You are on page 1of 8

1

A Roadmap for Distribution Energy


Management via Multiobjective Optimization
Bharadwaj R. Sathyanarayana, Student Member, IEEE, and Gerald T. Heydt, Life Fellow, IEEE

Abstract — Multiobjective optimization has been applied to the the present that have been focusing on the same. One such
problem of optimal energy management in power distribution initiative for the control of power distribution energy man-
systems. Future distribution systems with increased deployment agement systems is presented in this paper.
of distributed energy storage devices and distributed energy re- In 2008, the U. S. National Science Foundation inaugurated
sources are being designed under the aegis of a number of sectors an engineering research center denominated the Future Re-
of the power engineering community. This includes the National newable Electric Energy Distribution Management
Science Foundation supported Future Renewable Electric Ener- (FREEDM) center. The essence of the FREEDM energy
gy Distribution Management (FREEDM) center. A technical management controls relates to the balancing or management
roadmap for the FREEDM center distribution system is pro-
of various objectives including: peak demand reduction from
posed that is based on multiobjective optimization. Diverse objec-
tives in the optimal control of the distribution energy manage- the legacy system, maximal penetration of renewable re-
ment system are handled through a Pareto optimal approach. sources, low cost to the consumer, the ability to support at
The normal boundary intersection method is used to develop a least part of the demand in the case of loss of power supply
Pareto optimal front and solving the multiobjective optimization from the conventional distribution system, and reduction of
subroutine has been approached through separable program- greenhouse gas emissions.
ming techniques. Uncertainty in load demand and distributed Several efforts to resolve some of the issues relating to
resource power output are incorporated in the test cases studied managing the energy needs of distribution systems with in-
to formulate a robust algorithm for distribution energy manage- creased penetration of distributed resources have been pro-
ment. The paper also makes brief reference to the Smart Grid
posed and implemented. Lasseter in [3], [4] discusses auto-
and its objectives – and how the proposed distribution energy
management fits into this initiative. nomous control strategies for the microgrid systems using
active power versus frequency droop characteristics. Lehtonen
Index Terms — Distribution engineering, multiobjective opti- et al., in [5] present a distribution energy management system
mization, energy management, renewable generation, ‘smart in view of power system markets. The distribution energy
grid’, Pareto optimization, robust optimization, linear program- management problem is one of managing multiple objectives
ming, normal boundary intersection, separable programming.
some of which may be inherently conflicting. It can be viewed
as an optimal utilization of distributed energy resources and
I. THE SMART GRID AND ENERGY MANAGEMENT IN distributed energy storage devices in the power distribution
DISTRIBUTION SYSTEMS system to minimize several objective functions, and subject to

F UTURE electric power distribution systems are likely to


have a much greater range of automated features and
renewable resource penetration as compared with con-
temporary ‘legacy’ systems. The U. S. Department of Energy
energy balance and load balance constraints. Large system
solutions introduce high dimensionality and there is the added
possibility of issues relating to non-convergence. In addition,
generating a Pareto optimal solution can prove to be difficult
has indicated the following main features of the Smart Grid for non-convex problems.
Initiative [1], [2] as: The application of multiobjective optimization [6], [7] tools
• efficient and methods has been made in literature in various scientific
• accommodating disciplines and interdisciplinary research initiatives. This pa-
• motivating per details the application of normal boundary intersection
• opportunistic method [8], [9] in solving the distribution energy management
• quality focused problem. The computer solution for the multiobjective optimi-
• resilient zation is developed via a technique known as separable pro-
• green. gramming [10].
There is no doubt that automatic control theory will play an A key element to a smart distribution system is the devel-
important role in primary distribution systems of the future, opment of a smart controller to manage energy. One such
and there have been several research initiatives of the past and controller integrates the voltage transformation functions of
conventional transformers and the high speed control characte-
Manuscript received November 2009. ristics of electronic switches. The solid state transformer is
The authors acknowledge the support of the Future Renewable Electric such a controller. The subject of solid state transformers suit-
Energy Distribution Management Center; a National Science Foundation able for distribution energy management systems (EMS) is
supported Engineering Research Center, under grant NSF EEC-0812121. discussed in [11], [12]. The design of an energy management
Authors Sathyanarayana and Heydt are with the Department of Electrical,
Computer, and Energy Engineering at Arizona State University in Tempe, strategy for distribution systems is addressed in a number of
Arizona. They can be reached at brsathya@asu.edu and heydt@asu.edu, re- venues. The literature also indicates other approaches with
spectively.

