IEEE TRANSACTIONS ON SMART GRID, VOL. 4, NO.

4, DECEMBER 2013 2109

Advanced Distribution Management System

A. P. Sakis Meliopoulos, Fellow, IEEE, Evangelos Polymeneas, Student Member, IEEE,
Zhenyu Tan, Student Member, IEEE, Renke Huang, Student Member, IEEE, and Dongbo Zhao, Student Member, IEEE

Abstract—This paper proposes an advanced distribution man- Functions:

agement system (DMS) that a) monitors each component and per-
forms protection functions using a dynamic state estimation, b) Device state equations.
the estimated states are transmitted to the DMS where the real
time model of the entire feeder is synthesized, c) uses the real time Network flow equations.
model to perform upper level optimization (operations planning) Device binding constraints.
and lower level optimization (real time control) via a hierarchical
optimization procedure; and d) applies proper controls to operate Total active power consumed during [ ,
the system at optimal points. The proposed approach for protec- ). It is a function of the node voltages.
tion, operations planning, and real time control of the system pro-
vides the infrastructure for additional important applications. As
an example, the paper presents a novel application for monitoring I. INTRODUCTION
available reserves from all resources in the system. We propose the

T
concept of Reserve-O-Meter that monitors in real time the avail-
HE proliferation of customer owned smart devices and
able reserves from all resources (utility and customer owned).
active resources such as smart appliances, plugin hybrid
Index Terms—DMS (distribution management system), DSE electric vehicles (PHEVs), rooftop PVs, etc. makes the electric
(distributed dynamic state estimation), ESS (energy storage
system), GPS (global positioning system), HAOP (hours ahead
load highly controllable. One can take advantage of the con-
operations planning), RTO (real time optimization), setting-less trollability of the electric load to enable higher efficiency (loss
protection, spinning reserve, TCL (thermostatically controlled minimization, load levelization, etc.) and to improve the relia-
loads). bility of the system by responding in cases of need.
NOMENCLATURE: The potential of smart loads to improve the reliability and
Variables: the operating cost of the electric power system is undisputable.
However, the approaches to enable these benefits are debated.
There are two schools of thought: a) market approach based
Device controls at time .
on incentives to customers [1]–[6] and b) centralized approach
Device active power controls at time . using coordinated control of participating customers [7], [8].
Remaining (other than active power) device Hybrid approaches are also possible. The infrastructure require-
controls at time . ments of the market approach are comparably simpler, for ex-
ample two way smart meters can provide the required commu-
Device stochastic parameters at time . nications. However the design of the market approach could
Device internal state variables at time . be challenging especially if unwanted consequences are to be
avoided such as shifting of peak loads, unpredictability of de-
Node across variables (voltages) at time . mand response in cases of need, etc. The centralized approach
Device state and node voltage initial needs a much more sophisticated infrastructure of metering,
estimates. communications, analytics, controls, and participation (consent)
of customers. The centralized approach enables maximum ben-
Energy price ($/MWh) during [ , ).
efits and avoidance of unwanted side effects.
Voltage phasor of phase p at node i. Substantial literature exists on the topic of optimal central-
ized distributed energy resource (DER) operation. Ngand and
Angle phasor of phase p at node i.
Sheblé [9] discuss an optimal direct load control scheme based
Voltage magnitude at bus i. on maximization of the utility’s profit, while Ruiz et al. [10]
adopt a similar approach, but make an important contribution
Rated voltage magnitude at bus i.
in the modeling of thermostatically controlled loads (TCLs)
within this optimization problem. Celli et al. [11] focus on
Manuscript received October 02, 2012; revised February 08, 2013; accepted optimization of distributed storage operation, and in [12] Pilo
April 25, 2013. Date of publication November 14, 2013; date of current version et al. present an optimization scheme considering a wide range
November 25, 2013. Paper no. TSG-00680-2012.
of controllable DERs. Similarly, Molderink et al. [13] use a
The authors are with the School of Electrical and Computer Engi-
neering, Georgia Institute of Technology, Atlanta, GA 30332 USA. three level approach to optimally schedule house loads, without
(e-mail: [email protected]; [email protected]; [email protected]; inconveniencing the residents. Sathyanarayana and Heydt [14]
[email protected]; [email protected])
propose formulating distribution optimization as a multi-objec-
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. tive problem and finding a Pareto optimal policy for it. Thus,
Digital Object Identifier 10.1109/TSG.2013.2261564 existing literature covers the topic of optimization algorithm

1949-3053 © 2013 IEEE

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on June 04,2022 at 13:13:17 UTC from IEEE Xplore. Restrictions apply.
 2110 IEEE TRANSACTIONS ON SMART GRID, VOL. 4, NO. 4, DECEMBER 2013

