3rd International Hybrid Power Systems Workshop | Tenerife, Spain | 08 – 09 May 2018

Sizing and Optimization of Hybrid Mini-Grids

with micrOgridS - an Open-Source Modelling Tool
Sarah Berendes ∗ , Paul Bertheau, Philipp Blechinger
Reiner Lemoine Institut
Berlin, Germany 12489
∗ Email: sarah.berendes@rl-institut.de
∗ Telephone: +49 (0) 30 1208 434 - 42

Abstract—The transformation towards a sustainable, afford- to that, HMG are found the most cost effective option for
able and reliable energy system remains a challenging task for many island power supply systems [5] due to the drastic
many island communities. The electric power supply through decrease in mature RE technology cost, especially for PV,
Hybrid Mini-Grids (HMG), containing RE technologies and
energy storage systems in the mix, represent a cost-effective and and storage capacity cost [6]. As a result, many PICT
fuel saving option. In the planning phase of HMG, software- governments have defined ambitious RE targets within their
based sizing tools have become necessary to map the complex Nationally Determined Contributions (NDCs). For example,
interactions between the different sources of electricity. In this the kingdom of Tonga, of which oil imports accounted for
study we identify the requirements for such a software tool, 18 % of the GDP in 2015 [7], agreed on the implementation
and investigate on the advantages of open source modelling.
Based on our findings, the open source model micrOgridS was of 50 % RE until 2020 and 70 % in 2030 [8].
developed by applying the Open Energy Modelling Framework One of the challenging aspect that arise with the
(Oemof). As a result, we introduce the model in this paper and integration of RE technology is an increased uncertainty
show its performance via a case study by comparing it with associated with the planning and forecasting resulting from
the software tool Homer R Pro 3.11. Drawing on the results, the interrupted temporal availability of the natural renewable
we conclude that with micrOgridS, we provide a valid basis
for a open source HMG sizing tool. resources. RE in combination with diesel generator sets and
storage options lead to a complex system of interactions that
I. I NTRODUCTION has to be described and optimized (technically, economically
In the Sustainable Development Goal # 7 (SDG7) and environmentally) within the planning phase of HMG.
the United Nations agreed to assure universal access to Consequently, software based sizing tools become crucial
affordable, reliable and modern energy by 2030 [1]. Hence, to deal with the increased complexity of interactions
a transformation of the existing, predominately fossil-based through RE technology, provide briefings for decision
global energy system towards higher shares of renewable making by calculating a optimized set of solutions, and to
energy (RE) is required. Looking at the global electricity develop scenarios to energy transition pathways that play
generation in 2015, fossil energy sources accounted for a key role for policy making and project implementation [9].
66.3 % (39.3 % coal, 22.9 % natural gas and 4.1 % oil) of
the 24,255 TWh of total generated electricity [2], whereas As political decision making is mostly closely tight to
RE accounted for only 23.1 %. An even higher share of public participation, Morrison points out to open source
fossil-based electricity generation is widely obtained in energy models as being notably valuable due to such models
remote regions with low economic capabilities, such as the can increase transparency and public trust [10]. Following an
Small Island Developing States (SIDS). Due to the fact that argumentation on open source energy models in the context
for many of island communities the conventional approach of scientific research by [11], it is stated that proprietary
of grid extension is unprofitable, the energy supply on software tools are inapplicable as the scientific standards,
islands is widely realized through a set of diesel generation such as transparency and reproducibility can only be met
units that operate a decentralized and isolated distribution by open source software tools. Drawing on that, possible
grid. For example, referring to the Pacific Islands Countries advantages that open source tools could impart in energy
and Territories (PICT), as a subcategory of SIDS, these system planning and operation management of HMG are: an
regions are marked by a strong dependency on imported increased transparency in scientific electrification pathway
petroleum goods, a costly energy infrastructure and a high studies, a reduction of barriers associated with bottom up
threat towards effects of climate change [3]. planning resulting from high license cost, foster inclusive
planning approaches though collaborative modelling and
The introduction of additional power generation sources, community-based model-development. This supports both
preferrably renewable based, into diesel powered grids is public and private decision makers in convincing all involved
called hybridization. Hybridization is considered successful stakeholders, for example the local community, financing
in case the specific fuel consumption of a mini-grid can institutions, and regulatory authorities, on the validity of the
be reduced [4]. Consequently, further advantages can be planned hybridization projects and strategies.
achieved including lower greenhouse gas emissions (GHG) Currently, there is a lack of sufficient open-source soft-
as well as a decreasing dependency on oil price volatility, ware tools for mini-grid sizing in the global modelling
and a shift towards modern energy technologies. In addition community. The most commonly applied tool is Homer R
 3rd International Hybrid Power Systems Workshop | Tenerife, Spain | 08 – 09 May 2018