978-1-4244-6551-4/10/$26.00 ©2010 IEEE


2

various interpretations of the most desirable smart grid fea- in any computations, as it is a reasonable assumption in view
tures [13], [14]. of the nature of the study undertaken.
This paper presents a roadmap of distribution energy man- At this point, the focus is on the multiobjective optimization
agement via a robust multiobjective optimization formulation, problem. In the following discussion, the generalized problem
as part of the research initiatives on ‘distributed controls’ by shall be explained, and an application to the FREEDM system
the FREEDM system center. The ‘green hub’ system is a test given. For this purpose, consider several (e.g., m) objective
bed implementation of the FREEDM system currently being functions labeled fi. The parameter m could be quite large in
developed in North Carolina State University, one of the insti- the case of many conflicting as well as agreeing objectives –
tutions participating in FREEDM. but for the discussion below, the case of m = 4 is illustrated.
With the primary constraint of load balance established, the
II. DISTRIBUTION ENERGY MANAGEMENT objective for the optimization problem can be formulated.
A. Multiobjective Optimization For purposes of illustration, consider the following four
objective functions:
The optimal management of the energy needs of an electric 1. The entire time horizon over which the optimization is
power distribution system with distributed energy resource carried out is split into n equally spaced time intervals of
(DER) and distributed energy storage device (DESD) deploy- duration ΔT. The objective relating to peak power demand
ment is termed as distribution energy management (DEM). from the distribution system is captured by f1,
Traditional distribution systems are radial systems that feed . (4)
the end use customers through three phase primary feeders 2. The peak period energy supplied from the distribution
and single phase laterals [15]. In a conventional understand- network is,
ing, load management [16] refers to the management of the
end-use load shape to satisfy a number of objectives includ- ∆ , (5)
ing: reduced demand during peak system load periods, accept-
able system reliability, acceptable customer convenience and
where the peak period is user defined as the interval in which
incentives, and compatibility with existing system designs.
the index i ranges from , … , .
With the presence of DERs and DESDs in the distribution
3. The total cost of energy taken from the distribution supply
system, the optimal management is performed on the basis of
for the entire duration of the time horizon is expressed as
several diverse objective functions, hence resulting in a mul-
tiobjective optimization problem. In general, the distribution
energy management problem can be represented as a multiob- ∆ ∆ , (6)
jective minimization problem – namely the minimization of a ,…,
scalar aggregate function J, where cp and cop are the cost of energy in $/kWh for the peak
and off-peak periods, respectively.
min , ,…, (1) 4. The total power loss in the power converters for the DESD
is a function of the square of the current. As | | 1 . ., f4
subject to, can be approximated by,
, (2)
, (3) 1 , ∆ / , (7)

where m ≥ 2, and a is a vector of weights yet to be determined where is the efficiency of the power converter and battery
and F(X) represents the objective vector of objective func- combined, Xs is the rate of charging (kW) of the energy
tions, and , ,…, is the control vector of optimi- storage element.
zation. Equations (2) and (3) represent the constraints, lower Note that these are offered as examples. It may be that
and upper bounds on X, respectively. The inequality (3) is a there are other objectives that should be included (e.g.,
term by term inequality. reduces CO2 emissions, voltage regulation, reactive power use
B. Distribution Energy Management Objectives and production, harmonic distortion, factors that relate to the
cost of installation and equipment, and potentially a panoply
There are several possible objective functions that can be of factors relating to cost, use, and supply). The cited
considered for the optimal management of the distributed objectives are captured in the form of the fi functions. Let X
energy infrastucture. The objective functions chosen in the be a vector of power levels supplied by the distribution system
present formulation are: optimizing the peak demand from the at ΔT time intervals. The elements of X are xi. Also note that
distribution substation, optimizing the total cost of energy for several of the cited fi(X) functions could be interrelated (and
the demand interval, optimizing the energy demand during even nearly the same objective / mutually inclusive).
peak period, and optimizing the power loss in power Function f1(X) is nonlinear and cannot be represented in a
converters. Each of the above objectives can be represented conventional linear programming routine for the solution of
emperically as a functions of the instantaneous active power the multiobjective optimization. The function f1 is linearized
demand from the distribution feeder. Therefore, the control by augmenting the control variable with an additional variable
vector X could be chosen to be the active power demand from z such that,
the utility grid. The voltage fluctuations at the load end for (8)
variations in the active power has not been taken into account 1,2, … , . (9)
3

