Evaluating planning strategies for prioritizing projects in su

Evaluating planning strategies for prioritizing projects in
sustainability
improvement programs
Amir R. Hessamia , Vahid Faghihib , Amy Kimc and David N.
Fordb
aDepartment of Civil and Architectural Engineering, Texas
A&M University–Kingsville, Kingsville, USA; bZachry
Department of Civil
Engineering, Texas A&M University, College Station, USA;
cDepartment of Civil and Environmental Engineering,
University of
Washington, Seattle, USA
ABSTRACT
Programs to improve the sustainability of building
infrastructures often consist of project portfolios
that need to be prioritized in an appropriate chronological
fashion to maximize the program’s
benefits. This is particularly important when a revolving-fund
approach is used to leverage savings
from the initial projects to pay for later improvements. The
success of the revolving-fund approach
is dependent on the appropriate prioritization of projects.
Competing performance measures and
scarce resources make this task of project prioritization during
the planning stage a complex and
challenging endeavour. The current study examined the impact
of different project prioritization
strategies for revolving-fund sustainability program
performance. A novel modeling approach for

sustainability decision-analysis was developed using the system
dynamics method, and the model
was calibrated using a campus sustainability improvement
program at a major university. The
model was applied to evaluate the effects of five common
project-prioritization strategies on three
program-performance measures, across a wide range of initial
investment levels. For the university
case study, we found that the strategy of prioritizing projects
according to decreasing benefit/cost
ratio performed best. The research demonstrated that using a
system dynamics model can allow
sustainability program managers to make better-informed
sequencing decisions, leading to a
financially and environmentally successful program
implementations.
ARTICLE HISTORY
Received 15 August 2018
Accepted 11 April 2019
KEYWORDS
Project prioritization; system
dynamics; sustainability
improvement; revolving
fund; energy efficiency
Introduction
The development of sustainable infrastructure is of vital
concern in a world of limited resources. Currently, in the
United States, the residential and commercial sectors
account for about 40% of the country’s total consumed
energy (U.S. Energy Information Administration 2016).
Meanwhile, electricity generation, the industrial sector,
and the residential sector generate over 45% of the

country’s CO2 emissions (U.S. Environmental Protection
Agency 2016a). Reducing energy consumption in these
sectors through sustainability improvement programs
can provide great benefits – both in the form of imme-
diate monetary savings for owners and in the overall
context of a better living environment for the public.
A large amount of existing infrastructure was built
prior to the adoption of current sustainability design
and construction practices. For example, in the United
States, the electricity consumed in energy-efficient
buildings accounts for only 30% of the country’s total
building electricity consumption (Syal et al. 2013).
Upgrading older buildings to current energy standards
can thus help tremendously in reducing energy use.
The impact of energy efficiency improvements on the
economy has been thoroughly quantified, and these
numbers can be used as a basis for public policies
(Hartwig and Kockat 2016). In recent years there has
been a shift in policy and practice toward implement-
ing retrofit projects for broad portfolios of buildings,
rather than upgrading single buildings individually.
This approach creates a more efficient overall upgrade
process and has been supported through programs
such as the United States Department of Energy’s
Better Buildings Challenge (DoE 2018). In a similar
fashion, the Connecticut Energy Efficiency Program
facilitated the access of low-income households to
efficiency improvement opportunities by bundling
CONTACT Amir R. Hessami [email protected] Department of
Civil and Architectural Engineering, Texas A&M University–
Kingsville, 700
University Blvd, Kingsville 78363, Texas, USA
� 2019 The Author(s). Published by Informa UK Limited,

trading as Taylor & Francis Group.
This is an Open Access article distributed under the terms of the
Creative Commons Attribution-NonCommercial-NoDerivatives
License (http://creativecommons.org/licenses/by-
nc-nd/4.0/), which permits non-commercial re-use, distribution,
and reproduction in any medium, provided the original work is
properly cited, and is not altered, transformed,
or built upon in any way.
CONSTRUCTION MANAGEMENT AND ECONOMICS
2020, VOL. 38, NO. 8, 726–738
https://doi.org/10.1080/01446193.2019.1608369
http://crossmark.crossref.org/dialog/?doi=10.1080/01446193.20
19.1608369&domain=pdf&date_stamp=2020-05-21
http://orcid.org/0000-0001-7618-8159
http://orcid.org/0000-0002-6264-1378
http://orcid.org/0000-0001-8877-3777
http://orcid.org/0000-0003-3511-1360
http://creativecommons.org/licenses/by-nc-nd/4.0/
http://creativecommons.org/licenses/by-nc-nd/4.0/
http://www.tandfonline.com
similar retrofit components together across many indi-
vidual houses and forming collective project portfolios
(Cluett et al. 2016).
Identifying the optimal energy retrofit measures for
a building is essential to the success of sustainability
programs. One meta-analysis of the research literature
on building upgrades determined that these programs
could benefit from better energy-use modelling, eco-
nomic evaluation, and risk assessment to help select
the most cost-effective retrofit measures (Ma et al.
2012). These concerns become even more important

when considering collective portfolios. The limited
available studies in this area have focused on develop-
ing tools to quantify energy saving opportunities for a
portfolio of buildings (Lee et al. 2011) and to identify
buildings heterogeneities across portfolios (Pacheco-
Torres et al. 2016). The majority of research on the
cost-effectiveness of energy upgrades for a portfolio of
buildings is focused on identifying end-of-life positive
net present value opportunities (Granade et al. 2009),
rather than attempting to maximize the performance in
different dimensions. Carli et al. (2017) indicated that
there is a clear gap in the research literature for defi n-
ing optimal energy retrofit strategies for a portfolio of
buildings based on performance outcomes. The current
research study contributed to filling this gap by analyz-
ing strategies for the optimal allocation of sustainability
upgrade resources in a portfolio of buildings based on
both financial and environmental performance goals.
Improving energy efficiency in collections of build-
ings requires major capital investments when trad-
itional financing approaches are used. Access to
capital has been identified as a key barrier to initiating
energy efficiency retrofits (Hiller et al. 2011).
Innovative financing strategies such as the revolving-
fund mechanism can ameliorate the concern of inad-
equate capital. The revolving-fund financing mechan-
ism has gained widespread popularity in programs
focused on retrofitting existing structures and promot-
ing energy-conservation practices. In the revolving-
fund approach, the savings from the reduced operat-
ing costs achieved early in the sustainability program
are used to fund subsequent improvements, leading
to even greater savings. Thus, a relatively small initial
investment can leverage savings from energy-effi-
ciency improvements to fund many more projects

than the initial funding could support alone. Revolving
funds allow sustainability programs to be initiated
with far less than the total anticipated investment that
will be needed to complete their mission
(Peckinpaugh 1999). This approach has been adopted
by many university systems, as well as a variety of
other organizations (Indvik et al. 2013). According to
the Association for the Advancement of Sustainability
in Higher Education, over 80 higher-education institu-
tions now use a revolving-fund approach to promote
energy conservation, with a total investment of over
118 million dollars (AASHE 2016).
Despite the demonstrated value of revolving funds,
the lack of research on strategies to maximize the per-
formance of energy retrofits in building portfolios
makes it difficult to implement this approach effect-
ively. In some cases, there may be trade-offs between
the goal of maximizing early financial returns (and
thus having more funding to implement further proj-
ects) versus the goal of quickly implementing projects
that will maximize building performance (which may
require larger investments with slower returns, thus
reducing the amount of available capital). Analyzing
these factors to maximize the overall energy perform-
ance of the entire revolving-fund program over time
can be a daunting task. Project managers will need to
determine the best order in which to implement
desired projects to ensure that the maximum benefits
are obtained. If loan interest rates are low enough,
then it may be feasible to use the maximum amount
of capital possible from loans and improve all of the
facilities as soon as possible. However, moving too
quickly can also overwhelm the capital assets with
debt if the rate of financial savings cannot keep up.

Many sustainability improvement programs start with
very limited resources, which makes project sequenc-
ing a critical driver of performance.
Choosing an optimal project implementation
sequence is thus a complex program-design challenge.
In many cases, the data needed for a complete opti-
mization analysis is not available. Nonetheless, during
the project planning stages, managers must make
decisions about program sequencing. There is a sig-
nificant need for rapid, practical and reasonably accur -
ate methods to evaluate the feasibility of investments
and the sequencing of projects. The objective of the
current study was to evaluate common sequencing
heuristic strategies and identify their effect on the
overall performance of revolving-fund sustainability
improvement programs, considering a variety of differ-
ent program sizes and initial funding levels for a port-
folio of buildings. To achieve this objective a system
dynamics model was developed for sustainability pro-
gram decision-analysis, and several commonly used
heuristic strategies were tested to evaluate the effects
of the sequencing choices on the overall success of
the sustainability programs. The analysis carried out in
the current paper and the scope of research was
CONSTRUCTION MANAGEMENT AND ECONOMICS 727
limited to university campus building retrofit pro-
grams, but the methods described here can also be
generalized and applied to evaluate a variety of differ -
ent types of infrastructure.
Methods

This section discusses the research approach that was
used to analyze sustainability project sequencing. The
general method for solving sequencing problems is
defined, the applicability of the system dynamics
model is explained, and the specific design of the
model is described in detail.
Project sequencing strategies for sustainability
improvement programs
Project sequencing in a sustainability improvement
program can be viewed as a scheduling problem.
Methods of determining the optimum sequence of
activities in scheduling problems are categorized into
three major classes: (a) exact solutions, (b) approxima-
tions, and (c) heuristic algorithms (Shakhlevich 2004).
The best method for use in a particular context
depends on the level of accuracy that is needed and
the input parameters of the specific problem. Exact
solutions give more precise answers, but these meth-
ods also require more precise inputs, and the analysis
can often be very resource-intensive. Linear program-
ming (Mingozzi et al. 1998) and branch-and-bound
analyses (Lomnicki 1965) are examples of methods
used in exact mathematical scheduling solutions. In
contrast, approximation methods are designed to find
solutions that may not be the perfect optimum, but
that can be shown to be within an acceptable range
from the actual optimum. This approach can also be
complex, but it allows more flexibility and ease of
application compared to finding exact solutions.
Approximate methods have been successfully applied
in a variety of complex problems such as pavement
rehabilitation scheduling (Ouyang and Madanat 2004),
resource-constrained construction project scheduling

