A Decision Support System for Sustainable Energy Supply Combining
Multi-Objective and Multi-Attribute Analysis: An Australian Case Study
Mattiussi Alessandro, Rosano Michele, Simeoni Patrizia
PII:
DOI:
Reference:

S0167-9236(13)00226-1
doi: 10.1016/j.dss.2013.08.013
DECSUP 12395

To appear in:

Decision Support Systems

Received date:
Revised date:
Accepted date:

8 March 2013
23 August 2013
23 August 2013

Please cite this article as: Mattiussi Alessandro, Rosano Michele, Simeoni Patrizia, A
Decision Support System for Sustainable Energy Supply Combining Multi-Objective and
Multi-Attribute Analysis: An Australian Case Study, Decision Support Systems (2013),
doi: 10.1016/j.dss.2013.08.013

This is a PDF le of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its nal form. Please note that during the production process
errors may be discovered which could aect the content, and all legal disclaimers that
apply to the journal pertain.

ACCEPTED MANUSCRIPT
A Decision Support System for Sustainable Energy Supply Combining Multi-Objective and MultiAttribute Analysis: An Australian Case Study.

Dipartimento di Ingegneria Elettrica, Gestionale e Meccanica, University of Udine, via Delle

IP

Mattiussi Alessandroa, Rosano Micheleb, Simeoni Patriziaa

SC
R

Scienze 208, Udine, Italy.

Sustainable Engineering Group, Curtin University, Sarich Road, Bentley, Perth, Western

MA

NU

Australia.

ABSTRACT

A framework for an energy supply decision support system (DSS) for sustainable plant design and

TE

production is presented in this paper, utilising an innovative use of multi-objective and multi-attribute

CE
P

decision-making (MODM, MADM) modelling together with impact assessment (IA) of the emission
outputs. The mathematical model has been applied within an eco-industrial park (EIP) setting and includes
three steps. First, an assessment of the total EIP emissions inventory and impacts is conducted; the second

AC

step, focusing on the sustainability benefits of combined heating and power (CHP) plants and photovoltaic
technologies, developed a multi-objective mathematical model including both economic and environmental
objectives in a Pareto-frontier optimisation analysis. Four different scenarios involving combinations of CHP
plants (internal combustion engine, gas turbine, micro-turbines and fuel cells) and two types of PV plant
(monocrystalline and polycrystalline) were evaluated. The third step utilises a MADM methodology - the
analytic hierarchy process (AHP) - for selecting the best alternative among the Pareto-frontier efficient
solutions. This model has been applied to a case study of an EIP located in Perth (Kwinana Industrial AreaKIA), Western Australia.
Keywords Eco-Industrial Park; MCDA; AHP; CHP

1. Introduction

ACCEPTED MANUSCRIPT
Day-to-day decision making requires both objective and subjective perspectives, utilising the former for
rational, constrained modelling and the latter for adapting specific problem issues to the decision-making
process. The combination of both formal and informal information in the decision-making process is the

IP

main focus of this paper, referring to typical multi-criteria issues such as energy production. A decision
support system (DSS) is defined as a software-based tool assisting in the decision-making process by

SC
R

interacting with both internal/external users and databases while utilising standardised or specific algorithms
for problem solving [5].

NU

Power, D. [21] identified four main types of DSSs, depending on the main drivers guiding the decisional
process:

Model-driven DSSs: such DSSs require a limited amount of data because of the intrinsic

MA

composition of the system, used to evaluate quantitative data in a tailor-made structure that can be

adapted to other external requirements. Initially developed for financial planning, this category of

TE

DSS was later used for multi-criteria decision making and spatially driven decisions such as logistics
or distribution modelling;

Data-driven DSSs: the database structure behind the DSS is emphasised, and the operations of data-

CE
P

warehousing and manipulation are the most relevant for such DSSs. Online meaning interactive

AC

(such as the OLAP) and offline applications can be found, and web-based data-driven DSSs
currently represent the natural evolutions of such models;
-

Communication-driven DSSs are used for exploiting the network and communicating capabilities of
the system, which includes the use of groupware, conferencing or other computer-based
communications. This category is directly related to group DSSs, developed to promote a
participatory approach to the decision process, and their relation with model-driven DSSs has been
studied, aiming to include the shared approach of the former with the structured modelling of the
latter.

Document-driven DSSs, also called text-oriented DSSs, are used for document retrieval, especially
in large groups/organisations, to support the decision-making process. The advent of a Web-based
system increased the possibility of such DSSs, allowing rapid access of documents distributed in
worldwide databases;

ACCEPTED MANUSCRIPT
-

Knowledge-driven DSSs: these are specific, tailor-made systems used in a particular domain and
developed for a particular person or group of people. Power [21] acknowledged the relationship with
Artificial Intelligence systems, in which the DSS follows a series of rules to evaluate and eventually

IP

make decisions on the problem to be analysed.

SC
R

Arnott & Pervan [2] reported a framework for DSS classification and sub-classification, identifying personal
DSSs, group support systems, executive information systems, intelligent DSSs and knowledge-managementbased DSSs. Each of such DSSs presents sub-branches depending on their specific features and temporal

NU

evolution. In particular, model-driven DSS represents the focus of this study. The modellisation stage,
focusing on multi-criteria modelling, will be investigated in the following paragraphs.

