IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 52, NO.

3, MAY/JUNE 2016 1945

Optimal Energy Management in an Industrial Plant

Using On-Site Generation and Demand Scheduling
Moein Choobineh, Student Member, IEEE, and Salman Mohagheghi, Senior Member, IEEE

Abstract—Fierce global competition and reduced profit margins Pi Workstation i rated power, i.e., power consumed
are forcing many manufacturing plants to maximize their opera- by workstation i when working at full capacity
tional efficiency. This includes the consumption of energy across (kW).
the plant as well as the inventory buildup. To reduce the cost of
energy purchased from the electric utility, the plant operator may P b,max Battery capacity (kW).
use the available on-site generation or reschedule the operation of Pts Power available from solar resource at time
one or more workstations so that the plant is less heavily opera- period t (kW).
tional during peak-load hours when the price of electricity is high. P u,max Maximum power available from the utility (kW).
All this must be done while considering the interdependencies rtb Revenue from selling battery power to the utility
between various workstations as well as the target production level
(and possible penalties in case it is not met). These factors turn the at time period t ($/kWh).
energy management problem into a constrained optimization one rtDR Revenue from participating in demand response
where the operational constraints of the workstations, the capacity (DR) at time period t ($/kWh).
limits of available energy resources, and the financial information rts Revenue from selling solar power to the utility at
are all taken into account. Developing such a solution is the focus of time period t ($/kWh).
this current paper. A nonlinear mixed-integer optimization prob-
lem is formulated here that tries to optimize the performance of SOCb,max Maximum acceptable state of charge (SOC) for
an industrial plant subject to the above operational and techni- battery.
cal constraints. The problem is solved for two scenarios, when the SOCb,min Minimum acceptable SOC for battery.
reverse power injection by the plant is allowed and when it is not. t Time index.
I des Desired production level of the plant (number of
Index Terms—Demand response (DR), demand scheduling, dis-
tributed energy resources (DERs), energy management, energy units, volume, or weight produced); this can be
storage, factory automation, industrial plant operation. viewed as the postprocess inventory of the final
workstation.
vij Binary parameter indicating whether workstation
N OMENCLATURE i is directly supplied from workstation j, i.e.,
Parameters uses its product (= 1: if i is connected to j, 0:
cbt Cost of power provided by the battery at time otherwise).
period t ($/kWh). αi Factor relating the number of
cut Cost of power provided by the utility at time units/volume/weight produced by workstation i
period t ($/kWh). to the power consumed by it
i, j Workstation index. (number of units/kW, m3 /kW, or kg/kW).
Iimax Maximum acceptable postprocess inventory for δ Battery discharge rate (p.u.).
workstation i (number of units, volume, and η b,c Charging efficiency for battery.
weight). η b,d Discharging efficiency for battery.
Iimin Minimum acceptable postprocess inventory for π Penalty due to unfulfilled order
workstation i (number of units, volume, and ($/unit, $/m3 , $/kg).
weight). ρ Revenue of end product ($/unit, $/m3 , $/kg).
n Number of workstations, and the index for the
last workstation that produces the final product. Variables
Ii,t Postprocess inventory at the output of
Manuscript received June 22, 2015; accepted November 3, 2015. Date of workstation i at time period t (number of units,
publication December 22, 2015; date of current version May 18, 2016. Paper volume, and weight).
2015-ESC-0457, presented at the 2015 IEEE Industry Applications Society Ptb,c Power provided to charge the battery at time
Annual Meeting, Addision, TX, USA, October 18–22, and approved for pub- period t (kW).
lication in the IEEE T RANSACTIONS ON I NDUSTRY A PPLICATIONS by the
Energy Systems Committee of the IEEE Industry Applications Society. Ptb,d Power discharged by the battery at time period
The authors are with the Electrical Engineering and Computer Science t (kW).
Department, Colorado School of Mines, Golden, CO 80401 USA (e-mail: Ptb,p Battery power provided to the plant at time
mchoobin@mines.edu; smohaghe@mines.edu).
period t (kW).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. Ptb,u Battery power sold to the utility at time period
Digital Object Identifier 10.1109/TIA.2015.2511094 t (kW).
0093-9994 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
1946
Pts,b PV power provided to charge the battery at
time period t (kW).
Pts,p PV power provided to the plant at time period t
(kW).
Pts,u PV power sold to the utility at time period t
(kW).
Ptu,b Utility power provided to the battery at time
period t (kW).
Ptu,p Utility power provided to the plant at time
period t (kW).
SOCbt SOC of battery at time period t.
ub,c Binary variable indicating that the battery is
t
being charged at time period t (= 1 : charged,
0: not being charged).
ub,d Binary variable indicating that the battery is
t
being discharged at time period
t(= 1 : discharged, 0: not being discharged).
ui,t Integer variable indicating the operational level
of workstation i at time period t,
ui,t ∈ 0, . . . , 4 which is used to indicate
operation at 0%, 25%, 50%, 75%, and 100% of
rated capacity, respectively.