Red lines represent control command flow, blue lines rep-

resent data flow and grey lines represent the single line dia-
gram of the power system. The basic component of the pro-
posed system is the universal monitoring protection and control
unit (UMPCU) which performs distributed dynamic state esti-
mation (DSE) at the local (component) level and provides the
real time model of the component, consisting of: a) estimated
state, b) connectivity, and c) mathematical model of the com-
ponent. These results are used to perform component protection
(setting-less protection). Then the real time model is commu-
nicated to the distribution management system (DMS) where it
is utilized to synthesize the real time model of the entire dis-
tribution feeder [15], [17]. The UMPCU also receives control
commands from DMS through red lines and applies these con-
trols to individual components.
It should be emphasized that the functional specifications of
the UMPCU can be met by current top-of-the-line numerical
Fig. 1. Illustration of overall approach. protective relays. Specifically, the computational power of these
relays is adequate to perform the analytics of the UMPCU, i.e.,
formulation to a great extent, but from the simulation/analysis the state estimation based protection function and the extraction
point of view and ignores the issue of real time applications. of the real time model of the component by appropriately pro-
This paper addresses the issues associated with real time data gramming. These relays are also able to receive control signals
acquisition, filtering the data with state estimation procedures, from the centralized DMS and control the component as appro-
formulation of the coordination problem in real time, and priate. In current practice, a centralized data acquisition system
implementation of the controls in real time. Specifically, we linking these relays already exists. The feasibility of such a data
propose a distributed state estimation procedure that extracts acquisition system for execution of a system-wide state estima-
the real time high fidelity model of the system several times per tion is demonstrated in [15]–[17]. Other data acquisition sys-
second. The model is used to formulate the optimal operation tems, such as the home-area networks discussed in [18], [19],
of utility and customer owned resources. This combination of can be integrated into proposed system. Relays and data acqui-
high fidelity monitoring and real time optimized control for sition system are ubiquitous and therefore the proposed system
the components in the distribution system provides the funda- can be constructed with small additional investments.
mental block of an advanced distribution management system. The system described above provides the detailed state of the
A number of byproducts can be developed and integrated into entire distribution feeder (DF) and all resources attached to this
this system. One such example is presented in Section VII. system. The system enables a number of operational functions
Although some models of the individual components of the such as voltage control, reconfiguration for loss optimization,
system have been proposed in peer research and by the authors operations planning, etc. Here we will focus on the use of this
of this paper before, an integrated distribution-level real-time system to a) optimize and control the system, and b) to provide
management system is a new development and the focus of this in real time the available capacity for demand response in terms
paper. By enabling the unique system level real-time control of of the existing reserves in the various resources and to exer-
components, the proposed system can achieve not only better cise these reserves, when needed, without inconvenience to end
performance with higher accuracy but also extended and new users. This application is described in Section VII in the paper.
functionalities.
The paper is organized as follows: Section II provides the III. STATE ESTIMATION BASED SETTING-LESS PROTECTION
general method description; Section III discusses the compo- The proposed advanced DMS is based on a fast, reliable, and
nent-level distributed dynamic state estimation that enables Set- distributed state estimator (DSE) that provides the real time
ting-less protection of each component; Section IV describes the model of the system at rates of 10 times per second. This is
synthesis of the distribution feeder state; Section V describes achieved by operating on each device independently using the
the two-level optimization scheme used to optimally control all instrumentation for protection of the device. The DSE extracts
controllable components and smart loads; Section VI presents the real time model of the device which consists of a) connec-
sample results for the approach; Section VII presents an addi- tivity information, b) estimated states in terms of waveforms,
tional application for this system, a real time reserve capacity and c) mathematical model of the device. The method requires
meter. And Section VIII provides conclusions and comments. that at least one data acquisition system is GPS synchronized.
In this case the real time model of the device is computed on
a common time reference (UTC time). The overall approach
II. METHOD DESCRIPTION
is illustrated in Fig. 2. A more detailed implementation of the
A centralized coordination/optimization approach is pro- setting-less protection is illustrated in Fig. 3. The protection
posed for the integrated system of utility and customer owned function described in this system is referred to as “setting-less
resources. The overall structure is shown in Fig. 1. protection” because it simplifies the settings of the protection

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on June 04,2022 at 13:13:17 UTC from IEEE Xplore. Restrictions apply.
 MELIOPOULOS et al.: ADVANCED DISTRIBUTION MANAGEMENT SYSTEM 2111

Fig. 2. Functional diagram of setting-less protection unit.

