Professional Documents
Culture Documents
Abstract -- Electric demand side management (DSM) focuses provide efficient system capacity relief at a lower cost
on changing the electricity consumption patterns of end-use compared to many peaking power plants, while the service
customers through improving energy efficiency and optimizing can be offered at a faster rate [2]. Successful deployment of a
allocation of power. Demand response (DR) is a DSM solution
DR program can lead to substantial financial returns [3],
that targets residential, commercial and industrial customers,
and is developed for demand reduction or demand shifting at a benefiting both the utility and the customers.
specific time for a specific duration. In the absence of on-site Demand response programs can be roughly classified into
generation or possibility of demand shifting, consumption level three groups according to the type of action required and the
needs to be lowered. Whereas non-criticality of loads at the party initiating the action. These are incentive-based DR, rate-
residential and commercial levels allows for demand reduction based DR and DR bids [4]. The first category is the main
with relative ease, reducing the demand of industrial processes
topic of this paper. Here, a set of demand reduction signals
requires a more sophisticated solution. Production constraints,
inventory constraints, maintenance schedules and crew are issued by the utility or the DR Service Provider
management are some of the many factors that have to be taken (aggregator), and transmitted to the participating customers in
into account before one or more processes can be temporarily the form of voluntary demand reduction requests or
shut down. Some of these constraints can be viewed along the mandatory commands. Various types of resources can be
overall performance of the system, while others need to be utilized under this program such as directly controllable loads
analyzed and evaluated in real-time. A system is proposed in this that may be directly dispatched by the utility based on the
paper that dynamically ranks loads and workstations within an
industrial site as candidates for demand reduction. Constraints balance between the load and generation [5], e.g., Direct
on the daily operation of the industrial system are taken into Load Control (DLC), and loads that can be interrupted or
account in conjunction with real-time assessment of inventory reduced upon receipt of a signal from the utility, e.g.,
buildup. A fuzzy/expert-based system combined with an Interruptible and Curtailable Load (I&C). DLC loads can be
optimization module verifies whether and, if applicable, by how remotely cycled or turned off by the utility, and can normally
much the plant can participate in a utility-initiated DR event, be deployed within a relatively short period of time, whereas
while satisfying its local operational constraints. A case study is
proposed for a sample production line to further explain the I&C loads require an advance notification time that may vary
design concepts. from a few minutes to a few hours [6]. While typical
residential loads may fit well in the DLC program, I&C
Index Terms—Demand side management, demand response, scheme is more suitable for larger scale commercial and
energy management, industrial automation system, inventory industrial customers.
management. Much focus on DR has been on residential and commercial
customers. Here, demand reduction could lead to
I. INTRODUCTION inconvenience experienced by the end-users; however, the
Electric demand response (DR) is becoming one of the decision making –comply or opt-out– is relatively non-critical
main pillars of the Smart Grid paradigm. It refers to the since the types of loads to be interrupted are often lighting
changes in the electricity usage by the end-use customers loads, or heating, cooling and air-conditioning (HVAC) [7]. A
from their nominal consumption patterns. These changes in DR module will check the characteristics of the DR event
consumption are often in response to changes in the price of (time, duration, and amount) and by examining local policies
electricity over time, or to incentive payments that are and history of past events makes a decision accordingly [8],
designed to induce lower electricity use at times of high [9]. While this class of customers is easier to manage by the
wholesale market prices or when the system reliability is utility, it does not reflect the whole energy resource that can
jeopardized [1]. In addition to improving the reliability of the be potentially made available for DR event. In fact, the
power system, and making short-term impacts on the consumption level of the industrial sector is often the bigger
electricity markets, DR can make considerable environmental portion of the total load served by the utilities. In many cases,
impacts by reducing the system peak load in the long-term the industrial customers (2-10% of total customers) may
and therefore postponing the need for building new power account for up to 80% of the electricity usage [10]. This
plants and other related capacity expansion projects. DR can
further emphasizes the importance of the role of the industrial concluding remarks appear in section V of the paper.
