Roadmap for Adopting New Technology in the Utility Industry
Lianjun An, Dharmashankar Subramanian, Jayant R. Kalagnanam
IBM T.J. Watson Research Center
Sashidhar Pemmasani, Mark Welch
IBM Global Business Services
{alianjun,dharmash,jayant,pemmasani,mark.a.welch}@us.ibm.com
Abstract
In this paper, we develop a framework to optimally manage the time-phased deployment
planning of a new technology, namely Advanced Metering Infrastructure, in the Utility industry.
Advanced Metering Infrastructure enable two-way communication between residential and
commercial customers and the Utility grid, so that energy consumers may respond to price
signals during stressful conditions in the grid in a price-sensitive fashion, and reduce the overall
stress in the grid. It also enables consumers to shift discretionary energy loads from stressful,
peak periods to off-peak periods. We present an end-to-end solution framework for addressing
the various analytical challenges that are involved in developing an optimal deployment plan,
from a business case development perspective. Our solution framework uses a judicious
combination of system dynamics modeling, econometric modeling and mathematical
programming based optimization modeling. A system dynamics model is used to estimate the
dynamics of user adoption of the new technology, relative to deployment, which results from
marketing effectiveness for the new technology, as well as the viral effect of word-of-mouth
interactions among users. The model is also used to estimate the lag in benefits realization from
the new technology deployment, arising from the above dynamics of user adoption, coupled with
a lag in the maturity of the supporting Information Systems that enable effective functioning of
the new technology. These estimates are then subsequently used in a mathematical programming
model, which solves a multi-period, resource-constrained optimal deployment planning problem
that is subjected to the lags in user adoption and benefit realization, which are estimated by the
system dynamics model. The mathematical programming model also needs estimates of demand
response benefits that are obtainable from effective adoption of the new technology. An
econometric model is used to estimate the price-sensitive response of user demand, which is
facilitated by the new technology. Lastly, we incorporate a parameter estimation routine that
acts as a feedback loop between the output decisions of the planning model, which get
implemented, and the system dynamics model. This is so that we may reconcile the parameters in
the systems dynamics model, with respect to observations from the market place. Such an end-
to-end approach will enable us to use both external market information as well internal financial
information and to steer business success.
Keywords: System Dynamics, Market Analysis, Programming Model, Optimization
1. Introduction
Energy Delivery Companies are planning to deploy smart meters for its residential and
commercial customers. Such an Advanced Metering Infrastructure (A MI) would enable two-way
communication. Specifically, it includes a communication information system and distribution
automation system. Companies can use the system to monitor and collect the hourly or sub-
hourly electricity usage from each customer and to diagnose occurring problems (trouble
shooting). More important, companies can use the system to pose the whole network load and
pricing information to customer in real time and expect that price-sensitive customers change
their usage pattern to reduce the peak demand and to increase non-peak demand. These features
provide the customer feedback mechanisms that encourage economic investment and rational
market behavior.
There are several papers that are available to evaluate such a program from demand
response perspective [1], [2]. They analyze usage pattern changes for different user groups and
different seasons through pilot program in a southern Califomia region. Their models are used to
predict the impact of dynamic pricing on demand. Some potential opportunity costs and risks in
investing in AMI and related system are identified and investigated.
As mentioned in [1], business cases in support of system-wide AMI implementation has
not proven successful, since the scope of issues related to cost-justification of AMI is too narrow
and the methodology mainly focuses on minimizing cost. It is a great challenge to predict the
profitability from such investment. We face future uncertainty from market demand and resource
availability during implementation. Some of benefits, like say, the economic benefits from
demand response, depend on customer adoption rate of AMI technology and time-lagged
economic effects, which are difficult to estimate due to the dependency on intangible measures,
such as marketing effectiveness, word-of-mouth, etc.