Fig. 4. Synthesis of subsystem or full system state estimate from component

state estimates.

IV. SUBSTATION/DISTRIBUTION MODEL SYNTHESIS

The frequency domain component states are used to directly
synthesize the state of subsystems as well as the state of the
Fig. 3. Illustration of setting-less protection scheme via dynamic state estima- entire distribution feeder. The process is illustrated in Fig. 4.
tion. The figure illustrates how the DMS synthesizes the model of
the entire distribution feeder. Each component’s connectivity
function and it does not require coordination with protection data is used to synthesize the topology of the feeder. Using that
functions of other devices. The basic idea here is that the “set- topology, state estimates from each component that have the
ting-less protection” continuously monitors the validity of the same GPS timestamp are immediately combined (with no addi-
mathematical model of the component under normal operations tional calculations) to obtain the overall system estimate. Sim-
in a systematic way by the use of the DSE. The method is de- ilarly, the component models are combined with the feeder’s
scribed in detail in [20]. topology data to obtain its real time model.
It is important to recognize that while the protection function
operates on waveform data (point on waveform) the state esti- V. TWO-LEVEL DISTRIBUTION SYSTEM OPTIMIZATION
mate for the entire feeder is needed in terms of phasors. For this The model synthesis procedure outlined above provides the
reason, the system computes the phasors of the various states real time model of the entire distribution feeder that may include
from the estimated waveform data (virtual PMU). The proposed D components and N nodes. The real time model is updated
system is designed to compute the frequency domain real time 10 times per second (this capability has been already demon-
model 10 times per second. Thus the phasors are computed once strated in practice [16]) and is used to formulate and solve a
per 0.1 seconds using waveform data of the last 0.1 seconds (ap- two-level optimization procedure. This procedure generates the
proximately 6 cycles). The phasors with the GPS time stamp and control signals transmitted by the DMS to each individual com-
real time model of each component is transmitted to the DMS. ponent. While the system state is updated 10 times per second,
It is emphasized that the proposed approach facilitates fast we propose to run the optimization once per minute. Thus the
communications. Specifically, each device sends to the DMS control signals are recalculated once per min. This frequency is
only its real time model which comprises a very small number achievable with the developed technology.
of data. When connectivity changes, then connectivity data are The distribution feeder real time model is in the form:
transmitted by exception. Similarly if model changes occur, the
new mathematical model will be transmitted by exception. The
end result is that while the instrumentation may be collecting
data at rates of hundred thousands of data points per second,
the frequency domain state (phasors) are only a few tens of data
points per second. Only the frequency domain component state
is transmitted to the DMS. (1)

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on June 04,2022 at 13:13:17 UTC from IEEE Xplore. Restrictions apply.
 2112 IEEE TRANSACTIONS ON SMART GRID, VOL. 4, NO. 4, DECEMBER 2013

The variables are defined in the nomenclature section. Note should be minimized. When formulating the HAOP optimiza-
that the state equations F were derived from each component’s tion problem, however, the proposed DMS uses the expected
dynamic model (see Appendix), hence their algebraic form in values of , thus solving a deterministic optimization problem
(1) is obtained through an appropriate integration method. Also and ensuring computational tractability. It can be argued that
note that is a stochastic process modelling uncertain param- this choice does not adversely affect the feasibility and opti-
eters including solar irradiation and ambient temperature that mality of the problem because: i) the uncertain parameters in
affect component behavior. In (1) the binding constraints question can be forecast with a relatively low degree of un-
model the following type of considerations: a) component rat- certainty when the time horizon is short, ii) the deterministic
ings must be observed, and b) component owner must not be in- optimization problem is solved frequently with renewed initial
convenienced. The latter constraints ensure that any control se- estimates for the DF state ( and ) from
quence applied to device i will not violate the “convenience” the distributed state estimator which runs at very high rates, iii)
tolerance limits provided by the device customer. For a TCL the proposed DMS operates with quadratic models [21], [22] of
those are maximum and minimum temperature set points, for a the functions F, G, H. Hence, the resulting nonlinear program is
PHEV this is a device charging deadline, etc. A detailed com- quadratically constrained with a linear objective, and is solved
ponent model including this type of constraints can be found in using mature methods of second order cone programming which
Appendix A. can be solved with great efficiency by interior-point solvers, as
The proposed DMS is responsible for specifying the device discussed in [23], [25], iv) the RTO problem (Section V-B) in-
controls in an optimal fashion. However, to ensure optimality, corporates relaxation of soft equality constraints expressed in
certain controls must be optimized considering a horizon of sev- terms of the control directives to ensure feasibility.
eral hours (e.g., PHEV charging, TCL operation, storage device The deterministic HAOP that minimizes the distribution
charging/discharging etc.) while others must be optimized con- feeder’s total cost of energy is:
sidering their instantaneous effect (e.g., load tap changers (LTC)
setting, power electronics (PE) converter reactive power injec-
tion, etc.). For this purpose, the controls vector is decomposed
into two components:

(2)

where are the controls that specify a component’s active

power generation/consumption (and hence are longer-horizon
controls) and are the remaining controls mentioned above. (3)
To address the issue mentioned above, the proposed DMS
specifies the controls using a two-level optimization scheme. Note that the HAOP computes the optimal controls, ,
The upper level, referred to as Hours Ahead Operation Plan- and , , for all the time intervals in the
ning (HAOP), specifies the optimal for the planning horizon [ ) but only the controls ,
planning time interval [ ), spanning several hours, and con- are used as equality constraints for the real time optimization
sidering the forecast for the stochastic parameters. The lower (RTO) described next.
level, referred to as Real Time Optimization (RTO), is solved at
each instant and uses the quasi-steady state version of (1) as a B. Real Time Optimization
model, and the optimal specified by the HAOP as equality The RTO specifies the actual control vector to be trans-
constraints, where . Given the uncertain nature mitted as a command to the components controlled directly by
of the problem, these equality constraints are soft constraints al- the proposed DMS during [ ). The RTO algorithm uses
lowing the optimization method to compute an optimal solution the most recent DF model synthesized by the DMS. Since the
that will minimize deviations of these constraints in case a fea- RTO is solved at a given instant of time, it uses a static model of
sible solution satisfying all constraints does not exist. Physical the DF instead of the dynamic used by the HAOP. In this model
constraints are hard constraints and constraints such as end-user the dynamic state equations in (1) are replaced by their al-
convenience, have the highest priority resulting in a solution that gebraic counterparts , assuming a steady-state equilibrium.
satisfies the physical constraints as well as soft constraints based Note that at time the uncertainty is realized, and this is no
on a priority order. Most of the times all constraints are satis- longer a stochastic program. The objective function is mini-
fied. Note that the time step of the HAOP is user defined and is mization of total DF losses. The resulting RTO problem is:
a multiple of the time interval between two successive solutions
of the RTO.

A. Hours Ahead Operations Planning (HAOP)

As suggested in (1), in order to optimize the DF’s cost during
a time interval [ ) the stochastic process
must be considered. This would result in a stochastic optimiza-
tion program, in which the expected value of the cost function

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on June 04,2022 at 13:13:17 UTC from IEEE Xplore. Restrictions apply.
 MELIOPOULOS et al.: ADVANCED DISTRIBUTION MANAGEMENT SYSTEM 2113

(4)

In (4), the optimal choice of active power controls of the

HAOP is enforced as a soft constraint for the RTO through the
relaxation variables (L is positive-definite matrix).
Note that the objective of the real time control can also be
the minimization of the voltage magnitude variations of all the
buses of the DF by changing the formulation of the objective
function in (4) to:

(5)

VI. EXAMPLE RESULTS

The test system shown in Fig. 5(a) is utilized to demonstrate
simulation results with the proposed advanced distribution
management system. A typical feeder is shown connected to a
substation. A 200 kW/1 MWhr battery energy storage system
(ESS), a 600 kW wind turbine, and a 500 kW solar array are
connected to the feeder, in addition to several clusters of loads,
whose maximum capacity is 2.4 MW. For the purposes of
this study the system is simulated during a summer day with
peak load of 2150 kW. To illustrate the approach only TCLs Fig. 5. Sample test system. (a) Single-line diagram. (b) Hourly load and dis-
are considered controllable and their maximum capacity is tributed generation data.
400 kW. The hourly data used is summarized in Fig. 5(b).
The simulation was executed in Matlab using the optimization
toolbox to solve the problems described in Section V. As
stated in Section VI, all problems are in second-order cone
programming (SOCP) format.
The HAOP is able to solve the day-ahead optimization
problem for this feeder and obtain the optimal controls for the
TCLs and the ESS. The objective function used in this case is a
linear one, shown in (3), and the vector c has been appropriately
chosen to achieve peak reduction. Specifically for TCLs, it is
assumed that customers have a preferred temperature of
but are willing to tolerate a fluctuation of providing a
margin for control. To demonstrate the usefulness of the pro- Fig. 6. Optimal versus baseline TCL control. (a) Total TCL load in the feeder.
posed optimal TCL control, it is compared against a baseline (b) Temperature inside a residential building.
case, in which all TCLs are controlled via a traditional “dead-
band-type” control, where the temperature fluctuates between to a simple principle of operation: charge during the off-peak
and . Our controller uses limits that are one degree hours and discharge during the peak hours. The battery’s stored
above the thermostat, i.e., the band [ to ]. As shown energy, which is a vital constraint for its operation, is shown in
in Fig. 6(a) the optimal TCL control achieves a peak reduction Fig. 7(b).
of ( of peak air-conditioner load for that The combined effects of the ESS and the TCL control are
day). shown in Fig. 8, where the baseline, corresponding to the hourly
The proposed HAOP formulation ensures customer comfort, load curve shown in Fig. 5(b), is compared with the hourly ac-
as verified by Fig. 6(b). The temperature inside the customer’s tive power consumed by the entire feeder when it is under the
house is kept within the tolerance bounds, increasing slightly proposed DMS control.
during the peak hours. However, this reduced usage of the TCLs As outlined in Section V, the HAOP is only the first stage
is enough to cause a sizeable overall peak reduction of the proposed control scheme. RTO is equally vital, to ensure
Control of the ESS charge/discharge schedule is also de- a smooth voltage profile across the feeder. It receives the op-
termined by the HAOP. Fig. 7(a) shows the battery system’s erating schedule determined by the HAOP and upholds it as a
charge/discharge schedule during the day which corresponds constraint, which can be relaxed if needed, as shown in (4). To