Equations (8) and (9) relax the requirements of f1 into a variable might be the reactive power injected into the primary
constraint. Note that, n is the total number of time intervals in distribution supply. It is believed that the multiobjective
the demand horizon. Since this optimization formulation is a approach taken would accommodate most of these
minimization problem, it is permissible to include (9) in the formulations (a generic discussion appears in [19]).
set of all inequality constraints represented by (3) as n C. Constraints
additional constraints 1,2, … , ; and the updated X is
now represented by, The basic control mechanism for energy management of a
distribution system is facilitated by a solid state transformer
, ,…, , . (10) (SST) [11], an electronic controller, and the primary objective
of an EMS controller is to effectuate a balance of load, storage
Given a forecast for load and DER power output over the power, and distributed generation. This is constrained by the
demand interval, the optimal strategy to manage the energy power ratings of the electronic controllers, energy storage
demand during the demand interval can be calculated. The medium, and other system components. The load balance
general description of the system is shown in Fig. 1. equation has been shown in (11) where the net power demand
from the distribution primary network is denoted by [x1, x2, …,
xn]T, the power output of the distributed energy resource is
XDG and the rate of charging (kW) of the energy storage
element by XS,

, ,…, (11)
where,
Xload, is the SST user load at the SST low voltage AC
terminals (kW, for every time interval in the study)
XDG is the power output of the DER (kW)
XS is the rate of charging of the DESD (kW).
Note that (11) is an instantaneous balance of power. Since
Xload and XDG are known to within a certain level of error
tolerance, the control variable for multiobjective optimization
can be changed from X to XS. For simplicity, in the remainder
of this paper the control vector which is the rate at which the
storage element is charged, will be represented by X, the
standard notation for control variable. Substituting (11) in (4) -
(7), the objective functions can be represented appropriately.
The balance of energy within the distributed storage device
Fig. 1 Distribution energy management strategy and system description (e.g., a battery) introduces additional constraints on the
optimization. To maintain the state of charge of battery device
A detailed description of the strategy for energy within its permissible limits,
namagement has been presented progressively in the , (12)
forthcoming sections. The optimal solution seeks to utilize the 0 1 (13)
distributed resource infrastructure and the installed storage where,
capacity to minimize the several objectives. The distribution xi is the appropriate control(s) for the energy storage
energy management problem can hence be viewed as not element (where, 1)
distant from the much more traditional unit commitment s is the state of charge (SOC), assuming that the energy
problem which seeks to optimally schedule power generation storage element is a battery
units to minimize the costs associated with power generation. Es,max is the maximum installed storage capacity (kWh).
The time horizon of solving unit commitment algorithms is Since the value of xi depends on the states corresponding to
one day ahead or one week ahead, typically. Distribution the previous time intervals, (12) and (13) result in formulating
energy management has been studied along similar lines as an a portion of the inequality constraints that is represented by
algorithm for optimal deployment of the distributed storage (2). The remainder portion of (3) is constituted by the n
capacity in a day ahead approach. equations depicted in (9). The maximum power rating of the
In the foregoing, the several multiobjectives have been converters Pmax, introduces the lower and upper bounds on
defined in terms of a control variable X. The control variable control variable X (see (3)). Note that the formulation
selected is the charging power for the energy storage device. indicated is specific to the FREEDM configuration. It is
However, the concept indicated is general enough to possible to envision other formulations in which different fi
accommodate other objectives and other control variables. are selected and different constraints are present. However,
For example, there may be cases in which some elements of the generic approach of multiobjective optimization is still
the demand are controllable; there may be communication valid.
between one load and another, and there could be exchange of It is straightforward to develop the solution that individually
power in some way at the secondary level; if reactive power minimizes each of the stated objective functions –
were incorporated into the model, an additional control individually, and not simultaneously. For the case of
4