In this paper, we address the management issues that are pertinent to implementing an
AMI program from a business case development and deployment planning perspective, and
propose an integrated end-to-end approach. Firstly, we uses a market diffusion model based on
System Dynamics [3],[4] to create a time-varying user adoption profile. It includes marketing
effectiveness and word-of-month effect on customer behavior. The model is also used to estimate
the lag in benefits realization from the new technology deployment, arising from the above
dynamics of user adoption, coupled with a lag in the maturity of the supporting Information
Systems that enable effective functioning of the new technology. Secondly, we also need
estimates of demand response driven economic benefits that are obtainable from effective
adoption of the new technology. An econometric model is used to estimate the price-sensitive
response of user demand, which is facilitated by the new technology. Thirdly, a mathematical
programming model is developed to capture the above time-lagged benefits and costs as well as
operational constraints for a chosen planning horizon, which is typically order of 15-20 years. By
maximizing profit through the model, we achieve an optimized meter deployment plan, across
various jurisdictions and over the chosen planning horizon. Lastly, a re-planning step is
introduced during the plan implementation process. There are two reasons for introducing this
step. One is related to the change of resource and/or budget, which may force re-planning. The
other reason is to enable a parameter estimation functionality, which can be invoked to close the
plant-model mismatch, in a manner similar to model predictive control in control theory. It
allows periodic updating of parameter settings for the market diffusion model based on observed
data so far. In other words, the re-planning would close the loop and iteratively calibrate the
market diffusion model.
The paper is organized as the following. In section 2, we discuss information flow and
program management framework. Section 3 gives formulation detail of the integrated model.
Section 4 demonstrates some simulation results for certain scenarios. Section 5 concludes the
paper and discusses further research direction.
2. Program Management Framework
Program Management on AMI implementation is about how to develop and execute a
deployment plan. Our proposed solution will address problems for the following two phases:
planning phase before implementation and re-engineering during the implementing process.
2.1. Information flow model
Figure 1 shows the information flow among the various modeling components. There are
three models in the system (top three boxes). The market diffusion model takes parameters like
marketing effectiveness and customer contact rate. These values should be boot-strapped in the
first phase and might be obtained from historical data in other technology adoption processes,
such as, data from market penetration of natural gas vehicles [6] and observed values from some
pilot program of deploying AMI in electricity industry [2]. These values can be calibrated from
partially available actual data in the second phase. This model will generate customer adoption
rate, as well as the time lag between deployment and benefits realization, as inputs into the
optimization model.
Market Penetration
Calibratior
Figure 1: Information flow in proposed framework
The demand response model uses the hourly load and dynamic pricing information, as
well pricing elasticity to asses benefit (cost saving) and generate unit benefit (per meter) as a
function of penetration rate. The optimization model takes outputs from the other two models as
well as value factors, like labor unit cost, material unit cost, etc. and schedule factors, like start
date, labor duration (see Figure 2 for detail). This model results in a meter deployment plan that
satisfies the specified budget, material supply and workforce constraints. When situation occurs,
like cost change, supply shortage and workforce unavailable, during the deployment plan
execution, the model can be rerun for the remaining period with new information including the
calibrated penetration profile.
Program Drivers
Figure 2: Program Drivers
2.2. Portfolio consideration
There is another issue that we did not address in the last subsection. In fact, the company
manages multiple jurisdictions. Due to budget, supply and workforce constraints, it makes sense
to deploy smart meter in a certain order among the jurisdictions. There are scenarios in which
deploying labor cost are different due to the difference of residential density. Our framework
also addresses this issue by allowing user to specify different cost structure and schedule
requirements. The optimization model takes all there factors into its formulation and generates an
optimal plan that balances the cost and benefit among multiple jurisdictions.
3. Model Formulation
We describe mathematical formulation of three models in this section.
3.1. Market diffusion model
We use the System Dynamics methodology [3, 4] to model market diffusion. System
Dynamics gives a capability to visually describe casual relationship among factors (tangible and
intangible) and to quantify the relationship in a rational manor (dynamic hypotheses - physical
law to drive the system evolution). The model can be calibrated through parameter estimation
techniques based on observable data. The market diffusion has been studied and applied to
analyze market penetration for other processes of new product and/or technology development,
see [5] for case study in new energy technologies and [6] for natural gas vehicles in Switzerland.
In our case, we need to predict customer adoption percentage from time to time among all
customers that have smarter meters installed. There are two driving forces to change adoption
rate. One is through marketing effort. In fact, after installing smart meter, customers might not be
aware of the benefit of that infrastructure and need time to become familiar with the system. The
other would be “word of month” effect through customer contact. The following System
Dynamics model is shown in the top part of Figure 3 with both influence factors being included.
