Distributed Generation Placement for

Optimal Microgrid Architecture

Mallikarjuna R. Vallem, Student Member Joydeep Mitra, Senior Member Shashi B. Patra, Student Member
Klipsch School of Electrical and Klipsch School of Electrical and Klipsch School of Electrical and
Computer Engineering Computer Engineering Computer Engineering
New Mexico State University New Mexico State University New Mexico State University
Las Cruces, NM 88003 Las Cruces, NM 88003 Las Cruces, NM 88003
Email: [email protected] Email: [email protected] Email: [email protected]

Abstract— This paper describes a method for siting of economic benefits of DER deployment.
Distributed Energy Resources (DER) within the DG systems are quite different from the conventional
framework of an optimal microgrid architecture. An generation systems in operation. They are relatively small in
optimal microgrid architecture is characterized by output capacity today, but they are expected to increase in the
minimum cost interconnection, sizing, and siting of DER near future. Some examples are fuel cells, microturbines,
subject to stipulated global and local reliability criteria. solar panels, wind, geothermal, biomass, Ocean Thermal
This paper addresses the siting aspect of optimal Energy Conversion (OTEC), storage devices etc. Most of
microgrid architecture. The siting problem considers them are pollution free and have the potential to supply the
factors like deployment costs and savings gained by the deficit power in the energy market. Combined Heat and
use of Combined Heat and Power (CHP). The problem is
Power generation (CHP) is one of the major applications of
formulated as one of Nonlinear Programming and
DER as the efficiency of DERs can often be more than
Simulated Annealing optimization is applied. This paper
doubled when used in CHP mode, where possible [4].
presents the development and implementation of the
method, and demonstrates it using a six bus test system. DER systems are subject to planning considerations that
are also quite different from conventional generation. For
Index Terms—distributed energy resources, distributed example, fuel cells and microturbines give out heat as by-
generation, microgrid, optimal siting, simulated annealing. product which can be used for space heating. Geothermal
systems are extremely noisy which might preclude them from
I. INTRODUCTION being installed in residential areas; they are also subject to
availability of geothermal energy. Photovoltaic systems have
E LECTRICAL power demand is increasing steadily and
there is a great need to carefully design the future grid for
reduced costs and increased reliability [1]. System planning
high installation costs but almost zero running costs and
would best fit in places where there is high insolation. In view
must consist of an integrated approach to both location of of these constraints it becomes extremely important to build
resources and their interconnection, rather than a series of ad an optimal strategy for deploying such units within a
hoc and incremental responses to load growth. In this context, microgrid. As we are dealing with a microgrid, we assume the
distributed energy resources (DER) have gained increasing running cost of all the DERs is constant with respect to their
attention as an alternative to conventional generation systems operation anywhere in the microgrid.
[2]. The US Department of Energy (DoE) has developed and Ref. [5] has used a Mixed Integer Programming (MIP)
promulgated a roadmap that envisions widespread deployment formulation, with branch and bound optimization for an
of DERs by the year 2020. It has also been shown to be industrial power plant. Ref. [6] deals with a genetic algorithm
beneficial in many ways to deploy more DG rather than to optimization for deploying a DG resource in a distribution
increase the conventional centralized generation both in terms system. Ref. [7] computes capacity based on risk factor and
of environmental and economic considerations [3]. addresses about the reliability issues which arise as a
In order to exploit the reliability, security and economic consequence of transmission bottle-necks. In this paper the
benefits of DER, it is necessary to design DER networks that cost function has been modeled as a Nonlinear Programming
are potentially self-sufficient. In other words, it is necessary problem and Simulated Annealing (SA) optimization has been
to develop effective tools for microgrid design. A microgrid used to achieve global optimum.
should not only network DERs with load points, but should
also have the ability to operate in both grid-connected and II. PROBLEM FORMULATION
islanded modes. Microgrid architecture consists of optimally System architecture has many aspects like network layout,
designing the interconnections, as well as optimally sizing and capacity estimation system reliability and resource
siting the DER units to maximize reliability, security and deployment. It is important to design the architecture keeping
the future needs in view so as to achieve a global Mathematically the problem can be expressed as follows.
minimization with respect to cost, also addressing issues like
environmental degradation and reliability. Minimize
+ +
Consider a system with NB nodes, NG generators, and NT Cost = ( H − F × h ) + J power × U N × ( H − F × h ) (4)
transmission lines. The capacity addition to this system is + U N × averageC × J int
determined, and is divided into some DG units. Statistics Subject to:
show an average increase of 10% energy demand in a span of EIR > R0 (i)
5 years [1]. A load increase of 50% at every node is assumed C=0 (ii)
in the test case considered. The choice of the number of DG where
units is based on the minimal increase in the load. The EIR = Energy Index of Reliability [8]
number of DG units is NDG. R0 = minimum reliability desired
Further the energy demand is divided into heat demand and C = NB vector of bus power curtailments.
electrical demand. Then it is possible to model the use of CHP
from the DERs. In practice, separating heat demand and Constraint (i) specifies the minimum reliability the system
electrical demand in a system is not that easy, there is a is required to meet. Constraint (ii) eliminates solutions which
responsibility and necessity of estimating them to achieve may never be able to meet the complete demand of the system
better efficiency of the system. even when enough heat and electrical capacity is available
The costs arising from the strategy of deployment are
modeled and a cost optimization is done. The total cost C. Reliability Evaluation
involved can be classified as deployment costs and heat To obtain the energy index of reliability, a DC load flow is
compensation costs which will be detailed below. solved to minimize the curtailment [8]. The DC load flow
model is as follows.
A. Deployment Cost
⎛ N bus ⎞
An NDG × NB deployment cost matrix is developed such that Loss of Load = Min. ⎜
⎜ ∑C i
⎟⎟
its ijth element is the cost of deploying the ith DG unit at the jth ⎝ i =1 ⎠
node. The solution space is the selection of a node for each subject to:
unit which is represented as a binary NB × NDG matrix. A Bˆ θ + G + C = D
linear depreciation is assumed where annual depreciation rate G ≤ G max
will be r. Then the cost for deploying the resource in our C ≤ D (5)
system is,
b Aˆ θ ≤ F m ax
r × Trace ( F × J ). (1)
θ unrestricted
where
where,
r = rate of depreciation per year
D = NB vector of bus load
F = NB × NDG binary solution matrix where the ijth element
Gmax = NB vector of available generation at buses
is unity (1) if DG j is at node i.
Fmax = NT vector of flow capacities of transmission lines
J = NDG × NB deployment cost matrix.
b = NT × NT primitive (diagonal) matrix of
B. Heat Compensation Cost transmission line susceptances.
If the heat demand at any node is not supplied by the CHP Â = NT × NB element-node incidence matrix.
at that node, electrical energy is consumed to generate that B̂ = NB × NB augmented node susceptance matrix.
heat. So in addition to the electrical demand an excess energy = Aˆ T b Aˆ .
is to be supplied to those nodes which would be equal to the
θ = NB vector of bus voltage angles.
heat curtailment. The heat curtailment in the system is
( H − F × h) + (2) EIR is calculated then by
NC

