and
reciprocating engines were the preferred prime movers.
VI. CONCLUSION
A System Architecture model for a microgrid has been
constructed and presented. The needs of the stakeholders were
identified and based on them the goals and metrics of the
microgrid were determined. The grid was optimized in a two
stage optimization, using a NSGA II genetic algorithm and
brute-force non-dominated sorting. The results were analysed
using data mining and sensitivity analysis. The results show the
capability of the presented method to model and optimize a
microgrid based on the needs of the stakeholders. As each
microgrid is very site-specific and is dependent of the needs of
the associated stakeholders, this method can be applied to any
microgrid, anywhere in the world.
Although the final number of architectures on the Pareto
front is quite large, the number of feasible architectures can be
further narrowed down by introducing constraints. For example,
the NY state price has a fixed amount of money to dispense, this
will definitely place a cap on budget. Similarly, environmental
and emission laws can be used to limit the amount of emissions.
One of the main drawbacks of this model is the decoupling
of cost, performance, emission etc. per unit energy and the total
size of the system. This does not take into account economies of
scale of varying reliability among other aspects. Additionally,
the demand is taken to be fixed over the lifespan of the project,
which is a simplification. The fidelity of the model can be further
improved by introducing variable demand based on load
forecasting. Thus, the next step in this modelling process would
be to introduce economies of scale, as well as variable load
forecasting.
VII. REFERENCES
[1] B. Ki-Moon, “Sustainable energy for all,” A Vis. Statement. New
York, United Nations, no. april, 2011.
[2] H. Jiayi, J. Chuanwen, and X. Rong, “A review on distributed
energy resources and MicroGrid,” Renew. Sustain. Energy Rev., vol.
12, no. 9, pp. 2465–2476, 2008.
[3] C. Marnay and G. Venkataramana, “Microgrids in the Evolving
Electricity Generation and Delivery Infrastructure,” Contract, no.
February, 2006.
[4] Executive Office of the President, “Economic benefits of increasing
electric grid resilience to weather outages,” no. August, 2013.
[5] A. Yadoo and H. Cruickshank, “The role for low carbon
electrification technologies in poverty reduction and climate change
strategies: A focus on renewable energy mini-grids with case
studies in Nepal, Peru and Kenya,” Energy Policy, vol. 42, pp. 591–
602, 2012.
[6] D. Lee, J. Park, H. Shin, Y. Choi, H. Lee, and J. Choi, “Microgrid
village design with renewable energy resources and its economic
feasibility evaluation,” Transm. Distrib. Conf. Expo. Asia Pacific, T
D Asia 2009, pp. 1–4, 2009.
[7] C. Pappas, C. Karakosta, V. Marinakis, and J. Psarras, “A
comparison of electricity production technologies in terms of
sustainable development,” vol. 64, pp. 626–632, 2012.
[8] A. D. Hawkes and M. A. Leach, “Modelling high level system
design and unit commitment for a microgrid,” vol. 86, pp. 1253–
1265, 2009.
[9] S. A. Arefifar, Y. a R. I. Mohamed, and T. H. M. El-Fouly,
“Optimum microgrid design for enhancing reliability and supply-
security,” IEEE Trans. Smart Grid, vol. 4, no. 3, pp. 1567–1575,
2013.
[10] Z. Zhou, J. Zhang, P. Liu, Z. Li, M. C. Georgiadis, and E. N.
Pistikopoulos, “A two-stage stochastic programming model for the
optimal design of distributed energy systems,” Appl. Energy, vol.
103, no. January, pp. 135–144, 2013.
[11] G. Abdollahi and M. Meratizaman, “Multi-objective approach in
thermoenvironomic optimization of a small-scale distributed CCHP
system with risk analysis,” Energy Build., vol. 43, no. 11, pp. 3144–
3153, 2011.
[12] F. Zwicky, “The Morphological Approach to Discovery, Invention,
Research and Construction,” 1967.
[13] T.L. Saaty, “Decision Making with the Analytic Hierarchy
Process,” Int. J. Serv. Sci., vol. 1, no. 1, pp. 83–98, 2008.
[14] B. G. Cameron, E. F. Crawley, G. Loureiro, and E. S. Rebentisch,
“Value flow mapping: Using networks to inform stakeholder
analysis,” Acta Astronaut., vol. 62, no. 4–5, pp. 324–333, 2008.
[15] T. a Sutherland, “Stakeholder Value Network Analysis for Space-
based Earth Observations,” 2009.
[16] F. Zwicky, “The Morphological Approach to Discovery, Invention,
Research and Construction,” pp. 273–297.
[17] E. Crawley, B. Cameron, and D. Selva, “Systems Architecture:
Strategy and Product Development for Complex Systems,” Prentice
Hall, 2015.
[18] R. Billinton and C. Singh, “System reliability modelling and
evaluation,” Microelectron. Reliab., vol. 17, no. 2, p. 271, 1978.
[19] K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, “A fast and elitist
multiobjective genetic algorithm: NSGA-II,” IEEE Trans. Evol.
Comput., vol. 6, no. 2, pp. 182–197, Apr. 2002.

Microgrid paper camera copy final draft

More Related Content

What's hot (20)

Similar to Microgrid paper camera copy final draft (20)

Microgrid paper camera copy final draft