cumulative optimization of an aggregate objective function, put power. The generation of the entire set of Pareto optimal
the individual functions will need to be weighted suitably so points with respect to different weighting options is of interest
as to obtain the Pareto optimal solution X*. The definition of in the problem under study.
Pareto optimality, weights for the objective functions and their B. NBI based Distribution Energy Management
application in the distribution energy management problem
has been described and developed in the forthcoming section. NBI based solutions to the multiobjective optimization shall
be developed in this subsection. Considering the specific
III. NORMAL BOUNDARY INTERSECTION application to the FREEDM system, for the DEM problem the
number of objectives, m is four. The length of the control
A. Pareto Optimality vector X is 1, where n is the total number of time intervals
The solution for traditional linear optimization problems is in the demand interval. Let, F* denote a vector of individual
easily faciliated through the use of linear programming tools global minima (that do not occur at the same state X) for each
which make use of the simplex method. In the case of multiple of the objective functions in F, and Xi* denote the vector of
objective to be optimized simultaneously, with the functions best minima for the minimization of fi, then a 4 x 4 matrix φ, is
themselves not strictly linear, the use of superior techniques is defined as
nessecitated. The calculation of a global optimal solution for
the multiobjective optimization, which is also the global
(16)
optimal solution for each of the individual objective functions,
is highly improbable, as different objective functions might be
at least partly conflicting and incommensurable. This lead to
the development of Pareto optimality [6]-[9]. Pareto
optimality is defined thus [7].
“A control vector is Pareto optimal, if there
does not exist another control vector , such that . (17)
1,2, … , , and
for at least one index j, where S is a nonempty subset The subscript 4 x 4 reminds that the indicated matrices are
of called the feasibility region.” four by four in this formulation. It can be observed that the
The concepts of multiobjective optimization are often diagonal elements of φ will always be zero. The off-diagonal
viewed in terms of an approach that manages several objec- elements have to necessarily be non-negative as this
tives that are integrated into an aggregate objective. This is requirement ensures that the each Xi* is the best possible
has been presented in literature in [7]. The aggregate objective solution for the individual minimization of function f1, and the
vector is generally taken to be a convex linear combination of Pareto optimal solution (X*) can never be better than the
individual objective functions with weights forming the linear individual minima (Xi*). The set of all convex linear
combination coefficients, combination of the columns of φ is referred to as a convex hull
(14) of individual minima (CHIM),
∑ ,
| , ∑ 1, 0. (18)
∑ 1. (15) The vector W, therefore are the weights of the NBI method for
The optimization of (14), results in X* called a Pareto optimal solving the multiobjective problem. The convex hull can be
solution. A trace of the curve of Pareto optimal points is called viewed as a set of all optimal solutions to any given
the Pareto optimal frontier. The convex linear combination of multiobjective problem and will be a four dimensional
objectives is one technique for obtaining Pareto optimal solu- hyperplane. Let u represent an unit normal to the CHIM
tions. For problems with more than two objectives, the devel- pointing towards the origin. Therefore, (φW + ud) represents
opment of Pareto optimal solutions is significantly complex all points on the normal for different values of the variable d.
[7]. The work done in this paper makes use of an efficient The variable d is a distance from a hyperplane of the Pareto
algorithm called normal boundary intersection (NBI), for the optimal solution. The point of intersection between the normal
same. In understanding the application of NBI to DEM, it is norm and the vector of objective functions F that is a point on
essential to note the following: the Pareto optimal front can be obtained by solving the
• The NBI method for generating a set of Pareto optimal following optimization problem,
points is different from the linear combination of objectives max (19)
,
that has been described earlier.
• The weights used to describe the NBI subproblems in the subject to,
(20)
forthcoming section are different from the linear combina-
tion weights ai discussed above.
For the DEM problem, each of the objectives corresponds to .
separate objectives of load management for a distribution sys- Given a weight vector W, the optimization above is a
tem with the presence of DER and DESD. The knowledge of subproblem in generating the Pareto optimal front. When the
how each objective is affected with changes to every other optimization subproblems are repeated for different sets of W
objective will prove to be useful for the development of an terms, a set of Pareto optimal points for the multiobjective
algorithm that is robust, i.e., which accommodates for the er- optimization problem are obtained. These points provide us
rors in modeling the characteristics of the load and DER out- the Pareto optimal frontier. The Pareto optimal front therefore
5