(Liu and Wang 2008), and vehicle routing (Novoa and
Storer 2009).
Finally, heuristic algorithms are designed to find a
good solution, but they do not necessarily guarantee
that it is within a specific range of accuracy. With the
right set of knowledge and experience, heuristic analy-
ses can provide viable solutions for complex problems
in a very short amount of time. Heuristic approaches
are particularly useful during the earliest phases of
program development, when precise design-level data
about the projects may not yet be available. The rela-
tive simplicity of heuristic algorithms also makes
them particularly suitable for supporting decisions at
higher levels of management. Examples of commonly
used heuristic methods include the Bottleneck
Dynamics approach (prioritizing in order of decreasing
benefit–cost ratio) (Morton et al. 1995), and the Tabu
Search (solution neighbourhood searches with worsen-
ing moves permission) (Glover and Laguna 1998). In
the current work, the researchers examined the most
applicable heuristic strategies for sequencing projects
in sustainability programs and assessed their effects
on program performance.
System dynamics
Critical decisions in planning sustainability improve-
ment programs can be evaluated by developing a sys-
tem dynamics model of the program. System
dynamics is one of several established and successful
approaches to systems analysis and design (Flood and
Jackson 1991, Lane and Jackson 1995, Jackson 2003).
It shares many fundamental concepts with other sys-
tems approaches, including emergence, control, and

layered structures, which are intended to help the
model address issues such as risk in large, complex
systems (Lane et al. 2004). The system dynamics
method uses a control-theory approach to study the
non-linear behaviour of complex systems. Since this
approach represents systems using interacting feed-
back loops, it is suitable for and widely used in policy
analysis (Flood and Jackson 1991, Lane and Jackson
1995, Jackson 2003). Forrester (1961) described the
original philosophy behind the system dynamics
method, and Sterman (2000) developed the modelling
process in detail and described several practical appli -
cations. When applied to engineered systems such as
improvements in building infrastructure, system
dynamics simulates the interactions within the causal
structure of the system (e.g. project progress rates),
along with system design and management strategies
(e.g. different project sequences), and base conditions
(e.g. the initial funding level). The model then predicts
how the system performance will change as various
parameters are adjusted.
Examples of system dynamics applications for pro-
ject planning and management issues can be found
throughout the research literature, including project
fast-tracking failure (Ford and Sterman 1998), undesir-
able schedule performance (Abdel-Hamid 1988),
change impacts (Cooper 1980, Rodrigues and Williams
1997), and assessing rework impacts on project
728 A. R. HESSAMI ET AL.
performance (Ford and Sterman 2003). System dynam-
ics has also been applied in the construction industry,

to topics including engineering economics and invest-
ment analysis (Senge 1980), bidding competition ana-
lysis (Kim and Reinschmidt 2006), project risk
management (Nasirzadeh et al. 2008), project cash-
flow management (Cui et al. 2010), market fluctuation
analysis (Mbiti et al. 2011) and managing the complex-
ity of information flow (Khan et al. 2016). The wide-
spread applicability of system dynamics modelling in
these fields provides strong support for the use of this
method in the current research.
In the system dynamics method causal feedback
and the accumulations and flows of materials, people,
and information are combined with behaviour-based
representations of managerial decision-making. This
approach is unique in its integrated use of stocks and
flows, causal feedback, and time delays to model sys-
tem processes. Stocks represent accumulations that
change over time, and flows represent the movement
of commodities into, between, and out of stocks. The
system components are linked with causal arrows that
indicate the direction of influence, helping to identify
feedback loops and cascading effects. Initial condi -
tions, time/speed factors, and managerial decisions
affect the overall balance of the system, allowing for a
model that has a strong predictive capability. System
dynamics is an ideal approach for modelling the
impacts of project sequencing on sustainability
improvement program performance due to its capabil-
ity to track the diverse set of features, characteristics,
relationships, and strategies that may affect the pro-
gram outcomes. Several core components of revolv-
ing-fund sustainability programs grow and shrink over
time (e.g. the total sustainability fund and the total
energy savings), with significant program implications;
these factors are well suited to modelling with the

stocks and flows of a system dynamics approach.
In this paper, project prioritization was investigated
by building a system dynamic model of a sustainabil-
ity improvement program. The model was based on
an actual building retrofit program (the case study) at
a major university campus. The validated model was
then used as an experimental tool to simulate the per-
formance of five project-sequencing strategies in
terms of monetary, temporal, and environmental
objectives (performance dimensions), using a wide
range of initial funding levels. The results were ana-
lyzed to identify preferred strategies in the case study
and to demonstrate how program managers can use a
system dynamics approach to draw conclusions about
program design.
The case of a sustainability improvement program
To demonstrate the impact of program managers’
decisions on the success of sustainability improvement
programs, a system dynamics model of such programs
was developed. The model was then calibrated based
on the specific case of a sustainability program carried
out at Texas A&M University (TAMU). The data used
for calibration were from the first phase of the pro-
gram, carried out in 2011 (Siemens and TAMU 2011).
This phase was a $10M upgrade for 17 existing facili-
ties at the university, including 13 research and teach-
ing facilities and 5 parking garages. The TAMU Utilities
and Energy Management Department oversaw the
sustainability improvement program, which mainly
involved increasing lighting efficiencies, improving
building automation systems (BAS), and improving the
heating, ventilation and air-conditioning (HVAC) sys-
tems. The total area covered under the program,

including all of the buildings, was slightly more than
four million square feet. The individual parking
garages had the largest areas, ranging from about
200,000 ft2 to about 1 million ft2. The 13 research and
teaching buildings had a much smaller square foot-
age, less than 200,000 ft2 each. Lighting retrofits,
which comprised the bulk of the work, involved
switching inefficient light bulbs and lamps with more
efficient equivalents. The BAS optimization consisted
of installing better automated climate-control equip-
ment for HVAC systems in each facility. For example,
sensors for detecting occupancy were mounted and
wired to HVAC controllers to reduce airflow while an
area is unoccupied. The installation of these sensors
allowed for automatically turning off lighting and cli-
mate conditioning when the areas were not being
used. Facility reset and hold up/setback plans were
also applied to further decrease energy usage. These
plans involved programming building environment
technology according to anticipated usage, for
example by adjusting temperatures in such a way as
to maintaining users’ comfort while minimizing cool-
ing and heating energy charges. The enhancement of
the parking garages involved only lighting retrofits,
while the 13 research and teaching buildings had a
combination of different types of improvements.
The funding required for this improvement pro-
gram was made available under the federal American
Recovery and Reinvestment Act (State Energy
Conservation Office 2010) and was supplied by the
Texas State Energy Conservation Office (SECO) to
TAMU at an annual interest rate of 2%. TAMU and
Siemens, a large energy-service company, participated
in a guaranteed performance contract to complete the

project (Siemens Industry US 2011). This means that a
specific total of confirmed savings was guaranteed for
the 17 buildings annually over the 10-year term of the
contract. To help achieve these guaranteed savings,
an approach called the “cyclical process of action” was
used (Gottsche et al. 2016). In this process, first, the
existing condition of each building was reviewed to
identify areas for improvement. The possible improve-
ments were then prioritized using a hierarchical strat-
egy. As the sequence of improvements was carried
out, the performance of the program was periodically
evaluated to ensure that it was on course to achieving
the anticipated savings. The cyclical nature of this pro-
cess can be appropriately represented with a system
dynamics model.
Data from 2009 building energy consumption
records were used as the baseline to calculate future
energy savings. To determine the Actual Realized
Savings this baseline energy usage was considered as
the reference point and was compared against the
actual energy consumption during the Performance
Guarantee Period. Heating water, chilling water, and
electricity were the three basic energy consumption
sources that were identified for determining the total
energy consumption affected by the sustainability
improvements. Heating and chilling water were meas-
ured in millions of British Thermal Units (MMBTU) and
electricity was measured in kilowatt-hours (kWh). The
total annual energy consumption was calculated by
converting the kWhs to MMBTUs (1 kWh
¼ 0.0034 MMBTU).

Expected annual savings were defined in a Utility
Assessment Report, which was carried out and pro-
vided by Siemens to TAMU. The greatest energy sav-
ings were predicted for the parking garages; these
predictions ranged from about 30% to almost 50%
reduction compared to the baseline. All but one of
the teaching and research buildings were expected to
have yearly savings ranging from 10–30%. One build-
ing, the Zachry Engineering Center, was predicted to
have only about a 5% reduction in energy use.
Overall, the total predicted (and guaranteed) annual
cost savings for the project was about $1.126M. This
included $45K in operational savings and $1.08M in
utility savings.
Model structure
The conceptual basis of the system dynamics model
was the revolving fund structure (Like 2009). In this
structure, the costs of initial improvement projects are
covered by taking out loans from the revolving fund.
As a result of those improvement projects, the system
uses less energy and generates savings, which are
then used to repay the loan back into the revolving
fund. The system dynamics model was developed to
simulate the accumulations and flows of money and
the causal feedback that drive program behaviour and
performance (Figure 1). This general conceptual model
was extended to simulate the specific TAMU sustain-
ability improvement program, specifying the 17 TAMU
buildings and their particular characteristics (energy
usage, improvement cost, etc.) (Kim et al. 2012). The
model was developed in VensimV

R
DSS software and
used an arraying function to reflect facility and project
data that was stored in a MicrosoftV
R
Excel file.
The three main stocks in the system dynamics
model are the Sustainability Fund, Savings, and
Investment. External funds, as well as the monetary
savings of the program, gradually pool in the
Sustainability Fund over time. When the available
Sustainability Fund reaches the amount needed to
start the next project (the next building’s improve-
ment), as determined by the sequencing strategy, the
model triggers the project’s start and removes funds
equal to the defined project budget from the
Sustainability Fund (loop B2 in Figure 1). As a result of
implementing the projects, the amount of energy and
operating expenditure decreases in a manner defined
by the guaranteed contract, resulting in savings that
are added back into the Sustainability Fund (loop B1
in Figure 1). Loan payments are also processed by
removing them from the Sustainability Fund (loop B3
in Figure 1). Taken all together, these interactions cre-
ate the Revolving Fund Loop (R1 in Figure 1), a rein-
forcing feedback loop that maintains the Sustainability
Fund and then eventually increases it after all of the
projects have been completed. A more detailed
description of this model structure was published by
Faghihi et al. (2015).
Model testing and calibration

