water

Article
Evaluation of Groundwater Remediation
Technologies Based on Fuzzy Multi-Criteria
Decision Analysis Approaches
Hao Wang 1,2 , Yanpeng Cai 1,2,3, *, Qian Tan 3,4 and Yong Zeng 5
1 State Key Laboratory of Water Environment Simulation, School of Environment, Beijing Normal University,
Beijing 100875, China; wanghao_bnu@163.com
2 Beijing Engineering Research Center for Watershed Environmental Restoration and Integrated Ecological
Regulation, School of Environment, Beijing Normal University, Beijing 100875, China
3 Institute for Energy, Environment and Sustainable Communities, University of Regina, Regina,
SK S4S 0A2, Canada; 13260119689@wo.cn
4 College of Water Resources and Civil Engineering, China Agricultural University, Beijing 100083, China
5 State Key Laboratory of Petroleum Resource and Prospecting, College of Geosciences,
China Petroleum University, Beijing 102249, China; 201621180018@mail.bnu.edu.cn
* Correspondence: yanpeng.cai@bnu.edu.cn; Tel.: +86-10-5880-0830; Fax: +86-10-5880-2795

Received: 30 March 2017; Accepted: 15 June 2017; Published: 20 June 2017

Abstract: Petroleum is an essential resource for the development of society and its production is huge.
There is a great risk of leakage of oil during production, refining, and transportation. After entering
the environment, the oil pollutants will be a great threat to the environment and may endanger
human health. Therefore, it is very important to remediate oil pollution in the subsurface. However,
it is necessary to choose the appropriate remediation technology. In this paper, 18 technologies are
evaluated through constructing a parameter matrix with each technology and seven performance
indicators, and a comprehensive analysis model is presented. In this model, four MCDA methods are
used. They are SWA (Simple Weighted Addition Method), WP (Weighted Product Method), CGT
(Cooperative Game Theory), and TOPSIS (Technique for Order Preference by Similarity to Ideal
Solution). Mean ranking and Borda ranking methods are used to integrate the results of SWA, WP,
CGT, and TOPSIS. Then two selection priorities of each method (mean ranking and Borda ranking)
are obtained. The model is proposed to help decide the best choice of remediation technologies. It can
effectively reduce contingency, subjectivity, one-sidedness of the traditional methods and provide
scientific reference for effective decision-making.

Keywords: MCDA; groundwater remediation technologies; Shengli oil field

1. Introduction
Petroleum is fundamental for human and social development. Since the Industrial Revolution,
petroleum refinery products (PRPs) have been widely used as fuels and industrial raw materials.
Approximately 2 to 3 billion tons of crude oil are produced every year across the world (2016).
At the same time, over 100 million tons of oil and associated PRPs are entering the environment [1].
Along with extensive utilization of petroleum and related products, a large amount of leakage and
spillage get into the soil and groundwater, causing significant pollution to the environment [2].
This may consequentially lead to pollution to surface water, air, and agricultural crops, affecting
human and ecosystem health [3]. Such pollution is causing increasing concerns across the world. More
recently, many tools and technologies have been developed for dealing with oil pollution in both soil
and groundwater [4]. However, for many petroleum production managers, as well as decision makers,
it is of great difficulty to identify and evaluate cost-effective tools and technologies for mitigating oil

Water 2017, 9, 443; doi:10.3390/w9060443 www.mdpi.com/journal/water

Water 2017, 9, 443 2 of 20

pollution and remediating contaminated soil and groundwater due to their diversities in many factors
such as cost, occupational area, and efficiency [5]. Therefore, it is a necessity to advance effective
tools for facilitating decision-making and technology identification/evaluation of mitigating and
remediating technologies for oil pollution in soil and groundwater.
Generally, a number of processes and factors may affect the generation of oil pollution in soil and
groundwater, such as waste from petroleum drilling, leakage of underground tanks and pipelines,
and pollution discharge of many accidents [6]. For example, under the United States Environmental
Protection Agency and its designated State Administration, approximately 2 million storage tanks
are in operation in the USA. According to the technical management rules of underground storage
tanks, the United States has monitored 210 thousand gas stations in operation and the results show
that gas stations constructed before 1970s are most prone to leakage [7]. Approximately 510 thousand
oil storage tanks have leaking problems and 130 thousand leaking points need to be cleaned in the
USA. Sewage irrigation, oil leakage sludge, and garbage piling up would also cause oil to leak into
the groundwater [1]. Thus, since 1984 (the implementation of a United States federal government
underground storage tank program), the government has disabled 1.7 million underground storage
tanks [8]. In 1993, Shell Oil Company made a survey of 1100 gas stations in the United Kingdom and
found that 33% of the sites had contaminated soil and groundwater [9]. In Canada, underground
storage tanks and pipelines at the gas station constructed before 1990 almost all appeared to leak,
causing serious pollution of the groundwater [10,11]. In 2011, the Peru oil pipeline was leaking
and approximately 1100 barrels of crude oil caused serious pollution to the original forest of the
Amazon [12]. Groundwater is extremely vulnerable since the corresponding water circulation system
is relatively closed. Once contaminated, the water may endanger the environment not only for a long
time [3].
Particularly in China, groundwater pollution is serious in many gas stations, oil and gas fields, as
well as oil and chemical plants [13]. Comparatively, China has over 100 thousand gas stations, including
over 1000 in Beijing. There are approximately 6000 underground storage tanks in Shanghai [14].
According to the survey, in the 1980s the establishment of gas station underground storage tanks and
pipelines have caused corrosion and leakage. According to groundwater pollution survey data (2015)
in gas stations in Beijing and Tianjin, approximately 50% of gas station areas in Beijing exceeded the
standard (i.e., Groundwater Petroleum Hydrocarbon Pollutants) and the detection rate of petroleum
hydrocarbons in the groundwater of Tianjin gas station area was approximately 85% [15]. Due to the
leakage of oil storage tanks and pipelines, a large area of the soil and groundwater in Zhongyuan
Oilfield of Henan Province was polluted, and the content of petroleum hydrocarbons in contaminated
soil varied from 1% to 10% [16]. Petroleum hydrocarbons have become common pollutants in the
organic pollution of groundwater in China. Because of the good quality, wide distribution, and
convenient access of groundwater, it represents an ideal source of water supply. In China, groundwater
accounted for 20% of the total water supply, 70% of drinking water supply, 40% of irrigation water
demand, and 38% of industrial water demand [17]. Therefore, groundwater plays an important role in
China’s national production and life. The sustainable use of groundwater resources has become an
increasingly concerning issue.
At present, the treatment of oil pollution in soil and groundwater mainly consists of physical,
chemical, and biological methods, including: (a) physical treatment technologies that are mainly
employing physical means to control contaminated groundwater through various physical methods,
such as technologies of shielding, passive collection, and extraction treatment; (b) chemical
treatment technologies that mainly comprise technologies of dosing, permeable treatment bed
and soil modification technology; and (c) bioremediation to stimulate the growth of indigenous
microorganisms through artificial measures including the injection of oxygen and nutrients, thereby
enhancing the natural biodegradation process of pollutants. Usually, bioremediation technologies
should be combined with a well system to accelerate the diffusion of oxygen and nutrients under
the combined action of pumping wells, which can shorten the recovery time. Bioremediation
Water 2017, 9, 443 3 of 20