u∗i,t Integer variable indicating the operational level
of workstation i at time period t under nominal
operation (i.e., no DR).
β Unfulfilled orders; shortage in production level
of the last workstation compared to the desired
level (number of units, m3 , kg).
I. I NTRODUCTION
E NERGY efficiency is an integral part of the operation
of an industrial plant. Ideally, the plant operator would
like to maximize the production level (and hence the revenues
gained) while minimizing the energy consumption, the inter-
ruption in production, the material waste, and the inventory
buildup, to name a few. These may at times behave as contra-
dictory objective functions in a sense that optimizing one could
result in suboptimal solutions for one or more others. As such,
optimal operation of the plant requires solving a comprehensive
multiobjective optimization model that takes various elements
into account.
Many researchers have focused on using on-site distributed
energy resources (DERs) to improve the reliability of the plant,
while reducing the operational costs [1]–[4]. These DER units
can shield the plant from unexpected loss of service or power
quality issues originated on the utility side [5]. In the most
general case, DERs can be distributed generation (DG) units
or energy storage systems (ESSs). The former can be further
classified as dispatchable units, e.g., diesel engines that can fol-
low generation set-point commands, or nondispatchable ones,
e.g., solar and small scale wind turbines that typically oper-
ate based on maximum power point tracking (MPPT) [6]. In
either case, the power available through these units can be
used to supplement the power purchased from the utility dur-
ing the hours of peak energy prices or to sell power back to
IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 52, NO. 3, MAY/JUNE 2016
the utility during peak-load hours when the utility may be fac-
ing shortage of supply. This changes the notion of power flow
from the traditional unidirectional scheme (i.e., from the utility
to the load) to one that allows a bidirectional exchange of power
between the utility and the plant. Plant energy management sys-
tem (EMS) would then have to determine the most economical
and efficient course of action at each point in time.
In addition to on-site DER, plant operators can take advan-
tage of demand scheduling to increase the efficiency of the
plant. Here, the objective is to schedule the operation times of
various workstations in order to minimize energy consumption,
wasted labor, and/or wasted materials. In the literature, different
objective functions have been defined for real-time schedul-
ing of loads in an industrial plant. El-Metwally et al. [7] used
actions such as demand shifting and peak shaving to improve
the load factor of the plant. Others have tried to optimally allo-
cate resources to loads [8], [9]. For instance, Subramanian et al.
[8] modeled deferrable loads as tasks with attributes such as
arrival time, departure time, and energy requirements, and pro-
posed algorithms for resource allocation to these tasks based
on their energy needs and/or deadlines. Matching the overall
load profile with a specific target load profile has also been
defined as an objective for load shifting [10]. From a differ-
ent aspect, Dobrin et al. [11] combined offline scheduling and
fixed-priority scheduling to achieve flexibility while coping
with complex timings of different tasks. Industrial load schedul-
ing has also been addressed with the goal of minimizing the
total cost of energy [12], [13].
In addition to demand scheduling, DR is becoming an impor-
tant feature of the modern power and energy systems [14]. DR
can be initiated by the utility in the form of demand reduction
signals/commands, or can be initiated by the plant itself, i.e.,
in the form of capacity relief to be sold to the utility. It is a
voluntary form of load shedding that is performed in exchange
for financial incentives. Even though much of focus in DR is
on the residential and commercial sectors, many utilities are
gaining more interest in engaging the industrial sector in this
opportunity. This is mostly due to the potentially large amount
of congestion relief that could be made available through these
customers. In many utilities, industrial loads account for around
2%–10% of the total number of customers, but consume up to
70%–80% of the total power generated by the utility [15], [16].
To determine the effect of demand scheduling on the opera-
tion of an industrial plant, different constraints will have to be
considered. One of the most critical constraints is the inven-
tory buildup across the plant, which is directly translated into
wasted material. Naturally, demand scheduling can cause an
increase in the work-in-process inventory. In order to determine
the capital value of this inventory buildup, the products can be
divided into three different categories: A, B, and C [17]. The
products in the A category have the highest value in the orga-
nization. The B category also consists of important products,
but is less strictly monitored. And the third category C is rel-
atively less important than the other two; although, it usually
has larger quantities. This categorization can facilitate the selec-
tion of workstations for demand curtailment [18]. Another set
of constraints that should be considered prior to implementa-
tion of demand scheduling relates to the characteristics of the
CHOOBINEH AND MOHAGHEGHI: OPTIMAL ENERGY MANAGEMENT IN AN INDUSTRIAL PLANT 1947