Standard model-testing methods for system dynamics
(Sterman 2000) were applied to validate the model,
including a comparison of the model structure to
actual system structures, verifying unit consistency,
testing behaviour under extreme conditions, and com-
parison of model behaviour to known or expected sys-
tem behaviour. Partial model testing was also used to
develop confidence in the model’s fidelity with the
system being modelled. For example, the major rein-
forcing loop of investment in energy efficiency and
generating savings (R1) was isolated from the rest of
the model, so that it could be tested and calibrated
independently.
The model was calibrated to the TAMU case study
using data from the project’s Utility Assessment
Report, Texas A&M University utility records for each
building, the details of the contract between TAMU
and Siemens, and informal discussions with represen-
tatives of the involved parties. The behaviour of the
calibrated model was used to further validate its
applicability. After the model was tested and cali-
brated to the case study conditions, a few adjustments
were made so that the calibration would be more
realistic for a wide range of sustainability programs.
These changes included the addition of increases in
utility prices (assumed to be 2% per year). A negative
Sustainability Fund was allowed in the model as long
as it subsequently became positive again within one
fiscal year. The researchers assumed that in such a
case the owners would borrow funds to cover these

temporary deficits, paying an additional 2% interest
per year on the extra funds. This version of the model
is hereafter referred to as the “base case”. More details
about the model are available from the authors
upon request.
Simulation design
The most applicable heuristic strategies for sequencing
projects in sustainability programs were evaluated
using the system dynamics model. First, two heuristics
were set as benchmarks for comparative purposes (H1
and H2). Then an exhaustive list of heuristic schedul-
ing rules from the literature (Panwalkar and Iskander
1977) was carefully examined to select the approaches
that are most applicable for use in sustainability
improvement programs. Three common heuristic strat-
egies (H3, H4 and H5) were identified based on
Panwalker’s approaches of the highest dollar value
and shortest implementation time.
� Benchmark Heuristic 1 (H1): Projects are regarded
as a hypothetical set of homogenous projects, all
of which have the same costs and generate the
same amount of savings (thus, the prioritization
Figure 1. The conceptual system dynamics model of revolving-
fund sustainability improvement programs.
order does not matter). This scenario provided a
baseline against which the other strategies were

compared. This strategy is referred to as “H1:
Homogenous Projects.”
� Benchmark Heuristic 2 (H2): Projects are initiated in
the order in which they were actually implemented
during the real-world program that was used as
the case study for this investigation. This strategy is
referred to as “H2: Case Study.”
� Heuristic 3 (H3): Projects are initiated in order of
decreasing improvement cost. This strategy reflects
a risk-management perspective based on the view
that delayed projects have a lower chance of being
successfully completed. Many factors can combine
to generate higher risk in postponed projects,
including the possibility of internal program mis-
steps and possible changes in external support.
The prospects of available funding in the near
future are almost always clearer than the prospects
of the far future. Therefore, program managers
may try to mitigate risks by prioritizing the most
expensive projects. This strategy is referred to as
“H3: Decreasing Cost.”
decreasing first-year benefit to cost ratio (B/C). In
this approach projects that will generate the high-
est first-year B/C are completed first. The first year
B/C for a project is the sum of total savings antici-
pated from improving the energy consumption of
a building during the first year after the project
implementation, divided by the project’s cost. Thus,
the first projects to be implemented are not neces-
sarily those that will generate the greatest immedi-
ate benefits, but rather those that will produce the

most benefits in comparison to the cost of their
implementation. This strategy is referred to as “H4:
Decreasing B/C.”
decreasing estimated savings. This strategy priori-
tizes projects that have the greatest total energy
saving potential (without concern for their relative
implementation costs). This strategy is referred to
as “H5: Decreasing Savings.”
The winnowing process for selecting these heuris-
tics included developing scenarios to assess how each
strategy would be applied in the context of a sustain-
ability program, and in some cases running simula-
tions to help eliminate strategies that consistently
underperformed in comparison to others. Examples of
strategies that were eliminated due to their clear
inapplicability include “increasing first-year B/C”
(where projects with the lowest first-year B/C are pri-
oritized) and “increasing savings” (where projects with
the smallest amount of savings are prioritized). Such
approaches would be entirely unsuitable for maximiz-
ing revolving fund returns.
The success of the tested heuristic strategies was
evaluated using program performance measures over
an anticipated 30-year life cycle. Choosing the
Figure 2. Total monetary value (NPV) using different project
sequencing strategies at different levels of initial funding.

performance measures was a delicate task. A system-
atic approach to defining these measures begins with
identifying the agency’s sustainability goals and
related objectives to achieve these goals. Then, precise
performance measures need to be established to
assess progress toward each of the objectives
(Zietsman et al. 2011). In this case, the Texas A&M
University 2018 Sustainability Master Plan identified 16
“Evergreen Goals” (TAMU Office of Sustainability 2018).
Among these goals, only two were directly related to
the sustainability improvement program that was
examined in the case study:
� Goal 1: Achieve a 50% reduction in greenhouse gas
emissions per weighted campus user by 2030;
achieve net-zero emissions by 2050.
� Goal 2: Deliver the lowest life-cycle-cost construc-
tion to build, operate, maintain, and decommission
high-performing facilities.
To evaluate the progress toward achieving these
goals, the researchers identified specific objectives and
performance measures. The first objective relates to
the program’s environmental performance, which was
quantified and measured as the per-unit cost of car-
bon footprint reduction. Carbon footprint is a widely
accepted and commonly used measure in environ-
mental life-cycle assessment (Matthews et al. 2008). To
calculate this environmental performance measure,
the total cost of improvements was divided by the
total decrease in energy use over the life-cycle of the
program (defined in comparison to pre-improvement
energy use). Based on models from the U.S.
Environmental Protection Agency (U.S. Environmental
Protection Agency 2016b), each kilowatt-hour of elec-

trical energy saved reduces carbon dioxide by 0.0007
metric tons, and each million British Thermal Units
(MMBTU) of natural gas energy saved reduces carbon
dioxide by 0.005 metric tons.
The second performance measure focused on the
economic efficiency objective of the program (as
related to Goal 2) in terms of financial savings to the
university. There are several economic analysis meth-
ods that can be used to assess the economic feasibil -
ity of building-efficiency improvement projects. The
more credible methods are based on the concept of
the time value of money (Park 2013). These methods
include net present value (NPV), internal rate of return
(IRR), benefit-cost ratio (B/C), and discounted payback
period. A comparison of these economic analysis
methods is beyond the scope of this paper. However,
the most widely used economic analysis method in
energy retrofit projects is NPV (DeCanio 1998, Jackson
2010, Morrissey and Horne 2011, Ma et al. 2012), and
this approach was also selected as the economic per-
formance measure in the current study. The basic
engineering economics method was used to calculate
the NPV, assuming a 5% interest rate. It was assumed
that the interest rate reflects the market interest rate
(covering the earning power and effect of inflation),
and cash flows were indicated in actual dollars (includ-
ing inflation) (Park 2013).
In addition to the environmental and economic
performance measures discussed above, the research-
ers also introduced a third, temporal performance
measure. This measure was simply the total duration
of the program implementation phase (in months),
with shorter durations being preferable. University

administrations are always concerned about the dur-
ation of ongoing construction projects, and eager to
see these improvements completed as quickly as pos-
sible. Construction creates inconveniences and aes-
thetic impacts for students and campus visitors, and
may even jeopardize the quality of education if it
interrupts classroom activities. Thus, chronological per-
formance in the sense of minimizing implementation
time was also considered as a relevant measurement.
Results and discussion
Using the system dynamics model, each project
sequencing heuristic (with the exception of H2 as
noted below) was simulated over a range of initial
funding – from 15% of the total program costs to
100% of the total program costs, in 5% increments.
Program performance, as measured in the environmen-
tal, economic, and temporal dimensions, was plotted
over the range of initial funding levels (Figures 2–4).
Each line in these graphs, therefore, represents the per-
formance of a single project sequencing strategy in the
context of a single performance measure. Strategy H2,
which describes the actual case study as implemented
at TAMU, is shown in the graphs as a single “X” rather
than a series of points. This is because in the actual
case study the improvements were all fully funded at
the beginning of the program.
The sequence of improvement projects for Strategy
H2–H5 are provided in Table 1. Projects in Strategy H1
were assumed to be homogenous, and are therefore
indifferent to sequencing strategy. For this reason, H1
is not included in Table 1. H2 is the original case
study, wherein projects were categorized into four
groups, with the projects in each group implemented

at the same time.
A number of important observations for the plan-
ning of revolving-fund sustainability improvement pro-
grams can be made on the basis of these results. First,
performance varied widely across the sequencing heu-
ristics, and this was true for all three of the perform-
ance dimensions. Comparing the three non-
homogenous heuristics (H3, H4 and H5) with 50% ini-
tial funding, the program net present value varied up
to 100% ($3.0M vs. $1.5M). The schedule performance
varied up to 36% (160 months vs. 250 months), and
the environmental performance varied up to 25%
($60/ton CO2 vs. $80/ton CO2). The scale of these
Figure 3. Total program duration using different project
sequencing strategies at different levels of initial funding.
Figure 4. Per-unit cost of carbon footprint reduction using
different project sequencing strategies at different levels of
funding.
performance variations is much larger than those pro-
duced by many other program performance improve-
ment means. This demonstrates that project
sequencing is an important, high-leverage factor in
sustainability improvement programs using a revolving

fund approach and that such decisions should be
made with care based on a good understanding of
the program’s feedback structure.
Second, the financial returns and schedule perform-
ance generally improved for all strategies as initial
funding levels increased. The reason for this is that
regardless of the project sequencing strategy chosen,
partial funding will delay the start of some projects
and thereby delay the capture of their benefits. In
contrast, the environmental performance of the vari-
ous strategies was generally worse when initial fund-
ing was higher (i.e. the cost per unit of carbon
reduction was higher with greater initial funding, in all
but the baseline homogenous project sequence, H1).
This is because programs with more initial funding do
not exploit the maximum cost savings that can be
obtained from the revolving fund financing approach.
A third general observation is that all of the com-
petitive strategies (H3, H4 and H5) performed about
the same in all three performance dimensions if at
least 60% of the total improvement costs are provided
as initial funding. This suggests that the program per-
formance is fairly insensitive to the differences among
these three sequencing variations when the initial
funding level equals or exceeds 60% of the total
improvement costs.
Fourth, all of the competitive strategies (H3, H4 and
H5) performed noticeably better than the
Homogenous Projects strategy (H1) in all three per-
formance dimensions. Strategy H1 is the only
approach that generated a negative NPV (when initial
investment was less than 75% of the total improve-
ment costs). The relatively poor performance of H1

can be explained as a failure to take advantage of the
impacts of diversity in project characteristics.
Assuming that all of the projects are the same elimi-
nates most of the advantages that can be leveraged
from the revolving fund approach, as it is no longer
possible to prioritize more effective projects and then
roll these benefits over to the less effective projects.
Therefore, as expected, the resulting performance
curves of H1 are much smoother than those of other
strategies in all three performance dimensions.
Fifth, there is one major “kink” in the performance
curves that occurs at about 85% initial funding. This is
a result of a meaningful shortage of funds. When the
available funding falls below a certain percentage of
full funding (90% in the case study program), the lack
of funds begins to delay the initiation of projects. This
funding shortage pushes multiple improvement proj-
ects later in time while the program managers wait to
collect the needed funding from energy savings in
previously improved buildings.
Sixth, the results indicate that with few exceptions,
the Decreasing B/C strategy was best in all three per-
formance dimensions, followed by the Decreasing
Savings strategy, followed by the Decreasing Cost
strategy, followed by the Homogenous strategy. This
suggests that incorporating both benefits and costs in
decision-making improves performance when com-
pared to approaches that consider only benefits or
only costs. If only one factor, benefits or costs, can be
considered, then these results indicate that benefits
(savings) should be used to prioritize the sustainabil-
ity projects.