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on June 04,2022 at 13:13:17 UTC from IEEE Xplore. Restrictions apply.
 2114 IEEE TRANSACTIONS ON SMART GRID, VOL. 4, NO. 4, DECEMBER 2013

VII. ADDITIONAL APPLICATIONS: EXAMPLE:

RESERVE-O-METER
The proposed system enables a wide range of new applica-
tions, since the utility now has enhanced visibility of all the
components, customer or utility owned, along the distribution
feeder. As an example, an application is described in [24] which
utilizes state estimation of aggregate TCLs to manage energy/
power imbalances in the power system.
In this paper we describe another application, named Re-
serve-O-Meter (RoM), which uses the component real time
model, as obtained by the DSE, to calculate how much reserve
Fig. 7. ESS charge/discharge cycle. (a) Battery consumed power. (b) Battery capacity is available from that component (e.g., a responsive
stored energy.
load). This application enhances the overall system’s reliability
and reduces the cost of reserve, given that it secures additional
reserves from intelligent monitoring of assets without incurring
additional fuel or monetary costs, unlike fossil fuel spinning
reserves.
The proposed RoM module calculates the reserve available
from components with controllable active power generation
(i.e., distributed generation or storage plants) or active power
demands (i.e., controllable loads), subject to the binding con-
straints outlined in Section V. Hence the proposed approach
guarantees that the reserve value calculated can be called
upon during power system emergencies without violations
of these constraints, and thus it is superior to load shedding
schemes, which impose curtailment of load inconveniencing
the end user. To ensure no constraint violation, the amount of
reserve power available from one component is not enough
Fig. 8. Hourly feeder consumption: baseline versus proposed approach.
information: the time for which it can be sustained before the
local constraints are violated should also be calculated. One
should consider the simple example of a house air-conditioner:
if the state estimation reports that the air conditioner is on at
time , it is feasible to turn it off only as long as the house
temperature, also given by the DSE, is below the maximum at
. If it is turned off, it can remain turned off only for as long as
the temperature of the house remains within the desired limits.
This can be calculated by simulation using the house thermal
model, as shown in Appendix A.
Fig. 10 shows the combined hourly available reserve from the
TCLs connected to the sample feeder of Fig. 5(a). The total TCL
load in the feeder for the same day is shown in Fig. 6(a). The
Fig. 9. Voltage profile and reactive power injection of wind turbine. (a) Hourly importance of the underlying reserve power versus reserve du-
voltage profile at pcc. (b) Hourly reactive power injection.
ration tradeoff is evident: less reserve power is available during
the off peak hours, simply because less TCLs are operational at
demonstrate the purpose of the RTO, the daily voltage profile the time, but the reserve is available for longer periods of time.
at the point of common coupling (PCC) of the wind turbine is On the contrary, during the peak hours more reserve power is
shown in Fig. 9(a). The case where the ESS, the wind turbine available, but for much smaller time, since the temperature of
and the solar array are operating with unit power factor (dashed the cooled space tends to be closer to the highest acceptable
line) is considered the baseline. The objective function used for limit during that period as shown in Fig. 6(b). The proposed
the RTO in this case is (5). RoM module is able to capture that tradeoff, and ensures that
The proposed RTO scheme is proven capable to prevent the the true reserve providing capabilities of the DN are captured.
over voltages during the early hours when the wind turbine is
almost at full capacity and the load is minimal, by determining VIII. CONCLUSIONS AND COMMENTS
the wind turbine must consume reactive power. The opposite This paper has presented a new approach for optimizing, con-
decision is made during the peak hours, during which the wind trolling and protecting a distribution system with distributed re-
farm must generate reactive power to mitigate the low voltages. sources, utility and customer owned. The available instrumen-
Overall, a healthy voltage profile is achieved. The situation is tation for protection of the various components of the system
similar for the rest of the feeder. is utilized to provide the measurements which are utilized in a