technologies of groundwater oil pollution mainly include methods of biological injection, organic
clay, and biological reactors. There are 18 commonly used remediation technologies included in this
research [1,2,5,10,12,13,15]: passive collection, shielding, hydrodynamic and control, pump and treat,
light non-aqueous phase liquid recovery, in situ air sparging, enhanced bioremediation, permeable
reactive barrier, in situ chemical oxidation, organic clay, electrochemical dynamics, groundwater
circulation well, monitoring natural attenuation, soil vapor extraction and in situ aeration, biological
aeration in situ aeration, dual phase extraction single pump system, dual phase extraction double
pump system, and surfactant enhanced remediation.
Obviously, there are many ways to remediate groundwater. However, different methods should
be adopted under different pollution conditions and geographical features. At the same time, such
technologies might be performed at different efficiencies and expenses. Therefore, identification and
evaluation of groundwater remediation technologies is desired. In order to deal with identification of
pollution mitigation and remediation technologies, many conditions and criteria need to be considered.
About the selection of remediation technologies, we should take into account many features such
as pollutant characteristics, hydrogeological conditions of the contaminated sites, and costs of the
remediation technologies. Due to the complexities of actual ground water pollution sites, only through
comprehensive analysis and evaluation can the most scientific repair technology or combination
of technologies be determined. Such analysis and evaluation processes are normally subjective,
and may lead to multiple results based on the varying opinions of relevant experts. There are no
fixed patterns for the evaluation of remediation techniques. For example, Zhao (2012) presented a
screening method based on Standard Guide for Remedy Selection Integrating Risk-based Corrective
Action and Non-Risk Consideration of America [18]. Firstly, through eliminating of unsuitable repair
technologies, he selected and evaluated alternative repair techniques. He also gave an evaluation
matrix for remediation technologies and indicators (e.g., technical acceptability, site availability,
effectiveness, time and costs) [19]. Li (2016) used PROMETHEE (Preference Ranking Organization
Methods for Enrichment Evaluations) to identify desired remediation technologies based on indicators
such as pollutants migration, degradation, human health risk, and characteristics of technologies [20].
Bai (2015) established the method and index system of soil remediation, and carried on the gradation
of remediation technologies [21]. Li (2016) used the PROMETHEE method to select remediation
technologies for the ruins of a chemical plant [20]. Also, a few studies focused on the selection of
indicators but covered few technologies.
Previous studies have the following limitations: (a) the integrated approach to remediation of
groundwater pollution is usually analyzed unilaterally, but the technology selection is a multi-attribute
problem that includes a number of discrete variables and fuzzy factors. Thus, the quantitative
and qualitative description of the problem needs further consideration. (b) A single method is
accompanied by strong randomness, and the result can lead to serious uncertainty, which brings
about erroneous conclusion. In order to remedy such limitations, the objective of this research is to
propose a comprehensive approach for supporting identification and evaluation of a number of soil
and groundwater remediation technologies. Firstly, four different multi-attribute evaluation methods
will be used to evaluate a series of alternative schemes and decisions on the basis of irrelevant and
inconsistent rules. The problem of quantitative and qualitative description will be effectively solved.
In this way, the assessment of remediation technologies will be transformed into a multi-attribute
decision-making problem. Secondly, different multi-attribute evaluation methods will be integrated
through mean ranking and Borda ranking methods to avoid the accidental nature of different MCDA
methods. This represents an improvement upon stability and accuracy. Then, the developed evaluation
method will be applied to various forms of pollution treatment technology. The results will be useful
to the scheme selection problem in other cases. In this research, 18 technologies are included and
four MCDA methods are used. The results of different MCDA methods are integrated through the
introduction of mean ranking and Borda ranking methods to get the priority order. A desired solution
will then be generated by giving the ranked results of each scheme related to remediation technologies.
Water 2017, 9, 443 4 of 20

2. Methodology

2.1. Development of Evaluation Indicators

A systematic and reasonable evaluation index system is desired in the evaluation of a large
number of remediation technologies. Thus, evaluation indicators need to be identified and screened
out. The indicators should cover pollution types, and intensity, geological survey as well as social and
economic research. Therefore, the establishment of evaluation index system should use a variety of
indicator selection methods. Indicators are used to describe the advantages and disadvantages of the
methods in terms of cost, efficiency, application range, and technical maturity [19–21]. To carry out
the selection of indicators and establish of index system, we must refer to certain principles [22–28].
The general principles of establishing index system include: (a) the indicator system should be
scientifically based on concept and calculation method of the theory of sustainable development;
(b) any developed index systems should include not only the index of technical applicability, but also
the indexes that reflect the management cost, technology maturity, and efficiencies; there must be
indicators that can reflect the degree of mutual coordination between the above system indicators;
(c) because of the difference between the natural environment and social economy, the selection of
indicators should be consistent with the regional characteristics and local conditions of the study area;
(d) terminology, concepts, and calculation methods should be made as standard as possible to achieve
comparability with other regional indicators; and (e) a combination of qualitative and quantitative
indicators. Quantitative indicators should be selected as far as possible. Important indicators that
are difficult to quantify can be described qualitatively. (f) Simplicity, i.e., the availability of data and
the feasibility of statistical computation, should be taken into account. The index system should be
easy to be understood and should be convenient to use through the adoption of certain mathematical
evaluation models.
Therefore, to establish a scientific, reasonable and systematic evaluation index system for
groundwater remediation technologies, based on the basic concept of water resources security and
theoretical analysis method combined with expert consultation and system analysis, we scientifically
select indicators to reflect the safety of groundwater, and establish an evaluation index system. We have
collected a wide range of research results in recent years on water pollution, oil pollution, groundwater
treatment and oil pollution control at home and abroad. Through theoretical analysis, we select
indicators that can represent the features of petroleum and the associated products. The initially
established evaluation index system contains a wide range of indicators and the correlation degree
of some indexes is relatively large. Therefore, we need to choose certain indicators from all. Expert
consultation method is a commonly used index screening method that is conducted by means of
a questionnaire. We used a 0–1 scale design questionnaire from the “Oil Polluted Groundwater
Remediation Technologies Evaluation Index System”. In the questionnaire, 0 ≤ aij < 0.5. This means
that j is more important than i, and the smaller aij is, the more important j is than i. We solicited
the opinions of 18 domestic experts in the fields of water resources, water environment, ecological
environment, oil pollution remediation, and other research areas and ensure that the indicators can
fully reflect the views of the experts.