Seventh, the strategies diverge more extensively in
effectiveness as the initial funding level decreases (i.e.
the lines move further apart toward the left-hand side
of the graphs). This is true for all three performance
dimensions. When initial funding levels are low, any
inefficiency in the prioritization strategy is amplified
because this inefficiency creates a more significant
drag on future funding levels. The slower accumula-
tion of funds from energy savings when there is poor
prioritization, combined with lower starting funding,
leads to a slow-programs-become-slower behaviour
mode. This divergence in strategies at very low initial
funding levels can become quite significant. For
example, the schedule performance difference
between the Decreasing B/C strategy and the
Decreasing Savings strategy exceeds 40% when the
initial funding level is 25% of the total improvement
costs. When the initial funding level is reduced to 15%
of the total improvement costs, the difference in
effectiveness between these two strategies
exceeds 100%.
Table 1. Sequence of improvement projects for H2–H5.
Heuristic
Building ID H2 H3 H4 H5
1501 1 2 1 1
1507 3 1 8 2
378 4 7 2 6
388 2 6 3 5
1559 3 4 4 3
1194 1 5 5 4
469 1 11 6 9
379 3 10 7 8
392 2 8 9 10

463 3 9 10 11
518 3 3 11 7
1508 3 12 12 12
Conclusions
This study demonstrates the application of system
dynamics models in successfully planning and manag-
ing revolving-fund sustainability improvement pro-
grams. Designing sustainability improvement
programs is a complex and challenging task due to
the interactions among diverse system components,
the variety of potential performance measures, the
effects of limited funding, and the different trajectories
that the programs can take over time. Revolving fund
financing can leverage relatively small initial invest-
ments into large program benefits, but this approach
can only be used successfully when it is combined
with careful program management and informed pro-
ject-prioritization strategies.
The system dynamics model developed in this
research was calibrated and tested using a sustainabil -
ity improvement program at a major university. Three
program performance measures (net present value,
program duration, and per-unit carbon dioxide reduc-
tion) were evaluated to reflect the values of a diverse
set of program goals. Three program-sequencing heu-
ristics, based on cost, savings, and benefit/cost ratio,
were tested over a wide range of initial funding condi-
tions and compared against two benchmark heuristics.
As noted earlier, this approach addressed an import-

ant gap in the existing research literature in regard to
defining optimal energy retrofit strategies for a port-
folio of buildings based on performance outcomes.
The combination of a revolving-fund financing
approach with complex program performance meas-
ures creates an extremely complex scheduling prob-
lem. Identifying optimal heuristic approaches for
tackling this scheduling program is vital to help pro-
ject managers make reasonably good decisions. The
use of the developed approach supports the use of
dynamic planning of portfolios rather than static plan-
ning. There are several secondary contributions of the
paper. The paper demonstrated the application of a
previously developed structured method for defining
performance measures in energy retrofit programs.
This study revealed that the program performance is
more sensitive to the choice of sequencing strategies
when the initial seed funding levels decreases. The
paper also confirmed that the use of both cost and
savings in the sequencing of projects will result in the
best sequencing strategy.
The large variation in results among the scheduling
heuristics verifies that project sequencing policies are
a high-leverage component of the design and man-
agement of revolving-fund sustainability programs.
With lower initial funding levels, scheduling decisions
have increasingly pronounced effects on overall out-
comes. The simulation results for the university cam-
pus case study indicated that the decreasing benefit/
cost ratio heuristic performed best, followed by the
decreasing savings, decreasing cost, and the finally the
homogenous project sequencing strategies. Additional
applications of the model are needed to generalize
these results to broader classes of projects and pro-

grams, but the results of the current work can be
used as hypotheses in future investigations of similar
systems. The simulation model produced in this
research provides a formal causal structure that is
widely applicable to sustainability improvement pro-
grams using revolving funds.
The results and conclusions of the current work are
limited by the assumptions used in the analysis. The
current work looks only at a single program and not
its environment. Some sustainability program contexts
(e.g. those conducted by profit-driven organizations)
may need to address competing uses of financial,
managerial, and other resources, as well as various
macro-economic factors that are not considered here.
Different measures of program success may be used
by some decision-makers. Broader issues such as the
socio-environmental impact of construction activities
on the local community may need to be included for
some projects. To address these concerns, the model
used in the current study can be extended and recali-
brated to develop additional insights into program
design and optimization. The model can potentially be
adapted to investigate a much larger array of financ-
ing approaches, as well as other types of infrastructure
improvement programs (beyond sustainability
improvements). More nuanced versions of the model
may be developed that can incorporate different pro-
gram conditions, such as particular kinds of infrastruc-
ture or additional financial variables. The focus of the
current work was on improving sustainability through
physical changes to built infrastructures, but the
model can potentially also be expanded to incorpor-
ate the impacts of facility user behaviours, and the
combined effects of infrastructure upgrades with
behavioural energy-conservation efforts.

Increasing the sustainability of existing building
infrastructure is, and will continue to be, an important
part of responsible infrastruc ture ownership and man-
agement. Improvements in our understanding of sus-
tainability program design can tremendously enhance
the programs’ effectiveness, efficiency, and thereby
their attractiveness. The current research contributes
to this goal by showing how a system dynamics
modelling approach can be used to analyze the effect-
iveness of different project scheduling heuristics.
Acknowledgements
The authors are grateful to the Texas A&M University
Utilities and Energy Management group for sharing valuable
information that made this research possible.
Disclosure statement
No potential conflict of interest was reported by the authors.
ORCID
Amir R. Hessami http://orcid.org/0000-0001-7618-8159
Vahid Faghihi http://orcid.org/0000-0002-6264-1378
Amy Kim http://orcid.org/0000-0001-8877-3777
David N. Ford http://orcid.org/0000-0003-3511-1360
References

AASHE., 2016. Campus sustainability revolving loan funds
database [online]. Available from: http://www.aashe.org/
resources/campus-sustainability-revolving-loan-funds/
[Accessed 29 January 2019].
Abdel-Hamid, T., 1988. Understanding the ‘90% Syndrome’
in softwareproject management: a simulation-based case
study. Journal of systems and software., 8, 319–330.
Carli, R., et al., 2017. A decision making technique to opti-
mize a buildings’ stock energy efficiency. IEEE transactions
on systems, man, and cybernetics: systems, 47 (5), 794–807.
Cluett, R., Amann, J., and Ou, S., 2016. Building better energy
efficiency programs for low-income households.
Washington, DC: American Council for an Energy-Efficient
Economy (ACEEE).
Cooper, K.G., 1980. Naval ship production: a claim settled
and a framework built. Interfaces, 10 (6), 20–36.
Cui, Q., Hastak, M., and Halpin, D., 2010. Systems analysis of
project cash flow management strategies. Construction
management and economics, 28 (4), 361–376.
DeCanio, S.J., 1998. The efficiency paradox: bureaucratic and
organizational barriers to profitable energy-saving invest-
ments. Energy policy, 26 (5), 441–454.
DoE., 2018. Better buildings challenge [online]. Available
from:
https://betterbuildingsinitiative.energy.gov/challenge
Faghihi, V., Hessami, A.R., and Ford, D.N., 2015.
Sustainability

improvement program design using energy efficiency and
conservation. Journal of cleaner production, 107, 400–409.
Flood, R., and Jackson M.C., 1991. Creative problem solving:
total systems intervention. Chichester, UK: Wiley.
Ford, D., and Sterman, J., 1998. Modeling dynamic develop-
ment processes. System dynamics review, 14 (1), 31–68.
Ford, D., and Sterman, J., 2003. The liar’s club: concealing
rework in concurrent development. Concurrent engineer-
ing: research and applications, 111 (3), 211–219.
Forrester, J.W., 1961. Industrial dynamics. Waltham, MA:
Pegasus Communications.
Glover, F., and Laguna, M., 1998. Tabu search. In: D. Du &
P.M. Pardalos, eds. Handbook of combinatorial optimiza-
tion. Boston, MA: Springer, 2093–2229.
Gottsche, J., Kelly, M., and Taggart, M., 2016. Assessing the
impact of energy management initiatives on the energy
usage during the construction phase of an educational
building project in Ireland. Construction management and
economics, 34 (1), 46–60.
Granade, H.C., et al., 2009. Unlocking energy efficiency in the
U.S. economy. Milton, VT: Villanti&Sons.
Hartwig, J., and Kockat, J., 2016. Macroeconomic effects of
energetic building retrofit: input-output sensitivity analy-
ses. Construction management and economics, 34 (2),
79–97.
Hiller, J., Mills, V., and Reyna, E., 2011. Breaking down
barriers

to energy efficiency. New York, NY: EDF Climate Corps.
Indvik, J., Foley, R., and Orlowski, M., 2013. Green revolving
funds: a guide to implementation & management
[online]. Available from: http://greenbillion.org/wp-con-
tent/uploads/2015/07/GRF_Full_Implementation_Guide.pdf
[Accessed 15 May 2018].
Jackson, J., 2010. Promoting energy efficiency investments
with risk management decision tools. Energy policy, 38 (8),
3865–3873.
Jackson, M.C., 2003. Systems thinking: creative holism for
managers. Chichester, UK: Wiley.
Khan, K.I.A., Flanagan, R., and Lu, S.-L., 2016. Managing
infor-
mation complexity using system dynamics on construc-
tion projects. Construction management and economics, 34
(3), 192–204.
Kim, A., et al., 2012. Designing perpetual sustainability
improvement programs for built infrastructures. St. Gallen,
Switzerland: System Dynamics Society.
Kim, H.J., and Reinschmidt, K.F., 2006. A dynamic competi-
tion model for construction. Construction management
and economics, 24 (9), 955–965.
Lane, D.C., and Jackson, M.C., 1995. Only connect! An anno-
tated bibliography reflecting the breadth and diversity of
systems thinking. Systems research, 12, 217–228.
Lane, M.B., McDonald, G.T., and Morrison, T.H., 2004.
Decentralisation and environmental management in
Australia: a comment on the prescriptions of the