Adoption from
Panamera Merketing per Unit Tine
Intensity per Unit Time
Adoption fom Word of
Mouth per Unit Time
Marking Fune Makes Bicones Contact Eectiveness
“per Uk Tine path Tie
\
\ Ital Adaptation
Peccentaze
Potential: a a
Adoeton Adopion sal
Percentage Growth Rate Percentage} Attition Rate
at
a Inti! Benefit
ae 54 DR Delay
jure ——~ rd
Frequency Benefit aie
Benefit Change Schedule
Rate
Figure 3: Diffusion & Benefit Realization Model
If write it as a differential equation, we have
OA A)+c-A-(-A).
dt
Where A corresponds to “Adoption Percentage” in Figure 3, m is for “Marking Effectiveness per
Unit Time”, and c is for “Contact Effectiveness per Unit Time”. The first term in RHS accounts
for the effect from marketing and the second is for the effect resulting from the contact of un-
adopted customers with adopted customers. We can solve Equation (1) explicitly and obtain the
following solution
(1)
(cAy +m)-e""" ~m(1— Ay)
cA, +m)-e™* + c(1—A,) ”
Ay Ay
Its value varies from the initial A, to 1 (see Figure 4).
A(t;m,c, A,)=
1234567 8 9 WUD I6I7 19202
Year
P22eS STR VUUREUEWINBIA
Year
Figure 4: Adoption Percentage and Benefit Realization
Note that, since A is for the adoption percentage, its value is always less than one. In the
beginning when the adoption is small (A close to zero), company needs to invest money on
marketing and educate customesr to accept and utilize the new technology. The rate increases
mainly determined by the first term of Equation (1). As time elapses, the “word of month”
influence would become dominant force to convince people to adopt the new technology. The
rate increases mainly determined by the second term of Equation (1).
We take A as a multiplier of the number of customers with smart meters installed to
figure out the number of adopted customers. The other point we want to make here is that the
solution, for the dynamics of customer adoption, applies relative to the point in time when the
smart meter is installed. Since the meters can be deployed in different time, current total adopted
customers would be the summation of all adopted customers with different installation points in
time. Suppose that we use M(s) to represent the number of meters installed at time s. Then the
total adopted customer until time t, MA(t), would be equal to
MA(t;m,c, Ay) = EM A(t—s;m,c, Ay) .
The effectiveness of marketing investment and word-of-month will vary, and these
parameters are difficult to estimate accurately to begin with. The contact rate is different for each
jurisdiction due to different population density. Sometime, the word-of-month could have
negative effect [7]. Some customer groups might not be sensitive to pricing signals and would
not be motivated to use the system. Also the initial value for adoption rate depends on marketing
condition or technology developing stages (emerging, mature and saturated). The impact from
intangible variables related to soft factors, psychological influence is difficult to measure. In
order to capture correct adoption rate, we introduce a parameter calibration step in our solution
framework. When we are in the process of implementing the AMI deployment, the number of
the actual adopted customers is known until current time, say denoted by RA. We can find out
proper (m*, ck, A ) by using regression with the least squared error,
(m*,c*, A’) = argmin| SIMA, 0,8) Rao.
In the case when no real data is available, we can use data from other new technology adoption
process to bootstrap the process.
Similarly, we use the first order delay to model benefit realization. The corresponding
equation can be written as
4B _((-a)-B) (2)
dt ti
where B is for Benefit realization, a is the smart meter IT failure rate and r, is time delay
parameter for delay in maturity of information systems. The equation (2) has the following
solution
B(t;@,7,B)) =(1-@) -(l-a@—B,)-e™",
where B, is the initial value for B. The behavior of the solution is shown in the second of Figure
4. The solution will asymptotically close to (l-a) for 0<B,<1-—a. The function is a
multiplier and is applied relative to the deployment time of meter, similar to the adoption
multiplier A. The (a, Tp, B,) can be calibrated based on observable data in the same manner for
A.
