Professional Documents
Culture Documents
special monte
special monte
Peer-Refereed
Applied Papers
Keyword Search
Electricity
Energy
Research
The Official Electronic Publication of The Association of Technology, Management, and Applied Engineering • www.atmae.org
© 2009
Journal of Industrial Technology • Volume 25, Number 3 • July 2009 through September 2009 • www.nait.org
2
Journal of Industrial Technology • Volume 25, Number 3 • July 2009 through September 2009 • www.nait.org
tricity to control their peak demands Electrical Load Representation Equations (2) and (3) simulate high and
and reduce purchased electricity costs One metric for classifying industrial low load factor industrial loads respec-
when RTP rates spike (Sarimveis et al., electric load patterns is load factor tively. These demand functions gener-
2003; Coffy and Kutrowski, 2005). In (Turner, 2001). Load factor compares ate times series load data normalized to
general, industrial load control would average power demand to peak power peak power demand.
require additional investment in energy demand for a defined period such as
management systems and customer- a month or year. Equation (1) defines
owned generators to reduce consump- load factor mathematically. In this d H (nTm ) = a 1H sin(2πf 1 nTm ) + a 2 H sin( (2)2πf 2 nTm
tion during times of high prices. equation, Etotal is the
Recovering the cost of this equipment d H (nTm ) = a 1H sin(2πf 1 nTm ) + a 2 H sin(2πf 2 nTm + φ 2H + ε φn ) + d baseH + ε bn
depends on the potential savings,
d (nT )if=any, a sin(2πf 1 nTm ) + a 2 H sin(2πf 2 nTm + φ 2H + ε φn ) + d baseH + ε bn
realized from adopting the HRTP mtariff. 1H E
Shifting or curtailing industrial opera- LF = total (1)
tions is another method of load control Ppeak N
but maybe too costly or impractical to d L ( nT m ) = − a 1L d base sin( 2 πf 1 nT m +(3)
φ 1L ) + a 2 d
implement. total period electric energy usage in
d L ( nT m ) = − a 1L d base sin( 2 πf 1 nT m + φ 1L ) + a 2 d base sin( 2 πf 2 nT m + φ 2 L + ε φn ) + d
kilowatt-hours, Ppeak is the maximum
Previous researchers produced deter-
d L ( nT m ) = − a 1L d base sin( 2 πf 1 nT m + φ 1L ) + a 2ind base
period power demand kilowatts
sin( 2 πf nT + φ 2 L + ε φn ) + d baseL + ε bn
ministic dynamic models of industrial during the period, and N is the period2 m
customer types using linear and non- length in hours. Customers with high
linear programming to test the impact load factors present a nearly constant These equations represent industrial
of RTP tariffs (David and Lee, 1989). power demand over time to suppli- operations that exhibit daily periodic
The models generate optimal operating ers while customers with a low load changes. They include cycles for shift
schedules for plant production under an factor have more cyclic power demand work changes reflecting three eight
RTP with production constraints. These patterns. Low load factor industrial hour work periods. The variable, Tm,
theoretical models are for evaluating operations exhibit cyclic electricity represents the demand meter totalizing
supply and demand interactions, and consumption due to daily production period and n is the period index over
have value in industrial site planning and changes and shift operations. Industrial the total study interval N. The other
process expansion. Hughes and Bailey customers that operate at full capacity parameters are:
developed a scheduling methodology continuously have higher load factors.