Wentworth Group. Australian Geographical Studies, 42 (1),
103–115.
Lee, Y.M., et al., 2011. Modeling and simulation of building
energy performance for portfolios of public buildings. In:
Proceedings of the 2011 Winter Simulation Conference
(WSC), 11–14 December, Arizona. Phoenix, AZ: WSC.
Like, R.V.D., 2009. The paid-from-savings-guide to green exist-
ing buildings. Washington, DC: U.S. Green Building
Council, Inc.
Liu, S.-S., and Wang, C.-J., 2008. Resource-constrained con-
struction project scheduling model for profit maximization
considering cash flow. Automation in construction, 17 (8),
966–974.
Lomnicki, Z.A., 1965. A “Branch-and-Bound” algorithm for
the exact solution of the three-machine scheduling prob-
lem. Journal of the operational research society, 16 (1),
89–100.
Ma, Z., et al., 2012. Existing building retrofits: methodology
and state-of-the-art. Energy and buildings, 55, 889–902.
http://www.aashe.org/resources/campus-sustainability-
revolving-loan-funds/
http://www.aashe.org/resources/campus-sustainability-
revolving-loan-funds/
https://betterbuildingsinitiative.energy.gov/challenge
http://greenbillion.org/wp-
content/uploads/2015/07/GRF_Full_Implementation_Guide.pdf
http://greenbillion.org/wp-
content/uploads/2015/07/GRF_Full_Implementation_Guide.pdf

Matthews, H.S., Hendrickson, C.T., and Weber, C.L., 2008. The
importance of carbon footprint estimation boundaries.
Environmental science & technology, 42 (16), 5839–5842.
Mbiti, T.K., et al., 2011. System archetypes underlying the
problematic behaviour of construction activity in Kenya.
Construction management and economics, 29 (1), 3–13.
Mingozzi, A., et al., 1998. An exact algorithm for the
resource-constrained project scheduling problem based
on a new mathematical formulation. Management science,
44 (5), 714–729.
Morrissey, J., and Horne, R.E., 2011. Life cycle cost implica-
tions of energy efficiency measures in new residential
buildings. Energy and buildings, 43 (4), 915–924.
Morton, T., Narayan, V., and Ramnath, P., 1995. A tutorial on
bottleneck dynamics: a heuristic scheduling methodology.
Production and operations management, 4 (2), 94–107.
Nasirzadeh, F., et al., 2008. Integrating system dynamics and
fuzzy logic modelling for construction risk management.
Construction management and economics, 26 (11),
1197–1212.
Novoa, C., and Storer, R., 2009. An approximate dynamic
programming approach for the vehicle routing problem
with stochastic demands. European journal of operational
research, 196 (2), 509–515.
Ouyang, Y., and Madanat, S., 2004. Optimal scheduling of
rehabilitation activities for multiple pavement facilities:
exact and approximate solutions. Transportation research

part A: policy and practice, 38 (5), 347–365.
Pacheco-Torres, R., Heo, Y., and Choudhary, R., 2016. Efficient
energy modelling of heterogeneous building portfolios.
Sustainable cities and society, 27, 49–64.
Panwalkar, S.S., and Iskander, W., 1977. A survey of schedul -
ing rules. Operations research, 25 (1), 45–61.
Park, C.S., 2013. Fundamentals of engineering economics. 3rd
ed. London, UK: Pearson.
Peckinpaugh, C., 1999. How does a revolving fund work?
[Online]. Available from: https://fcw.com/articles/1999/09/
05/how-does-a-revolving-fund-work.aspx [Accessed 14
December 2018].
Rodrigues, A., and Williams, T.M., 1997. System dynamics in
project management: assessing the impacts of client
behavior on project performance. Journal of the oper-
ational research society, 49, 2–15.
Senge, P.M., 1980. A system dynamics approach to invest-
ment-function formulation and testing. Socio-economic
planning sciences, 14 (6), 269–280.
Shakhlevich, N., 2004. Heuristic algorithms: dispatching rules
[online]. Available from: http://web-static.stern.nyu.edu/
om/faculty/pinedo/scheduling/shakhlevich/handout09.pdf
[Accessed 9 March 2017].
Siemens & TAMU., 2011. A detailed account of how one
univer-
sity is improving its energy efficiency and campus environ-
ment through effective management and performance
contracting [online]. Available from: https://w3.usa.siemens.

com/buildingtechnologies/us/en/consulting-engineer/engi-
neeradvantage/Documents/texas-a-and-m-energy-improve-
ments.pdf [Accessed 29 January 2019].
Siemens Industry US., 2011. Answers for Texas A&M
University
[online]. Available from: http://www.youtube.com/
watch?v¼xIa8Ix91_rk [Accessed 14 March 2012].
State Energy Conservation Office., 2010. Building efficiency
and retrofit program round I: awarded projects to date
[online]. Available from: http://seco.cpa.state.tx.us/arra/
sep/building/ber_awards.php [Accessed 14 March 2012].
Sterman, J., 2000. Business dynamics: systems thinking and
modeling for a complex world. Irwin, USA: McGraw-Hill.
Syal, M., et al., 2013. Information framework for intelligent
decision support system for home energy retrofits. Journal
of construction engineering and management, 140 (1),
04013030-1–04013030-15.
TAMU Office of Sustainability., 2018. Sustainability master
plan [online]. Available from: http://sustainability.tamu.
edu/Data/Sites/1/downloads/2018SMP.PDF [Accessed 29
January 2019].
U.S. Energy Information Administration., 2016. Energy con-
sumption by sector [online]. Available from: http://www.
eia.gov/totalenergy/data/monthly/pdf/sec2_3.pdf
U.S. Environmental Protection Agency., 2016a. U.S.
Greenhouse gas inventory report: 1990–2014 [online].
Available from: https://www.epa.gov/ghgemissions/us-
greenhouse-gas-inventory-report-1990-2014 [Accessed 29

January 2019].
U.S. Environmental Protection Agency., 2016b. GHG equiva-
lencies calculator - calculations and references [online].
Available from: https://www.epa.gov/energy/ghg-equiva-
lencies-calculator-calculations-and-references [Accessed 29
January 2019].
Zietsman, J., et al., 2011. A guidebook for sustainability per -
formance measurement for transportation agencies.
Washington, DC: The National Academies Press.
https://fcw.com/articles/1999/09/05/how-does-a-revolving-fund-
work.aspx
https://fcw.com/articles/1999/09/05/how-does-a-revolving-fund-
work.aspx
http://web-
static.stern.nyu.edu/om/faculty/pinedo/scheduling/shakhlevich/h
andout09.pdf
http://web-
static.stern.nyu.edu/om/faculty/pinedo/scheduling/shakhlevich/h
andout09.pdf
https://w3.usa.siemens.com/buildingtechnologies/us/en/consulti
ng-engineer/engineeradvantage/Documents/texas-a-and-m-
energy-improvements.pdf

http://www.youtube.com/watch?v=xIa8Ix91_rk
http://www.youtube.com/watch?v=xIa8Ix91_rk
http://seco.cpa.state.tx.us/arra/sep/building/ber_awards.php
http://seco.cpa.state.tx.us/arra/sep/building/ber_awards.php
http://sustainability.tamu.edu/Data/Sites/1/downloads/2018SMP
.PDF
http://sustainability.tamu.edu/Data/Sites/1/downloads/2018SMP
.PDF
http://www.eia.gov/totalenergy/data/monthly/pdf/sec2_3.pdf
http://www.eia.gov/totalenergy/data/monthly/pdf/sec2_3.pdf
https://www.epa.gov/ghgemissions/us-greenhouse-gas-
inventory-report-1990-2014
https://www.epa.gov/ghgemissions/us-greenhouse-gas-
inventory-report-1990-2014
https://www.epa.gov/energy/ghg-equivalencies-calculator-
calculations-and-references
https://www.epa.gov/energy/ghg-equivalencies-calculator-
calculations-and-references
Copyright of Construction Management & Economics is the
property of Routledge and its
content may not be copied or emailed to multiple sites or posted
to a listserv without the
copyright holder's express written permission. However, users
may print, download, or email
articles for individual use.
AbstractIntroductionMethodsProject sequencing strategies for
sustainability improvement programsSystem dynamicsThe case
of a sustainability improvement programModel structureModel
testing and calibrationSimulation designResults and
discussionConclusionsAcknowledgementsDisclosure
statementReferences
EMA 630 Risk Matrix

Three Risks and Mitigations: Define, quantify, and mitigate the
following risks using the risk matrix. Note: There are multiple
correct answers possible in this assignment.
1. You hear on the news that there could be political unrest in
the country in which one of your key international suppliers
resides. The opposition party’s major platform is higher wages
for workers in order to gain popularity with the majority of
voters. The upcoming national elections are nine months away.
What are the risks that this political unrest could possibly affect
the production line?
2. Management wants to reduce costs by moving some
production facilities to a country where labor costs are cheaper.
Even though this move is consistent with current trade
agreements, the move is counter to the policies of the new
executive administration of the federal government, which is
promoting made-in-America products. The move would be legal,
but the federal government is threatening increased tariffs and
taxes to companies that do not comply with the buy-American
initiative. What are the risks to the company and the production
line if management decides to move the production facility?
3. The production line is staffed with union workers. The four -
year union agreement is coming to an end in 12 months.
Previous negotiations have been cooperative, but there have
been discussions of union discontent with current worker
benefits. On the other hand, raising worker benefits could affect
VALID’s profits, which would cause the company’s stock value
to drop and adversely affect stockholders. If there were a union
strike, the production line would be directly affected. What are
the risks?
Likelihood (L)
5

4
5
Consequence (C)
Figure 1: Risk Matrix
Table 1: Risk Likelihood Criteria
Level
Likelihood
Probability of Occurrence
5
Near Certainty
90%
4
Highly Likely
70%
3
Likely
50%
2
Low Likelihood
30%
1
Not Likely
10%
Table 2: Risk Consequence Criteria
Level
Consequence
5
Will jeopardize project success
4
May jeopardize project success

3
Limited impact to project
2
Can be tolerated with little or no impact to project
1
Minimal to no consequence
Directions: Study the three risks in the scenario. Then, address
the following:
1. Develop an “If this risk occurs, then this could happen to the
project” statement in Table 3.
2. Quantify the risk using the Likelihood Criteria (Table 1) and
Consequence Criteria (Table 2), and insert corresponding
likelihood and consequence numbers in the columns in Table 3.
3. Write a corresponding mitigation plan in Table 3. This step
aligns with critical element III.B in Final Project II.
4. Insert the risk number in the corresponding box in the risk
matrix in Figure 1. This step aligns with critical element III.A
in Final Project II.
Table 3: Risk ID, Quantification, and Mitigation
Risk #
L#
C#
Risk If Then Statement & Mitigation
(Example)
(1–5)
(1–5)
(If this risk happens, then this will happen to the project)
Mitigation: (State mitigation plan here.)