MA

Multi-criteria, multi-attribute and multi-objective analysis while similar in their ultimate purpose of
assisting with the final decision-making process differ in their defining concepts. Multi-criteria decision

making (MCDM) deals with a general class of problems that involves multiple attributes, objectives and

TE

goals [32]. Although MCDM represents the major class in decision-making support systems, multi-attribute
(MA) and multi-objective decisionmaking (MODM) represent its subclasses, [20] relate to more specific

CE
P

approaches in the decision-making model. Life cycle assessment (LCA) and impact assessment (IA) are tools
used in industrial ecology [3] to quantify and evaluate the emissions (air, water and soil) from various parts

AC

of a production process and then evaluate their impacts on different elements of the ecological system (e.g.,
human health, ecosystem damage and resources) depending on the IA methodology chosen. Optimisation
with multiple conflicting objectives has no single best solution, but a set of solutions, named the Pareto-set
for Villfred Pareto (1848-1923), who first studied them, which can be applied to social science, economy and
game theory. Multi-objective optimisation techniques therefore identify a set of non-dominated solutions
which represent the optimums for a given problem. The concept of domination can be illustrated as follows:
an alternative a is non-dominated by b if a is better than b for at least one objective while not being worse
than b for all of them. Identifying the Pareto-frontier means also satisfying the following requisites for the
solutions identified while minimising the total elaboration time, as reported in [1]:

Spread: To find a set of solutions that capture the whole spectrum of the true Pareto front;

Accuracy: To find a set of solutions as close to the real Pareto front as possible;

ACCEPTED MANUSCRIPT

Diversity: To find a set of solutions as diverse as possible.

Weise [30] provided a broad taxonomy of evolutionary algorithms, defined as population based

metaheuristic optimization algorithms that use biology-inspired mechanisms like mutation, crossover,

IP

natural selection and survival of the fittest in order to refine a set of solution candidates iteratively. First,

SC
R

metaheuristics is defined as a method for solving general problems, combining objective functions in an
abstract way, treating problems as a black box.

NU

According to [30], the five main stages of evolutionary algorithms involve the following:
Initial population, which allows the initial sample for analysis to be created from the possible set of

MA

candidate solutions;

Design evaluation, which computes the objective value from the candidate solution;

Fitness assignment, which, depending on the objective, determines the fitness of the candidate

TE

solution relative to a fitness criterion (weighed sum of objective values, Pareto ranking, etc.) which
evaluates the suitability of the candidates to the optimisation required;
Selection: based on the fitness of the candidate solution, at this stage the population (the group of

CE
P

candidate solutions) to be maintained is selected, while the remaining solutions are discarded.
Reproduction: selected candidate solutions are reproduced by different mechanisms such as partial

AC

mutation, crossovers, or complete change.

As a class of MO techniques, the family of evolutionary algorithms includes, among others, evolution
strategies (ES), genetic algorithms (GA), genetic programming (GP) and learning classifier systems (LCS)
[30]. Among GA techniques, the non-dominated sorting algorithm (NSGA) represents a increasingly used
method for the design stage. NSGA and its variant NSGA-II, first developed by Srinivas and Deb [25], are
population-based metaheuristics encompassing seven steps for design optimisation [30], i.e., population
initialisation, non-dominated sorting, crowding distance, selection, genetic operators, recombination and
selection. Having defined the initial population based on problem constraints or user design of experiments
(DOE), sorting is performed by assigning a priority value (rank) to non-dominated designs, selecting
designs for further explorations based on rank and crowding distance, i.e., higher fitness is assigned to

ACCEPTED MANUSCRIPT
individuals located on a sparsely populated part of the front [16]. Genetic operators, mainly recombination,
crossover and mutation, are used for exploring the design space, which is then selected, maintaining a
range of best-performing designs (elitism) for the next fitness assessment, until the last generation of

IP

designs is assessed or the end criterion is reached.

SC
R

To overcome the shortenings of lateral diversity in Pareto front determination of NSGA-II, Jeyadevi et al.
[12] developed a modified NSGA (MNSGA-II) including a controlled version of elitism for improving the
exploration stage and the lateral distribution of the Pareto Front, used in reactive power dispatch modelling.

NU

Guo et al. [10] used a modified version of NSGA-II to solve scheduling issues in production planning,
relating scheduler utilisation to a production process simulator. Panda [19] used NSGA-II for electrical noise

MA

reduction in controller designs.

Yusoff et al. [31] reviewed the application of NSGA-II in machining design, concluding that such an

algorithm represents a reliable and popular tool in MO machining set up, allowing the inclusion of multiple

TE

performances and variables. There are numerous published applications of MCDM in plant design. Multi-

CE
P

objective (MO) analysis has been broadly used in designing product components, but limited research has
considered the environmental impacts of the process. Vince et al. [29] assessed the design installation of a
Reverse Osmosis plant for desalinated water production, including both economic and environmental

AC

criteria. However, in this analysis, the environmental impact was limited to a quantitative environmental
assessment of water discharges, considering electricity production and the water recovery rate as the
environmental criteria. Mirzaesmaeeli et al. [15] treated environmental emissions as a constraint in a mixed
integer linear programming (MILP) optimisation of an Canadian power producer, while the optimisation
model proposed in [22] included environmental emissions considered as externalities, i.e., those externally
generated but unaccounted for in the costs. Harkin et al. [11] used MO optimisation to design CO2 capture
systems retrofitted in coal power stations. They took into account the percentage of CO2 captured
(maximised) and the energy input to the process (minimised), evaluating results as a function of the input
parameters. Guilln-Goslbez [9] applied MO optimisation, discussing its validity when assessing multiple
objectives such as environmental outcomes, and introduced a mixed MILP-MO model, which they then
applied to heat exchanger designs and petro-chemical supply chains. Environmental impacts within LCA

ACCEPTED MANUSCRIPT
typically include acidification, eutrophication, global warming and eco-toxicity. Bernier, Marchal, &
Samson [4] used both thermo-economic and environmental objectives for a carbon dioxide capture plant
design, integrating LCA (in terms of global warming potential) into the optimisation model. Applied

IP

evolutionary algorithms were used in [7] for power plant capacity estimation, considering only technical
(maximise exergy efficiency) and economic (minimising total costs) criteria in identifying Pareto-optimal

SC
R

solutions.