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on June 04,2022 at 13:13:17 UTC from IEEE Xplore. Restrictions apply.
 MELIOPOULOS et al.: ADVANCED DISTRIBUTION MANAGEMENT SYSTEM 2115

where:

component internal state vector;

component external state vector (across
variables);
component control vector;
component through variable;
component uncertain parameters;
state equations;
component constraints.
Fig. 10. Hourly available TCL reserve and reserve duration for feeder being
studied. Consider a residential house with two TCLs: An electric
water heater (EWH) whose water temperature is state and
distributed state estimation that provides the real time model of an air conditioner (AC), with representing the temperature
the system at speeds of 10 times per second. The distributed inside the house. For this model, the ambient temperature is
state estimation is enabled with a new approach for protection an exogenous random variable.
of the various components of the distribution system referred to The standard form model of the two TCLs is:
as setting-less protection. Setting-less protection uses synchro-
nized measurements and a dynamic state estimation that pro-
vides the real time model of the system in the time domain. A
virtual PMU converts the time domain real time model into a
frequency domain real time model and transmits it to the dis-
tribution management system where the real time model of the
entire system (feeder) is synthesized.
The real time model enables the optimization and control of
the system via a two level hierarchical optimization procedure.
The two optimization problems have been formulated and ex-
ample results have been presented.
(7)
The proposed system enables development of additional ap-
plications. One such application is described in a paper listed in
the references and another is described in this paper. This ap-
thermal capacitance of the EWH and the
plication has been named “Reserve-O-Meter,” or RoM. It com-
house;
putes in real time available reserves from customer owned as
well utility owned resources that can be used in case of emer- thermal conductance between house and
gency. The RoM is a real time tool that reduces the required re- EWH;
serves from large fossil units that normally must be on standby. thermal conductance between house and
Example results of the proposed approach are provided for an ambient;
example test system.
The paper is focused on the technical aspects of the approach. duty cycle of the EWH (control variable);
Since customer participation is important for realizing the ben- duty cycle of the AC (control variable);
efits of this system, incentives to entice customer participation
are very important. These issues have not been discussed in the coefficients of performance of EWH and AC.
paper but will be addressed in subsequent papers. In (7), the inequality constraints on the EWH and house tem-
peratures are binding constraints for the model and ensure the
APPENDIX “no customer inconvenience” requirement.
COMPONENT MODEL EXAMPLE Fig. 11 is an example illustrating how the directives from
two-level optimization would coordinate with the local air con-
This Appendix presents the standard form of each compo- ditioner control, for a 2 kW AC unit. The data shown in the
nent model and an example of a model used in the proposed figure are taken from the example test system, which includes
approach, that of a residential house with thermostatically con- this 2 kW AC unit.
trolled loads (TCLs).
The general format of the component model is:
ACKNOWLEDGMENT
(6.a)
The work presented in the paper has been supported by NSF,
(6.b) DoE, EPRI, PSERC, NEC, and GE. Their support is greatly
(6.c) appreciated.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on June 04,2022 at 13:13:17 UTC from IEEE Xplore. Restrictions apply.
 2116 IEEE TRANSACTIONS ON SMART GRID, VOL. 4, NO. 4, DECEMBER 2013