sector in DR. Nevertheless, implementation of DR for the
industrial customers requires a different approach. For these II. DEMAND RESPONSE FOR INDUSTRIAL PLANTS
customers, inconvenience would not be the only outcome of Due to the dynamics of general loads, it takes a certain
DR. Instead, interruption of service may lead to interruption amount of time (ramp period) until the consumer demand is
of production and/or violating the daily operational and reduced to the requested levels [15]. Upon receiving the
production constraints of the plant. demand reduction signal from the utility, many industrial
In general, an industrial customer may reduce its plants can switch to backup generation units. These may
consumption through using on-site generation, energy storage, include gas-fired or diesel-fired generators or small scale
shifting the demand to a different time, curtailment of wind turbines. In the extreme cases if deployed as spinning
noncritical loads such as lighting, or through temporary shut- reserves, many of these units have a time response of a few
down of one or more processes (a more detailed discussion seconds to about 5 minutes [16], [17]. Regardless of the time
along with the review of the state of the art are provided in it takes for the unit to come online, once they start generating
section II of the paper). The last approach could potentially power, the overall demand of the industrial plant –from the
lead to a considerable reduction in demand; however, it poses utility’s standpoint– will be reduced.
some unique challenges. Depending on the criticality of the On the other hand, introduction of energy storage systems
processes shut down, the number of workstations available, for power grid applications adds to the flexibility of the
the required throughput, the maintenance schedule, the crew industrial plant operator, as it will now be possible to store
constraints and inventory restrictions, shutting down a process the energy available from non-dispatchable resources of
may create long-term impacts on the overall process line, energy such as wind and solar, and use it at a later time when
hence making DR non-justifiable economically. In fact, the DR event is initiated.
among other things, the plant operator would need to If on-site generation or energy storage are not available,
investigate the effect of shutting down a workstation on demand shifting may be used as another viable option.
meeting the customers’ demand with the given runtime [11], Maintenance scheduling, crew constraints and other
the capital value of the resulting inventory buildup [12], and operational requirements of the plant can be analyzed to see if
the workstation interrelations (series stations, several it is feasible for the production to be shifted to a later time,
workstations feeding into a common workstation, etc) [13]. e.g., when the electricity rates are low or when the DR event
In an earlier paper [14], the authors proposed an intelligent terminates. Some examples of demand shifting have been
system that, using the concepts of fuzzy systems and graph reported in [18], [19].
theoretic approaches, identified whether loads/workstations Finally, demand reduction/curtailment could be considered
existed within the plant whose demand could potentially be as an alternative. In fact, efforts have been made in the
curtailed. In order to achieve this, the system would literature to develop intelligent schemes for load shedding in
investigate the various operational constraints of the industrial an industrial plant [18], [20]-[22]. Much of the focus in load
process as well as the interrelations of the workstations with shedding is to maintain the security and stability of the
one another. In the approach proposed in [14], inventory isolated industrial system –following a disturbance in the
(both pre-process and post-process) was viewed as a binary grid– when using backup generation. Methods have been used
variable. In other words, only existence of inventory was based on priority ranking of the loads [18], expert systems
considered in the analysis, and not its amount. Also, the level [20], [21], and neural networks [22]. Also, solutions have
of energy consumed by each workstation was not included in been proposed for energy efficiency and demand conservation
the analysis. This paper extends the work in [14] by looking through DSM techniques [18], [23].
into these additional variables and constraints, namely, the However, efficient demand reduction for a DR event
dynamic variations of different post-process inventories, the requires a more detailed analysis and thorough evaluation of
overall demand of the plant, and the energy consumption the industrial process. Through analyzing the operational
level of different workstations. This would present a dynamic constraints of the plant, processes can be prioritized as
DR solution (compared to the quasi-dynamic approach candidates for being shut down or for operating at reduced
proposed in [14]). The additions to this modified approach capacity. The revenue losses incurred by the plant must be
are essential for integrating the proposed solution into a larger compared with the direct gains, i.e., incentive payments
factory automation framework. and/or discounted rates, as well as indirect gains, e.g.,
The rest of the paper is organized as follows. Section II concurrence of the process shut-down with an upcoming
summarizes some considerations and constraints for the maintenance scheduling or crew bottleneck. This can be
industrial plant operation as it relates to demand response. either used as a standalone DR tool, or in conjunction with
The description of the proposed system along with the design other means such as demand shifting and on-site generation.