is the locus of all points that is traversed by the Pareto optimal


solution, when the weights of the subproblem are modified. , , (22)
When the ||Wi - Wi+1||2, the euclidean norm of the difference in
the weights for two successive subproblems i and i +1 are less , , , , (23)
than an acceptable value, the solution of the subproblem
numbered i can be used as a starting point, or initial condition
for the subproblem numbered i + 1. Fig. 2 is a sample where for i = 1…N,
graphical representation of the NBI technique. Actual graphs
are much more complicated than Fig. 2 and actual graphs lie , , , , , , (24)
in multidimensional space. A detailed development of the 1 , , , . (25)
NBI method, from fundamental principles can be found in [8],
[9]. The points (xi,1….xi,M) are in the range of xi and are called
breakaway points, i.e., the points between which the functions
f1(X) fs,i are approximated as being linear. The variables denoted by
λ are the weights of the the breakaway points and they are
convex (i.e., they add to one) and non-negative. The
f 1* mathematical formulation presented above was for any general
function fs that is separable.
The principle described is applied to the objective functions
in F of the DEM problem. The variable N in case of DEM is n
+ 1. Using (7), f4 can be written as

∑ | |. (26)

Note that f4 is not a function of z, hence the summation in (26)


is terminated in n and k’ can be expressed as a function of ε.
Functions f2 and f3 can also be represented in the form shown
in (26) using separable programming. Three breakway points
Solution are chosen on 1,2, … , as,
to subproblem i f 2* f2(X)
Fig. 2 Graphical representation of the solution to NBI subproblems for the xi,1 = - Pmax (27)
case of two functions xi,2 = 0 (28)
xi,3 = + Pmax. (29)
The objective function f4(X) represents the power lost in
the DESD power converters. Since the values of xi can be Therefore,
positive or negative depending on charging or discharging (30)
xi = -Pmax. λi,1 + 0.λi,2 + Pmax.λi,3.
operation respectively, f4 introduces a absolute value function
in the optimization which introduces a nonlinearity. To handle where, Pmax is the maximum power rating of the DESD power
the nonlinearity introduced by f4(X) a method called separable converter. The selection of the breakaway points were
programming (SP) is employed. motivated by the shape of the function f4(X) (v-shaped) which
C. Separable Programming is also piecewise linear in the the intervals generated between
the breakaway points. Corresponding to each xi for i = 1,2,…n;
A function is a separable function, if it can be written as a
transformation of variables using the breakaway points yeilds
sum of N single variable functions. When seperable functions
λi = [λi,1, λi,2, λi,3]. Additional conditions for the λi,j are,
are applied in linear programming (LP) problems the resulting
class of optimization problems are classified in literature as
λi,1 + λi,2 + λi,3 = 1 (31)
separable programming (SP) problems [10]. Further, in case a
0 ≤ λi,j ≤ 1. (32)
function of N variables is nonlinear, it can be written by an
approximate form in SP by separating the functions into N
The variables of optimization have effectively been
single variable functions such that the function is
modified from xi to λi,j. Functions f2, f3, and f4 are functions of
approximately linear in each segment of the seperation.
x1 - xn; and f1 is a function of z. Hence, to obtain the Pareto
The described separation is illustrated in the following
optimal solution using NBI, f2, f3, and f4 will need to be
equations (21) - (25), for any separable N variable function fs
represented as a function of the λi,j. Table I shows the value of
objective functions computed at the breakaway points. Using
(21) (23) and the results depicted in Table I, it is possible to
,
represent the f’s in their final form for the optimization. Cp,
Cop, k’ are emperical constants.
where, X = [x1 x2 …. xN]T. If the ith function fs,i is nonlinear it
can be approximated as
6

TABLE I where subscripts with ‘f’ indicate forecast values and index i
OBJECTIVE FUNCTIONS COMPUTED AT BREAKAWAY POINTS indicates that the correction term introduced is due to i-1th
interval). Variable γ can represent load power demand or DER
Function Time Interval j=1 j=2 j=3 power output. In this fashion, the NBI - SP is solved in a day-
xi,j demand interval -Pmax 0 Pmax ahead format at the beginning of each hour and the forecast
f1(xi)* - - - - load and DER power output are progressively updated based
peak period -Pmax 0 Pmax on forecast errors for the previous time interval, at the end of
f2(xi,j)
off-peak period 0 0 0 each hour,
peak period -CpPmax 0 CpPmax ,
. (33)
f3(xi,j)
off-peak period -CopPmax 0 CopPmax
f4(xi,j) demand interval k’Pmax 0 k’Pmax
*
f1 is not computed at the breakaway points as it is not a
function of x1, x2, …, xn.