[11] G. Celli, F. Pilo, G. Pisano, and G. G. Soma, “Optimal operation of

active distribution networks with distributed energy storage,” in Proc.
2012 IEEE Int. Energy Conf. Exhib. (ENERGYCON), Sep. 2012, pp.
557–562.
[12] F. Pilo, G. Pisano, and G. G. Soma, “Advanced DMS to manage active
distribution networks,” in Proc. 2009 IEEE Bucharest Power Tech, Jul.
2009, pp. 1–8.
[13] A. Molderink, V. Bakker, M. G. C. Bosman, J. L. Hurink, and G. J. M.
Smit, “Management and control of domestic smart grid technology,”
IEEE Trans. Smart Grid, vol. 1, no. 2, pp. 109–119, Sep. 2010.
[14] B. R. Sathyanarayana and G. T. Heydt, “A roadmap for distribution
energy management via multi-objective optimization,” in Proc. 2010
IEEE PES Gen. Meet., Minneapolis, MN, USA, Jul. 2010, pp. 1–8.
[15] S. Choi, B. Kim, G. J. Cokkinides, and A. P. Meliopoulos, “Feasibility
study: Autonomous state estimation in distribution systems,” IEEE
Trans. Power Syst., vol. 26, no. 4, pp. 2109–2117, Nov. 2011.
[16] A. P. Meliopoulos, G. J. Cokkinides, C. Hedrington, and T. L. Conrad,
“The supercalibrator — a fully distributed state estimator,” in Proc.
2010 IEEE PES Gen. Meet., Minneapolis, MN, USA, Jul. 2010, pp.
1–8.
[17] R. Huang, E. Farantatos, G. J. Cokkinides, A. P. Meliopoulos, B. Far-
danesh, and G. K. Stefopoulos, “Object-oriented 3-Phase distributed
dynamic state estimator,” in Proc. 2011 IEEE PES Gen. Meet., Detroit,
MI, USA, Jul. 2011, pp. 1–8.
[18] I. Choi and J. Lee, “Development of smart controller with demand re-
sponse for AMI connection,” in Proc. 2010 Int. Conf. Control, Autom.,
Syst., Korea, Oct. 2010, pp. 752–755.
[19] V. C. Gungor, D. Sahin, T. Kocak, S. Ergut, C. Buccella, C. Cecati,
and G. P. Hancke, “A survey on smart grid potential applications and
communication requirements,” IEEE Trans. Ind. Inf., vol. 9, no. 1, pp.
28–42, Feb. 2013.
[20] A. P. Meliopoulos, G. J. Cokkinides, Z. Tan, S. Choi, Y. Lee, and P.
Myrda, “Setting-less protection: Feasibility study,” in Proc. 2013 46th
Hawaii Int. Conf. Syst. Sci. (HICSS), Maui, HI, USA, Jan. 2013, pp.
2345–2353.
Fig. 11. Air conditioner control example: (a) optimal control directives; (b) [21] A. P. Meliopoulos, G. J. Cokkinides, and G. K. Stefopoulos, “Symbolic
original local control; (c) local control interfered by optimal control directives; integration of dynamical systems by collocation methods,” in Proc.
(d) room temperature with local control; (e) room temperature with optimal con- 2005 IEEE PES Gen. Meet., pp. 817–822.
trol; (f) power consumption with optimal control directives; (g) ambient tem- [22] A. P. Meliopoulos, G. J. Cokkinides, and G. K. Stefopoulos, “Quadrati-
perature of a day. zation integration method,” in Int. Conf. Power Syst. Transients, Mon-
treal, QC, Canada, 2005.
[23] Y. Nesterov and A. Nemirovsky, Interior-Point Polynomial Algo-
REFERENCES rithms in Convex Programming. Philadelphia, PA, USA: SIAM,
1994.
[24] J. L. Mathieu, S. Koch, and D. S. Callaway, “State estimation and con-
[1] M. Negnevitsky, D. T. Nguyen, and M. de Groot, “An agent-based trol of electric loads to manage real-time energy imbalance,” IEEE
market-clearing scheme for the exchange of demand response,” in Trans. Power Syst., vol. 28, no. 1, pp. 430–440, Feb. 2013.
Proc. 2011 IEEE PES Gen. Meet., San Diego, CA, USA, Jul. 2011, [25] R. A. Jabr, “Optimal power flow using an extended conic quadratic
pp. 1–6. formulation,” IEEE Trans. Power Syst., vol. 23, no. 3, pp. 1000–1008,
[2] H. Saele and O. Grande, “Demand response from household customers: Aug. 2008.
Experiences from a pilot study in norway,” IEEE Trans. Smart Grid,
vol. 2, no. 1, pp. 102–109, Jan. 2011.
[3] D. T. Nguyen, M. Negnevitsky, and M. de Groot, “Walrasian market
clearing for demand response exchange,” IEEE Trans. Power Syst., vol.
27, no. 1, pp. 535–544, Feb. 2012.
[4] Z. Zhou, F. Zhao, and J. Wang, “Agent-based electricity market simu-
lation with demand response from commercial buildings,” IEEE Trans.
Smart Grid, vol. 2, no. 4, pp. 580–588, Dec. 2011. A. P. Sakis Meliopoulos (M’76–SM’83–F’93) was
[5] C. Su and D. Kirschen, “Quantifying the effect of demand response born in Katerini, Greece, in 1949. He received the
on electricity markets,” IEEE Trans. Power Syst., vol. 24, no. 3, pp. M.E. and E.E. diploma from the National Technical
1199–1207, Aug. 2009. University of Athens, Greece, in 1972; the M.S.E.E.
[6] F. Rahimi and A. Ipakchi, “Demand response as a market resource and Ph.D. degrees from the Georgia Institute of Tech-
under the smart grid paradigm,” IEEE Trans. Smart Grid, vol. 1, no. nology, Atlanta, GA, USA, in 1974 and 1976, respec-
1, pp. 82–88, Jun. 2010. tively. In 1971, he worked for Western Electric in
[7] S. Lu, N. Samaan, R. Diao, M. Elizondo, C. Jin, E. Mayhorn, Y. Zhang, Atlanta. In 1976, he joined the Faculty of Electrical
and H. Kirkham, “Centralized and decentralized control for demand Engineering, Georgia Institute of Technology, where
response,” in Proc. 2011 IEEE PES Innov. Smart Grid Technol. (ISGT), he is presently a Georgia Power Distinguished Pro-
Jan. 2011, pp. 1–8. fessor. He is active in teaching and research in the
[8] H. Salehfar and A. D. Patton, “Modeling and evaluation of the system general areas of modeling, analysis, and control of power systems. He has made
reliability effects of direct load control,” IEEE Trans. Power Syst., vol. significant contributions to power system grounding, harmonics, and reliability
4, no. 3, pp. 82–88, Jan. 1989. assessment of power systems. He is the author of the books Power Systems
[9] K. Ngand and G. B. Sheble, “Direct load control-A profit-based load Grounding and Transients (Marcel Dekker, June 1988), “Lightning and Over-
management using linear programming,” IEEE Trans. Power Syst., vol. voltage Protection,” section 27, Standard Handbook for Electrical Engineers
13, no. 2, pp. 688–694, May 1998. (McGraw Hill, 1993). He holds three patents and he has published over 220
[10] N. Ruiz, I. Cobelo, and J. Oyarzabal, “A direct load control model for technical papers. In 2005 he received the IEEE Richard Kaufman Award.
virtual power plant management,” IEEE Trans. Power Syst., vol. 24, Dr. Meliopoulos is the Chairman of the Georgia Tech Protective Relaying
no. 2, pp. 959–966, May 2009. Conference, a Fellow of the IEEE and a member of Sigma Xi.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on June 04,2022 at 13:13:17 UTC from IEEE Xplore. Restrictions apply.
 MELIOPOULOS et al.: ADVANCED DISTRIBUTION MANAGEMENT SYSTEM 2117