2.2. Uncertainties of Evaluation

Many uncertainties are associated with the evaluation process. For example, fuzzy judgments
might be given for any criteria, which can be dealt with through the introduction of fuzzy sets theory.
Normally, the first stage is fuzzy impact transformation, which includes two steps: (a) transformation
of descriptive linguistic variables, which means transforming a descriptive language index into fuzzy
sets; and (b) transforming the fuzzy sets into crisp values. The result of this stage is to generate a new
index matrix that contains only numerical data. The second stage is to use the classical MCDA method
to sort the various decision schemes. In the third stage, we use mean ranking and Borda ranking
methods to integrate the results of MCDA methods to obtain a more accurate priority.
Water 2017, 9, 443 5 of 20

Water 2017, 9, 443 5 of 20

Chen proposed a numerical approximation system that can systematically convert linguistic
variables Chen
intoproposed a numerical fuzzy
the corresponding approximation
sets [29].system that can
According to systematically convert linguistic
Chen, this transformation required
variables into the corresponding fuzzy sets [29]. According to Chen, this transformation
eight conversion scales, as shown in Figure 1a–h. These conversion scales were presented in the required
eight conversion scales, as shown in Figure 1a–h. These conversion scales were presented in the
synthesis and modification of Baas (1977), Bonissone (1982), Chen (1988), Efstathiou (1979, 1982),
synthesis and modification of Baas (1977), Bonissone (1982), Chen (1988), Efstathiou (1979, 1982),
Kerre (1982) and Wenstop (1976) [30–36] and applied widely by Wang (2016), Xue (2016), Mardani
Kerre (1982) and Wenstop (1976) [30–36] and applied widely by Wang (2016), Xue (2016), Mardani
(2016), and Karsak (2015) [37–40]. In general, the scale given in the graph is sufficient to cover all
(2016), and Karsak (2015) [37–40]. In general, the scale given in the graph is sufficient to cover all
the the
representations
representations of each feature,
of each “high”“high”
feature, and “low”. When given
and “low”. When a given
specifica variable
specific representation,
variable
onerepresentation,
of them can be used
one to analyze.
of them Eventoifanalyze.
can be used the same variable
Even is used
if the same as the
variable is “high”, the“high”,
used as the membership
functions are not thefunctions
the membership same in are
the not
different graphs.
the same Chen
in the arguedgraphs.
different that this phenomenon
Chen argued thatwas thiscaused
phenomenon
by the fact that was causedlinguistic
the same by the factvariables
that the same
were linguistic
expressedvariables were expressed
differently differently
in different situationsin [29].
different situations [29]. We used the membership function set to express the
We used the membership function set to express the meaning of language data. For example, in meaning of language
data.1a,
Figure For example,
the red peak in Figure 1a, the
indicates the red peak indicates
membership the membership
degree of “high”; the degree of “high”;
farther away ittheis farther
from 0.8, the
away it is from 0.8, the less it belongs to “high”.
less it belongs to “high”.

medium high low medium high

1 1

μ(X) μ(X)

0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
X X

(a) (b)

very low low medium low medium low

high very high medium medium high
1 high
1

μ(X) μ(X)

0 0
0 0.10.20.30.40.50.60.70.80.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
X X
(c) (d)

very low low very low low

more or less low more or less high more or less low medium
high very high more or less high high
1 very high
1

μ(X)
μ(X)

0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
X X

(e) (f)

Figure 1. Cont.
Water 2017, 9, 443 6 of 20
Water 2017, 9, 443 6 of 20

very low low to very low none very low

low medium low low to very low low
medium medium high medium low medium
high high to very high midium high high
very high high to very high very high
1 excellent
1

μ(X)
μ(X)

0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
X X

(g) (h)
Figure 1. Scale (a–h) for the graph of membership function [22].
Figure 1. Scale (a–h) for the graph of membership function [22].
The second step in the transformation of fuzzy variables is to convert the fuzzy values into
crisp
Thevalues.
secondAstep
lot of scholars
in the have made relevant
transformation of fuzzyresearch
variables such
is toasconvert
Hipel (1982) [41] and
the fuzzy Cheng
values into crisp
(2000)
values. A[42]. Generally
lot of scholarsspeaking,
have madethis relevant
kind of conversion can beasregarded
research such as a method
Hipel (1982) to calculate
[41] and the [42].
Cheng (2000)
fuzzy average
Generally value.
speaking, thisThe fuzzy
kind mean valuecan
of conversion is not
be necessarily
regarded as the highest degree
a method of membership.
to calculate the fuzzyAaverage
left–right scoring approach based on Jain (1977) [43] and Chen (1985) [44] was used. The score of
value. The fuzzy mean value is not necessarily the highest degree of membership. A left–right scoring
fuzzy sets M can be obtained through the following steps. In order to get the score value, fuzzy sets
approach based on Jain (1977) [43] and Chen (1985) [44] was used. The score of fuzzy sets M can be
need to be compared with the maximum fuzzy sets (fuzzy maximum) and the minimum fuzzy sets
obtained through the following steps. In order to get the score value, fuzzy sets need to be compared
(fuzzy minimum). These two fuzzy sets can be defined as:
with the maximum fuzzy sets (fuzzy maximum) and the minimum fuzzy sets (fuzzy minimum). These
two fuzzy sets can be defined as: μ (x ) = x ,0 ≤ x ≤ 1
max  (1a)
(0, otherwise
x, 0 ≤ x ≤ 1
µmax ( x )= ,0 ≤
1 − x0, x ≤ 1 (1a)
μ min(x ) =  otherwise . (1b)
( 0, otherwise
The score on the right is obtained 1 − x, 0 ≤ x ≤ 1
µmin (by
x ) the
= intersection of fuzzy sets. M and fuzzy maximum. The (1b)
0,
right score can be obtained by the following equation: otherwise