using discounted cash flow modeling f1 = daily frequency = 1/T1, T1 in hours,
for optimally allocating co-generating Electric demand meters totalize energy
resources in a nylon plant using RTP consumption over periods varying from f2 = shift frequency = 1/T2, T2 in hours,
information (Hughes and Bailey, 2004). five minutes to one hour. These inter-
In this work, a Monte Carlo simulation val values provide industrial custom- a1H, a2H = daily and shift component
evaluated a number of fuel cost and ers with power demand information amplitudes for high load factor,
electricity price scenarios to aid in the and energy consumption data. Simple
decision process and risk analysis of calculations convert shorter interval a1L, a2L = daily and shift component
optimal co-generator schedules. data into equivalent hourly values. amplitudes for low load factor,
Graphing demand meter data produces
This paper examines demand func- time series plots of consumer electricity φ2H = phase delay of shift load compo-
tions to determine what factors influ- consumption. A Fast Fourier Transform nent for high load factor,
ence RTP tariff savings by introducing (FFT) decomposition of power demand
a benefits index to calculate savings time series values identifies significant φ1L , φ2L = phase delay of daily and shift
potential. The nominal values of the cyclic load components that can be load components for low load factor,
demand functions derive from analysis modeled mathematically using sinusoi-
of actual plant electricity consumption dal terms. The FFT decomposition pro- dbaseH, dbaseL = average power demands
data. The analysis uses a power demand vides amplitude, frequency and phase for high and low load factors,
measure, load factor, to character- parameters for the load model (Cohen,
ize industrial load types. The analysis 1995). Random load variations occur in εbn = stochastic average load compo-
also compares the potential benefits of industrial loads also. Stochastic phase nent,
adopting the RTP tariff to load factors and amplitude parameters represent this
with high and low values. The analysis part of customer load. The sum of the εφn = stochastic phase delay component.
assumes industrial customers purchase periodic and stochastic load compo-
no additional load control equipment. nents gives the total customer power
demand function.
3
Journal of Industrial Technology • Volume 25, Number 3 • July 2009 through September 2009 • www.nait.org
Figure 1. Phase Relationship between Demand and RTP. rate, pnf. Equations (4) and (5) compute
1
normalized customer cost for a fixed
rate tariff over the interval [0, N] using
24 hours normalized power demands from (1)
and (2). The formulas assume a one
Per Unit Price and Demand
0.8
hour metering interval, Tm=1. The vari-
ables FECH and FECL are the fixed rate
0.6 electricity cost indexes for high and low
load factor time series respectively.
N
0.4
FECH = pnf ∑d ( n )
n =0
H (4)
0.2
� N
0
FEC L = pn f ∑d
n =0
L (n ) (5)
0 10 20 30 …….. N
n
Equations (6) and (7) compute normal-
Normalized Power Demand ized costs for RTP tariffs with RECH
Normalized RTP Tarriff and RECL representing the customer
electricity cost index for high and low
load factor consumption patterns.
Figure 1 shows the phase delay rela- reduced weekend activity and building
tionship between the price and load. indoor environmental control. N
Phase delay is the angular difference
between the real-time price daily peak Fixed rate tariffs remain constant over
REC H = ∑ pn d
n =0
n H (n ) (6)
and load series peak with price as defined contract periods. Fixed rate
reference. These parameters relate the tariffs account for utility fuel costs and N
temporal difference between the two
time series. The equations represent
capital investment in their systems,
and are based on average customer
REC L = ∑ pn d
n =0
n L (n ) (7)
random changes in operations with consumption. Fixed rate tariffs include
two stochastic variables: one for power cross-subsidies between customer
A benefit index quantifies potential
demand and another for phase delay. classes such as industrial, commercial,
industrial customer savings from adopt-
Using load models based on actual con- and residential (Borenstein, 2006).
ing a RTP tariff. The indices shown in
sumption data allows energy managers
equations (8) and (9) are the differences
to correlate operation parameters such Most industrial tariffs charge for both
between the fixed and RTP costs for
a work start times to the price series electric energy and power demand.
both high and low load factor power
and examine the impact of parameter Demand charges help suppliers re-
demand patterns.
changes quantitatively. cover costs associated with peak power
generation, transmission, and delivery. N N
Electricity Cost And Benefit
Calculation
Electric system owners must construct
facilities to handle peaks that may oc-
BH = pn f ∑
n =0
d H (n ) − ∑ pn d
n =0
n H (n ) (8)
Suppliers provide real-time price cur for only a single hour annually. The
information to customers at regular demand charge penalizes customers N N
intervals ranging from five minutes
to one hour. Some markets publish
with lower load factors since their peak
power demand is large relative to their B
L = pn f ∑
n =0
d L (n ) − ∑ pn d (n) (9)
n =0
n L
4
Journal of Industrial Technology • Volume 25, Number 3 • July 2009 through September 2009 • www.nait.org
than the cost of the fixed rate, so a cus- mand series. Highly correlated prices by comparing the benefits from the two
tomer would realize savings by switch- and demands increase the magnitude different demand patterns.
ing to the RTP tariff. of this term. Large demand swings also
increase it. For each experiment, N=648 hours.