NU

Most MCDM methodologies provide a unique utilisation of MO or MA analysis. This paper aims to

MA

establish a framework for including both MO and MA decision-making modelling and introduces a general
methodology for DSS in sustainable energy plant design ( Section 2.x). The proposed framework has been
developed to assess both individual companies and EIPs, in which the summations of each individual

companys emissions can be aggregated to provide an emissions figure for the entire EIP. Such a framework

TE

has then been applied to a specific case study in the Kwinana Industrial Area (KIA, Perth, Western

CE
P

Australia): the three specific stages of data assessment ( 3.1) are related to the whole industrial park, and the
two stages of energy alternatives identification ( 3.2) and choice ( 3.3), refer to a specific case study of a
company in the industrial area.

AC

2. Objective and methodology

The major objective of this work is to develop a DSS aimed at helping the decision maker in identifying and
choosing the better energy generation options among a range of many feasible solutions. The starting point
for building this model was initially identified by Simon [25], as shown in Figure 1. It consisted of a general
DSS structure and three major decision-making steps: a) problem classification/definition (intelligence
phase), b) alternative generation/evaluation (design phase) and c) alternatives negotiation/selection and
action determination (choice phase). The proposed methodology uses the tools of LCA, MODM and
MADM in each of these stages of the DSS.

NU

SC
R

IP

ACCEPTED MANUSCRIPT

In the proposed methodology, attributes are used in ex-post decision making i.e., referring to a known

MA

situation while objectives are antecedent to the decision-making process. In other words, attributes are used
to analyse a goal towards which the objectives have been chosen. Considering the three-step decision-

making approach previously described in Figure 1, the following can be stated:

MODM provides support for the decision maker in identifying a range of alternatives;

Multi-attribute decision making (MADM) supports the decision maker in the latter stages (the

TE

CE
P

Choice Phase) of selecting from a range of feasible alternatives.

AC

The three stages of the research methodology are noted in Figure 1 and are discussed below.

2.1 Stage I: Impact Assessment of an Industrial Areas Emissions

The data-intensive stage of life cycle inventory (LCI) was used for problem identification, focusing on an
assessment of emissions from three main contributors to air, water and soil impacts in the industrial park.. A
two-step approach has been used. The first step ranks the companies in the industrial park for macrotoxicants toxicants whose main impacts are associated with the volumes emitted such as NOx, SOx, PMs
and NMVOCs. The LCI is then followed by an IA (LCIA) to evaluate the micro-toxicant impacts i.e.,
toxicants whose impact is due to the quality of the toxicant itself. The LCIA methodology chosen for the
study was Eco-Indicator 99, a damage-oriented method for LCIA [8] that takes into account human health,
eco-system and land use impacts. Two main impacts have been considered for the following LCIA stage,

ACCEPTED MANUSCRIPT
namely Human Health and Eco-system Damage. In Eco-Indicator 99, human health impacts are measured
in Disability Adjusted Life Years (DALY), a variable that measures the number of life years lost due to
the inhalation/ingestion of pollutants p via different sources (e.g., air, food, water); eco-system impacts, are

IP

measured by the Potentially Disappeared Fraction (PDF) of animal and natural species per square meter per
day. Human health impacts (HHIs) and ecosystem impacts (ESIs) Impacts have been calculated according to

SC
R

eq. 1 and eq. 2:

MA

NU

(Eq. 1)

(Eq. 2)

thus summing up the emissions of each pollutant p to each compartment c (IEp,c), multiplied for its impact

factor (HHIp,c, ESIp,c) on the two considered impact categories of human health (HHIp,c) and ecosystem

TE

damage (ESIp,c).

CE
P

2.2. Stage II: MODM for Combined Heat and Power Plant Design
Having assessed the outputs and impacts of the EIP, the focus is then shifted to single companies inside the
EIP. An MO model has been developed for the design of a gas-fired CHP plant integrated with PV

AC

(photovoltaic) solutions. The model accounts for both economic and environmental impacts. The Economic
assessment indicator is a Net Present Value (NPV) cash flow assessment, while the environmental analysis
considered local impacting pollutants such as NOx, CO, SO2, NMVOCs and PMs. The optimisation process
relies on a number of factors including the following:
There is a discrete choice among plants. The four different types of CHP technologies being assessed
(internal combustion engines, turbines, micro-turbines and fuel cells) have been selected from the
most relevant manufacturers worldwide, with technical performance (e.g., power conversion
efficiency) data taken from manufacturers catalogues and implemented in a Microsoft Excel
spreadsheet. Similarly, data were collected from PV manufacturers. Environmental performance
measures including emission rates have been taken from [27] and [28];

ACCEPTED MANUSCRIPT
Life cycle assessment has been used for assessing the environmental impact of the life cycle
emissions, and the impact area considered is an algebraic sum of the emission outputs from the CHP
plants and the emissions avoided from the heating and power produced. Human health impacts, as

IP

defined in the previous section, have also been considered to evaluate and quantify the damage
produced by each pollutant, resulting in the calculation of the human health impact reduction

SC
R

(HHIR) coefficient considered in the MO assessment;

Technical constraints have been used to avoid excessive waste heat while still satisfying at least 70%

NU

of companys power needs on a monthly basis;

The economic analysis relies on a traditional cash flow analysis but also considers the current trends

MA

in power and fossil fuel prices in all revenues and costs, resulting in both revenue and cost

adjustments during the medium-term investment.