TABLE I
C OMPONENT- SPECIFIC REQUIREMENTS FOR A SUFFICIENT SOFTWARE B. Tool design based on the Open Energy Modelling Frame-
TOOL ( STAKEHOLDER WORKSHOP FINDINGS ) work (Oemof)

high priority medium priority low priority

The Open Energy Modelling Framework is a generic,
solar PV wind turbine geothermal component
open-source toolbox that holds a range of useful functions
generic energy source biomass tidal sources
to describe and optimize energy systems. By calling it
hydro-turbines
generic, we mean that Oemof is not programmed for specific
Lead-Acid BSS thermal storage Redox-Flow BSS applications that follow one specific mathematical approach.
Li-ion BSS fuel cell high temperature BSS Furthermore it can be utilized for various optimization tasks.
generic storage pumped hydro storage Omeof’s developer community is embedded in the Open En-
ergy Modelling Initiative, which follows strict open-source,
controller strategies grid open-data and open-science policies. Oemof is programmed
multiple DGs in the object-oriented programming language Python, and its
inverters development follows scientific standards.
AC- and DC coupling To describe an energy system in Oemof, a graph based
approach is used in which energy system components are,
basically, represented by N odes-objects that are connected
Pro but it is a commercial software and the code is not open. via F low-objects. Furthermore, the N odes are classified into
Thus we elaborate the requirements for a comprehensive subclasses, including Sources, Sinks GenericStorage. This
open-source software tool for the purpose of optimized HMG structure allows to separate the energy system model from
system sizing and operation management in this work. On the mathematical parameterization and the solving process.
the basis of our findings the open-source tool micrOgridS Following the description in energy system modelling with
was developed and validated. In this study the basic approach Oemof, the paramterization of the system components is
behind the tool is introduced and analyzed with respect to mainly done by adding attributes to the F low-objects. Other
the defined requirements as well as a case study of a Pacific aspects to considered Oemof as generic are represented by its
Island energy system is shown. generic objects (subclasses of the N odes) that can be used
to describe various system components as well as no units
are associated within the energy system model. Drawing on
II. M ETHODS that, we assume an increased modularity and flexibility in
A. Requirements for a comprehensive software tool the modelling practise. Taking an example of describing a
PV system that supplies direct current on the DC-bus, in
A stakeholder workshop was conducted to determine the that case both the PV system and the DC bus would be
requirements for a software sizing tool for the purpose of described as N odes that are connected via directed F lows.
optimal HMG design. Together with nine micro-grid experts The F low holds various attributes, such as actual value
coming from academic institutions, as well as from private (series, gen. specific power output) , nominal capacity (gen.
companies the requirements were elaborated. Results show rated capacity) , variable cost (gen. specific cost per rated
that a comprehensive software tool for optimal HMG design, capacity), fixed (True means that the output is determined by
should be capable of: the actual value and is not variable). In the given example
• system sizing and identification of optimal operational the value of the flow would be interpreted as the power
strategies output of the PV plant, and could therefore be written as:
• multi-objective optimization (cost of energy, emissions,
flow.value(t) = flow.actual value(t) · flow.nominal value
capacity shortage, RE share)
(1)
• adjustable time resolution (15 minutes to hourly time
increments)
• grid stability functions Ppv (t) = ppv (t) · Ppv,rated (2)
• modelling different HMG components (high-, medium- Following this approach, a set of equations is generated
and low priority) that represent the graph-based model of the energy system.
• secondary functionality (couple with resource data and The energy system model can than be transformation into
load profiles, update uncertain parameters over the an optimization problem by adding it to a Model-object
course of optimization horizon). and defining the optimization horizon (also referred to as
In Table I the required component models are depicted prediction horizon PH). Throughout this step an objective
and characterized according to the resulting prioritization, function representing the sum of all attributed costs is
that is drawn from the workshop results. Following the generated, that has to be minimized, is initialized:
discussion on requirements for a comprehensive tool, it is
found that applicability of a tool would increase if the results min cT · x (3)
are validated and if easy-to-use capabilities are integrated in
the software tool. In addition to that, a modular structure where c describe the cost and x a set of decision variables.
is preferred by the stakeholder group, that incorporate the Once the energy system model is described and the
possibility to extend and link the tool to other use cases, optimization task is set up by defining the Model, the
modelling libraries and databases. problem then can be solved. To achieve this, four solvers
 3rd International Hybrid Power Systems Workshop | Tenerife, Spain | 08 – 09 May 2018

class
Node 1 Flow Node 2 DC-bus balance:

0 =Ppv (t) + Pbss,out (t) − Pbss,in (t) − Pinv,in (t)

graph model
DC- (5)
PV
Bus AC-bus balance:
dg
Node1 (…) Flow(…)
X
0 =Pinv,out (t) − Pload (t) − Pex (t) + Pdg (t)
Label ‚PV‘

object attributes
actual value [0.3,0.5,0.98] 0
Outputs Node 2 kW/kWp (6)
for ∀dg
nominal value 200 kWp
Node2 (…)

Label ‚DC-Bus‘
variable_costs 0 USD/kWp Fuel bus balance:
Inputs Node 1 fixed True dg
X
0 =f uelsource (t) − f ueldg,in (t) for ∀dg (7)
Fig. 1. Simplified schematic model of a PV system contributing to the
DC-bus Due to the generic character of Oemof, no units are
assigned to the Flows by default. Moreover, the user is
requested to determine the units and to ascertain their
are linked with Oemof, namely Coin-OR (CBC), Glpk, consistency. For the micrOgridS model, all Flows are
Gurobi R , Cplex that can be utilized [12]. regarded as power flows in kW, except the Flows being
connected to the Bus-object representing the Fuel bus.
III. R ESULTS These Flows represent actual fuel flows and are therefore
given in Liters.
Considering a set of simplifications and assumptions,
the micrOgridS tool was developed to support the project
development of a PV-diesel-battery HMG by providing both In analogy to Oemof, the model formulation is pro-
optimized sizing of system components and optimized dis- grammed in Python. At the moment the application of
patch of power flows. This is achieved by minimizing a micrOgridS is suitable for sizing of PV and a BSS. Thus,
single objective function representing the total cost of the sizing capabilities for DGs are not integrated due to their
system T C, as shown in (4), mathematical formulation result in nonconvex bound con-
straints (or integrality constraints), which can yet not be
X integrated in combination with sizing parameters of DG
min T C = (CCj + OP EXj ) (4) representing a decision variable. Further components such
j as loads (including aggregated appliances), inverters, AC and
X DC buses as well as a fuel commodity bus, and a dump for
+ AW C + F Cdg for ∀j ,∀dg
excess or surplus power flows, can also be modelled with
dg
micrOgridS.
where CCj are the component cost, and OP EXj The modelled power supply system is solved for an ad-
represent the operation and maintenance cost for all system justable number of time steps, and with respect to a set of
components j, respectively; AW C are wear costs associated described constraints including:
with the BSS; and F Cdg represent fuel cost for all diesel • minimum and maximum power flow constraints asso-
generator components dg. ciated with the dispatch of DGs
• spinning reserve provided by DGs or BSS
In Fig. 2 the schematic illustration of the energy system • rotating mass, which describes a defined share of load
model is depicted as it is implemented in micrOgridS. It that has to be provided by accelerating power genera-
is shown that the energy system model is composed of the tion units (in this case referred to DG) and BSS
following components : • a set that regulates generator order
• a DG set including 3 units (DG1, DG2, DG3) Drawing on the possibility to calculate a optimized dis-
• a PV system patch of power flows, comprehensive operational strategies
• a battery storage system can be identified by the practitioner. Calculation times for
• an inverter unit one reference year (of 8760 time increments) range from
• an aggregation of loads minutes to hours which opes up the debate upon the difficulty
• a dump for excess energy between describing the system’s behaviour in a sufficient
• a fuel source level of detail, and assuring applicability through acceptable
Each component is connected to one of three Busses via computational times. However it has to be mentioned that
Flows. The Busses can be classified into a DC-, an AC- and the computational time achieved is strongly dependent on
a Fuel bus, respectively. Therefore, the following balance the hardware applied and the solver utilized. Hence, its
equations can be derived for each Bus that have to be met quantification represents not always the global optimum but
at every single time step of the optimization. we found results within close range to this optimum.
 3rd International Hybrid Power Systems Workshop | Tenerife, Spain | 08 – 09 May 2018