workstations. These may include the capacity of the worksta-

tion (products processed per unit of time), the layout of the
workstations (series or parallel), and the number and skill lev-
els of crew members working on the workstation. For instance,
shifting the operation of a workstation in a series layout to a
later time may have more adverse effects on the plant operation
than in a parallel layout [19]. Likewise, rescheduling a worksta-
tion with a large number of specifically trained skilled workers
can lead to wasted labor.
Finally, demand curtailment could be considered as an alter-
native. Intelligent load shedding in an industrial plant has been
addressed using priority ranking of the loads [7], expert sys-
tems [20], [21], and neural networks [22]. Also, solutions have
been proposed for demand conservation through demand side
management (DSM) techniques [23].
The objective of this paper is to devise a solution for optimal
energy management in an industrial plant using all poten-
tial resources of energy. To achieve this, the solution will
take advantage of any available on-site generation, as well as
scheduling of different workstations. The ultimate objective is
to maximize the financial profits by the plant while the target Fig. 1. Schematic diagram of the proposed solution.
production is met as closely as possible and the operational con-
straints are all maintained. A case study is presented using a and a battery system as energy storage. Naturally, other types
simplified layout for a vehicle cockpit assembly plant in order of DER units can be added without changing the nature of the
to demonstrate the applicability of the proposed methodology. problem.

B. Problem Formulation
II. P ROPOSED M ETHODOLOGY
The objective function of the optimization problem can be
A. Overall Scheme expressed as follows:
Fig. 1 illustrates the schematic diagram of the proposed solu- ⎧ T
⎪
⎪ ρ t=1 un,t αn P4n − π · β
tion for energy management. The energy management module ⎪
⎪ T T
⎪
⎪ b b,u s s,u
needs access to the plant configuration and the layout of the ⎨+ t=1 rt Pt + t=1 rt Pt 
T n ∗
workstations. This information is needed in order to model Max + t=1 rtDR i=1 4 (ui,t − ui,t ) +
Pi
(1)
⎪
⎪ T u u,b
the interdependencies between the workstations. Financial data ⎪
⎪− t=1 ct (Pt + Pt )
u,p
⎪
⎪
are also necessary, and could be provided by the plant opera- ⎩ T b b,p
− t=1 ct (Pt + Ptb,u )
tor as well as the electric utility. These may include the cost
of purchasing electricity from the utility, revenue from poten- where the first row represents the profit made due to plant’s
tial injection of power into the utility’s network, revenue from production and the possible penalty due to failure in meeting
participation in DR, and the O&M cost of using the on-site the target production level, the second row represents the rev-
generation and storage. Other relevant information could be enue from selling battery power and solar power to the electric
the profit gained from selling each unit of final product (unit, utility, the third row represents the revenue due to participating
weight, or volume), and the possible penalties due to failure to in the DR program and is nonzero only if the plant reduces its
fulfill the desired production. It is also crucial for the module to consumption compared to the nominal operation, the fourth row
receive updated data from the field. These could be the genera- is the cost of purchasing power from the utility, and the last row
tion status of the plant energy resources, the charge/discharge indicates the operational cost of discharging the battery power
status of the ESS, the inventory buildup status at the output either to provide power to the utility or to the plant. This is a
of each individual workstation, and suchlike. This information constrained optimization problem that is solved subject to the
can be obtained from the remote terminal units (RTUs) that are following constraints.
installed across the plant working as a communication gateway 1) Power balance for the workstations: This constraint
to the industrial automation system. ensures that the power provided to the plant from vari-
Upon receiving all necessary information, the energy man- ous resources is sufficient for running the workstations at
agement module determines the most cost-effective operation the intended capacities
schedule by solving an optimization problem. This output will n Pi
be provided to the plant operator who then decides whether or ∀t : Ptu,p + Ptb,p + Pts,p = ui,t . (2)
i=1 4
not to implement it.
In Section II-B, without loss of generality, it has been 2) Power balance for the power available from PV: The
assumed that the plant has a PV panel as the generation resource power available from this resource may be used to
1948 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 52, NO. 3, MAY/JUNE 2016