TE

The two MO objectives are to maximise the Net Present Value (NPV) and the HHIR of the combined
solutions (CHP + PV). The simulations combined a Microsoft Excel worksheet for flexible modelling and

CE
P

simulation and ModeFrontier software (developer: Esteco, Trieste, Italy) for advanced scheduling and
optimisation; the programs were then run in an iterative process to identify the Pareto-frontier, i.e., the group

AC

of non-dominated solutions, defined as the group of alternatives which cannot be improved in one of the two
objectives without reducing the other. The design space of the optimisation model the total number of
candidates solutions to be assessed consisted of 256 CHP plants, 48 PV module types, four variables (0.7
to 1, step 0.1) assessing the percentage of heat and four parallel variables for the percentage of power to be
covered by the plant, four variables assessing the number n of similar CHP plants to be considered (1 to 4,
step 1) and 11 variables (0 to 1, step 0.1) assessing the percentage of total available roof area covered by the
PV plants.
The procedure for the optimisation process is summarised in Figure 2.

ACCEPTED MANUSCRIPT
OBJECTIVE

TOOLS

DONE BY

Define the initial generation

of candidate solutions to be
assessed

DOE Algorithms

Output Calculation

Calculate the results from

the single simulation of
each generation of
candidate solutions

Mathematical Model

Scheduling

Variates the candidates

solutions depending on
simulation results

Scheduling Algorithm

Pareto Front

Identifies the nondominated solutions

Mode Frontier/
User

DOE Initial Set

of Variables

IP

STAGES

NU

SC
R

Microsoft Excel

Mode Frontier

MA

Optimizer

Mode Frontier

The identification of the initial set of candidate solutions, DOE, i.e., the combination of variables to be

TE

simulated by Microsoft Excel, represents a major stage in the whole optimisation process. Among the

CE
P

advantages of a consistent DOE, the rapidity of the optimisation process for rapidly converging to optimal
solutions is most likely the most relevant.

AC

The design space is defined by the number of variables multiplied by the number of variable levels (i.e., the
allowed values for each variable). Given a problem with five variables, each with three possible levels, the
maximum number of design combinations is 53 = 125 designs. Obviously, increasing the number of variables
and levels increases the complexity of the problem. Over the hypercube defined as the polyhedron with a
number of dimensions equal to the number of variables (e.g., three variables with equal level size are
represented by a cube), the initial design space can be selected in various ways.
The software used for the optimisation process (Esteco Mode Frontier 7.0) allows the following DOE
algorithms: user-defined sequences, random DOE, Sobol sequences, constraint satisfaction problem (CSP)
models, Latin HyperCube/Montecarlo, full factorial DOE, reduced factorial DOE, cubic face centred DOE
and Box-Behnken DOE.

ACCEPTED MANUSCRIPT
Having fixed the initial set of solutions for which the output values have been assessed, the optimiser has to
decide how to move from one set of candidates to the next, to identify best-performing candidates. This

process, known as scheduling, depends on many variables, such as the following:

Variable type: continuous vs. discrete;

Search operators: mutation/selection/crossover;

Relation among generations (keeping best solutions, i.e., elitism, or completely changing the

SC
R

generation set);
-

IP

NU

Computing capabilities: varying the number of contemporary assessed variables.

MA

2.2.1 The mathematical model

The mathematical optimisation model considers the following objective functions:

Max(NPV) =y (ACF)y - Io

(eq. 3)
(eq. 4)

TE

Max(HHIR) = AEI - CEI

CE
P

NPV, as expressed in equation 5, represents the sum of the cash flows (ACF) actualised and considered
throughout the whole duration (y) of the initial investment (Io) on a monthly basis (m) for each of the
technologies assessed (t). ACF is equal to the difference between the actualised revenues (TAR) and costs

AC

(TAC), in turn depending on the revenues from selling excess power (Rexc), the avoided purchasing of
electricity and heat (Rav), incentives (I) and associated variable costs such as fuel consumption (Cf),
maintenance (Cm), heat and power integration (Cint) and excessive heat disposal(Cexc).

15

15

y 0

y 0

NPV I o ACF y TAR y TAC y R el R el R h I

C C C C C
y

int
el

int
h

exc
h

exc

av

av

y , m ,t

(eq. 5)

y , m ,t

The HHIR, as reported in equation 6, has been calculated as the sum of the impacts from the avoided
emissions from electricity generation (AEIe, positive), heat (AEIh positive) production and current emissions
(CEI, negative) from the simulated design.

ACCEPTED MANUSCRIPT
HHIR AEI e AEI h CEI
5

p 1

p 1

EL ER pel IFp Eh ER ph IFp FCt ERt , p IFp

(eq. 6)

p 1 t 1

IP

AEIe is calculated by multiplying the total electricity production (EL) by the emission rates of each pollutant
p from the national power system and the relative specific impact factor (IF) of each pollutant in terms of

SC
R

specific human health impact; the latter is expressed in DALY/kg, using the LCIA Eco-Indicator
methodology described in 2.1 . Emission Rates (ER) for each pollutant p were used to calculate the

NU

avoided power production and associated emissions; these rates were taken from the IPPC/Corinair Emission
Factor database, taking into consideration the energy mix of the specific country. Similarly, AEIh depends on

MA

the heat saved by the company (Eh) multiplied by the pollutants emitted by traditional natural-gas boilers
again from the Corinair database and the pollutant-related impact factor of LCIA Eco-Indicator

methodology. Eventually the CEI considered plant-specific emission rates, expressed in kg per input MJ, fuel

TE

consumption (FC) of the simulated design considering each CHP technology t and the relevant impact factor.