procedure is presented in section III. Section IV presents the This feature is the main focus of this paper.
results of the case study on a typical industrial plant. Finally,
III. DYNAMIC DEMAND RESPONSE SCHEME the plant, and will often not be modified in real-time. For a
The system proposed in this paper is created and expanded detailed description of the ORA and the functions of different
based on the original design presented in [14]. The objective sub-modules, the reader is referred to [14]; nevertheless,
here is to determine the statuses of the workstations (fully some overall aspects are briefly described below.
operating, partially operating, or shut-down) in such a way In this paper, a fuzzy cognitive map (FCM) based design
that the industrial plant is able to participate in a DR event [24] is used for implementing the ORA (Fig. 2). The
while satisfying all real-time operational constraints. Figure 1 workstations are defined as the nodes (vertices) of the FCM,
illustrates the structure of the proposed solution. In short: while the links (edges) denote the interrelations between the
• Offline Ranking Analysis (ORA) module ranks the workstations. Within this framework, the impact of a
workstations based on their priorities for demand workstation is defined as the total effect of the workstation
curtailment during a DR event. The module uses offline node on the end product node. The ranking of the
information (as explained below) to achieve this goal. workstation, i.e., priority for demand reduction, is defined as
the logical inverse of the impact.
• Real-Time Ranking Modification (RTRM) modifies the
Without the loss of generality, the weights of the FCM
rankings received by the ORA based on the daily
links have been assigned fuzzy terms low, medium and high,
operational constraints such as maintenance and crew.
which are determined based on how critical the corresponding
• Integer Programming (IP) module then receives the
node would be. For instance, if node A represents a critical
modified rankings from the RTRM, the workstation
workstation, the weights wjA (j: rest of the nodes) of all the
connections from the ORA, as well as the inventory
links leaving A will be denoted high. This is due to the fact
constraints; and in return, determines the optimal schedule
that the impact A has on the final product of the plant would
of operation for the workstations (refer to section III.C for
be considered high. This scheme would ensure that the
details).
workstations that are classified as non-critical and time-
flexible would become natural candidates for receiving higher
priority of demand reduction, whereas being critical or
needing a long startup time can be grounds for assigning
lower priorities to the workstation's demand to be curtailed.
The rules in Table I are used for identifying critical and non-
critical workstations.
TABLE I
Fig. 1. Proposed system for dynamic demand response.
WORKSTATION CLASSIFICATION
Criticality
Examples
A. Offline Ranking Analysis Module Classification
Workstations that highly affect the completion of the
ORA module is designed to prioritize the existing Critical (high) end product; workstations that process the product of
workstations as candidates for demand reduction. The module multiple other workstations
performs the ranking based on the information received on Workstations with a large number of crew;
Semi-critical
workstations with a long set up time; workstations that
the offline classification of the workstations, critical (medium)
carry high value inventory
performance requirements, and process interrelations. This Noncritical Workstations not included in above categories; those
information is associated with the long-term performance of (low) with flexible schedules, and/or low number of crew
B. Real-Time Ranking Modification Module TABLE III
RULE BASE FOR REAL-TIME RANKING MODIFICATION
RTRM module updates the rankings determined at ORA Priority Derived from
Time to Maintenance Crew Constraint
based on the information on maintenance scheduling and crew RTPM
management. This module is expected to receive updated High Low
information on a regular basis; therefore, it can reflect the Long Medium Low
latest operational requirements of the plant. Low Medium
High Low
1) Maintenance Scheduling
Medium Medium Low
This sub-module looks into the scheduled maintenances
Low Medium
coming up for different workstations. Clearly, stations that are High Low
scheduled for maintenance soon enough would normally Short Medium Medium
move higher in the priority list to be interrupted. This way, Low High
the demand reduction target may be met while the workstation
undergoes its scheduled maintenance. For this purpose, a 4) Ranking Adjustment
fuzzy variable time to maintenance is defined on the basis of The output of RTRM (i.e., priority based on real-time
the time remaining to DR event: operation) will be combined with that of the ORA (i.e.,
Time to Scheduled Maintenance priority based on long-term operation) in order to achieve an
Tm = (1)
Time to DR Event overall DR priority index for the loads/workstations. This
Without the loss of generality, fuzzy terms short, medium data would then be used by the optimization module to assign
and long are defined for Tm (see [14] for details). weights to different workstations (see section III.C). A simple
2) Crew Schedule rule base is used for this purpose as shown in Table IV.