supply the workstations, charge the battery, or sell to the 7) Postprocess inventory constraints: These indicate the
utility accumulation of postprocess inventory based on the inter-
relationships between workstations. In addition, individ-
∀t : Pts,u + Pts,p + Pts,b = Pts . (3)
ual workstations will have upper and lower limits for the
3) Upper limit for the power available from the utility: The acceptable amount of postprocess inventory. A nonzero
power that can be purchased from the electric utility may value for I min moves the process further away from a just-
be constrained by the physical limitations of the feeder in-time (JIT) manufacturing strategy. It should be noted
supplying the plant (e.g., rating of the distribution line, that depending on the units defined for the output of each
rating of the transformer, etc.) or due to the existing product, one can consider the postprocess inventory to be
contract with the utility a discrete or a continuous variable
∀t : Ptu,p + Ptu,b ≤ P u,max . (4) Pi
n
Pj
∀(i, t) : Ii,t = Ii,t−1 + αi ui,t − αj uj,t vji
4) Battery operational constraints: Constraints (5) and (6) 4 4
j=1,=i
indicate that the battery can be charged by the power (13)
available through the utility or the PV resource, and its
discharge power can be used to supply the workstations ∀(i, t) : Iimin ≤ Ii,t ≤ Iimax . (14)
or to sell power to the utility. In addition, the discharge
8) Unfulfilled orders: Ideally, the plant manager would like
power of the battery is limited by its capacity [see (7)].
to meet the target production level. However, to allow
The battery can be either charging or discharging at any
the plant to take advantage of selling power to the util-
point in time, but not both [see (8)]. Constraint (9) pro-
ity, DR, and/or demand shifting, this constraint is relaxed.
vides an equality constraint for the SOC of the battery,
Of course, the shortage in meeting the demand leads to
and (10) expresses the desired range for the SOC of the
penalties, as indicated in the objective function
battery
T Pn
∀t : Pts,b + Ptu,b = ub,c
t Pt
b,c
(5) αn un,t = I des − β (15)
t=1 4
∀t : Ptb,p + Ptb,u = ub,d
t Pt
b,d
(6) β ≥ 0. (16)
∀t : Ptb,p + Ptb,u ≤ P b,max (7)
9) Non-negativity and integrality constraints
∀t : ub,c b,d
t + ut ≤ 1 (8)
⎧ ∀t : Ptb,p , Ptb,u , Pts,b , Pts,p , Pts,u , Ptu,b , Ptu,p ≥ 0 (17)
⎪
⎨(1 − δ) · SOCt−1
b
∀(i, t) : ui,t ∈ {0, 1, 2, 3, 4} (18)
∀t:SOCt = +ut (Pt η /P b,max )
b b,c b,c b,c
(9)
⎪
⎩ b,d b,d b,d b,max ∀(i, t) : Ii,t ≥ 0 (19)
−ut (Pt /η P )
∀t : ub,c b,d
t , ut ∈ {0, 1}. (20)
∀t : SOCb,min ≤ SOCbt ≤ SOCb,max . (10)
5) Reverse power flow: It has been assumed here that the
plant can either purchase power from the utility or sell III. C ASE S TUDY
power to the utility, but not both at the same time. In order The proposed methodology is applied to a simplified version
to ensure both conditions do not occur simultaneously, of an industrial manufacturing plant for luxury vehicle cock-
(11) has been added as an additional constraint. It should pit assembly [18]. The schematic diagram of the plant layout
be noted that all power terms in (11) are non-negative is illustrated in Fig. 2. The plant consists of multiple worksta-
∀t : (Ptb,u + Pts,u ) · (Ptu,b + Ptu,p ) = 0. (11) tions that are arranged in series. The problem is solved under
two scenarios: 1) the utility does not allow the plant to inject
6) Workstation interdependencies: In a series configuration, power back into the utility’s network and 2) reverse power flow
if workstation i depends on workstation j’s product (i.e., is allowed. The problem is solved for 10 one-hour time steps,
if vij = 1), then at any point in time, its production level assuming a workday starting from 8:00 A . M . to 6:00 P. M. It is
would be limited by how much product workstation j has solved using the LINDO solver in the GAMS software.
produced until that time, and how much of it has been
consumed by workstation i during the previous time steps
(see Section V-A for a general discussion on this issue). A. Input Data
This constraint allows for modeling demand shifting to a Workstation data and battery data are provided in Tables I
later time if necessary, i.e., dynamic demand scheduling and II. The forecasted solar power available at the plant is
α i Pi assumed to be following the diagram in Fig. 3. It is assumed
∀(i, τ ) : ui,τ