CE
P

Each simulation has been subjected to constraints to promote the following:

Correct plant sizing (avoiding excessive heat waste), considering the LT index (eq. 7), stating that at
least one third of the thermal output must be exploited;

AC

LT = Eth / (Eth + Ee) > 0.33 (eq. 7)

Satisfy a percentage (PC) of the heating and power needs of the company c on a monthly basis m;
Eeem,t > PCeec,m Eeec,m

(eq. 8)

Ehm,t > PChc,m Ehc,m

(eq. 9)

Respect site-specific limits: the occupied area of PV plants (SPV) has to be inferior to the available
roof area (ARA) for PV plants multiplied by the percentage of covered area (PCPV).
SPV < PCPVARA (eq. 10)

2.3 Stage III: MADM

ACCEPTED MANUSCRIPT
Having identified the technical feasible solutions from both an economic and environmental perspective, the
last step of the methodology involves the development of a decision-making model to assist with the final
choice among the identified solution sets. AHP methodology (Saaty, 1980) has been selected for this

IP

purpose. AHP has been widely used for MADM, particularly for correlated sustainability issues, as reported
by [6], [13] and [17]. The method allows the inclusion of both quantitative and qualitative criteria in the

SC
R

assessment and involves three stages: problem structuring: pairwise comparison, aggregation and results, as

NU

outlined in Figure 3.

Criteria n

Criteria 2

Criterias weights (to

the main goal) are
calculated by
pairwise comparison

Alternative 2

Alternative 3

Alternative m

Alternatives weights
(to each criteria) are
calculated by
pairwise comparison

AC

CE
P

Alternative 1

TE

Criteria 1

MA

Goal

3. The Case Study: Kwinana Industrial Area

The Kwinana Industrial Area (KIA) is approximately 40 km from the city of Perth (Western Australia) and is
characterised by a significant presence of heavy industries such as refineries (crude oil, alumina, titanium
dioxide, nickel) and chemical production, together with regional utility plants for power production and
water treatment. Over 50 companies are present in the Kwinana Industrial area (KIA). The KIA was assessed
during the period January to June 2011, by directly visiting the companies, interviewing personnel and
collecting data.

3.1 Life Cycle Impact and Life Cycle Impact Assessment

ACCEPTED MANUSCRIPT
LCI was first performed using data from the Australian National Pollutant Inventory (NPI) database, using a
1% cut-off percentage therefore considering only relevant emitters in the Area (C1 to C27) and allowing

CE
P

TE

MA

NU

SC
R

IP

the ranking of companies regarding their NOx, PM, SOx and NMVOC emissions, as shown in Figure 4.

AC

The micro-toxicant IA considered the wide range of impacts in the Eco-Indicator 99 methodology, but was
limited to Eco-System and Human Health Impacts, due to the lack of Australian-specific impact factors for
the other impact categories. The human health and ecosystem-specific impacts (per unit of pollutant
emitted/discharged) were computed by [14] and have been used in this study. The results shown in Figure
5 allow us to identify the most significant companies in the area regarding the two impact categories
considered.

MA

NU

SC
R

IP

ACCEPTED MANUSCRIPT

Figure 5: Impact Assessment of KIA's micro-toxicant emissions

TE

3.2 MODM for a selected case study company

While the first stage of the research addressed the entire EIP (Kwinana Industrial Area), in this next analysis

CE
P

we focus on a specific company within the KIA. This company is a chemicals producer and has specific
energy requirements in its productive cycle of both hot water and electricity generation. The MO
methodology has been used at this stage. A combination CHP and PV plant was selected as the energy

AC

source for the individual company given the requirement for continuous loads of both hot water and
electricity and the availability of high solar radiation in the immediate region, which averages some 19.6
MJ/m2 (Australian Bureau of Meteorology). The companys monthly energy and water consumption data
have been collected via direct interviews with company personnel to gather the necessary information
regarding the productive cycle, power and natural gas consumption and the costs to be used in the
mathematical model in the MO model mentioned in section 2.3. An initial set of 100 feasible designs
randomly generated with feasible solutions manually inserted constituted the first generation of modelled
runs.
An NSGA-II algorithm was then selected for the scheduling and optimisation of these 100 generations, thus
covering approximately 10,000 potential solutions, which is extremely limited compared to the size of the
design stage.

ACCEPTED MANUSCRIPT
Simulation results are shown in Figure 6, showing the value of economic profitability (to be maximised) on
the y-axis and the best environmental performing solutions (reducing the impact of its emissions) on the xaxis. As shown, the simulator assesses the feasible points, which in this case study are related to

IP

combinations of gas-fuelled-ICEs or fuel cells, as well as PV plants. Other solutions (gas and micro turbines)
could not conform to the current limit on excess waste heat output while simultaneously satisfying at least

SC
R

70% of companys power needs. The top left corner of Figure 6, represents the most profitable solution for
the optimisation results but presents a NPV that is barely positive (AU$113,000) and an environmental

NU

impact reduction that is negative (more pollutants are emitted than the effective emission savings). In the top
right corner, the most environmental impact-reducing solution presented has an HHIR of 19.7 DALY (the

MA

unit of measurement for human health impact) and an economic profitability of AU$110,000, while in the
bottom right corner the Pareto-frontier presents a comparable value of HHIR, but at a reduced profitability.