With the change of variables from x to λ, the existing


constraints of optimization (20) will need to be modified. The
energy balance constraints are inequality constraints. In
addition, the boundary conditions on x terms will disappear
due to their inherent inclusion in the separable program.
Additional constraints on the new control variables (i.e., the λ)
will apply now in the SP (see (31), (32)). The exact form of
the constraint equations are shown in the Appendix. The NBI
subproblems are resolved into the separable programming
form and solved for a given set of weight vector to obtain the
Pareto optimal point. Fig. 3 illustrates the normal boundary
intersection algorithm for distribution energy management and
its solution using separable programming. The forthcoming
sections highlight the DEM application through an example.

IV. ILLUSTRATIVE EXAMPLE


The description of the mathematical concepts of NBI and
SP and their application to the DEM problem has been de-
tailed in the previous sections. The example studied in this
section, takes into account the system configuration for a sin-
gle household application of an energy management controller
(in this case a SST, for the FREEDM system application) and
the application of NBI and SP based energy management
strategy to it. Actual and forecast values for DER power out-
put and load demand are obtained for the study in this exam-
ple from [17], [18] and they have been shown in Fig. 4. The
load data and the DER power output have been obtained for
the location profile of southwest United States for a typical
winter day. The DER considered in this example is a PV ar-
ray.
The demand interval in this example, has been chosen as
one day and the number of time intervals (n) chosen is 24, i.e.,
with one variable for each hour of the day. After resolving the
objective functions and the constraints equations into the SP
form, the total number of variables in the control vector is 74
(= 3x24 + 2). The exact form of the constraint equations can Fig. 3 Normal boundary intersection algorithm with separable programming
be seen in the Appendix.
A ‘sliding technique’ has been incorporated to accommo- The selection of weights is performed such that the Eucli-
date the errors between the forecast values and the actual val- dean norm of the deviation between weight vectors for succes-
ues of load and DER output. In this method, the NBI based SP sive subproblems is less than a predefined tolerance level, ζ,
is solved at the beginning of each time interval (~ 1h) of the with W0 as the weights for the first subproblem. Table II de-
demand interval using forecast values for the load and DER tails the values of important system parameters used in the
output. At the completion of each interval, the forecast values case study.
for the next n intervals are updated based on the error between The Euclidean norm of XS has been used as an index to se-
the actual and forecast values for the past interval (see (33) lect a ‘model’ Pareto point on the Pareto front as having the
least norm of XS. Note that, several other indices could be used
7

e.g., the Euclidean norm of the vector of deviations of the ob- 4


jective functions computed at a Pareto point from their global Load Pow er

Power (k W)
minima. Pow er Demand from Grid
3
Battery Output
4
DER Pow er Actual
Power (k W)

3.5 DER Pow er Forecast 2


Load Actual
Load Forecast
3
1

2.5

0
2

1.5 -1

Time (h)
1
-2
2 4 6 8 10 12 14 16 18 20 22 24
0.5
Fig. 5 Results of DEM strategy applied to example case study. Note that the
Time (h) battery charge / discharge is shown discharging for 0500 – 1000 h. The bat-
0 tery starts at t = 0 with a 50% charge from the previous day.
2 4 6 8 10 12 14 16 18 20 22 24