This sub-module investigates whether the crew scheduled
to work on a specific workstation can be assigned to work on TABLE IV
COMBINED PRIORITIES BASED ON ORA AND RTRM OUTPUTS
a different workstation for the duration of demand reduction. Priority Derived from Priority Derived from
Although the number of crew working on a workstation is a Overall DR Priority
ORA RTPM
key aspect to avoid wasted labor, it is also important to High Very High
distinguish between single-skilled workers versus cross- High Medium High
trained crew. The former may not be dispatched to another Low Medium
High High
workstation whereas the latter provides more flexibility in Medium Medium Medium
reassignment. Low Low
Without the loss of generality, three fuzzy terms, i.e., low, High Medium
medium and high, have been considered for the number of Low Medium Low
crew working at each workstation (see [14] for details). Based Low Very Low
on the number of crew working on each workstation, and their
skill levels, the rule-base in Table II is used to derived the so C. Integer Programming Module
called crew constraint measure. A low crew constraint would
The objective of this module is to determine the statuses of
be grounds for crew reassignment, whereas a high crew
the workstations (fully/partially operating or shut-down) that
constraint would indicate low priority for crew reassignment.
provides maximum amount of curtailable load possible, while
TABLE II
ensuring the dynamic constraints of the overall plant are met.
RULE BASE FOR CREW MANAGEMENT This module takes into account the real-time dynamics of
Crew Number Crew Skill Crew Constraint inventory accumulation after each workstation (i.e., post-
process inventory) as well as the interrelations between the
Specific-Skilled Medium
Low workstations.
Cross-Trained Low It is assumed that the problem is solved and revisited every
Specific-Skilled High ∆t units of times, e.g., every hour or every half-hour. In
Medium addition, the problem is solved over multiple time periods in
Cross-Trained Low
order to incorporate the concept of accumulation of
Specific-Skilled High inventories over time, as well as the capacity impact of the
High
Cross-Trained Medium loads.
The workstation loading decision is formulated as an
3) RTRM Ranking integer programming problem provided in (2). The objective
Based on the time to maintenance and the crew constraint, is to find the integer variables ui for each time period,
the real-time operational rankings are derived as in Table III. indicating the level at which the ith unit is operational for each
∆t period. In this paper, ui is assumed to be capped at 4,
which means every workstation can operate at 25%, 50%, can be viewed as the output of the last workstation, i.e.:
75% or 100% of the rated capacity. Clearly, ui = 0 indicates a ∀t = 1LT : I n (t ) ≥ d (t ) (8)
unit that is shut down. The objective function is defined as: The parameters and variables used in the formulation of
T n
the optimization function are defined as follows:
Min ∑∑α
t =1 i =1
i ⋅ ( Pi ,base / 4 × ∆t ) ⋅ ui (t ) (2)
B. Simulation Results
To demonstrate the proposed algorithm, without the loss of
generality, it is assumed here that at time 0 all workstations
(except for workstations 1, 7 and 8 that utilize external
resources) have a starting pre-process inventory of 10 units. It
should be noted that changing this number would change the
final dispatch of the workstations, but does not affect the
generality of the solution proposed. The problem has been
solved for three time step, i.e., T = 3. Also, it has been
assumed that the minimum demand of the plant for these three
time periods is 15, 14 and 20 units respectively.