where EPNS = ∑ LOL

i =1
i × prob i
H = NB vector of heat demand at the nodes
EPNS
h = NDG vector of heat output from the DG units. EIR = 1 − (6)
a+ = a, if a>0, else a+=0. D Tot
For simplicity we take the units for heat as power units. The where
cost of energy is taken a flat rate of 0.1 $/kWh. The additional EPNS = expected power not served.
costs for the heat compensation on the system will be DTot = total power demand.
NC = number of contingencies.
J power × U N × ( H − F × h ) + (3)

where UN is a unit vector of dimension NB.

 III. SIMULATED ANNEALING OPTIMIZATION START
The optimization is a Nonlinear programming type and this
section demonstrates how simulated annealing technique is Set Initial Parameters
applied to solve the problem. Initial Solution i
Temperature Tstart
A. Simulated Annealing Neighborhood N start
Simulated Annealing is a stochastic global optimization
technique which has an asymptotic convergence to the global
Change Neighborhood
minima [10]. The concept is based on the manner in which
liquids freeze or metals recrystalize in the process of
annealing. In an annealing process a melt, initially at high Count = 0
temperature and disordered, is slowly cooled so that the
system at any time is approximately in thermodynamic
Generate
equilibrium. This method can get over the local minima as we New Solution
not only consider downhill moves but also uphill moves with a
probability based on the temperature ‘T’ using the metropolis
Check
criteria [11]. At higher temperatures a lot of uphill moves will F(i’)<F(i)
be accepted but their acceptance gets reduced with the
temperature decrease and the search asymptotically converges
to the global minima. A proper choice of neighborhood and
i=i’
cooling schedule will give near optimal solution in a i=i’ with probability
reasonable time. A=exp{−[F(i’)−F(i)]/T}