3.2. Demand response model
The approach followed here is based closed on the RAND report [8]. Assume that the customer
with an installed smart meter will receive real time electricity prices and is price sensitive. The
focus here is to estimate the cost saving associated with demand shifting from peak periods to
off-peak periods (see Figure 5 for illustration).
System
Load
123 4 5 6 7 8 9 1011 12 13 14 15 16 17 18 19 20 21 22 23 24
Hour of the Day
Figure 5: The electricity demand profile for a typical day
In order to assess the cost savings related to demand shaping from market, we need to
estimate the following quantities:
* Market penetration(M , ). Percentage of customers that have smart meter installed would
take advantage of this feature and change their energy usage pattern. In fact, this is the
same as the adoption percentage discussed in the last subsection.
ePrice elasticity of demand (77). Percentage change in the demand for a 1% change in price.
We have estimated this using an empirically determined coefficient for price elasticity of
demand [9]. Typically price sensitivity is negative (demand reduce with price increases).
We have chosen a nominal value of -0.1 with a spread of [-0.15, -0.05].
¢ The supply curve is approximated by using the hourly price and load for each day. This
data can be plotted as a curve from demand to clearing price (we need to sort the price-
demand pairs in ascending order of price). For any given price, the load can be
determined by interpolating over this curve.
Demand shaving L, is determined by the following
L, = =P). (L-1,)-M,,
P,
where L load, L, base load, p price, p, the fixed price. The second factor on the RHS is the
percentage change of price. The price p, corresponding to the update load L + L, is interpolated
for the price within the supply curve that is approximated using hourly loads and clearing prices.
Note that we smooth out the variability in the data locally (using a 3-4 hour time window). The
saving is given by
(L-L,)-p-(L+L,-L,)-p,.
That value is positive for p> p, (increase price during peak hour) since L, <0 andp, <p.
Note that, the savings for even low penetration (20%) of smart meters is substantial. However,
many of the inputs, like fixed price, price elasticity are based on guesstimates and hence need
further calibration. This can also be accommodated through some Monte Carlo based uncertainty
analysis.
3.3. Optimization model
Mathematical programming techniques have been used to address the time-tabling problem, to
generate manufacturing plan and schedule to meet demand. We use the technique to create smart
meter deployment plan in the next 20 year horizon. The output from the models described in the
last two subsections is used in our benefit estimates during the process of deploying smart meters.
We only include benefit and cost related deploy schedule.
Our objective is to find a deployment plan such that it maximizes benefit and minimizes
cost. Suppose that the variable for deployment plan is written asY[j,t], subscript j is for the
jurisdiction index, and t is for time period. Formally, the objective can be expressed as
N T
(0): mac 5° (Benet —DeployCost -OMCost — ExtraWF —h* ExtraM tial .
jal tal
Where h is for holding cost of extra meter in hand, N is the number of jurisdictions and T is the
time horizon. The expression sums over jurisdictions and over multiple time periods. Its solution
would address both the portfolio concern as well budget distribution among time horizon. The
last two terms are used to penalize the extra meter supply and extra workforce. In fact, without
them, the solution tends to allocate the enough resources at the first period.
The value of Y[j,t] varies from 0 to 1, and represents the percentage of meters that get
installed, relative to the number of customers. Under the constraint of early start ES[ j] and late
finish LF[j], we have the following constraint, for each jurisdiction j
(C1): DYE t=1 Y[j,t]=0 fort ¢[ES[j], LF[j]].
tel
3.3.1. Benefit formulation
Note that benefits are associated with the cumulative number of meters that have been
installed, and depend on how long has any individual meter been installed (to account for the lag
in the benefit realization profile mentioned in the subsection 3.1). We assume availability of
estimates of the unit benefits BU and percentage growth BG related to various benefit categories
(customer contact benefit, demand response benefit, meter reading benefit, asset optimization
benefit etc.). For instance, meter reading benefit is formulated as
mrBenefit{ j,t] = BU, [j]-(1+BG,,,{jJ)"-M[j]-
Y[B,e-s{ ri steoatn vt.)