Load

Excess
PV

DG1
Battery Storage

Inverter
DG2
Fuel source

DG3

DC-Bus AC-Bus Fuel Bus

Fig. 2. Schematic overview of the Energy System Model

TABLE III
A. Application example of micrOgridS I NPUT PARAMETERS : C APACITY OF DG AND COST OF SYSTEM
COMPONENTS
The micrOrgridS model was tested in a case study for
parameter unit PV BSS DG1=DG2 DG3
the Pacific island Lifuka, Tonga, located at latitude = - installed capacity kW free free 186 320
19.814545◦ and longitude = -174.35049◦ . In the case study CAP EX U SD
kW
2500 300 5002
the model was thoroughly analyzed and compared to results U SD U SD
O&M kW
, kW hop
2.53 3.88 0.022
achieved with the software Homer R Pro 3.11. The project lifetime yrs 20 10 20
lifetime is estimated to 20 years and the assumed WACC
amounts to 0.094. Considering input data used in this case
study, most of the values are derived from project specific Additional input parameters are attributed to the battery,
values of the Outer Island Renewable Energy Project. Hence, such as the round-trip efficiency of 85 % and a constant
load profiles show an overall annual energy demand of 1.35 C-rate of 0.546 h1 . The input parameters were used analo-
GWha−1 , a peak load of 236 kW and an average load of 154 gously in both models, micrOgridS and Homer R . Following
kW. In Table II the resource data are depicted that is used this, the models were solved according to lowest power
as an input to the model comparison. Hence, the Clearness generation costs and results are achieved and compared,
Index k, the global horizontal irradiance GHI as well as as illustrated in Table IV. From there, the resulting PV
the ambient temperature Tamb and the average wind speed Pr,PV and BSS capacities Pr,BSS as well as levelized cost
vwind are presented. of electricity LCOE, the RE share, the consumed fuel
and the excess energy can be obtained. It can be seen
TABLE II that the micrOgridS optimization show lower values for
W EATHER DATA SET FOR L IFUKA I SLAND 1 all compared parameters. The highest absolute deviation is
k GHI Tamb vwind found in the BSS size (-392 kW, which equals 46.9 %
◦
kW h C m relative deviation) and excess energy (-20,739 which equals
m2 day day s
Jan 0.574 6.69 25.81 6.34 35.7% relative deviation), respectively. We assumed that
Feb 0.566 6.3 26.4 6.18 one explanation for the lower values can be found in the
Mar 0.559 5.62 26.24 6.14 optimized operational strategy, which is another results of the
Apr 0.543 4.65 25.56 7.5
May 0.56 4.04 24.17 7 micrOgridS optimization. This was validated by specifically
Jun 0.547 3.58 23.17 6.91 simulating the optimized capacities of HOMER optimization
Jul 0.554 3.78 22.27 6.91 in the micrOgridS tool which showed better operational
Aug 0.557 4.43 21.96 6.95
Sep 0.554 5.23 22.1 6.17 performance of the system. The Homer R model applies a
Oct 0.587 6.28 22.59 6.74 per-hour time step decision on the optimal dispatch of power
Nov 0.583 6.69 23.65 6.75 flows. Consequently, the hourly dispatch of power flows and
Dec 0.57 6.7 25.03 7.17

1 These data were obtained from the NASA Langley Research Cen-

The cost assumptions and installed capacities of the DGs ter Atmospheric Science Data Center Surface meteorological and So-
lar Energy (SSE) web portal. URL: https://eosweb.larc.nasa.gov/cgi-
are illustrated in Tab. III. It is shown that three diesel bin/sse/sse.cgi?+s01+s04 (Accessed: 05.03.2018)
generators are modelled, of which DG1 and DG2 are being 2 [13]

considered identical types. 3 1 % of CAP EX based on [14]

 3rd International Hybrid Power Systems Workshop | Tenerife, Spain | 08 – 09 May 2018

TABLE IV
R ESULTS

model Pr,pv Pr,batt LCOE RE share fuel excess

/ kW / kWh / USD kWh−1 /% /l / kWh
micrOgridS 264 338 0.31 33.1 212,419 37,415
Homer R 288 730 0.34 34 266,012 58,154

Power / kW 200

150

100
DG3
50
DG2
DG1
0
BSSin
1.0
PV
0.9 excess
BSSout
Battery SOC

0.8 load
BSSSOC
0.7

0.6

0.5
01-07 00 02-07 00 03-07 00 04-07 00
Datetime / DD-MM hh

Fig. 3. Hourly dispatch of power flows: micrOgridS simulation for scenario a) (PH=8760)