In order to analyze the impact of the proposed dynamic
demand reduction scheme on the overall performance of the
system, a base case has been established where the objective
Fig. 5. Hourly electric power consumption under two scenarios: with and
of the plant is to maximize production subject to the without demand reduction.
operational constraints. In this case, the optimization problem
can be formulated as:
C. Discussion at an industrial site through demand reduction. The proposed
The simulation results showed that with the proposed solution first considers the offline and quasi-dynamic
solution, it is possible for the industrial plant to maintain the operational constraints of the plant in order to assign weights
requested demands while minimizing the electric power (priority ranking) to the workstations targeted for demand
consumption. Obviously, the decision to proceed with the reduction. Then, through an integer programming based
reduced demand reduction regime requires the plant operator optimization approach, it incorporates dynamic constraints
to perform a financial analysis. Under the base case scenario, such as post-process inventory buildup in order to finalize the
the plant manufactures more products which is directly workstations that can operate under demand curtailment or
translated into higher revenue. However, this could also mean demand reduction, while ensuring the overall demand of the
a possible inventory buildup at the output of the plant (if all plant is met. Such a scheme can be used by itself, or in
the products cannot be sold), possible need for maintenance conjunction with other alternatives such as on-site generation
of the workstations that are operating at higher capacities, and and energy storage.
higher electricity bill. On the other hand, reduced demand
means that less number of products are manufactured APPENDIX
(although adhering to the minimum required demand), which TABLE VII
can lead to lower revenue. However, lower electricity bill and OPERATIONAL DATA FOR WORKSTATIONS
Workstation Ii,max (units) Pi,base (kW) ci (units/time)
incentive payments as a result of participating in a demand 1 20 100 8
reduction can further incentivize this strategy. A financial 2 30 120 5
cost-benefit analysis would determine which scenario the 3 30 200 4
plant operator would choose, but this would be a case-by-case 4 20 80 5
analysis and falls outside the scope of this paper. 5 25 80 5
6 22 340 4
7 20 360 3
8 25 80 5
9 25 60 4
10 25 60 5
11 25 60 4
REFERENCES
[1] “Assessment of Demand Response and Advanced Metering,” Staff
Report, Federal Energy Regulatory Commission, Aug. 2006.
[2] Thomas Weisel Partners, “A Primer on Demand Response,” White
Paper, Oct. 2007.
[3] S. Kiliccote, M.A. Piette and D. Hansen, “Advaned Controls and
Communications for Demand Response and Energy Efficiency in
Commercial Buildings,” in Proc. Carnegie Mellon Conf in Electric
Power Systems, Pittsburgh, PA, USA, Jan. 2006, pp. 1-10.
[4] S. Mohagheghi, J. Stoupis, Z. Wang, Z. Li and H. Kazemzadeh,
“Demand Response Architecture- Integration into the Distribution
Fig. 6. Potential demand reduction capacity available to be deployed at the Management System,” in Proc. IEEE SmartGridComm, Gaithersburg,
MD, USA, Oct. 2010.
plant using the proposed solution.
[5] A. Brooks, E. Lu, D. Reicher, C. Spirakis and B. Weihl, “Demand
Dispatch,” Power & Energy Magazine, vol. 8, no. 3, May/Jun. 2010,
pp. 21-29.
V. CONCLUDING REMARKS [6] “Demand Response: Design Principles for Creating Customer and
Market Value,” Technical Report, Peak Load Management Alliance,
Demand response is an integral part of the Smart Grid Nov. 2002.
paradigm. By reducing the demand during peak times or [7] J. Han and M.A. Piette, “Solutions for Summer Electric Power
Shortages: Demand Response and its Applications in Air Conditioning
shifting the demand from peak to non-peak hours, DR helps and Refrigeration Systems,” Journal of Refrigeration, Air
stabilize the market and provides financial benefits for both Conditioning and Electric Power Machinery, vol. 29, issue 1, pp. 1-4,
the utility and the customers. In addition, by postponing the Jan. 2008.