ral
s=1
where M[j] denotes the number of current customers who are considered to convert into smart
meters. GR[j] is customer growth rate in that jurisdiction j and B,(t) is benefit realization
profile created from system dynamics model. The demand response benefit is slight different
from others, since it also related to customer adoption rate A(t) that was estimated in system
dynamics model. The unit of benefit BU ,, for it is obtained from the demand response model in
subsection 3.2.
drBenefit{ j,t]=BU ,,[j]-(1+ BG,,[ jl)" -M[jl-
t
3 ate-s)-B {YE GRE) DY}
s=l
3.3.2. Deployment and operation maintenance cost
Deployment cost is associated with activities to install/or replace the meter. It includes both
equipment cost and labor cost. There is difference between the initial deployment cost and later
replacement cost, since the latter only includes the incremental portion. We also include the
inflation associated to labor and equipment into cost structure. For simplicity, cost difference
related to types of customer is ignored in the description of formulation. Mathematically, the
deploy cost is written as, for jurisdiction j at the time t,
DeployCostj,t}=[Cgl j]- (+14, )7 + Cul i)-Q +1, 2) MCi)-YE 0)
+ [Cul f)- (+ 1,4)? +C,Ci1-C+ 1, MCG) GRLT+ NRE) YC sI.
s=l
Where the parameters used in the formulation are listed here
C44: The equipment cost for deployment, per meter
C,, : The labor cost for deployment, per meter
C,,, : The equipment cost for replacement, per meter
C,, : The labor cost for replacement, per meter
C,,: The communication cost during operation, per meter
C,, : The labor cost for service, per meter
1, : The inflation rate for equipment, per year
I,, : The inflation rate for labor, per year
NR: The normal replacement rate, per year
Note that first term is for deployment cost and the second term is for growth and normal
replacement cost. Similarly, operation maintenance cost is given as
OMCost{j,t]=|C,,Ci)-(+1,) 2 +C,Li-0+h.)"]
MLL s}+GRL)- DVL.
e4. pet
The difference with deployment cost results from that of the maintenance cost is associated with
the total number of meters being installed over time from 1 to t.
3.3.2. Penalty related to resources
If we include additional variables for hiring, releasing workers and meter acquisition in our
formulation, then we can address issues associated to supply delay and workforce shortage. Note
that both deployment and operation maintenance includes the labor cost that is formulated based
on the unit cost per meter. If we assume average cost per workerC,,, then this cost can be
translated into deploy and service workforce capacity requirement.
w?
deployWC[j,t]=C,0j]-(1+1,)*-MLj]-YUj,t]
+C,(j]-(+1,) 7 -MIjl-(GRIj]+ NREjl)- YU, sI/C,
s=l
serviceWC[ j,t]=C,[j]-Q+1,)~ MUD EYE 81+ GRE DVL A/C
= t=l
The available workforce capacity can be written as
t
availableWC[ j,t]=W,{ j]+ >“ (W,[j,s-LT,]- EL[t-s—LT, +1]-W,[j,s])
s=l
(C2): YW,Li,s—RTJ-W,Lj,s])>0
sl
(C3): availableWC[j,t] = deployWC[j,t] + serviceWCTj, t].
Where W, is the number of initial workers, W,]|.,t] is the number of acquired workers at time t,
W,L,t] is the number of released workers at time t. The parameter EL[t] is an experience level
multiplier and changes from some value less then 1 to 1. The new hired workers need time to
learn and catch up, and then become fully experienced. The parameter RT in (C2) is minimal
residence time for a new hired worker before which, the worker may not be released. The
constraint (C2) specifies the relationship between hiring and releasing workers. The constraint
(C3) means that we need enough workers to carry out tasks. Note that deployWC, serviceWC are
just intermediate variables.
Now we formulate ExtraWF, that is extra cost associated with hiring and learning activity,
sitting on bench (less utilized) due to worker release delay (minimal residence). It is given by
ExtraWF[j,t]=C,,-(1+1,)“{LT, -W,lj,t]
+ YWLi, s —LT,]-W,[j,s])—deployWC[j, t]- sevice CL.
s=l
Without the term in the objective function, we could get a solution to hire enough workers in the
first period and to release nobody in the later periods.