4
that the maximum power that can be delivered to the plant by

τ −1  
the utility is 650 kW. This can be a limitation enforced by the
α j Pj α j Pj α i Pi
≤ vij uj,τ + uj,t − ui,t . transformer or the feeder supplying the plant.
4 4 4
t=0 It is also assumed that the desired number of vehicle cockpits
(12) to be manufactured by the plant is 30. Each unit will be sold at
CHOOBINEH AND MOHAGHEGHI: OPTIMAL ENERGY MANAGEMENT IN AN INDUSTRIAL PLANT
Fig. 2. Schematic diagram of the luxury vehicle cockpit assembly line. The
external input inventory for the workstations is not considered here.
TABLE I
BATTERY DATA
TABLE II
W ORKSTATION DATA
a price of $30 000. The cost of each unfulfilled order to the
plant is $2000. Also, it has been assumed that at the start of the
day, all workstations, except for the final one, have an initial
postprocess inventory of 10 units.
Fig. 4 illustrates the cost and revenue of power exchanges.
These values are used for demonstration purposes only.
Changing the numbers may change the output of the problem,
but does not affect its structure. It can be seen that the cost
of electricity purchased from the utility follows a critical peak
pricing (CPP) scheme, whereas the revenue obtained from sell-
ing power to the utility follows a two-peak model. The utility
buys power at higher rates during the peak-load hours; however,
during low load conditions, it purchases power at a lower rate
than what it charges the customers. The cost of power provided
by the battery is assumed to be a constant 0.07$/kWh, as it
relates to the loss of life of battery due to each additional charge
and discharge cycle. No cost is assumed for the PV panel.
1949
Fig. 3. Forecasted solar power available at the plant.
Fig. 4. Cost of energy resources and revenue from selling power to utility.
B. Case Study 1: No Reverse Power Flow
In this case study, the plant is not allowed to sell power back
to the utility; therefore, any excess power must be used for inter-
nal consumption or to charge the battery. To find the optimal
schedule, (1) is solved subject to (2)–(20), where the following
constraint is also added to ensure reverse power flow does not
occur
∀t : Ptb,u = Pts,u = 0. (21)
Figs. 5 and 6 illustrate the results. It can be seen that the
energy management module uses the solar power to charge the
battery, to then use it at hour 7 when the price of electricity is
higher. The workstations are used throughout the day, although
with heavier loading during the times of lower electricity rates
(hours 1–6).
C. Case Study 2: With Reverse Power Flow
Here, the plant is allowed to sell power back to the utility.
To find the optimal dispatch, (1) is solved subject to (2)–(20).
Figs. 7 and 8 illustrate the results. It can be seen that under
energy buyback scenario, the energy management module shifts
all the loads to the early hours. From hour 7 onwards, the plant
1950
Fig. 5. Power provided by various generation resources under case study 1.
Fig. 6. Operational statuses of the workstations under case study 1.
starts selling all its available power from the battery and the PV
to the utility. This is expected since the power selling revenue
is very high during this period. This also helps the utility since
the load of the plant is taken off its network during peak-load
conditions.
Table III lists the total number of units processed by the last
workstation, which indicates the final product manufactured by
the plant. It can be seen that in both case studies, the plant ends
up manufacturing 15 units; however, with reverse power flow
allowed, the energy management module pushes most of the
production to the early hours. The total revenue gained by the
plant (for selling the product and, if applicable, selling power
to the utility) along with the costs to the plant (due to unful-
filled orders and cost of purchasing electricity) is also listed.
The difference in the revenue in the two cases is in part related
to the power sold to the utility. It should be noted that the dif-
ference between the target production (30) and final production
(15) is mostly determined based on the cost of producing each
additional unit and the penalties due to each unfulfilled order.
Variations in the cost of electricity, the profit gained from each
unit, and/or the penalties for production shortage could change
the final production.
IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 52, NO. 3, MAY/JUNE 2016
Fig. 7. Power provided by various generation resources under case study 2.
Fig. 8. Operational statuses of the workstations under case study 2.
TABLE III
P LANT P RODUCTION DATA
D. Case Study 3: Reverse Power Flow and DR
In the final case study, it has been assumed that the plant is
allowed to participate in a DR program with the utility. Upon
reducing the overall energy consumption compared to the nor-
mal operation (case study 1 or 2), the plant receives a revenue
[see (1)]. For demonstration purposes, the revenue from DR
rtDR is assumed to be 125% of rts and rtb .
It should be noted here that DR can be implemented and ver-
ified in variety of ways [14]. Here, it has been assumed that the
plant seeks ways to reduce its demand from nominal operation
(no DR) and upon finding the appropriate operation schedule,
will reduce its consumption and notify the utility. How to verify
that the plant has in fact reduced its demand and how to identify
the nominal (no DR) consumption for comparison are issues
CHOOBINEH AND MOHAGHEGHI: OPTIMAL ENERGY MANAGEMENT IN AN INDUSTRIAL PLANT 1951