Assumption: (Insert assumptions associated with your response
to the risk.)
1
If political unrest within the country occurs, then ---
Mitigation:
Assumption: (if needed)
2
If management moves some of the production facilities to a
foreign country, then ---
Mitigation:
3
If the union strike occurs, then ---

Mitigation:
EMA 630 Scenario
VALID Short-Circuited Battery: Three Months Later
Introduction and Context: This case study builds on the case
study presented in EMA 600: Introduction to Engineering for
Engineering Managers. We will revisit
the company VALID, Inc., three months after the scenario
presented in EMA 600. For the purposes of this final project,
remember that we do not assume that
you have an engineering or technical background. The focus is
on the big picture of project leadership and how to ask the right
questions and think critically to
solve problems. There are no clear answers to the issues in this
scenario, so think critically and creatively, and always keep the
rubric and your overall goal in
mind.
Background: In this scenario, we have adequately and
successfully tested the new design. The production line is in
full operation, and VALID is delivering
batteries to satisfied customers. Marketing is very happy
because there are new customers ready to place orders that will

bring the company close to full
production capacity. Human resources (HR) is in hiring mode to
meet the new demand. Management is thrilled with the new
revenue stream. Battery
subcontractors are in production and delivering raw materials
and subcomponents on time and within quality standards. This
is great, but what could go wrong?
Three Issues: Ishikawa Diagram
1. Instructions: Use the Ishikawa diagram to identify the
potential root causes of the following issues. Choose the best
root cause(s) and propose
appropriate corrective action.
a. The production line is currently yielding 100 units/day. Full
capacity with three shifts can yield 150 units/day. For some
reason, when the
production manager performed a test yield run to determine if
the production line could produce the 150 units/day, the
production line only
yielded 110 units/day. This is a big issue because marketing is
already signing orders with new customers, which will increase
production
demand. What could be the root cause of this issue?
b. There have been five cases of production line workers
experiencing headaches on the assembly line. The Occupational
Safety and Health
Administration (OSHA) is involved and investigating. OSHA
has threatened to close the production line unless a quick root
cause is found. The
union is investigating the incidents and is demanding a quick
resolution to the issue. What could be the root cause of this
issue?

c. At the final quality checkpoint in the production process, QA
engineers perform acceptance tests to determine if the batteries
are delivering
constant voltage and steady heat dissipation. QA has been
witnessing batteries with out-of-specification low voltage
measurements. The issue is
not constant across all production units. The result has been the
rejection of these batteries, which has adversely affected
production quantities.
What could be the root cause of this issue?
Three Risks and Mitigations: Define, quantify, and mitigate the
following risks using the risk matrix:
1. You hear on the news that there could be political unrest in
the country in which one of your key international suppliers
resides. The opposition party’s
major platform is higher wages for workers in order to gain
popularity with the majority of voters. The upcoming national
elections are nine months
away. What are the risks that this political unrest could possibly
affect the production line?
http://snhu-
media.snhu.edu/files/course_repository/graduate/ema/ema630/e
ma600_case_study.zip
2. Management wants to reduce costs by moving some
production facilities to a country where labor costs are cheaper.
Even though this move is
consistent with current trade agreements, the move is counter to

the policies of the new executive administration of the federal
government, which is
promoting made-in-America products. The move would be legal,
but the federal government is threatening increased tariffs and
taxes to companies that
do not comply with the buy-American initiative. What are the
risks to the company and the production line if management
decides to move the
production facility?
3. The production line is staffed with union workers. The four -
year union agreement is coming to an end in 12 months.
Previous negotiations have been
cooperative, but there have been discussions of union discontent
with current worker benefits. On the other hand, raising worker
benefits could affect
VALID’s profits, which would cause the company’s stock value
to drop and adversely affect stockholders. If there were a union
strike, the production line
would be directly affected. What are the risks?
RESEARCH ARTICLE
Expert system for selecting and prioritizing projects for
handling urban water
supply crises
Welitom Ttatom Pereira da Silvaa and Marco Antonio Almeida
de Souzab
aDepartment of Sanitary and Environmental Engineering,
Federal University of Mato Grosso, Cuiabá, Brazil;
bDepartment of Civil and
Environmental Engineering, University of Brasília, Brasília,

Brazil
ABSTRACT
The water supply crisis (UWC) has affected various cities
around the world. The variability of possible
causes, the many viable alternatives to UWC management and
methodologies for selecting these
alternatives, as well as local government’s economic and
technical constraints make the problem
complex. The aim of this paper is to help select a set of
alternative solutions suitable for the UWC
problem. The proposed methodology comprised the following
steps: (1) theoretical foundation, (2)
planning the expert system (ES) to be built, (3) formal
knowledge explicitation, (4) knowledge coding,
(5) evaluation and adequacy of ES and (6) application of ES to
real-life UWC cases. The main result was a
computational decision support system, called UWC-ES. The
conclusion was that UWC-ES behaved as a
computational tool that reasonably reproduces knowledge from
various human experts with accepta-
ble applicability, and considering the possibility of using it in
other cases.
ARTICLE HISTORY
Received 24 January 2018
Accepted 24 September 2018
KEYWORDS
Management strategies;
rule-based expert system;
water crisis
1. Introduction
The urban water supply crisis (UWC) is currently a significant

problem affecting many populations around the world.
Numerous UWC cases can be found in the literature, such
as the city of São Paulo (Brazil), the provinces of northern
and western China, California (USA), the city of Cape Town
(South Africa), and the western prairie provinces of Canada,
which have been described in Coutinho, Kraenkel, and Prado
(2015); Zheng et al. (2010); Pollak (2010); Ziervogel, Shale,
and Du (2010), and Schindler and Donahue (2006), respec-
tively. This specific problem has motivated researchers to
seek alternative solutions and methodologies to cope with
them adequately.
The alternative solutions are varied and may consider struc-
tural strategies (technological options to reduce water con-
sumption, such as using water-saving equipment), non-
structural (actions that influence demand, such as changes to
pricing policies) and the combination of structural and non-
structural strategies. A more detailed discussion of structural
and non-structural strategies is presented in Savenije and Van
der Zaag (2002). The analysis methodologies for handling the
UWC include traditional optimization methods, simulation and
scenario generation techniques, statistical models, multiobjec-
tive and multicriteria methods, among others. For example,
Zarghami, Abrishamchi, and Ardakanian (2008) carried out
studies aiming to select alternative water management mea-
sures in an environment with significant population growth
and frequent water supply failures (in the case of the city of
Zahedan, Iran). A multiobjective and multicriteria model for
the problem of water supply contemplating several variables
(losses in the water network, consumption measures and
others) was developed. Different criteria (costs, need for
water supply, etc.) were aggregated using the Compromise
Programming method. The results showed that demand man-
agement measures can delay water transfer projects to the
city of Zahedan for more than 10 years. Artificial intelligence

techniques have also been used (León et al. 2000; Tillman
et al. 2005; López-Paredes, Saurí, and Galán 2005).
To analyze this context where there are various alterna-
tive solutions and different methodologies, faced with situa-
tions of severe limitations of financial and human resources
that many Brazilian cities and cities throughout the world
commonly go through, the following question arises: how
to select alternative solutions for a given UWC problem? As
a response, using UWC classification techniques and con-
structing an expert system is suggested based on studies by
Silva and Souza (2017) and Liao (2005). Thus, this study aims
to help select suitable project alternatives for the UWC
problem. More specifically, it is hoped that a system can
be obtained to support the decision-making process of
selecting priority projects to solve the UWC problem in
urban environments with a significant limitation of financial
and human resources.
2. Methods
The proposed methodology comprised the following steps: (1)
understanding the real problem using a literature review and
a study of the theoretical foundations of possible solutions, (2)
planning the expert system to be developed, (3) formal knowl -
edge explicitation, (4) knowledge coding and development of
the expert system (ES), (5) evaluation and adequacy of the ES
and (6) application of the ES to real-life UWC cases to verify
the acceptability of the developed ES response.
CONTACT Welitom Ttatom Pereira da Silva [email protected]
URBAN WATER JOURNAL
2018, VOL. 15, NO. 6, 561–567
https://doi.org/10.1080/1573062X.2018.1529806
© 2018 Informa UK Limited, trading as Taylor & Francis Group

http://www.tandfonline.com
http://crossmark.crossref.org/dialog/?doi=10.1080/1573062X.20
18.1529806&domain=pdf
2.1 Theoretical foundation
In this section, topics such as UWC classification (typification)
and the expert system (ES) are presented, which are the basis
for this study.
Using typical cases (case classification) plays an important
role in decision-making, especially when the decision involves
a large number of indicators and/or influencing factors (López-
Paredes, Saurí, and Galán 2005; UNEP/UNESCO 1987). In this
study, the decision support model for crisis management in
urban water supply (UWC-MODEL), developed by Silva and
Souza (2017), was used to simplify the analysis and study
different UWC situations (classification of UWC cases).
The UWC-MODEL performs the following activities: (1) it
aggregates influential factors in the UWC into five levels
(socioeconomic, management, environmental, urban and cul-
tural), (2) it evaluates its intensity of contribution to the UWC
situation for each level and (3) based on this evaluation, it
classifies the UWC situation at each level (classes: very strong,
strong, moderate, weak and very weak). The UWC-MODEL can
classify and/or typify UWC cases and, consequently, the cause
of the UWC is identified, helping to select and prioritize pro-
jects handling UWC. For example, in a case that has a very
strong contribution from the cultural level, the measures to
restructure the urban water supply system should prioritize
projects related to the cultural level. Therefore, setting up an
environmental education program could be an appropriate
project for handling UWC. In Equations (1) and (2), results