Let MP[.,t] be procuring meters at time t. The ExtraM is equal to the accumulated
number of procured meters minus the accumulated number of required meters
t
ExtraM[j,t]=MI[j]+ MPI j,t-LT,.]
s=l
-MULY VU, s1+ RCI» nti) vt Ip.
s=l rl
(C4): ExtraM[j,t]>0
Where MI are initial available meters, LT,, is procuring lead time. So the first two terms
represent available cumulative number of meters at time t. The third term, the accumulated
number of required meters, includes deployment, growth and normal replacement portions.
Without the term (ExtraM) in objective function, we could get a solution to procure all meters in
the first period.
3.3.2. Constraints being considered
There are several resource constraints that can be imposed when we solve the optimization
problem, like budget upper bound, available workforce upper bound, procuring meter upper
bound as well as benefit lower bound (least benefit expectation). Mathematically, at time t,
N
(C5): }°(DeployCost{ j, t]+OMCostf j, t]+ ExtraWF[ j,t]) < investUBI[C],
ja
(c6): SwLit] < hireUBIt],
ja
N
(c7): > -MP[j,t] < procureUBIt],
jal
N
(c8): > Benefit{ j,t] > benefitLB[t].
jal
By imposing some combination of constraints (C5) to (C8), we can address different concern or
issues during the deploy process. We demonstrate some cases in the next section.
4. Simulation Scenarios and Results
In the case of excluding ExtraWF and ExtraM in the objective function, we would get a trivial
solution from minimizing the objective without any constraints. The best solution is to hire all
needed workforce in the first period and to procure all needed meters in the first period since it
does not involve any cost in the objective function, and to deploy all meters in the earliest start
period, since it would achieve the maximum benefit. By including ExtraWF and ExtraM in the
objective function, even without any constraints, the answer would be different: a) hiring would
spread among whole evaluation horizon with periodic releases (aligned with the optimal
deployment plan) due to hiring cost and on-bench wasted cost, b) procurement of meters would
be based on requirement that aligns with the deployment plan, c) the deployment plan wpuld
potentially spread across multiple periods between the specified earliest start and the latest finish
periods. From the optimized solution, we can obtain the benefit, cost, workforce hiring, and
meter procurement profile along considered timeline.
eee sor
= ato00
eee ronooo | FA
row] — || ||
100000000 | ae ct
crown |? 1 aonooo | 7 Tt
1 300000 4
coin — Ss
ao000000 | 2 = 0000
:
cs ° im
Figure 3: Cost and Benefit during 21 year periods
Figure 3 shows the cost and benefit as well as the corresponding meter procurement fora
time horizon of 21 years. Cost has a peak during deployment years (from 2 to 6) and benefit will
pick up gradually after meters being installed. The meter procurement (the right) has the same
figure as the meter deployment except it shifts to the left by one period due to procuring lead
time. Figure 4 shows the corresponding workforce profile. The span of peak is much wider due
to hiring and training lead time (time taken to become fully experienced), and the minimum
residence time constraint (worker can not be released immediately after hiring). The shape can
be different in the outsourcing case in which, an outsourced worker can be released right away
after finishing deployment. The right figure shows cash flows for two cases: one with demand
response benefit and the other one without it. The investment for deploying smart meters will
payoff around year 14. The utility company will achieve profit based on that schedule.
3000 ‘00000000
2500 t By ieee
7 | ct 9 $F
2000 " 00000000
Y >
1500 \ 20000000
\ y,
1000 \ ° =
ie TPAy 45 67 8 9 WUpe i yRt 119 20
- ~e oe x
é 0000000 wat
1224567 8 9 WIs 1617 2192021 || 500000000 —*
[Fe Wontoce Taner] [GRR Fo wa OR CHEN Fl WOR RTT CE
Figure 4: Workforce Change and Cash flow
Figure 5 shows the accumulative percentage deployment plan in different jurisdictions.
For specified earliest start and latest finish periods, the optimizer would create the deployment
plan in term of percentage of total current customers in that jurisdiction. Note that the solution
gives even distribution among the specified deployment periods under no budget constraints. The
solution would be different for mixed integer programming formulation, in which, we allow
users to specify, the early start, late start, duration and late finish, and we can get a solution with
different deploy cover periods.