Setting the benefit equations to zero This number represents 27 consecutive
finds the break-even normalized fixed Monte Carlo days of hourly usage. Each experiment
rate for adopting a RTP tariff with a Experimental Design computes results for 5000 iterations of
given load factor, pnf0. The RTP se- Monte Carlo simulation is a tool for the benefit indices using the fixed and
quence can be written as the sum of the analyzing systems with parameter random variables from Table 1. The
average RTP over N, pnave, and the se- variation or incomplete knowledge of experiments use historical real-time
ries residuals, pnrn. Setting the benefits data. This type of analysis is one way to price data (Ameren, 2009) normalized
equation to zero and replacing the RTP quantify uncertainty in data and model to the peak hourly price over a 27-day
series with the above sum gives: parameters. Monte Carlo simulation interval in August, 2007 for these com-
results in a probability distribution that putations. All experiments use this RTP
N N N time series.
∑ ∑ ∑
describes how uncertainty propagates
pn f 0 d (n ) = pn ave d (n ) + pn r n d (through
n) a system. Statistical analysis
n =0 n =0 n =0 of the resulting distribution describes The Ameren site archives an hourly RTP
N N
system performance (Wittwer, 2004). time series beginning on December 28,
∑ d(n) +∑ pn
2006 and running to the present. This
r n d(n ) Four statistical experiments examine data derives from the Midwest regional
=0 n =0 wholesale electricity market. Similar
the impact of load and price parameter
variations on customer benefit func- data is available for every region of the
Solving this equation for pnf0 gives the tions, equations (7) and (8), for differ- U.S. from the Federal Energy Regulato-
break-even normalized fixed rate for ing load factor demand patterns. Two ry Commission Website (Federal Energy
any load pattern and price time series. statistical experiments analyze the Regulatory Commission, 2009). The
impact of demand parameter variations Ameren site also posts day-ahead prices
N on the break-even fixed rate given by that are next day RTP forecasts. The site
∑ pn rn d(n ) (10). Table 1 shows the fixed and ran-
dom variables for the six experiments.
updates the day-ahead prices with RTP
prices at 4:30 pm CST each day. RTP
pn f 0 = pn ave +
n =0
N (10) The first four experiments study how tariff customers use the day-ahead prices
∑ d(n )
n =0
variation of average power demand, as indicators of the actual RTP to adjust
their load profile.
phase delay, and fixed rate prices im-
pact customer benefits when compared
The second term in (10) quantifies the to a real-time price tariff. The remain- Table 2 lists the parameter values and
variation in cost due to price changes ing two experiments examine how load probability distributions for the power
about the mean. It accounts for the function parameter variations affect demand functions used in the first four
correlation between the price changes the break-even fixed rate. The first four experiments. This analysis assumes
about the series mean and power de- experiments capture load factor effects a uniform probability distribution for
5
Journal of Industrial Technology • Volume 25, Number 3 • July 2009 through September 2009 • www.nait.org
phase delay, base power demand, and Table 2. Parameter Values and Probability Distributions for Load Functions.