μ R ( M ) = by
The score on the right is obtained x [ μ M ( x ) ∧ μ max( x )] .
supthe intersection of fuzzy sets M and fuzzy (2a)
maximum.
The right score can be obtained by the following equation:
Similarly, M left score can be calculated by using the following formula:

μµLR((M supx [xμ[µMM( (xx) )∧∧μµminmax

M)) ==sup ( x()]x.)]. (2b) (2a)

The total score of M can be calculated by left and right scores:

Similarly, M left score can be calculated by using the following formula:
μT(M ) = [μR(M ) + 1 − μL(M )]/ 2 . (2c)
µ L ( M ) = supx [µ M ( x ) ∧ µmin ( x )]. (2b)
As shown in Figure 2, μ max and μ min are the intersection points of the two diagonal with the
The total score
membership of M
function can be calculated
respectively, by left and right
and the membership scores:
functions are as follows:

µ ( M) =0.[µ3R-( M
x ) + 1 − µ L ( M)]/2. (2c)
μM (Tx ) = ， 0 ≤ x ≤ 0. 3 (3a)
1
0.3
As shown in Figure 2, µmax and µmin are the intersection points of the two diagonal with the
membership function respectively, and the membership functions are as follows:

0.3 − x
µ M1 ( x ) = , 0 ≤ x ≤ 0.3 (3a)
0.3
  4x, 0 ≤ x < 0.25

μ M 2(x ) =  0.5 − x . (3b)
 0.25 ,0.25 ≤ x ≤ 0.5
Water 2017, 9, 443 7 of 20
 x − 0.3.
,0.3 ≤ x < 0.5
( .2
μ M 3(x ) =  04x, 0 ≤ x < 0.25 (3c)
µ M2 ( x ) = 0.
 0.570.25
−-x.x 0.5 ≤ x ≤ 0.7 (3b)
 0.2 ,
, 0.25 ≤ x ≤ 0.5
(
x − 0.3.
, 0.3 ≤ x < 0.5
µ M3 ( x ) =4x0.70.2
-−x2,,0.0.5
5 ≤≤ xx ≤<0.70.75
(3c)

μ M 4(x ) = ( 1 -0.2x (3d)
 04x.2−5 2,, 0 ≤ x≤<x0.75
.75
0.5 ≤1
µ (x) =
M4 1−x (3d)
0.25 , 0.75 ≤ x ≤ 1
x - 0.7
μ M (x ) = x − 0.7 ，0.7 ≤ x ≤ 1 . (3e)
µ M5 ( x ) = 0.3 , 0.7 ≤ x ≤ 1.
5
(3e)
0.3

low medium low medium medium high high

μL(M1) μR(M5)
1
μL(M2) μR(M4)

μL(M3) μR(M3)

0.5
μR(M2) μL(M4)

μR(M1) μL(M5)

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure2.2.Illustration
Figure Illustrationof
ofdetermining
determiningcrisp
crispvalue.
value.

Then,using
Then, usingEquations
Equations(1)–(3),
(1) to we
(3),can
wegetcanthe
get thescore.
total total They
score.can
They can be
be used used of
instead instead of the
the original
original language
language to the
to describe describe
data. the data. Asinshown
As shown in0.8846,
Table 1, Table 1, 0.8846,
0.7000, 0.7000,
0.5000, 0.5000,
0.4333, 0.4333,
and and
0.1154 0.1154
take the
take the place of “high”, “medium high”, “medium”, “medium
place of “high”, “medium high”, “medium”, “medium low”, and “low”. low”, and “low”.

Table1.1.Determination
Table Determinationofofµμ total..
total

i μR (Mi) μL (Mi) μT (Mi)

i µR (Mi ) µL (Mi ) µT (Mi )
1 0.2308 1.0000 0.1154
1 0.2308 1.0000 0.1154
2 0.6667 0.8000 0.4334
2 0.6667 0.8000 0.4334
33 0.5833
0.5833 0.5833
0.5833 0.5000
0.5000
44 0.8000
0.8000 0.4000
0.4000 0.7000
0.7000
55 1.0000
1.0000 0.2308
0.2308 0.8846
0.8846

2.3.Multi-Criteria
2.3. Multi-CriteriaDecision
DecisionAnalysis
Analysis
Multi-criteria decision
Multi-criteria decision analysis
analysis (MCDA)
(MCDA) is is aa batch
batch ofof methods
methods that
that can
can evaluate
evaluate aa series
series of
of
alternativesonon
alternatives thethe basis
basis of irrelevant
of irrelevant and inconsistent
and inconsistent rules tothe
rules to identify identify
desiredthe desired
decision decision
alternatives.
Italternatives. It has theofcharacteristics
has the characteristics flexibility andof flexibility
clear judgment and clear
of the judgmentbetween
correlation of the correlation between
indicators. Over the
indicators.
past Over
20 years, MCDA the methods
past 20 years, MCDA amethods
have become powerfulhave tool become a powerful
for decision analysis,tool
andfor decision
developed
rapidly in management, engineering, and other fields. More recently, MCDA has been applied to many
research areas such as environmental resource management. Hipel (1982), for example, introduced a
fuzzy MCDA model in a sludge management (SWM) problem [41]. In order to broaden its scope of
application, decision support systems (DSS) and MCDA methods were combined in the 1980s to form
Water 2017, 9, 443 8 of 20

an integrated system, which has been applied in a variety of areas [45]. In Finland, PROMETHEE were
used as an assistant decision method to solve landfill site selection problems [46]. Maniezzo (1998)
applied it to the site selection of industrial waste management facilities in Italy [47]. Haastrup (1998)
developed a decision support system that combined optimization algorithms to solve facility location
problems [48]. Cheng (2000) established a DSS and applied it to support urban solid waste management
problems [49]. Normally, MCDA is a method for decision makers to evaluate the merits and demerits
of several schemes containing many attributes. In this research, four MCDA methods are introduced.
These four methods are selected because they can deal with the same type of data (the index value and
the weight).
(1) Simple Weighted Addition Method (SWA)
Simple Weighted Addition Method is believed to be the simplest MCDA method. Because it is
relatively easy for decision makers to understand, it is widely used in various fields. For each scheme,
the utility value of the index is obtained by the product of the normalized index value and the weight
of each index, and the sum of the scheme can be obtained:
n
Uj = ∑ wi rij , j = 1, 2, . . . , m. (4)
i =1