B. Algorithm
The Simulated Annealing algorithm is as shown in Fig. 1. Count = Count +1
Simulated annealing starts at a high temperature with an
initial feasible solution. Initially the neighborhood is assumed
to be the entire solution space. The temperature is then Check
Count = N No
decreased following a cooling schedule. At every temperature
a fixed number of solutions are visited. For each solution the Yes
new objective value is compared to the current one. If the new
Tnew = αxTold
solution has objective value lesser than the current one, it is
directly accepted. But if the new objective value is greater
than the current one, it is accepted with an acceptance Check
probability given by, T<T min No
Yes
⎛ ( f (i' ) − f (i )) ⎞ (7)
AT (i, i' ) = exp⎜ − ⎟
⎝ T ⎠ STOP

This is how it is able to escape from local minima. At Fig. 1. Simulated Annealing Algorithm.
higher temperatures a lot of uphill solutions are accepted, but
as the temperature is reduced, the uphill moves get reduced.
After N number of iterations at a temperature, the temperature IV. TEST SYSTEM AND RESULTS
is reduced using a geometric cooling schedule.
A. System Description
Tnew = α × Told (8)
The optimization strategy was applied on a 6 bus test
where α < 1 and it is better to choose a ‘α’ around 0.9. The system [12]. Four generators are available, two each at nodes
neighborhood for the next iteration is changed so that half the 1 and 5 which can supply a maximum power of 28 MW.
numbers of previous solutions are accepted. Then the cooling Nodes 2, 3, 4 and 6 are load buses with a total demand of 21
schedule is followed till we reach a final temperature or a MW. Further the load is separated into electrical and heating
temperature where we have consistent minimum in the last load. The system data is tabulated below.
few temperatures.
 G G the system which correspond to EIR = 95%.
6 5 4 Table 4. DG resource capacity
DG Electrical Capacity Heat Capacity
(MW) (×106 BTU/hr.)
1 2 3.412
2 1.8 4.094
3 2.5 5.118
1 2 3
4 2.4 5.459
G G Table 5. Deployment cost (in $ million)
Fig. 2. Six Bus Test System. DG Node Node Node Node Node Node
1 2 3 4 5 6
Table 1. Line Data 1 0.5 0.6 0.6 0.7 0.4 0.6
From To Z (p.u) Susceptance 2 0.55 0.65 0.6 0.65 0.5 0.4
(p.u) 3 0.7 0.8 0.8 0.85 0.6 0.65
1 2 0.2238 + j 0.5090 1.646 4 0.65 0.7 0.8 0.7 0.65 0.7
2 3 0.2238 + j 0.5090 1.646
3 4 0.2238 + j 0.5090 1.646 Twenty iterations were performed for each temperature and
4 5 0.2238 + j 0.5090 1.646 a geometric cooling schedule with α=0.9 is applied. The initial
5 6 0.2238 + j 0.5090 1.646 neighborhood was chosen to be the entire solution space. The
1 6 0.2276 + j 0.2961 2.123 neighborhood was the choice of moving a number of DG units
1 5 0.2603 + j 0.7382 1.205 from the current solution. Initially the choice is 4 which can
cover the entire solution space and after every 5 cooling
B. Selection and initialization of parameters routines, it was reduced. The convergence was achieved when
The transmission line capacities are kept constant even any of the three conditions were satisfied.
though the load has been increased. The reliability indices for Condition 1: Current global solution doesn’t change for
the generators and DG units are taken to be the same. Mean four consecutive cooling routines.
time to failure (MTTF) and mean time to repair (MTTR) are Condition 2: The neighborhood is very small which ever is
taken as 2940 and 60 hours correspondingly [13]. The average met early.
outage duration for the transmission lines is 35 hours. Condition 3: The end temperature is reached.