fixed tariff parameters. A normal dis- Parameter High Load Factor (H) Low Load Factor (L)
tribution represents the random phase
delay and base load variations in the a1 0.05 0.80
power demand functions. The analysis a2 0.08 0.30
considers a daily power demand period φ1 (rad) N/A PU(π/12,π)
of 24 hours and a production shift φ2(rad) PU(3π/12,3π) PU(3π/12,3π)
period of eight hours. Figure 2 shows
dbase PU(0.3,0.6) PU(0.3,0.6)
weekly high and low load factor time
series computed using nominal pa- T1 (hrs) 24.00 24.00
rameter values of φ2H=φ2L=π/4, φ1L=π/3 T2 (hrs) 8.00 8.00
dbaseH=0.8 for high load factor and f1 rad/hr π/12 π/12
dbaseL=0.5 for low load factor. The high f2 rad/hr π/4 π/4
load factor demand model is based on εφn PN(M=1, SD=2.5) PN(Μ=1, SD=2.5)
a case study of a blow-molding factory
running continuously at full capacity εbn PN(Μ=0.01,SD=0.025) PN(Μ=0.01,SD=0.025)
producing plastic soda bottles. The low pnf PU(0.10,0.80) PU(0.10,0.80)
load factor demand model is based on a
case study of a soda bottling facility op-
erating a 5-day week and a single shift. Figure 2. Power Demand Time Series Showing Weekly Patterns
of High and Low Load Factor Industrial Operations.
1.4
For price experiments 5 and 6, a1 and
a2 are uniformly distributed normalized High LF Demand
values. Table 3 lists the distributions 1.2
Low LF Demand
Normalized Power Demand
6
Journal of Industrial Technology • Volume 25, Number 3 • July 2009 through September 2009 • www.nait.org
Table 4 summarizes simulation results Figure 3. Cumulative Probability Distributions Comparing Low Load Factor Power
for the first four experiments. The low Demand Having Random Price, and Average Demand with a Case Having Random
load factor with random average loads Price, Average Demand, and Shift Phase Delay.
gives the lowest benefit index value. 1
In this case, the random time series Random Price and Demand
parameters are average power demand Random Price,Demand and Phase
and fixed rate tariff. If power demand 0.8
peaks when RTP peaks, the customer
experiences an increase in electricity
Cumulative Probability
costs compared to a fixed rate. Moving 0.6
the power demand series with respect
to the peak RTP series improves the
benefits index by making the peak
0.4
demand and price coincide less. High
values of daily phase delay correspond
to moving industrial operations out-
0.2
side the time of peak electricity usage,
which may increase production costs
due to increased labor expenses.
0
-150 -100 -50 0 50 100 150 200 250
The high load factor demand series
Benefit Index
produces the highest benefit index
values. The index does not improve
significantly with changes in shift phase Figure 4. Cumulative Probability Distributions Comparing High Load
delay. The low variation in the daily load Factor Power Demand Having Random Price, and Average Demand with a Case
pattern increases the savings potential Having Random Price, Average Demand, and Shift Phase Delay.
1
of the RTP tariff since most of the daily
price cycle is lower than the fixed rate. Random Price and Demand
The lower hourly rates produce savings Random Price,Demand and Phase
while the RTP is below the fixed rate. As 0.8
long as price spikes are of short dura-
tion, an industrial customer will realize
Cumulative Probability
7
Journal of Industrial Technology • Volume 25, Number 3 • July 2009 through September 2009 • www.nait.org
loads with respect to the RTP tariff Table 5.Mean Equality Test Results.
produces a statistically significant im- Experiment Means Z Statistically Different
provement in the benefit means. High Compared α=0.05
load factor experiments show no statis-
tically significant difference between 1-3 -16.88 Yes
benefit means when load parameters
vary. This result indicates that adjust- 2-4 -5.590 Yes
ing shift operations in high load factor 1-2 -11.16 Yes
demands will produce no statistically
significant benefits when compared to 3-4 -0.256 No
high load factor demands without shift
operation adjustment.
Figure 5. Break-Even Fixed Rate Distribution for Low Load Factor Power Demand.
Experiments 5 and 6 examine the im-
pact of power demand variations on the 350
break-even normalized fixed rate. The
M=0.3177 pu
histograms in Figures 5 and 6 show the SD= 0.0460 pu
300
simulation results.