In the formula, wi is the weight of the index I, rij is the index value after standardization.
After calculating the Uj , the scheme with the maximum value is the most desirable scheme for
the decision maker.
A basic assumption of the SWA method is that it is independent of the index. Therefore, the
weight of the index will not be affected by the weight of other indexes. Simplicity is the biggest
advantage of SWA, but its disadvantages are also obvious: there is usually contact or complementarity
between indicators. The basic assumption is not easy to accept, while ignoring the relevance between
the indexes may lead to incorrect results.
(2) Weighted Product Method (WP)
The weighted product method has been in use for a long time. The SWA method needs to first
standardize indicators of data to remove the impact of the unit, but the WP method does not have to
standardize the data. When the index value is multiplied, the index weight is the power of the index
value, and the Uj of each scheme is:

n
wj
Uj = ∏ xij , j = 1, 2, . . . , m. (5)
i =1

In the formula, w j is the weight of the index of number i. Forward index weights are positive
in power. Backward index weights are negative in power. When adopting this method, the scheme
with the maximum Uj is the most desirable for decision makers. Theoretically, due to the characteristic
of the product, the value may be infinite. The purpose of WP is to screen out the scheme with the
smallest Uj , and the difference between the largest and the second largest values is larger than that
given by the SWA method. WP has reasonable logic and a simple calculation method, but it has not
been widely used.
(3) Technique for Order Preference by Similarity to Ideal Solution (TOPSIS)
Technique for Order Preference by Similarity to Ideal Solution is proposed by Chen [29].
This author proposes that a MCDA problem can be regarded as a collection system. The m schemes
with n indicators that need to be evaluated are equivalent to m points in the n-dimensional space.
Therefore, the most desirable program should meet the “shortest distance” to the best scheme and
the “longest distance” to the worst scheme. Compared to the CGT, in which only the worst solution
Water 2017, 9, 443 9 of 20

is considered, MCDA can give more comprehensive consideration of the data when solving the
TOPSIS problem.
The specific process is as follows:
(a) The data need to be standardized; the dimensional effects need to be removed and thus it is
more convenient to make the comparison among the indicators. According to the given weight, the
index value matrix with weight needs to be calculated.

vij = wi rij , i = 1, 2, . . . , n; j = 1, 2, . . . , m. (6)

In the formula, wi is the weight of the index i.

(b) After vij is calculated, the best scheme A∗ and the worst scheme A− is defined as follows:

A∗ = (max j vij |i ∈ I ), (min j vij |i ∈ I 0 )| j = 1, 2, . . . , m = {v1∗ , v2∗ , . . . , vi∗ , . . . , v∗n }


(7a)

A− = (min j vij |i ∈ I ), (max j vij |i ∈ I 0 )| j = 1, 2, . . . , m = v1− , v2− , . . . , vi− , . . . , v−

 
n , (7b)

where I represents the number of positive indicators and I’ represents the number of reverse indicators.
(c) Calculate the value of each scheme, that is, the relative closeness of the best scheme.
s
n
S∗j = ∑ (vij − v∗j )2 , j = 1, 2, . . . , m (8a)
i =1
s
n
∑ (vij − v−j )
2
S−
j = , j = 1, 2, . . . , m (8b)
i =1

Uj = S− ∗ −
j / ( S j + S j ), j = 1, 2, . . . , m (8c)

In the formula, S∗j is the distance between the number j scheme and the best scheme. S− j is the
distance between the number j scheme and the worst scheme. At the same time, 0 < Uj∗ < 1.
Finally, we can sort all the schemes by the value of Uj . The scheme that has the biggest Uj is the
most desirable. The advantages of this method are the same as for SWP. It is easy to understand, but
in a situation, it cannot point to a clear decision. If an MCDA problem involves only two indicators,
it can be considered as a geometric problem of a two-dimensional space; the optimal and the worst
scheme are defined as P∗ and P− , assuming there are two schemes (P1 and P2 ) and they have the same
Uj . In this case, they are considered to be the same and there is no more desirable scheme. The decision
makers can only select a scheme based on their own judgment.
(4) Cooperative Game Theory (CGT)
Cooperative game theory is similar to the WP method and can also be considered as a combination
of WP and TOPSIS. It can also enlarge the distance among the schemes. So decision makers can choose
the scheme with the longest distance to the worst scheme. The method is designed to help decision
makers choose the scheme that has the maximum geometric distance to the worst case. To define the
worst scheme, the decision maker first defines a minimal acceptable set of indicator values. However,
not all of the indicators have a minimum acceptable level. For example, it is difficult to determine
the minimum cost to a decision maker, and when the cost is an indicator that must be considered in
the MCDA problem-solving process, the decision maker must give the minimum value. Therefore, in
order to avoid this situation, we can select the minimum value of the index before calculating. When a
set of decision schemes is given, the worst index set A− is defined as:

A− = (min j xij i ∈ I ), (max j xij |i ∈ I ∗ )| j = 1, 2, · · · , m = x1− , x2− , . . . , xi− , . . . xn−

 
(9)