Fig. 4 Actual and forecast values for load and DER output power • provide a roadmap of a generalized strategy for DEM
using multiobjective optimization
TABLE II • given a set of weights W, one Pareto optimal solution
SYSTEM INPUT DATA results, but if alternative weights are selected, different
solutions will be obtained
Parameter Value Parameter Value • In the example shown, if different weights or different
ε 0.98 ζ 0.01 load and forecast conditions were studied (e.g., admit-
[0.12, 0.06] ting uncertainty in DER power), a different Pareto op-
Es,max 5 kWh [cp, cop]
$/kWh timal result would occur.
Pmax 3.6 kW s0* 0.5 As indicated in earlier discussions, one important advantage
[p, k] [5, 5] W0 [0 0 0 1]T of the NBI method is that it generates the Pareto optimal solu-
*
is the initial state of charge. tions for the given DEM problem for different weighting op-
tions. Therefore, it is easy to assess the tradeoff on an objec-
The NBI based DEM algorithm provides a set of Pareto op- tive for changes in the weights of the other objective with re-
timal points for the given set of weight vectors. The results for spect to itself. With the knowledge of the tradeoffs seen in
a model set of weights that has the least ||Xs||2, i.e., correspond- individual objectives with respect to the others, it is possible
ing to WR = [0.049 0.1228 0.4949 0.3332]T are shown in Fig. to select weights based on the nature of operation of the sys-
5. The load demand curve, grid power demand curve and the tem and specific requirements imposed on system functions.
battery output curve have been portrayed. It can be observed This methodology is envisioned for future research efforts in
that the battery storage device is seen to discharge during the the distribution energy management.
predefined peak period of 0500 h to 1000 h. For simplicity, a
single peak period has been defined in this problem, where the V. CONCLUSIONS AND FUTURE WORK
effect of energy management is seen to reduce the energy con-
The application of multiobjective optimization concepts to
sumption from the distribution system during the peak hours.
solving the distribution energy management problem has been
It can be noted that for the min-norm solution corresponding
presented in this paper. The NBI method applied to the DEM
to WR, the peak demand minimization objective (f1) has been
problem has been found to be suitable for calculating the op-
weighted lower as compared to the peak period energy con-
timal energy management strategy. The formulation of such a
sumption objective (f2). Consequentially, the battery storage
strategy is dependent on the level of complexity and detail
element is seen to minimize the energy consumption during
envisioned and on the specific nature of its application. There-
the predefined peak period but at the cost of overall peak de-
fore, the formulation can be modified by appropriate changes
mand.
to the objectives in F and the constraints in AX ≤ b. The opti-
The example described in this section along with the prob-
mization concepts however, have been developed for a general
lem formulation developed in this paper, seek to highlight the
problem formulation and the effectiveness of the NBI algo-
following:
rithm in calculating the Pareto optimal solutions has been
• the verification of the concepts of Pareto optimality and demonstrated.
their application to the DEM problem,
8