Evangelos Polymeneas (S’12) was born in Athens, Renke Huang (S’09) was born in Wenzhou, China
Greece. He received a Diploma in ECE from Na- in 1984. He received the B.S. degree in electric
tional Tchnical University of Athens, Greece in 2010. power system and automation from the Shanghai
He is currently pursuing his Ph.D. in the Department Jiao Tong University, China, in 2006 and the M.S.
of Electrical and Computer Engineering, Georgia degree in E.C.E. from the Georgia Institute of Tech-
Institute of Technology, Atlanta, GA, USA. His nology, Atlanta, GA, USA, in 2009. He is currently
research interests include large-scale power system pursuing his Ph.D. at the School of Electrical and
estimation and control focusing on optimization of Computer Engineering of Georgia Institute of Tech-
power system operation under uncertainty. nology. His research interests include power systems
automation, control, and the application of digital
signal processing technologies in power system area.

Zhenyu Tan (S’12) received a B.E. degree in

electrical engineering from Tsinghua University in Dongbo Zhao (S’10) was born in Shenyang, China,
Beijing, China, in 2011. He is currently pursuing a in 1985. He received his B.S. and M.S. degree in elec-
Ph.D. degree, working in the Power System Control trical engineering from Tsinghua University, China,
and Automation Laboratory at the Georgia Institute in 2008 and from Texas A&M University, College
of Technology, Atlanta, GA, USA. His research Station, Tx, USA, in 2010 respectively. He is cur-
interests lie in power system generator protection rently a Ph.D student in Dept. of Electrical and Com-
and control, distributed power system automation. puter Engineering of Georgia Institute of Technology,
Atlanta, GA, USA. He has worked with ABB Inc.
US Corporate Research Center located in Raleigh,
NC from 2010 to 2011. His research interests include
power system control, protection, reliability analysis,
transmission and distribution automation, and electric market optimization.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY ROORKEE. Downloaded on June 04,2022 at 13:13:17 UTC from IEEE Xplore. Restrictions apply.