Among them, xij is the value of the index i; xi− is the minimum value (the worst level) of the index
i in all the schemes. Therefore, the Uj of each index can be calculated by the following formula [49]:
Water 2017, 9, 443 10 of 20

n
wi
Uj = ∏ xij − xi− , j = 1, 2, . . . , m, (10)
i =1

in which wi is the weight of each index. The most desirable scheme is the one with the maximum Ui .
At the same time, each program can be sorted by Ui .
It is not uncommon to solve the MCDA problem by Cooperative Game Theory. Lau believes that
CGT can produce scheme selection system with more security and less risk [50]. In fact, the scheme
sorting given by CGT is conservative for policy makers, and there are also some problems. Since any
number multiplied by 0 equals 0, the CGT method will automatically exclude all programs that contain
at least one minimum value (the worst level). They are not considered, even if the other indicators of
these schemes are larger (better level).
(5) Integration of MCDA Approaches
To avoid the accidental nature of different methods, two integration methods, mean ranking and
Borda Ranking, are used. The mean ranking method is one of the simplest methods. This method is
based on the concept and theory of statistical computation. According to the ranking of technologies
obtained by each MCDA method, we carry out mean ranking and get the final rank of each technology.
Based on the voting theory, the Borda method is used to construct the N × N matrix by comparing
every scheme to another. For each pair of schemes A j and A j0 , the number of votes is defined as the
number of support. For example, the first two rows in the table can determine the number of A j and
A j0 votes. According to M2 and M4 , A1 is better than A2 . However, according to M1 and M3 , A1 is
worse than A2 . Thus, compared to A2 ,A1 gets two votes by { M2 , M4 }. Similarly, compared to , A1 ,
A2 also gets two votes by { M1 , M3 }. So an N × N matrix X is established, and: x jj0 = 1, which means
that A j gets more votes than A j0 ; x jj0 = 0, and vice versa.
As Table 2 shows, A1 and A2 get the same number of votes. Thus, x12 and x21 both equal 0.
The last column S j indicates the degree to which A j is better than other scheme, which is the sum of
the same row. In this way, the scheme with the largest value of S j is considered the best and most
desirable scheme. For example, this sorting result is A3 > ( A2 , A5 ) > A1 > A4 , and the most desirable
scheme is A3 .

Table 2. N × N matrix used in the Borda ranking method.

Scheme A1 A2 A3 A4 A5 Sj
A1 0 0 0 1 0 1
A2 0 0 1 1 0 2
A3 1 1 0 1 1 4
A4 0 0 0 0 0 0
A5 0 1 0 1 0 2
S’ j 1 2 1 4 1

Thus, an integrated approach is proposed based on the combination of the abovementioned

indicator system, as well as uncertainty expression and MCDA methods. Also, in order to reflect
the uncertainties associated with weights for the evaluation, a fuzzy analytical hierarchy process
(FAHP) [51] is used. Therefore, a fuzzy multi-criteria decision analysis (FMCDA) system is established.

3. Application for Evaluation of Technologies

3.1. Overview of the Case Study

The northern oil fields of Shengli are located in the Yellow River Delta region and cover an area
of 8600 km2 , including Hekou, Lijin in Dongying, Bincheng, Zhanhua and Wuli in Binzhou, Linyi in
Dezhou, Shanghe, and Jiyang in Ji’nan and other counties.
Water 2017, 9, 443 11 of 20

The study area is the main production area of Shengli oilfield. There are many oil wells. In the
process of drilling and production, some of the crude oil will be scattered. During oil transportation,
due to unreasonable design, incorrect installation, and material failure, the pipeline will rupture,
causing oil leakage. A crude oil pipeline is generally buried 1 m deep or so, will be a direct threat
to the nearby shallow groundwater and soil. According to the distribution of organic compounds,
chlorinated organic compounds have the highest detectable rate of 72.7%, followed by halogenated
hydrocarbons and aromatic hydrocarbons, with detectable rates of 14.8% and 12.5% respectively.
In order to quantify subjective feelings, the fuzzy analytical hierarchy process (FAHP) is used to
calculate the weight of indicators. First, the priority relation matrix is established. It is then transformed
into a fuzzy consistent matrix. After priority relation ranking, we get the weight of the indicators [51].
The comparison value of importance between every two indicators is given in Table 3.

Table 3. Precedence relation matrix.

Indicator I1 I2 I3 I4 I5 I6 I7
Aquifer depth (I1 ) 0.5 0.5 0.5 0.5 1 1 1
Permeability of aquifer (I2 ) 0.5 0.5 0.5 0.5 1 1 1
Applicable scope of pollution (I3 ) 0.5 0.5 0.5 0.5 1 1 1
Applicable pollution level (I4 ) 0.5 0.5 0.5 0.5 1 1 1
Technology maturity (I5 ) 0 0 0 0 0.5 0 0
Governance cost (I6 ) 0 0 0 0 1 0.5 1
Pollutant removal rate (I7 ) 0 0 0 0 1 0 0.5

In this research, FAHP is used to transfer the precedence relation matrix into a fuzzy consistent
judgment matrix, as shown in Table 4. Table 5 shows the description of indicators in natural language.
As shown in Table 6, we transfer them into mathematical language using the rules in Table 1.

Table 4. Fuzzy consistent judgment matrix and weight list of evaluation indicators.

Indicator I1 I2 I3 I4 I5 I6 I7 Weight
Aquifer depth (I1 ) 0.5000 0.5000 0.5000 0.5000 0.8214 0.6786 0.7500 0.1735
Permeability of aquifer (I2 ) 0.5000 0.5000 0.5000 0.5000 0.8214 0.6786 0.7500 0.1735
Applicable scope of pollution (I3 ) 0.5000 0.5000 0.5000 0.5000 0.8214 0.6786 0.7500 0.1735
Applicable pollution level (I4 ) 0.5000 0.5000 0.5000 0.5000 0.8214 0.6786 0.7500 0.1735
Technology maturity (I5 ) 0.1786 0.1786 0.1786 0.1786 0.5000 0.3571 0.4286 0.0816
Governance cost (I6 ) 0.3214 0.3214 0.3214 0.3214 0.6429 0.5000 0.5714 −0.1225
Pollutant removal rate (I7 ) 0.2500 0.2500 0.2500 0.2500 0.5714 0.4286 0.5000 0.1020
Water 2017, 9, 443 12 of 20

Table 5. The description of indicators in natural language. (Aquifer depth: A—4.6 m; B—4.6~15.2 m; C—15.2~30.5 m; D—>30.5 m. Permeability of aquifer: a—better;
b—good; c—bad; d—worse.)