from the UWC-MODEL (RUWC-MODEL), the basis and
starting
point of this study, are presented in vector format, where
Cj=1, Cj=2, Cj=3, Cj=4 and Cj=5 are the classes of
socioeconomic,
management, environmental, urban and cultural levels,
respectively. More details about the UWC-MODEL can be
found in the study by Silva and Souza (2017).
RUWC�MODEL ¼ Cj¼1; Cj¼2; Cj¼3; Cj¼4; Cj¼5
� �
(1)
RUWC�MODEL ¼ fo; mfr; fr; mo; mfof g (2)
Another basis for this research was using the technique to
generate expert systems (ES). Artero (2009) defined ES as a
computational system designed to represent the knowledge
of one or more human experts on a particular domain and,
from the processing of the knowledge base, seek solutions to
problems that, in general, require a great deal of specialized
knowledge.
In an ES operation, it is assumed that the user feeds the ES
with factors or information and the system provides the user
with expert knowledge. Internally, ES consists of two main
components: the knowledge base and inference engineering.
The knowledge base stores knowledge and inference engi-
neering uses stored knowledge to construct the conclusions.
Some basic concepts refer to the problem domain, the domain
knowledge and the inference engineering. A problem domain
refers to a problem specific to an area (medicine, finance,
science or engineering) that the expert can solve. The expert’s
knowledge of how to solve a specific problem is called domain
knowledge. Inference engineering refers to the ability the ES
has to infer in the same way a human expert should infer

when faced with a problem.
The general strategies for ES development are shown by
Giarratano and Riley (2004). Briefly, the ES development pro-
cess consists of: (1) the ES developer establishes a dialogue
with the experts for the expert knowledge explicitation, (2) the
developer encodes the explicit knowledge (ES development),
(3) the experts evaluate and criticize the developed ES, the
developer makes adjustments and the process is repeated
until the ES is considered adequate by the experts. In practice,
the ES is an executable program that searches for the knowl-
edge about its domain in a separate file. This means that the
knowledge base can be completely changed and even then,
the program will work normally, adopting the knowledge from
the new base (Artero 2009). Some suggested references on
the subject are: Kim, Wiggins, and Wright (1990); Wright et al.
(1993); Nikolopoulos (1997); Resende et al. (2005); Artero
(2009); Giarratano and Riley (2004) and Liao (2005).
2.2 Expert system planning
The purpose of the ES planning stage was to produce a formal
plan for ES development called the UWC-ES. Thus, the feasi-
bility assessment, resource management and preliminary func-
tional layout tasks were performed based on
recommendations made by Giarratano and Riley (2004). For
the feasibility assessment task, the factors and returns sug-
gested by Giarratano and Riley (2004) were verified, in order to
decide if the ES approach would be adequate. The resource
management task was carried out by researching the compu-
ter resources (software and hardware), human resources and
financial resources to develop the UWC-ES. In order to do this,
a literature review of the resources used to develop precursor
ESs with similar objectives was carried out, and a comparison
was made with the resources available to develop the UWC-
ES. The preliminary functional layout task should define what

the system will achieve by specifying the system functions.
Thus, the objectives of the ES were carefully analyzed in order
to define the functions of the system, following recommenda-
tions by Giarratano and Riley (2004).
2.3 Formal knowledge explicitation
Knowledge explicitation refers to the process of acquiring the
knowledge needed to solve the problem (domain knowledge).
To do this, the activities used by Collier, Leech, and Clark
(1999); Tillman et al. (2005) and Patlitzianas, Pappa, and
Psarras (2008) were adapted. In this case, these activities
included: (1) defining the population universe of simulated
UWC cases, (2) defining the sample analyzed by the experts,
(3) identifying projects for handling UWC and (4) obtaining
domain knowledge. A total of 13 specialists (five with a mas-
ter’s degree and seven with a doctorate degree) were consid-
ered, of which six were working in the sanitation area, two in
the environment area and five in the water resources area, six
linked to water regulatory agencies, two to the environmental
protection agency and five to research institutions and
universities.
The population universe of simulated UWC cases is the
total possible number of combinations of the UWC-MODEL
classifications. Thus, 3125 (five levels and five classifications,
562 W. T. P. D. SILVA AND M. A. A. D. SOUZA
N = 55) individuals or typologies of simulated UWC cases were
observed that form the population universe. To define the
sample to be analyzed by the experts, the simple random
sample method was used. As justification, this method of
sampling leads to the sample in which each typology of the

sample population has the same probability of being selected,
not privileging specific situations or cases. The number of
sample units (n) was defined in 10% of the population,
which made a total of 313 typologies analyzed by the experts.
To identify the projects for handling UWC, a literature review
was carried out. Identifying priority projects (PP) for handling
UWC by experts for the ‘n’ sample units yielded the training
database, an initial part of the task of obtaining the domain
knowledge. For this purpose, the UWC (UWC-MODEL)
classifi-
cation and/or typology information, the identification of pro-
jects for handling UWC, the sampling technique used and the
samples to be analyzed were made available to the experts.
The experts were then asked to identify PP for handling UWC
(selection of five major projects for handling UWC) for each of
the typologies of the real-world/simulated cases analyzed by
them. For exemplification, from the process of obtaining the
training database, a graphical representation is illustrated in
Figure 1.
Having defined the training database, the final part of obtain-
ing the knowledge domain (obtaining the rules) was started.
Moreover, a machine learning technique was used for this pur -
pose, which automatically extracts information from the
training
database. More specifically, a decision tree was used as the
classification model, which is one of the most widely used
machine supervised learning methods in practice (Artero 2009).
The method is based on the decision tree construction, from the
training database to obtaining the production rules (domain
knowledge). For the construction of the decision tree, algorithm
J48, which is one of the most known and used algorithms for
constructing decision trees, was used (Artero 2009). To evaluate
the classification model (decision tree), the Confusion Matrix
and
Kappa Statistics (κ) were used, as recommended by Resende

et al. (2005). Furthermore, it was considered that the classifica -
tion model would be adequate if it presented Kappa Statistics
(κ)
values equal or above κ = 0.41 (moderate agreement), according
to Landis and Koch (1977). Otherwise, adjustments in the
classi-
fication model would be necessary.
2.4 Knowledge coding
For knowledge coding, a Pentium 2.13GHz microcomputer
was used, with 4GB of RAM in the Windows operating system
using CLIPS (C Language Integrated Production System) shell,
version 6.3. In this case, it was adopted as a robust and
efficient shell for ES development, one that: (1) presented
the ability to resolve conflicts between rules, (2) operated
satisfactorily with the forward chain, (3) was a free access
shell and (4) presented good answers (accuracy). This robust
shell definition considered the existence of conflicting opi -
nions among the experts consulted, the proposition of the
Modus Ponens type ‘if (condition) – then (action)’ as an appro-
priate form of inference, and the economic limitation for
commercial shell acquisition. Thus, the CLIPS shell can be
evaluated as robust to the problem in focus agreeing with
the works of Riley et al. (1987); Mettrey (1991), and Kuesten
and McLellan (1994).
2.5 Evaluation and adequacy
According to Giarratano and Riley (2004), at this stage, the
expert should evaluate and criticize the UWC-ES, passing on
this information to the ES developer, who in turn performs the
adjustments and again returns the ES to the expert for re-
evaluation. This process is iterative until the expert judges that
UWC-ES is adequate. Considering the characteristics of the

problem and the studies carried out by Spring (1997) and
Collier, Leech, and Clark (1999), the Turing test (a cl assic test
that aims to verify if a machine has the intelligence matching
that of a human). To implement the test, the methodologies
used by Spring (1997); Collier, Leech, and Clark (1999), and
Artero (2009) were adjusted.
The Turing test is based on forming three groups of differ-
ent experts, indicated here by G-1, G-2 and G-3. The test
basically consists of collecting a set of ‘m’ test cases,
previously
solved by experts from the G-1 group, solving these cases by
developed ES (G-2), carrying out the specific evaluation of
both solutions, S (G-1) and S (G-2) by other experts (G-3). In
the specific evaluation, two outputs were requested from the
G-3 group; the first output refers to the quality evaluation of
the G-1 and G-2 solutions, according to a scale ranging from 1
to 7 (1 = very bad, 4 = reasonable, 7 = very good). In the
second output, the identification of the solutions from the ES
was requested. If G-3 assigns a value greater than or equal to 4
to the quality of solutions presented by G-2 and cannot deter-
mine (with a minimum of 50% accuracy) which one of the two
(G-1 or G-2) is the group of experts, it is said that the machine
has passed the Turing test and therefore can simulate human
intelligence. In this case, the end of the UWC-ES development
is observed, and the ES is considered suitable to select the
Figure 1. Obtaining the knowledge domain.
URBAN WATER JOURNAL 563
best solutions for the UWC problem. Otherwise, adjustments
must be made in the UWC-ES.

2.6 Application of the expert system
The purpose of the application cases was to help evaluate
the results of the developed ES model. Considering the
prospect of possible water supply problems in the Federal
District, as mentioned by Conejo et al. (2009), some of the
Administrative Regions (AR) of the Federal District were
adopted as case studies. These AR included Brasília, Lago
Norte, Cruzeiro, Guará, Varjão, Estrutural and Park Way.
These AR were chosen according to the importance of
studying urban environments with different economic levels.
For ES application, secondary information was used, based
on data from Silva (2012). In addition, the Federal District
Government was considered as the decision-maker in the
case, with its respective competent institutions (Brazilian
Federal District’s Regulatory Agency for Water, Energy and
Sanitation – ADASA, Brazilian Federal District’s Water
Supplier and Sanitation Company – CAESB, Brazilian
Institute of Environment and Water Resources – IBRAM and
Secretary of State for the Environment – SEMA).
3. Results
Based on following the formal plan and setting the predefined
tasks for UWC-ES development (expert system planning stage),
responses about its viability were obtained. The result of the
feasibility assessment task, the verification of the factors and
returns suggested by Giarratano and Riley (2004), led to the
return of the viability response of the ES approach. The rea-
sons that led to this response refer to the fact that most of the
returns (factors 1, 3, 4, 5 and 6) showed a favorable return to
ES development, as shown in Table 1.
For the resource management task, the result indicated
that the available resources are comparable to the resources
used to develop other ES with equivalent functions, according