The major focus for future research efforts include: demon- [5] M. Lehtonen et al., “Distribution energy management in the environ-
ment of de-regulated electricity market,” Proc. Energy Management and
strating the application of the DEM strategy to a larger test Power Delivery, v. 2, 1995, pp. 516-521.
system, distributed control calculations among multiple [6] J. Branke, K. Deb, K. Miettinen, R. Slowinski, Multiobjective Optimiza-
neighboring energy management controllers, and considera- tion: Interactive and Evolutionary Approaches, Berlin, Germany: Sprin-
tion of several other objectives, e.g., the management of reac- ger-Verlag, 2008.
[7] K. M. Miettinen, Nonlinear Multiobjective Optimization, Boston, MA:
tive power. Also, the uncertainty in DER power needs to be Kluwer, 1999.
considered. It is also envisioned to apply the strategy depicted [8] C. Roman, W. Rosehart, “Evenly distributed Pareto points in multi-
in this paper to the FREEDM ‘Green Hub’ system model, objective optimal power flow,” IEEE Trans. on Power Systems, v. 21,
which is currently being developed at North Carolina State No. 2, May 2006, pp. 1011-1012.
[9] I. Das, J. Dennis, “Normal-boundary intersection: An alternate method
University, one of the participating institutions in FREEDM. for generating Pareto optimal points in multicriteria optimization prob-
The Green Hub is expected to be a demonstration project for lem,” Institute for Computer Applications in Science and Engineering
the general effort of the Smart Grid initiative. (ICASE) report, No. 96-62, Nov. 1996.
[10] S. M. Stefanov, Separable Programming: Theory and Methods, Boston,
MA: Kluwer, 2001.
VI. ACKNOWLEDGEMENT [11] T. Zhao, L. Yang, J. Wang, A. Q. Huang, “270 kVA solid state trans-
The authors acknowledge the input of Dr. R. Ayyanar of former based on 10 kV SiC power devices,” Proc. IEEE Electric Ship
Technologies Symposium, May 2007, pp. 145 – 149.
Arizona State University in this work.
[12] T. Zhao, J. Zeng, S. Bhattacharya, M. E. Baran, A. Q. Huang, “An aver-
age model of solid state transformer for dynamic system simulation,”
VII. APPENDIX Proc. IEEE Power and Energy Society General Meeting, Calgary AB,
GENERALIZED DISTRIBUTION ENERGY MANAGEMENT July 2009, pp. 1 – 8.
[13] J. A. Momoh, “Smart grid design for efficient and flexible power net-
CONSTRAINTS works operation and control,” IEEE Power and Energy Society Power
The general form of the multiobjective optimization con- Systems Conference and Exposition, Mar. 2009, pp. 1-8.
[14] R. E. Brown, “Impact of smart grid on distribution system design,”
straints for distribution energy management is presented here.
IEEE Power and Energy Society General Meeting, July 2008, pp. 1-4.
The control vector can be expressed as, [15] T. Gönen, Electric Power Distribution System Engineering, NY:
McGraw Hill Inc., 1986.
[16] R. Delgado, “Load management - A planner’s view,” IEEE Trans. on
Power Apparatus and Systems, v. PAS-102, No. 6, 1983, pp. 1812-1813.
The inequality constraints are (for all i in [1 2 .. n]), [17] Itron - Eshapes, Load profiles for electricity and gas usage, available at:
http://www.itron.com/pages/products_detail.asp?id=itr_000491.xml
[18] National Renewable Energy Laboratory (NREL) - Renewable Resource
Data Center, PVWatts Calculator, available at:
http://www.nrel.gov/rredc/pvwatts/
[19] A. Belegundu, T. Chandrupatla, Optimization Concepts and Applications
in Engineering, Englewood Cliffs, NJ: Prentice Hall, 1999.

IX. BIOGRAPHIES
Bharadwaj R. Sathyanarayana (StM ’07) is from
Bangalore, India. He received his B. Tech. in
Electrical Engineering from VIT University, India.
In December 2008, he received his M. S. E. E.
degree from Arizona State University working in
The equality constraints are, the area of power distribution systems. Mr. Sa-
thyanarayana is presently a student in the doctoral
program at ASU. He is working in the area of
control of renewable resources in power distribu-
tion systems.
The length of the control vector is nℓ + 2. The total number of
inequality and equality constraints is 5n and n + m respective- Gerald Thomas Heydt (StM ’62, M ’64, SM ’80,
ly where, F ’91, LF ‘08) is from Las Vegas, NV. He holds
n is the number of time intervals in the demand interval the Ph.D. in electrical engineering from Purdue
m is the number of objectives of optimization University, West Lafayette, Indiana (1970). His
industrial experience is with the Commonwealth
ℓ is the number of breakaway points chosen on the xi in Edison Company, Chicago, and E. G. & G., Mer-
the separable program. cury, NV. He is a member of the National Acade-
my of Engineering. Dr. Heydt is presently the site
VIII. REFERENCES director of a power engineering center program at
Arizona State University in Tempe, AZ where he
[1] 110th Congress of United States, “Smart Grid,” Title XIII, Energy Inde- is a Regents’ Professor. He is the recipient of the
pendence and Security Act of 2007, Washington DC, Dec. 2007. 2010 Harold Kaufmann IEEE Field Award.
[2] Office of Electric Transmission and Distribution, United States Depart-
ment of Energy, “Grid 2030: a National Vision for Electricity’s Second
100 Years,” Washington DC, Apr. 2003.
[3] R. H. Lasseter, “Dynamic distribution using (DER) distributed energy
resources,” Proc. IEEE Power Engineering Society Transmission and
Distribution Conference and Exhibition, 2006, pp. 932-934.
[4] H. Nikkhajoei, R. H. Lasseter, “Microgrid protection,” Proc. IEEE
Power Engineering Society General Meeting, June 2007, pp. 1-6.

You might also like