Aquifer Permeability Applicable Scope Applicable Technology Governance Pollutant

Remediation Technology
Depth of Aquifer of Pollution Pollution Level Maturity Cost Removal Rate
Passive Collection Method (A1) A, B c, d small heavier applied widely general high
Shielding Method (A2) A, B a, b, c, d smaller heavier applied widely higher /
Hydrodynamic and Control (A3) A, B, C, D a, b bigger heavier applied little general general
Pump and Treat (A4) A, B, C, D a, b, c bigger heavier applied widely higher high
Light Non-Aqueous Phase Liquid Recovery (A5) A, B, C, D a, b, c bigger heavier applied widely general high
In Situ Air Sparging (A6) A, B, C, D b, c, d small light applied widely low higher
Enhanced Bioremediation (A7) A, B, C, D a, b, c all light applied widely lower higher
Permeable Reactive Barrier (A8) A, B a, b, c, d small all applied widely low higher
In Situ Chemical Oxidation (A9) A, B, C, D a, b, c, d all all applied widely low higher
Organic Clay (A10) A, B, C, D d all heavy average low higher
Electrochemical Dynamics (A11) A, B, C, D a, b, c, d smaller all average low higher
Groundwater Circulation Well (A12) A, B, C, D a, b big all average general higher
Monitoring Natural Attenuation (A13) A, B, C, D a, b, c, d all lighter applied widely lower higher
Soil Vapor Extraction and In Situ Aeration (A14) A, B, C, D b, c small all applied little high higher
Biological Aeration In Situ Aeration (A15) A, B, C, D a, b, c small all average high higher
Dual Phase Extraction Single Pump System (A16) A, B, C, D c, d bigger heavier applied little higher high
Dual Phase Extraction Double Pump System (A17) A, B, C, D a, b, c bigger heavier applied little higher high
Surfactant Enhanced Remediation (A18) A, B, C, D a, b, c bigger heavier average low higher
Water 2017, 9, 443 13 of 20

Table 6. The description of indicators in mathematical language.

Aquifer Permeability Applicable Scope Applicable Technology Governance Pollutant

Remediation Technology
Depth of Aquifer of Pollution Pollution Level Maturity Cost Removal Rate
Passive Collection Method (A1) 0.1154 0.5000 0.4333 0.8846 0.8846 0.5000 0.7000
Shielding Method (A2) 0.1154 0.8846 0.1154 0.8846 0.8846 0.8846 0.1154
Hydrodynamic and Control (A3) 0.7000 0.5000 0.8846 0.8846 0.1154 0.5000 0.5000
Pump and Treat (A4) 0.7000 0.7000 0.8846 0.8846 0.8846 0.8846 0.7000
Light Non-Aqueous Phase Liquid Recovery (A5) 0.7000 0.7000 0.8846 0.8846 0.8846 0.5000 0.7000
In Situ Air Sparging (A6) 0.7000 0.7000 0.4333 0.4333 0.8846 0.4333 0.8846
Enhanced Bioremediation (A7) 0.7000 0.7000 0.5000 0.4333 0.8846 0.1154 0.8846
Permeable Reactive Barrier (A8) 0.1154 0.8846 0.4333 0.5000 0.8846 0.4333 0.8846
In Situ Chemical Oxidation (A9) 0.7000 0.8846 0.5000 0.5000 0.8846 0.4333 0.8846
Organic Clay (A10) 0.7000 0.1154 0.5000 0.7000 0.5000 0.4333 0.8846
Electrochemical Dynamics (A11) 0.7000 0.8846 0.1154 0.5000 0.5000 0.4333 0.8846
Groundwater Circulation Well (A12) 0.7000 0.5000 0.7000 0.5000 0.5000 0.5000 0.8846
Monitoring Natural Attenuation (A13) 0.7000 0.8846 0.5000 0.1154 0.8846 0.1154 0.8846
Soil Vapor Extraction and In Situ Aeration (A14) 0.7000 0.5000 0.4333 0.5000 0.1154 0.7000 0.8846
Biological Aeration In Situ Aeration (A15) 0.7000 0.7000 0.4333 0.5000 0.5000 0.7000 0.8846
Dual Phase Extraction Single Pump System (A16) 0.7000 0.5000 0.8846 0.8846 0.1154 0.8846 0.7000
Dual Phase Extraction Double Pump System (A17) 0.7000 0.7000 0.8846 0.8846 0.1154 0.8846 0.7000
Surfactant Enhanced Remediation (A18) 0.7000 0.7000 0.8846 0.8846 0.5000 0.4333 0.8846
Water 2017, 9, 443 14 of 20

3.2. Result Analysis

Based on the weight of indicators and standardized indicator data of different technologies, results
of SWA can be obtained via Equation (4). Table 7 shows the results obtained by the SWA method.
Light non-aqueous phase liquid recovery, surfactant enhanced remediation, and pump and treat are
the first three technologies according to this method.

Table 7. Ranking results of the Simple Weighted Addition, Weighted Product, Cooperative Game
Theory, and Technique for Order Preference by Similarity to Ideal Solution methods.

SWA WP CGT TOPSIS

Remediation Technology
Uj Ranking Uj Ranking Uj Ranking Uj Ranking
Passive Collection Method (A1) 0.139008 16 0.536495 17 0.093818 17 0.53 16
Shielding Method (A2) 0.104581 18 0.365298 18 0.020821 18 0.47 18
Hydrodynamic and Control (A3) 0.177789 7 0.679240 10 0.280765 11 0.68 4
Pump and Treat (A4) 0.198598 3 0.820635 4 2.875641 1 0.69 3
Light Non-Aqueous Phase Liquid Recovery (A5) 0.217416 1 0.880015 1 0.789412 4 0.8 2
In Situ Air Sparging (A6) 0.170366 9 0.716045 6 0.585903 9 0.61 10
Enhanced Bioremediation (A7) 0.190247 4 0.863145 3 0.567308 10 0.67 6
Permeable Reactive Barrier (A8) 0.147914 14 0.559180 14 0.094611 16 0.55 14
In Situ Chemical Oxidation (A9) 0.189545 5 0.783693 5 0.656443 6 0.68 5
Organic Clay (A10) 0.144997 15 0.556964 15 0.094752 15 0.54 15
Electrochemical Dynamics (A11) 0.153728 12 0.580043 13 0.099372 13 0.55 13
Groundwater Circulation Well (A12) 0.165646 10 0.705785 8 0.603190 8 0.63 8
Monitoring Natural Attenuation (A13) 0.182401 6 0.714569 7 0.098510 14 0.61 11
Soil Vapor Extraction and In Situ Aeration (A14) 0.127692 17 0.552921 16 0.250788 12 0.51 17
Biological Aeration In Situ Aeration (A15) 0.150386 13 0.660682 11 0.638466 7 0.57 12
Dual Phase Extraction Single Pump System (A16) 0.165036 11 0.655532 12 1.067405 3 0.63 9
Dual Phase Extraction Double Pump System (A17) 0.176863 8 0.694933 9 1.147820 2 0.66 7
Surfactant Enhanced Remediation (A18) 0.215410 2 0.875487 2 0.752292 5 0.81 1