TABLE IV
P LANT P RODUCTION DATA

Table IV lists the corresponding plant production data. For

better comparison, the results when no reverse power flow is
allowed have also been reported.

IV. P RACTICAL C ONSIDERATIONS

Fig. 9. Power provided by various generation resources under case study 3.
A. Workstation Configurations
A series configuration was assumed in this paper for demon-
stration purposes. However, many plants have more compli-
cated layouts in which a workstation i may use the products of
more than one workstation. Here, the production level of work-
station i would depend on the amount of products available
from those workstations j ∈ Ji , and would naturally be limited
by the amount of product of the workstation with the low-
est available postprocess inventory. Mathematically, this can be
expressed as
α i Pi
∀(i, τ ) : ui,τ
4

τ −1   
α j Pj α j Pj α i Pi
≤ min vij uj,τ + uj,t − ui,t .
Fig. 10. Operational statuses of the workstations under case study 3. j∈Ji 4 t=0
4 4
(25)
that are determined by the bilateral contract between the utility
Clearly, the nonlinear constraint (25) can be rewritten as
and the plant. However, this aspect falls outside the scope of our
multiple linear inequality constraints, one for each j ∈ Ji .
current work and it has simply been assumed that the plant is
financially incentivized for any demand reduction with respect
to the nominal (no DR) operation. B. Solar Data
To model the nonlinear term in the third row of (1), the
following reformulation has been applied: Throughout the simulations in this paper, it was assumed
  that solar data are known in advance and can be forecasted
n Pi ∗ with accuracy. During days that are heavily windy and cloudy,
∀t : (ui,t − ui,t )
i=1 4 the accuracy of solar power forecast reduces. In this case, the
+
  energy management module may be forced to resolve the opti-
n Pi ∗ st (st + 1)
= (ui,t − ui,t ) (22) mization problem as soon as it is determined that the available
i=1 4 2
solar power at a future time step is different from what assumed
subject to the additional constraints earlier.
 