200
load factor series. The break-even fixed
rate varies 0.18 per unit (pu) over the
range of simulation values. The load 150
variations add to the average RTP rate
and relate to the load changes. The 100
high load factor simulation results in a
very sharp price distribution, which is
consistent with the low demand varia- 50
tion about the mean in this model. The
break-even fixed rate is very near the 0
average RTP for the analysis period for 0.2 0.22 0.24 0.26 0.28 0.3 0.32 0.34 0.36 0.38 0.4
high load factor demands. Normalized Break -Even Rate (Low LF)
Conclusion 200
Introducing real-time price tariffs to
all customers creates opportunities
100
for electricity cost saving in industrial
operations. The shape and timing of
the load time series relative to the price 0
series impacts the potential benefits 0.27 0.275 0.28 0.285 0.29 0.295 0.3 0.305
a business realizes from adopting a Normalized Break -Even Rate (High LF)
8
Journal of Industrial Technology • Volume 25, Number 3 • July 2009 through September 2009 • www.nait.org
real-time price tariff. Results of Monte Laboratory, LBNL-542338. [Avail- Energy Information Administration.
Carlo simulations using power demand able Online] http://eetd.lbl.gov/ea/ (2009) Retrieved Jun 1, 2009 from
mathematical models with high and EMS/EMS_pubs.html. http://www.eia.doe.gov/emeu/mecs/
low load factors indicate that higher Borenstein, S. (2006, July). Customer predata/estimates.html.
load factor time series have a greater risk from real-time retail electricity Federal Energy Regulatory Commis-
economic benefit potential. The high pricing: bill volatility and hedgabil- sion. (2009) Retrieved June 1, 2009
load factor power demand series ity, University of California Energy from http://www.ferc.gov/market-
exhibits a lower break-even fixed price Institute, working paper CSEM WP oversight/mkt-electric/overview.asp
when compared to the low load factor 155. [Available Online] http://www. Hughes, P. D. & Bailey, W. F. (2004).
series. Industrial operations that oper- ucei.org. Industrial powerhouse optimization
ate around the clock and have a steady Borenstein, S. (2006, July). Wealth in the deregulated electricity mar-
process load can save electricity costs transfer among large customers ketplace, Energy Engineering, vol.
by adopting a real-time rate without from implementing real-time retail 101, no. 1, pp. 57-77.
further investment in energy manage- electricity pricing, University of Power Smart Pricing. (2008). Retrieved
ment or control systems. Using actual California Energy Institute, working February 20, 2009 from http://www.
data to compute benefits could provide paper CSEM WP 156. [Available powersmartpricing.org.
more accurate case studies but would Online] http://www.ucei.org. Sarimveis, H. K. et al., (2003). Optimal
eliminate analysis parameters. Future Coffey, B. & Kutrowski, E. (2005). energy management in pulp and
work will extend the analysis period Demand charge considerations in paper mills, Energy Conversion and
of the experiments to cover an entire the optimization of cogeneration Management, vol. 44, no. 10, pp.
summer. dispatch in a deregulated energy 1707-1718.
market, International Journal of Turner, W. C., (2001). Energy manage-
References Energy Research, vol. 30, no. 7, pp. ment handbook 4th ed. Lilburn, GA:
Ameren Corporation. (2009). Retrieved 535-551. The Fairmont Press, Inc.
January 21 2009 from https:// Cohen, L. (1995). Time frequency Wittwer, J. W. (2004, June). Monte
www2.ameren.com/RetailEnergy/ analysis 1st ed. Englewood Cliffs, Carlo Simulation Basics, Vertex42.
rtpDownload.aspx. NJ: Prentice Hall. com. [Available Online] http://ver-
Barbose, G, Goldman, C, & Neenan, B. David, A. K, & Lee, Y. C. (1989). tex42.com/ExcelArticles/mc/Monte-
(2004). A survey of utility experi- Dynamic tariffs: theory of utility- CarloSimulation.html.
ence with real time pricing, Ernest consumer interaction, IEEE Trans-
Orlando Lawrence Berkeley Nation actions of Power Systems, vol. 4, no.
3, pp. 904-911.