Most Profitable
Solutionof the group
(113.000 AU$)

TE
Optimizer Checking
feasible solution FC
and PV Plants

AC

CE
P

Optimizer Checking
feasible solutions of
ICE-G and PV Plants

Most Environmental
Friendly group of
Solutions (20.1 Ecop)

The Pareto-frontier is therefore made up of two main groups of solutions: first, a combination of gas-fuelled
ICE and PV plant, located in the top left of Figure 6 above, and second, a fuel cell coupled with a PV plant
(top-right and bottom right, respectively). The selected solutions have been oversized due to an expressed
request by the company, which is expecting to increase production output and therefore the required heat and

ACCEPTED MANUSCRIPT
power loads in the near future. The technical details associated with these two alternatives, exported from the
simulator outputs, are reported in Table 1.

kW
%
%
kWp
%
m2
kWh
kWh
%
kWh
%
AU$
AU$
AU$
years
AU$
%
t
t
t
t
t
t

AC

CE
P

TE

MA

NU

SC
R

CHP Plant Type

Power
Power Efficiency
Thermal Efficiency
PV Peak Power
PV Efficiency
PV Occupied Area
Power Produced by PV
Total Power Produced
Percentage
Total Heat Produced
Percentage
Economic Criteria
Capital Costs
Operating Costs at y=0
Operating Revenues at y =0
Payback
NPV
IRR
Environmental Criteria CO2 emissions
(avoided emissions)
CO emissions
SOx emissions
PM emissions
NMVOC emissions
NOx emissions

Solution 2
FC
300
54%
41%
190
20%
1,050
234,965
1,138,810
91%
685,431
125%
2,152,555
219,113
266,533
19
228,839
11%
365.8
0.64
2.78
0.08
0.00
1.52

Technical Criteria

Solution 1
ICE
370
39%
45%
120
14%
1,050
161,663
1,255,568
100%
1,259,469
230%
1,108,486
270,561
284,708
19
161,020
15%
193.82
-15.51
3.06
0.05
-0.11
-7.95

IP

UM

3.3 AHP analysis of selected alternatives

The previous stage involved only two-objective functions to identify trade-off ranges, but given the various
issues involved in the decision-making process, a more detailed problem-solving structure is considered here
for the more complex AHP analysis. A business-as-usual (BAU, i.e., continuing current plant management)
option has also been considered, and its economic/environmental performances have been calculated.
Specifically, four categories of criteria have been used, considering environmental, economic, technical and
social criteria. The hierarchical model for the AHP analysis is represented in Figure 7, together with the
criterias associated priorities to selected Equal Weights (EW), Environmental (En), Decision Maker (DM)
and Technical (Te) scenarios.

CE
P

TE

MA

NU

SC
R

IP

ACCEPTED MANUSCRIPT

The criteria used in the decision-making assessment involved both quantitative (efficiencies, emissions and
IRR) and qualitative (reliability, commercial viability and the decision makers attitudes) criteria.

AC

Quantitative criteria were directly translated into the 1-9 Saaty Scale for pairwise comparison using the AHP
reference software SuperDecisions 2.08, while the qualitative criteria indicating the decision makers attitude
towards the solutions has been considered by interviewing company personnel and identifying their
preferences among the selected alternatives. When assessing the weight of the four main groups of criteria in
terms of the main objectives (maximising NPV and maximising HHIR), three scenarios were considered, i.e.,
the technical/economic perspective, the environmental perspective and the DMs viewpoint. The evaluation
of these three criteria was performed by assigning maximum weight (9 in the Saaty Scale) to the respective
criteria in the three scenarios assessed, i.e., Technical/Economic in Scenario 1, decision maker (social
criteria) in Scenario 2 and Environmental criteria in Scenario 3, while assigning equal weight to intra-criteria
assessment (e.g., environmental and social criteria) in Scenario 1. Furthermore, a scenario with equal weights
for all four criteria has also been considered.

ACCEPTED MANUSCRIPT
Results are shown in Figure 8. The combination of the internal combustion engine and PV plants was the
most preferred alternative in terms of both the Techno-Economic assessment and the decision makers point
of view, largely due to its economic performance (NPV, IRR and PB) and the commercial/technological

IP

reliability of the CHP solution. From the environmental viewpoint, FC and PV plants were the mostpreferred solution, due to the low environmental impact of the FC plant and its higher power conversion

SC
R

efficiency, hence lowering fossil fuel consumption and increasing CO2 emission reduction.

NU

Normal Priorities for selected Scenarios

0.450
0.400

MA

0.350
0.300
0.250

0.100

ICE + PV

TechnoEconomic
0.148

Decision Maker

Environmental

Equal Weights

0.281

0.334

0.235

0.421

0.282

0.400

0.377

0.431

0.437

0.266

0.388

AC

FC + PV
ICE + PV

CE
P

0.050

BAU Scenario

FC + PV

TE

0.150

0.000

BAU Scenario

0.200

From the Environmental viewpoint, the business as usual (BAU) scenario i.e., maintaining the current
operation represents the second-preferred solution, given its poor environmental performance of ICE
(internal combustion engine) solutions, which feature high emissions of NOx and CO higher than the basecase scenario. Considering the DMs viewpoint, an almost identical priority has been assigned to the BAU
and ICE scenarios, with a small preference for the latter, with higher preference given to the absence of
capital investment a priority perhaps from the decision makers perspective.
The equal-weights scenario presented similar results for ICE and FC plants with a slightly higher
performance for the former whereas the BAU scenario ranked considerably behind the CHP technologies.

ACCEPTED MANUSCRIPT
The solutions presented for the various scenarios have been tested for rank reversal problems, as suggested
by Schenkerman [24]. This has been done by deleting the poorest-performing solution for each scenario
(BAU in Scenario 1, 2 and 4, FC+PV in Scenario 2) and re-assessing the rankings. The results are shown in

SC
R

IP

Figure 9.