the BSS State of Charge, that are both calculated with the ACKNOWLEDGMENT
micrOgridS model are depicted in Fig. 3 for an exemplified The authors would like to thank the Photovoltaik Institut
sequence of four days from 2017-06-30 until 2017-07-04. Berlin AG for project funding and cooperation.
It shows an optimized strategy with perfect foresight. In Drawing on the multiple advantages in terms of collab-
reality it is difficult to achieve these results but it can serve orative modelling and community based software develop-
as baseline for optimized operation strategies. ment, any kind of interest in collaboration in the future
development process of micrOgridS is warmly welcomed
IV. C ONCLUSIONS AND FURTHER STEPS and intended by the authors. If you are interested in col-
laboration or if any further information on the topic or data
We conclude that with the development of micrOgridS a used for this study are required, please contact the authors.
valid foundation was built for a HMG sizing tool that meets The source code of micrOgridS is available on Github:
the requirement list. However a few limitations are identified https://github.com/Py-micrOgridS/micrOgridS.git
that point out to future development regarding the enhanced
R EFERENCES
applicability of micrOgridS including:
[1] M. Howells and R. A. Roehrl, “Perspectives on sustainable energy
• clear definition of user/developer interfaces e.g. easy- for the 21st century,” Sustainable Development in the 21st Century
to-use GUI development, (SD21) project. New York: United Nations Department of Economi-
cand Social Affairs, Division for Sustainable Development, 2012.
• reduce model complexity in favour of shorter compu- [2] International Energy Agency, World Energy Outlook 2017. Organi-
tational time, e.g. investigate in time series aggregation zation For Economic Co-Operation & Development, 2017.
approaches or incorporate rule-based dispatch strate- [3] H. Lucas, S. Fifita, I. Talab, C. Marschel, and L. Cabeza, “Crit-
ical challenges and capacity building needs for renewable energy
gies, deployment in pacific small island developing states (pacific sids),”
• investigate in more efficient solver strategies, e.g. Renewable Energy, vol. 107, pp. 42–52, 2017.
heuristic algorithms [4] A. Gupta, R. Saini, and M. Sharma, “Modelling of hybrid energy sys-
tem—part II: Combined dispatch strategies and solution algorithm,”
Additionally, further integration of multiple objective Renewable Energy, vol. 36, no. 2, pp. 466–473, Feb. 2011.
functions and additional components is required as well as [5] P. Blechinger, C. Cader, P. Bertheau, H. Huyskens, R. Seguin, and
C. Breyer, “Global analysis of the techno-economic potential of
the validation of micrOgridS with real operation data is renewable energy hybrid systems on small islands,” Energy Policy,
desirable. As micrOgridS is closely tied to a community it vol. 98, pp. 674–687, Nov. 2016.
is assumed to foster for collaborative development, which [6] D. Gandini and A. T. De Almeida, “Direct current microgrids based
on solar power systems and storage optimization, as a tool for cost-
could broaden the perspectives on the modelling task, and effective rural electrification,” Renewable Energy, vol. 111, pp. 275–
for continuous maintenance support. 283, 2017.
 3rd International Hybrid Power Systems Workshop | Tenerife, Spain | 08 – 09 May 2018

[7] A. Simons and C. Hidalgo, “The economic complexity observatory:

An analytical tool for understanding the dynamics od economic
development,” Workshops at the Twenty-Fifth AAAI Conference on
Artificial Intelligence, 2011.
[8] The Government of Tonga, “Intended nationally de-
termined contribution (indc),” UNFCCC INDC Sub-
mission Page, Dec. 2015, accessed: 2018-04-04.
[Online]. Available: http://www4.unfccc.int/Submissions/INDC-
/Published%20Documents/Tonga/1/Tonga%20INDC.pdf
[9] Z. Abdmouleh, A. Gastli, L. Ben-Brahim, M. Haouari, and N. A. Al-
Emadi, “Review of optimization techniques applied for the integration
of distributed generation from renewable energy sources,” Renewable
Energy, vol. 113, pp. 266–280, Dec. 2017.
[10] R. Morrison, “Energy system modeling: Public transparency, scientific
reproducibility, and open development,” Energy Strategy Reviews,
vol. 20, pp. 49–63, Apr. 2018.
[11] F. Wiese, R. Bramstoft, H. Koduvere, A. P. Alonso, O. Balyk, J. G.
Kirkerud, Å. G. Tveten, T. F. Bolkesjø, M. Mnster, and H. Ravn, “Bal-
morel open source energy system model,” Energy Strategy Reviews,
vol. 20, pp. 26–34, Apr. 2018.
[12] Oemof Developer Group, “Open energy modelling framework (oemof)
- a modular open source framework to model energy supply systems.
version v0.1.4,” pp. –, 2017.
[13] P. Peerapong and B. Limmeechokchai, “Optimal electricity develop-
ment by increasing solar resources in diesel-based micro grid of island
society in thailand,” Energy Reports, vol. 3, pp. 1–13, Nov. 2017.
[14] J. Y. Tao and A. Finenko, “Moving beyond LCOE: impact of various
financing methods on PV profitability for SIDS,” Energy Policy,
vol. 98, pp. 749–758, Nov. 2016.