to the literature review (Cheng, Yang, and Chan 2003; Chau,
Chuntian, and Li 2002; León et al. 2000). Based on the pre-
liminary functional layout task, it was found that the proposed
ES must ensure compliance of the purpose of pointing out
priority projects for handling UWC. From the knowledge expli-
citness stage, the population universe (possible combinations,
which make a total of N = 3125) and the identification (Id.) of
the sample units (typologies, totalizing n = 313) were identi -
fied to be studied.
As a result of the task of identifyi ng projects for handling
UWC,
Table 2 shows a summary list obtained from a literature review.
As the problem was modeled to obtain five priority projects
(PP) for handling UWC from the experts, five classification
models (decision trees) were found, one for each priority
estimate (PP1, PP2, …, PP5). Part of the classification model
(decision tree) and respective production rules (domain knowl -
edge) obtained for PP1 are presented in Figure 3(a,b).
In total, 409 production rules were obtained that make up
the domain knowledge. Additional information on these clas-
sification models (decision tree) and production rules was
presented in Silva (2012).
As a result of the evaluation of the classification model
(decision tree), the Confusion Matrix and the Kappa Statistics
(κ) were obtained. The Confusion Matrix is shown in Figure
2(c).
The Confusion Matrix provides an effective measure of fit for
the classification model by showing the number of correct
classifications versus the number of classifications predicted
for each class, concerning a training database. Thus, the correct
classification of the model (coincidence of the response pre -
sented by the expert, shown in the lines, and the response

presented by the classification model, presented in the col -
umns) is given by the diagonal elements of the Confusion
Matrix. The total number of training data correctly classified
by the classification model for PP1 is given by the sum of the
elements in the diagonal of the Confusion Matrix, and all others
were incorrectly classified. Therefore, a reasonable fit of the
classification model (decision tree) of PP1 was observed in
Figure 2(c). Moreover, it should be mentioned that the other
classification models presented slightly better results.
For the average Kappa Statistics (κ), whose individual
values for each classification model (decision tree) are
κPP1 = 0.41, κPP2 = 0.49, κPP3 = 0.54, κPP4 = 0.49 and
κPP5 = 0.45, an average value of κ = 0.48 was found, consid-
ered adequate according to the adopted methodology. This
value indicates that the classification showed a moderate
agreement. The classification model presented a moderate
adjustment and, according to Landis and Koch (1977), can
represent, with moderate precision, the training data.
Table 1. Factors and returns considered in the ES viability
assessment.
Item Factora Returnb Evaluationc
1 Can the problem be solved efficiently by
conventional programming?
No No
2 Is the problem’s domain well defined? Yes No
3 Is there a need and interest for an ES? Yes Yes
4 Are there human experts willing to cooperate? Yes Yes
5 Can the experts pass on their knowledge? Yes Yes
6 Does the solution of the problem mainly involve

heuristics and uncertainty?
Yes Yes
Notes: a) Factors suggested by Giarratano and Riley (2004), b)
expected return
for the ES approach to be viable, c) return found after
feasibility assessment.
Table 2. Summary list of projects for handling UWC.
P Projects for handling UWC
P1 Loss reduction (S)
P2 Macro and micro-mediation implementation (S)
P3 Implementation of individualized measurement (S)
P4 Implementation of efficient bathrooms (S)
P5 Reduction in pressure in the hydraulic system in bathrooms
(S)
P6 Reduction in pressure in the water distribution network (S)
P7 Rainwater collection and use (S)
P8 Greywater collection, treatment and use (S)
P9 Setting up environmental education programs (NS)
P10 Application of fiscal stimuli for consumption reduction
(NS)
P11 Tax on inefficiency in water use (NS)
P12 Adjustment of tariff policy (NS)
P13 Regulation of the water consumption of household
appliances/savers (NS)
P14 Increase in production capacity (S)
P15 Intermittence/rationing in the supply system (S)
P16 Regulation of consumption (NS)
P17 Creating green roofs (S)
P18 Strengthening water supply operator (NS)
P19 Using good practices for water conservation (NS)
P20 Privatization/concession of the water supply services
operator (NS)

Note: (S) is structural measures and (NS) is non-structural
measures.
564 W. T. P. D. SILVA AND M. A. A. D. SOUZA
The knowledge coding step occurred satisfactorily. The tool
used was considered adequate as the production rules and
conflict resolution strategies were easy to implement. Figure 3
shows the CLIPS development environment and part of the
elaborated coding.
The results of the UWC-ES evaluation and adequacy stage
indicated that the first group, the G-1 group, was formed by
the 13 experts who effectively contributed to forming the
training database (domain knowledge). The second group
(G-2) was formed by the answers given by the ES, i.e. it refers
to the UWC-ES. Furthermore, the third group was the G-3,
formed by three experts who did not participate in obtaining
domain knowledge. The first output, given by the G-3, indi-
cated an average value of 4 for the quality of the solutions
presented by UWC-ES, on a scale ranging from 1 to 7 (1 = very
bad, 4 = reasonable, 7 = very good). When analyzing the
quality of the solutions presented by the G-1 human experts,
which was also 4, a similarity can be observed between G-1
and G-2. This also shows a reasonable divergence between the
opinions of the human experts of the G-1 group and the G-3
group. These divergences are also conveyed in the responses
given by the ES. It was observed that cases with similar
characteristics receive different solutions, depending predomi-
nantly on the training, experience and professional experience
of the expert who analysed the case. This fact requires careful
use of the results of the developed ES (UWC-ES) and proves
the complexity of the studied problem. Similar problems were

reported by Giarratano and Riley (2004) because even among
the experts there is no consensus.
The second result indicated that the G-3 was unable to
determine, with 67% accuracy, which of the two (G-1 or G-2),
is the group of human specialists, therefore UWC-ES was
approved by the Turing test. In other words, it can be con-
cluded that the UWC-ES is able to select the best solutions to
the problem of handling UWC.
The main results found for the case studies chosen, after
using the UWC-MODEL, are presented in Table 3. These were
the results used to feed the UWC-ES.
According to the UWC-MODEL, the environmental level
(j = 4) was the one that presented the greatest contribution
to the intensification of the studied UWC. For the second and
third level of greatest contribution, the urban dimension (j = 3)
and managerial dimension (j = 2) were found, respectively.
This suggests that the PPs selected by the UWC-ES for solving
the UWC case studies are targeted at reducing the contribu-
tion or collaboration, of the environmental, urban and man-
agerial levels. The results obtained for the case studies, after
using the ES (input of the results of the UWC-MODEL in the
UWC-ES) are presented in Table 3.
In summary, eight PPs were suggested for the solution of the
studied case of UWC, which are the following: loss reduction
(P1),
implementation of individualized measurement (P3), rainwater
collection and use (P7), greywater collection, treatment and use
(P8), application of fiscal stimulus for consumption reduction
Figure 2. (a) Part of the classification model (decision trees);
(b) production rules; (c) confusion matrix of PP1.

URBAN WATER JOURNAL 565
(P10), consumption regulation (P16), strengthening water
supply
operator (P18) and use of good practices for water conservation
(P19). When analyzing the results presented by the UWC-ES,
some
problems can be observed, such as the recommendation of the
guideline ‘implementation of individualized measurement’ (P3)
for
the Estrutural and Varjão AR, whose predominant housing
typol-
ogy is isolated single-family residences that already have
indivi-
dualized measurement; the non-recommendation of the
individualized measurement (P3) for regions (in the case of the
Brasília and Cruzeiro AR) in which there is predominance of
apart-
ment housing without individualized measurement, and the
recommendation to strengthen the water supply operator (P18)
to a well-structured company (CAESB). These problems suggest
the need for making adaptations to the UWC-ES since the model
responded reasonably to these cases. In contrast, the indication
of
the PP for rainwater collection and use (P7), greywater
collection,
treatment and use (P8), regulation of consumption (P16) and
using good practices for water conservation (P19) can be
consid-
ered appropriate for the case studies, as they try to solve the
UWC
problem by addressing its cause (j = 4, greater influence of the
environmental level). Thus, it can be considered that the devel -

oped ES presented acceptable results, in agreement with pre-
viously presented adjustment indicators.
4. Conclusions
A computational tool was developed to help select a set of
priority projects (PP) to solve the UWC problem. This tool was
called UWC-ES. The tool (UWC-ES) can replace human and
financial resources for decision making in UWC. Therefore, it
is especially suitable for urban environments where limitations
of human and financial resources are important.
The results of the UWC-ES indicated acceptable applicabil-
ity and the possibility of using it in other cases. The use of
UWC-ES is based on analyzing various pieces of information
about the urban environment by an (artificial) UWC expert,
which reasonably reproduces the knowledge of several human
experts. Thus, the resources required to use the UWC-ES con-
sist of efforts to obtain these various pieces of information
and, of course, without the experts’ full participation.
Although the characteristics of the problem are appropriate
to the approach of the expert system, some obstacles were
encountered during the development of the UWC-ES sub-
model, including the following: (1) the difficulty of finding
specialists willing to collaborate, (2) the existence of diver -
gence between the opinions of specialists and (3) the exis-
tence of problems in inference, mainly due to the existence of
divergence between the opinions of the specialists. Thus, new
studies are suggested focusing on changes in methodology in
order to minimize the divergence of expert opinions. One
possible modification, for example, could be the aggregation
of responses from experts with similar academic backgrounds
and the assignment of weights to each specialty class.

Acknowledgements
The authors would like to express their gratitude for the
financial sup-
port from the Brazilian agencies CNPq (Project Nº
556084/2009-8) and
CAPES. The authors would like to express their gratitude to the
following
institutions: Brazilian Federal District’s Regulatory Agency for
Water,
Energy and Sanitation (ADASA), Water National Agency of the
Brazil
(ANA), Brazilian Federal District’s Water Supplier and
Sanitation
Company (CAESB) and Brazilian Federal District’s Planning
Company
(CODEPLAN).
Disclosure statement
No potential conflict of interest was reported by the authors.
Figure 3. CLIPS development environment and part of the
developed ES coding.
Table 3. UWC-MODEL and UWC-ES results for the case
studies.
Classification of cases according
to the UWC-MODEL
UWC-ES results for the
case studies
Case Study j = 1 j = 2 j = 3 j = 4 j = 5 PP1 PP2 PP3 PP4 PP5

Evaluating planning strategies for prioritizing projects in su

Evaluating planning strategies for prioritizing projects in su

Recommended

Recommended

More Related Content

Similar to Evaluating planning strategies for prioritizing projects in su

Similar to Evaluating planning strategies for prioritizing projects in su (20)

More from BetseyCalderon89

More from BetseyCalderon89 (20)

Recently uploaded

Recently uploaded (20)

Evaluating planning strategies for prioritizing projects in su