Normal Priorities for selected Scenarios - Rank Reversal Test

0.7

NU

0.6
0.5

MA

0.4
0.3
0.2

FC + PV

ICE + PV

0.1

BAU Scenario

AC

CE
P

TE

As illustrated in Figure 9, no rank reversal occurred when testing the solutions. For each scenario, the rank
among the alternatives has been maintained, with variations in the raw priority values due to the removal of
the poorest-performing solution, which contributed to their original calculation.
Sensitivity tests have been conducted to evaluate the robustness of the Equal Weights scenarios. The
results show that a relevant variation in BAU priorities is required (+57%) suggesting that the alternative
solutions are significantly more valuable than the BAU option. Similar computations with analogous results
were also found for the FC and ICE solutions. A slight increase in the FC raw priority or a minimal decrease
in the ICE solution (4%) leads to a Pareto efficiency intersection, resulting in a non-optimal solution set.
4. DISCUSSION AND CONCLUSIONS

ACCEPTED MANUSCRIPT
This paper developed a DSS for addressing energy optimisation decisions in energy plant designs. The
potential value of this analysis is that it provides a DSS framework for assessing and benchmarking the
sustainability of energy production systems, considering both objective and subjective criteria throughout the

IP

analysis by exploiting the advantages of both MODM and MADM.

Following the traditional frameworks of DSS (problem identification, alternatives generation and alternatives

SC
R

selection), the developed DSS consisted of a three-step approach:

Problem identification has been conducted by emissions assessment, considering both the volumes and

NU

impacts of emitted substances. Particular emphasis has been given to the Eco-Indicator 99 methodology,
which calculates the environmental impacts (damage done) from a range of emission variables. Next, an MO

MA

optimisation was conducted to determine the possible energy alternatives by developing a mathematical
model considering economic and environmental objectives at the very beginning of the energy plant design,
involving mixed renewable/non renewable CHP (internal combustion engines, gas turbines, fuel cells) and

TE

PV technologies. The third and last step of the DSS involved using the input results of the optimisation
process (step 2), which included MADM in the selection of the best alternatives among the candidate

CE
P

solutions from the Pareto-Frontier.

While traditional energy efficiency trade-off studies relate to energy use, efficiency and cost, in this paper
two additional impacts are included environmental impacts and an allowance for conventional wisdom in

AC

decision making.

The methodology followed potentially aids in decision-making processes and the planning process for
energy plant designs that are both more eco-efficient and sustainable. This is particularly relevant for energy
investment in large industrial and manufacturing companies and increasingly important for those involved in
eco-industrial parks and the energy-intensive mining and resources sectors.
A more holistic DSS has been presented that assists in reviewing the emission footprint and environmental
impacts of energy production investment. The days of assessing energy production solely in terms of NPV
are coming to a close. Increasing regulatory pressures and community concerns about GHG production also
suggest a need to review plant energy design in terms of environmental impact and other emissions criteria.
Further research could also use this methodology in worldwide industrial park benchmarking assessments, to
promote lower emissions and higher industrial emission standards. The full methodology detailed in this

ACCEPTED MANUSCRIPT
paper could also be extended to other types of industrial plants, including water treatment plants and energy
plants integrated with enhanced pollution abatement technologies.

IP

5. REFERENCES

1. Alarcon-Rodriguez, A., Ault, G., & Galloway, S. (2010). Multi-objective planning of distributed

SC
R

energy resources: A review of the state of the art. Renewable and Sustainable Energy Reviews ,
13531366.

NU

2. Arnott, D., & Pervan, G. (2005). A critical analysis of Decision Support Systems research. Journal
of Information Technology , 67-87.

MA

3. Ayres, R. (1994). A Hanbook of Industrial Ecology. Edward Elgar.

4. Bernier, E., Marchal, F., & Samson, R. (2010). Multi-objective design optimization of a natural
gas-combined cycle with carbon dioxide capture in a life cycle perspective . Energy, 1121-1128.

TE

5. Burstein, F., & Holsapple, C. W. (2008). Handbook on Decision Support System 1. Springer.
6. Chatzimouratidis, A. I., & Pilavachi, P. A. (2009). Environmental, technologic and economic

CE
P

criteria are used for evaluating different power plants. Energy Policy , 778-787.
7. Dipama, J., Teyssedou, A., Aub, F., & Lizon-A-Lugrin, L. (2010). A grid based multi-objective

816.

AC

evolutionary algorithm for the optimization of power plants. Applied Thermal Engineering , 807-

8. Goedkoop, M., & Spriensma, R. (2011). The Eco-Indicator 99: A damage oriented method for Life
Cycle Impact Assessment. Pr Consultants.
9. Guilln-Goslbez, G. (2011). A novel MILP-based objective reduction method for multi-objective
optimization: Application to environmental problems. Computers & Chemical Engineering , 14691477.
10. Z.X. Guo, W.K. Wong, Zhi Li and Peiyu Ren; Modeling and Pareto optimization of multi-objective
order scheduling problems in production planning; Computers & Industrial Engineering 64 (2013)
972986;
11. Harkin, T., Hoadley, A., & Hooper, B. (2011). Using multi-objective optimisation in the design of
CO2 capture systems for retrofit to coal power stations . Energy 41 (2010), 228-235

ACCEPTED MANUSCRIPT
12. S. Jeyadevi, S. Baskar, C.K. Babulal, M. Willjuice Iruthayarajan; Solving multiobjective optimal
reactive power dispatch using modified NSGA-II; Electrical Power and Energy Systems 33 (2011)
219228

IP

13. Krajnc, D., & Glavi, P. (2005). A model for integrated assessment of sustainable development.
Resources, Conservation and Recycling , 189-208.