Table 2. Generator Data C. Results

Generator Capacity Location The simulated annealing approach was applied to the test
1 10 Node 1 system. The simulation converged after sampling 142
2 8 Node 1 solutions. The entire solution space consists of 1296 states.
3 5 Node 5 The cost of the optimal deployment scheme is found to be a
4 5 Node 5 total of $3.9292 million per year with the deployment cost
being $0.25 million and the heat compensation cost $3.6792
Table 3. Load Data million. The reliability of the system is 0.9702. The location
Bus Electrical Load (MW) Heat Load of DG both for the initial and the optimal solution are
(×106 BTU/hr.) tabulated below.
1 0 0 Table 6. DG Location
2 4 6.824
DG Initial Location Optimal
3 7.5 10.236
Location
4 5.5 6.824
1 Node 5 Node 3
5 0 0
2 Node 6 Node 6
6 5 8.53
3 Node 5 Node 3
4 Node 1 Node 2
An average cost of 200$/kW is taken as the deployment
cost. A running cost of 0.1 $/kWh is chosen.
The Initial configuration shows that two of the DG units to
Initial solution was chosen such that for each DG unit the
be placed at node 5 but none at node 3 where the load is
deployment cost is minimum at that node. Initial temperature
maximum. The optimization moved our DG location to the
was taken to be twice the objective value at the initial
load buses where the heat compensation cost had its
solution. A maximum load curtailment of 5% was chosen for
importance along with the deployment cost in the [3] T. R. Casten, M. J. Collins, “Optimizing future heat and
optimization process. The number of iterations taken to reach power production,” Executive Summary, Private Power
the optimal solution will be more in a small system compared LLC & The World Alliance for Decentralized Energy,
to a larger system. The technique promises an asymptotic Oct 2004.
reachability with probability ‘1’ of reaching the global [4] Consortium for Electrical Reliability Technology
minima. Solutions “Integration of distributed energy resources—
One important result of this scheme is that no additional the CERTS microgrid concept,” April 2002.
line capacities are introduced into the system even though the http://certs.lbl.gov/CERTS_P_DER.html
system had a 50% load increase yet achieving reliability. This [5] S. W. Illerhaus, J. F. Verstege, “Optimal operation of
shows how we will be able to save more on transmission line industrial CHP-based power systems in liberalized
installations by using DG. energy markets,” IEEE Industry Applications
Conference, 2000. vol. 2, pp. 901–908.
V. CONCLUSION [6] H. Shim, J. H. Park, I. S. Bae, J. O. Kim, “Optimal
This paper described a method for siting of DER units capacity and allocation of distributed generation by
within the framework of an optimal microgrid architecture. minimum operation cost in distributed systems,” KIEE
An optimal microgrid architecture is characterized by International Transactions on Power Engineering, vol.
minimum cost interconnection, sizing, and siting of DER 5-A, no. 1, p. 915, 2005.
subject to stipulated global and local reliability criteria. This [7] N. S. Rau, F. Zeng, “Adequacy and responsibility of
paper addressed the siting aspect of optimal microgrid locational generation and transmission-optimization
architecture. Deployment costs, including the economic procedures,” IEEE Transactions on Power Systems, vol.
benefits of CHP comprised the objective function of this 19, no. 4, pp. 2093–2101, Nov. 2004.
formulation. The method of Simulated Annealing was [8] J. Mitra, S. B. Patra, S. J. Ranade, “A dynamic
identified and demonstrated as a viable approach. This paper programming based method for developing optimal
presented the development and implementation of the method, microgrid architectures,” in Proceedings of IEEE Power
and demonstrated it using a six bus test system. Engineering Society General Meeting, San Francisco,
Further research efforts will attempt to integrate the sizing June 2005.
problem with that of location. This will then be integrated [9] J. Mitra, C. Singh, “Incorporating the DC flow model in
with network configuration within the framework of optimal the decomposition-simulation method of multi area
microgrid architecture subject to stipulated reliability criteria. reliability evaluation,” IEEE Transactions on Power
Progress of this research will be reported in due course. Systems, vol. 11, no. 3, August 1996.
[10] Tao Wang, “Global optimization for constrained
ACKNOWLEDGMENTS nonlinear programming,” Ph.D Thesis, University of
The work reported in this paper was supported in part by Illinois at Urbana-Champaign, 2001.
the National Science Foundation, under grants ECS-0134598 [11] N. Metropolis et al, “Equation of state calculations by
and ECS-0140092, and by the Electric Utility Management fast computing machines,” Journal of Chemical Physics,
Program. This support is gratefully acknowledged. Thanks are 21:1087–1092. 1953
also due to Kjell Eikland and Peter Notebaert, developers of [12] N. S. Rau, Y. H. Wan, “Optimum location of resources
the open source Linear Programming package used in this in distributed planning,” IEEE Transactions on Power
work. Systems, 1994, pp. 2014-2020.
[13] IEEE Committee Report, “IEEE Reliability test system”,
REFERENCES IEEE Transactions on Power Apparatus and Systems,
[1] EIA, Department of Energy, “Annual energy outlook vol. 98, no. 6, Nov./Dec. 1979.
2005,” Feb 2005.
[2] A. R. Wallace, G. P. Harrison, “Planning for optimal
accommodation of dispersed generation in distributed
networks,” 17th International conference on Electricity
Distribution, CIRED 2003, Barcelona, Spain.