[8] E. Koch and M.A. Piette, “Architecture Concepts and Technical Issues
need to grid capacity expansion, it can offer substantial for an Open, Interoperable Automated Demand Response
environmental benefits. DR for residential and most Infrastructure,” presented at the Grid Interop Forum, Albuquerque,
commercial customers is relatively straightforward to NM, USA, Nov. 2007.
implement, mostly due to the non-criticality of the loads [9] S. Kiliccote and M.A. Piette, “Advanced Control Technologies and
Strategies Linking Demand Response and Energy Efficiency,” in Proc.
involved. However, DR for industrial customers faces unique Intl Conf for Enhanced Building Operations, Pittsburgh, PA, USA,
challenges. Sept. 2005.
This paper proposed a scheme based on fuzzy systems and [10] J.G. Roos and I.E. Lane, “Industrial Power Demand Response Analysis
for One-Part Real-Time Pricing,” IEEE Transactions on Power
mathematical optimization techniques for implementing DR Systems, vol. 13, no. 1, pp. 159-164, Feb. 1998.
[11] J. Duanmu and K. Taaffe, “Measuring Manufacturing Throughput Management Technique,” in Proc. International Middle East Power
Using Takt Time Analysis and Simulation,” in Proc. Conference on Systems Conference, Dec. 2006.
Winter Simulation, 2007. [19] Y. Tang, F. Xu, and L. Chen, “Research into Possibility of Smart
[12] Y. Chen, K.W. Li and S.F. Liu, “A Comparative Study on Multicriteria Industrial Load Participating into Demand Response to Supply the
ABC Analysis in Inventory Management,” in Proc. IEEE Intl. Conf. on Power System,” in Proc. China International Conference on
Man, Machines and Cybernetics, Singapore, Oct. 2008, pp. 3280- Electricity Distribution, Nanjing, China, Sept. 2010.
3285. [20] B. Delfino, F. Croce, P.A. Fazzini, S. Massucco, A. Morini, F.
[13] S. Ghosh and R.J. Gagnon, “A Comprehensive Literature Review and Silvestro and M. Sivieri, “Operation and Management of the Electric
Analysis of the Design, Balancing and Scheduling of Assembly System for Industrail Plants: An Expert System Prototype for Load-
Systems,” International Journal of Production Research, vol. 27, no. Shedding Operator Assistance,” in Proc. IEEE IAS, Rome, Italy,
4, pp. 637-670, 1989. August 2000, pp. 3260-3268.
[14] S. Mohagheghi and N. Raji, “Intelligent Demand Response Scheme for [21] F. Shokooh, J.J. Dai, S. Shokooh, J. Tastet, H. Castro, T. Khandelwal
Energy Management of Industrial Plants,” in Proc. IEEE IAS, Las and G. Donner, “An Intelligent Load Shedding System Application in a
Vegas, NV, USA, Oct. 2012. Large Industrial Facility,” in Proc. IEEE IAS, Oct. 2005, pp. 417-425.
[15] S. Coe, A. Ott and D. Pratt, “Demanding Standards,” Power & Energy [22] C.T. Hsu, H.J. Chuang and C.S. Chen, “Artificial Neural Network
Magazine, vol. 8, no. 3, May/Jun. 2010, pp. 55-59. Based Adaptive Load Shedding for an Industrial Cogeneration
[16] I. Kuzle, D. Bošnjak and S. Tešnjak, “An Overview of Ancillary Facility,” in Proc. IEEE IAS, Austrailia, Oct. 2008.
Services in an Open Market Environment,” in Proc. Mediterranean [23] H.P. Inamdar and R.P. Hasabe, “IT Based Energy Management
Conference on Control and Automation, Athens, Greece, July 2007. through Demand Side in the Industrial Sector,” in Proc. International
[17] X. Yu and L.M. Tolbert, “Ancillary Services Provided from DER with Conference on Control, Automation, Communication and Energy
Power Electronics Interface,” in Proc. IEEE PES, Montreal, Quebec, Conservation, India, June 2009.
Canada, Oct. 2006. [24] B. Kosko, “Fuzzy Cognitive Maps,” International Journal Man-
[18] M.M. El-Metwally, M.S. El-Sobki, H.A. Attia and S.A. Wahdan, Machine Studies, vol. 4, pp. 65-75, 1986.
“Priority Ranking of Industrial Loads and Application of Demand Side