SC
R

14. Lundie, S., Huijbregts, M. A., & Rowley, H. V. (2007). Australian characterisation factors and
normalisation figures for human toxicity and ecotoxicity. Journal of Cleaner Production , 819-832.

NU

15. Mirzaesmaeeli, H., Elkamel, A., Douglas, P., Croiset, E., & Gup, M. (2010). A multi-period
optimization model for energy planning with CO2 emission consideration. Journal of

MA

Environmental Management , 1063-1070.

16. P. Murugan, S. Kannan, S. Baskar; NSGA-II algorithm for multi-objective generation expansion
planning problem; Electric Power Systems Research 79 (2009) 622628

TE

17. Nagesha, N., & Balachandra, P. (2006). Barriers to energy efficiency in small industry clusters:
Multi-criteria-based prioritization using the analytic hierarchy process . Energy , 1969-1983.

CE
P

18. Olson, D. (2008). Multi-Criteria Decision Support. In Burstein, & Holsapple, A Handbook on
Decision Support System, 1 (p. 299). Springer.
19. Sidhartha Panda, Narendra Kumar Yegireddy; Automatic generation control of multi-area power

AC

system using multi-objective non-dominated sorting genetic algorithm-II; Electrical Power and
Energy Systems 53 (2013) 5463
20. Pohekar, S., & Ramachandran, M. (2004). Application of multi-criteria decision making to
sustainable energy planning. Renewable and Sustainable Energy Reviews , 365-381.
21. Power. (2008). Decision Support Systems: A Historical Overview. In Burstein, & Holsapple,
Handbook on Decision Support System 1 (p. 121 - 140). Springer
22. Rong, A., & Lahdelma, R. (2005). CO2 emissions trading planning in combined heat and power
production via multi-period stochastic optimization. European Journal of Operational Research ,
1874-1895.
23. Saaty, T. (1980). The analytic hierarchy process. New York: McGraw-Hill.

24. Schenkerman S.; Avoiding rank reversal in AHP decision-support models; European
Journal of Operational Research, Volume 74, Issue 3, 5 May 1994, Pages 407-419

ACCEPTED MANUSCRIPT
25. Simon H. A. (1960) The new science of management decision, NewYork: Harper & Row
26. Srinivas N, Deb K. Multi-objective optimization using non-dominated sorting in genetic algorithms.
IEEE Trans Evol Comput 1994;2:22148.

IP

27. US-EPA. (2008). Technology Characterization: Reciprocating Engines, Gas Turbines, MicroTurbines and Fuel Cells,

SC
R

28. US-EPA. (2011). Emissions Factors & AP 42, Compilation of Air Pollutant Emission Factors:
http://www.epa.gov/ttnchie1/ap42/

NU

29. Vince, F., Marechal, F., Aoustin, E., & Brant, P. (2008). Multi-objective optimization of RO
desalination plants . Desalination, Volume 222 , 96-118.

MA

30. Weise, T. (2009). Global Optimization Algorithms. da Global Optimization Algorithms:

http://www.it-weise.de/

31. Yusliza Yusoff, Mohd Salihin Ngadiman, Azlan Mohd Zain; Overview of NSGA-II for Optimizing

TE

Machining Process Parameters; Procedia Engineering 15 ( 2011 ) 3978 3983

AC

CE
P

32. Zeleny, M. (1982). Multiple criteria decision making. McGraw-Hill.

ACCEPTED MANUSCRIPT
Biographical Note

NU

SC
R

IP

Dr. Patrizia Simeoni graduated in Industrial Technology Engineering and got a Ph.D. in Energetics (2005) at
the University of Udine (Italy). From 2006 shes ad-joint professor of Waste Treatment Plants and
Environmental Compatibility of Industrial Plants. From 2010 she is researcher of Mechanical Plants at the
University of Udine. Her main areas of research concerns Decision Support Systems (DSS) for energy and
Environmental planning, Life Cycle Assessment (LCA), Multi-Criteria Analysis (MCA), Risk management,
Industrial ecology. Shes actively involved in research project of national relevance and works as a
technical-scientific consultant in the areas of waste management and soil remediation, developing and
implementing decision support systems for enterprises and public bodies.

MA

Alessandro Mattiussi graduated (2008) in Industrial Engineering at the University of Udine (Italy) and where
he got the Ph.D. (2012) after a visiting period at Curtin University (Perth, Western Australia), with a thesis
on Decision Support Systems (DSS) for sustainable plant design. Currently working for the Italian Industrial
Association, hes researching on the Electricity and Gas Markets and Multi-Criteria Analysis in Energy
Planning focusing on Life Cycle Assessments (LCA) of energy-converting technologies, and energy recovery
in industrial contexts..

AC

CE
P

TE

Michele Rosano, PhD, Director of Centre of Excellence in Cleaner Production (University of Western
Australia, Perth), researching on Resource Economics, Sustainability Management, Life Cycle Assessment
and Waste Management. She leads the Industrial Ecology Group within the Centre of Sustainable Resource
Processing (CSRP). She worked internationally in the mining industry in a number of senior executive
positions, and as a Lecturer and Researcher in Australia. She is currently assisting with the establishment of
Australia's first Industrial Ecology Networking organisation.

ACCEPTED MANUSCRIPT
Highlights

A DSS for sustainable technologiess design and selection is presented in the paper;

LCI and Impact Assessment have been used for initial problem definition;

Both Multi-Objectives and Multi-Attributes methods are used for CHP plant design and selection;

DSS is applied to a case study in Kwinana Industrial Area (Perth, Western Australia)

AC

CE
P

TE

MA

NU

SC
R

IP