n Pi ∗
∀t : (ui,t − ui,t ) · st > 0 (23) V. C ONCLUDING R EMARKS
i=1 4

∀t : st ∈ {−1, 0, + 1}.
Figs. 9 and 10 summarize the results. Comparing Figs. 9 and
7 shows that the energy management module takes advantage
of both revenue options: selling power back to the utility and
reducing demand. Hence, workstation schedules are arranged
around the time when the incentives from DR or reverse power
flow are low.
(24) A solution was proposed in this paper for optimal energy
management across a manufacturing plant. It was assumed that
the plant could take advantage of on-site generation, demand
scheduling, and DR in order to maximize its profits. A non-
linear mixed-integer optimization problem was formulated that
balances the losses due to failure to meet the target produc-
tion level with the revenues gained from selling power back
to the utility and participating in DR. Several case studies were
1952 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 52, NO. 3, MAY/JUNE 2016
presented with and without DR, and with and without the pos-
sibility to inject excess on-site generation back into the utility’s
network.
R EFERENCES
[1] K. P. Papalexandri, E. N. Pistikopoulos, and B. Kalitventzeff, “Modelling
and optimization aspects in energy management and plant operation with
variable energy demands-application to industrial problems,” Comput.
Chem. Eng., vol. 22, no. 9, pp. 1319–1333, 1998.
[2] L. Merkert, I. Harjunkoski, A. Isaksson, S. Synevirta, A. Saarela, and
G. Sand, “Scheduling and energy industrial challenges and opportuni-
ties,” Comput. Chem. Eng., vol. 72, pp. 183–198, 2015.
[3] D. Q. Hung, N. Mithulananthan, and R. Bansal, “An optimal investment
planning framework for multiple distributed generation units in industrial
distribution systems,” Appl. Energy, vol. 124, no. 1, pp. 62–72, 2014.
[4] R. M. Ciric and N. Rajakovic, “A new composite index of reliability of
supply in the industrial systems with distributed generation,” Int. J. Elect.
Power Energy Syst., vol. 44, no. 1, pp. 824–831, 2013.
[5] T. Ackermann, G. Andersson, and L. Söder, “Distributed generation: A
definition,” Elect. Power Syst. Res., vol. 57, pp. 195–204, 2001.
[6] F. A. Farret and M. G. Simões, Integration of Alternative Sources of
Energy. Hoboken, NJ, USA: Wiley, 2006.
[7] M. M. El-Metwally, M. S. El-Sobki, H. A. Attia, and S. A. Wahdan,
“Priority ranking of industrial loads and application of demand side
management technique,” in Proc. Int. Middle East Power Syst. Conf.,
El-Minia, Egypt, Dec. 2006, pp. 341–346.
[8] A. Subramanian, M. Garcia, A. Domínguez-García, D. Callaway,
K. Poolla, and P. Varaiya, “Real-time scheduling of deferrable electric
loads,” in Proc. Amer. Control Conf., Montreal, QC, Canada, Jun. 2012,
pp. 3643–3650.
[9] M. L. Della Vedova and T. Facchinetti, “Real-time scheduling for indus-
trial load management,” in Proc. IEEE Int. Energy Conf. Exhib., Florence,
Italy, Sep. 2012, pp. 707–713.
[10] G. O’Brien and R. Rajagopal, “A method for automatically scheduling
notified deferrable loads,” in Proc. Amer. Control Conf., Washington, DC,
USA, Jun. 2013, pp. 5087–5092.
[11] R. Dobrin, G. Fohler, and P. Puschner, “Translating off-line schedules
into task attributes for fixed priority scheduling,” in Proc. IEEE Real-Time
Syst. Symp., London, U.K., Dec. 2001, pp. 225–234.
[12] K. Collins, M. Mallick, G. Volpe, and W. G. Morsi, “Smart energy moni-
toring and management system for industrial applications,” in Proc. IEEE
Elect. Power Energy Conf., London, ON, Canada, Oct. 2012, pp. 92–97.