120%
0% 4
«0% geet
am | 7
0%
1 2 3 4 5 6 7 8 9 1W
—+_JursdictionI —s—Jurisdiction2 Jurisdiction’
—»—Jurisdiction4 —x—J urisdiction5 —» J urisdiction6|
Figure 5: Accumulative percentage deployment among jurisdictions
From Figure 3, we know that the cost at deployment peak could be very high. It is
reasonable to impose annual budge and/or procurement constraints, and to relax late finish
requirement. Figure 6 shows the cost, benefit and deploy plan under budget limit to be $100M
during deploy period. Figure 7 shows the cost, benefit and procurement profile under both
budget limits to be 100M and meter procurement limit 380,000.
The model also allows user to specify predefine deploy plan (specify percentage of
meters deployed in each year for each jurisdiction). Then the optimizer would create cost, benefit,
workforce and meter procurement profile. User can further add meaningful constraints based the
above outputs and re-create new deploy plan. In the case of deployment being the half way of
process, the model can be used to re-plan based on observable data. User can specify deploy
percentage happened in the past and use the optimizer to create deploy plan for the rest of
periods.
Tacuoon 0
raoowon Po | 00 :
00000000 — 80% <i
stnotco cox Zak
60000000 4 40%
oe
seca \ om |
o%
20000000
al 1203 4 5 6 7 @ 9
Vea 37 9 no os Wo 2
[-sSeriest Series? Seres3 Series
[:e amet] -xe—Seriess “+ Series
Figure 6: Cost, Benefit and Deploy Plan under Budget Constraint
120000000
x |_| 450000
a 400000
100000000 350000 a amt
80000000 becca ie —
= 250000
ooo | real 2o0000 | 7 7 atl
000000 asoooo }é/ Vy
? me 100000 | 4
20000000 + 7 50000 f+ ~
i‘ “a Vee ‘
1
aos 7 9 1B ob Wow a 12.345 6 7.8 9 10311213 1435 1617 18192021
[Ee cost a Benett] [Ee tater Procurement e—Meter Deployment)
Figure 7: Cost, Benefit and Procurement under Budget and Supply Constraints
5. Conclusion
We propose an integrated model for AMI program management. It takes advantage of both
system dynamics and linear programming methodologies. System dynamics is used to estimate
intangible measure in the AMI deploying program management, like market penetration and
benefit delay and employee experience profile. The linear programming is used to create
deploying plan, meter procurement and workforce requirement. The model can address operation
issue related to budget constraint, workforce and supply shortage. The feedback loop is built into
the system in a general sense that the parameters in system dynamics can be calibrated from
observable data as the program progresses.
References
{1] A. Faruqui and S. George, Quantifying customer response to dynamic pricing, The
Electricity Journal, Vol. 18 (4), pp. 1040-6190.
[2] R. Levy, K. Herter and J. Wilson, Unlocking the potential for efficiency and demand
response through advanced metering, Lawrence Berkeley National Laboratory. Paper
LBNL-55673, http://repositories.cdlib.org/Ibnl/LBNL-55673, 2004
[3] J.W. Forrester, Industry dynamics, MIT Press, Cambridge, MA, 1961.
[4] J.D. Sterman, Business dynamics: system thinking and modeling for a complex world,
Irwin McGraw-Hill, Boston, 2000.
[5] P. Lund, Market penetration rates of new energy technologies, Energy Policy, Vol. 34, pp.
3317-3326, 2006.
[6] A. Janssen, S.F. Lienin, F. Gassmann and A. Wokaun, Model aided policy development
for the market penetration of natural gas vehicles in Switzerland, Transportation research
part A, Vol. 40, pp. 316-333, 2006.
[7] T. Eres, S. Moldovan and S. Solomon, Social anti-percolation, resistance and negative
word-of-mouth
[8] W.S. Baer, B. Fulton and S. Mahnovski, Estimating the benefits of the GridWise initiative
(phase I report), TR-160-PNNL, Rand Science and Technology, May 2004
[9] C.S. King and S. Chatterjee, Predicting California demand response, Public Utilities
Fortnightly, Jule 1, pp. 27-32, 2003