[13] S. Bahrami, F. Khazaeli, and M. Parniani, “Industrial load scheduling
in smart power grids,” in Proc. Int. Conf. Elect. Distrib., Stockholm,
Sweden, Jun. 2013, pp. 1–4.
[14] S. Mohagheghi, J. Stoupis, Z. Wang, Z. Li, and H. Kazemzadeh,
“Demand response architecture—Integration into the distribution man-
agement system,” in Proc. 1st IEEE Int. Conf. Smart Grid Commun.,
Gaithersburg, MD, USA, Oct. 2010, pp. 501–506.
[15] J. G. Roos and I. E. Lane, “Industrial power demand response analysis
for one-part real-time pricing,” IEEE Trans. Power Syst., vol. 13, no. 1,
pp. 159–164, Feb. 1998.
[16] Y. Tang, F. Xu, and L. Chen, “Research into possibility of smart industrial
load participating into demand response to supply the power system,”
in Proc. China Int. Conf. Electr. Distrib., Nanjing, China, Sep. 2010,
pp. 1–5.
[17] Y. Chen, K. W. Li, and S. F. Liu, “A comparative study on multicriteria
ABC analysis in inventory management,” in Proc. IEEE Int. Conf. Man
Mach. Cybern., Singapore, Oct. 2008, pp. 3280–3285.
[18] S. Mohagheghi and N. Raji, “Intelligent demand response scheme for
energy management of industrial systems,” in Proc. IEEE Ind. Appl. Soc.
(IAS) Annu. Meeting, Las Vegas, NV, USA, Oct. 2012, pp. 1–9.
[19] S. Ghosh and R. J. Gagnon, “A comprehensive literature review and anal-
ysis of the design, balancing and scheduling of assembly systems,” Int. J.
Prod. Res., vol. 27, no. 4, pp. 637–670, 1989.
[20] B. Delfino et al., “Operation and management of the electric system for
industrial plants: An expert system prototype for load-shedding operator
assistance,” in Proc. IEEE Ind. Appl. Soc. (IAS) Annu. Meeting, Rome,
Italy, Aug. 2000, pp. 3260–3268.
[21] F. Shokooh et al., “An intelligent load shedding system application in
a large industrial facility,” in Proc. IEEE Ind. Appl. Soc. (IAS) Annu.
Meeting, Oct. 2005, pp. 417–425.
[22] C. T. Hsu, H. J. Chuang, and C. S. Chen, “Artificial neural network based
adaptive load shedding for an industrial cogeneration facility,” in Proc.
IEEE Ind. Appl. Soc. (IAS) Annu. Meeting, Oct. 2008, pp. 1–8.
[23] H. P. Inamdar and R. P. Hasabe, “IT based energy management through
demand side in the industrial sector,” in Proc. Int. Conf. Control Autom.
Commun. Energy Conserv., Jun. 2009, pp. 1–7.
Moein Choobineh (S’15) was born in Mashhad, Iran.
He received the B.Sc. degree in electrical engineer-
ing from Ferdowsi University of Mashhad, Mashhad,
Iran, in 2008, and the M.Sc. degree in power sys-
tem engineering from Iran University of Science and
Technology, Tehran, Iran, in 2011. He is currently
working toward the Ph.D. degree in electrical engi-
neering at the Colorado School of Mines, Golden,
CO, USA.
His research interests include power system opera-
tion under extreme events, energy management opti-
mization, network restoration, microgrids, and electrical asset management.
Salman Mohagheghi (S’99–M’07–SM’14) received
the B.Eng. degree from the University of Tehran,
Tehran, Iran, in 1999, the M.Sc. degree from Sharif
University of Technology, Tehran, Iran, in 2001,
and the Ph.D. degree from Georgia Institute of
Technology, Atlanta, GA, USA, in 2006, all in elec-
trical engineering.
Currently, he is an Assistant Professor with
the Electrical Engineering and Computer Science
Department, Colorado School of Mines, Golden, CO,
USA. Prior to that, he was a Senior R&D Engineer
with the ABB Corporate Research Center, Raleigh, NC, USA. His research
interests include situational awareness, power system management under uncer-
tainties, communication networks in power systems, and automation systems.