Similarly, based on the weight of indicators and standardized indicator data of different
technologies, results of WP can be obtained by Equation (5). Table 7 shows the results obtained
by the WP method. Compared with the SWA method, WP increases the distance between every two
technologies. The results are a little different. Liquid recovery, surfactant enhanced remediation, and
enhanced bioremediation are the first three technologies according to this method.
The results of TOPSIS are obtained based on the weight of indicators and standardized indicator
data of different technologies and Equations (6) to (8). Table 7 shows the results obtained by the
TOPSIS method. The first three technologies are surfactant enhanced remediation, light non-aqueous
phase liquid recovery, and pump and treat according to this method.
As to the CGT method, based on the weight of indicators and standardized indicator data of
different technologies, the results of CGT can be obtained by Equations (9) and (10). Table 7 shows the
results obtained by the CGT method. CGT also increases the distance between every two technologies.
Because the minimum of indicators is from the value of one technology, there will be some invalid
values. To avoid invalid values, we make the positive indicator 0.00001 smaller and the negative
indicator 0.00001 larger in the worst sample. This will not influence the sorting result. The first
three technologies are Pump and Treat, Dual Phase Extraction Double Pump System, and Dual Phase
Extraction Single Pump System according to this method. Table 6 shows the results obtained by the
CGT method.

3.3. MCDA Aggregation

The MCDA method aims to sort the given scheme. There are two ways to further integrate the
results: the mean ranking method and the Borda ranking method. The mean ranking method sorts
Water 2017, 9, 443 15 of 20

the schemes according to the average value, and the Borda ranking is given by comparing every
two technologies.
The results are shown in Table 8. This table shows the scheme sorting index and final average
value of sorting. According to the final average sorting value, we can produce a final scheme order,
that is: A5 > A18 > A4 > A9 > A7 > A17 > A3 > (A6, A12) > A16 > A13 > A15 > A11 > A8 > A10 > A14 >
A1 > A2. The two in brackets have the same sorting value.

Table 8. Results of mean ranking method.

MCDA Methods
Remediation Technology Mean Rankings
SWA WP CGT TOPSIS
Passive Collection Method (A1) 16 17 17 16 16.5 17
Shielding Method (A2) 18 18 18 18 18 18
Hydrodynamic and Control (A3) 7 10 11 4 8 7
Pump and Treat (A4) 3 4 1 3 2.75 3
Light Non-Aqueous Phase Liquid Recovery (A5) 1 1 4 2 2 1
In Situ Air Sparging (A6) 9 6 9 10 8.5 8
Enhanced Bioremediation (A7) 4 3 10 6 5.75 5
Permeable Reactive Barrier (A8) 14 14 16 14 14.5 14
In Situ Chemical Oxidation (A9) 5 5 6 5 5.25 4
Organic Clay (A10) 15 15 15 15 15 15
Electrochemical Dynamics (A11) 12 13 13 13 12.75 13
Groundwater Circulation Well (A12) 10 8 8 8 8.5 8
Monitoring Natural Attenuation (A13) 6 7 14 11 9.5 11
Soil Vapor Extraction and In Situ Aeration (A14) 17 16 12 17 15.5 16
Biological Aeration In Situ Aeration (A15) 13 11 7 12 10.75 12
Dual Phase Extraction Single Pump System (A16) 11 12 3 9 8.75 10
Dual Phase Extraction Double Pump System (A17) 8 9 2 7 6.5 6
Surfactant Enhanced Remediation (A18) 2 2 5 1 2.5 2

The matrix is given in Table 9, and Sj is the votes the schedules get. As shown in Table 10, the
results are a little different from mean ranking: A5 > A18 > A4 > A7 > A9 > A17 > (A3, A6, A12) >
(A13, A16) > A15 > A11 > A8 > A10 > (A1, A14) > A2. The results of the mean ranking and Borda
ranking methods are conflated as following: A5 > A18 > A4 > (A7, A9) > A17 > (A3, A6, A12) > (A13,
A16) > A15 > A11 > A8 > A10 > (A1, A14) > A2. The two in brackets have the same sorting value.

Table 9. N × N matrix used in Borda ranking method.

Technologies A1 A2 A3 A4 A5 A6 A7 A8 A9 A10 A11 A12 A13 A14 A15 A16 A17 A18 Sj’
A1 0 0 1 1 1 1 1 1 1 1 1 1 1 0 1 1 1 1 15
A2 1 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 17
A3 0 0 0 1 1 0 1 0 1 0 0 0 0 0 0 0 0 1 5
A4 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 2
A5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
A6 0 0 0 1 1 0 1 0 1 0 0 0 0 0 0 0 1 1 6
A7 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 1 3
A8 0 0 1 1 1 1 1 0 1 0 1 1 1 0 1 1 1 1 13
A9 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 1 3
A10 0 0 1 1 1 1 1 1 1 0 1 1 1 0 1 1 1 1 14
A11 0 0 1 1 1 1 1 0 1 0 0 1 1 0 1 1 1 1 12
A12 0 0 0 1 1 0 1 0 1 0 0 0 0 0 0 0 1 1 6
A13 0 0 0 1 1 1 1 0 1 0 0 0 0 0 0 0 0 1 6
A14 0 0 1 1 1 1 1 1 1 1 1 1 1 0 1 1 1 1 15
A15 0 0 1 1 1 1 1 0 1 0 0 1 1 0 0 1 1 1 11
A16 0 0 1 1 1 0 1 0 1 0 0 1 0 0 0 0 1 1 8
A17 0 0 0 1 1 0 1 0 1 0 0 0 0 0 0 0 0 1 5
A18 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1
Sj 1 0 8 15 17 8 13 4 13 3 5 8 7 1 6 7 10 16
Water 2017, 9, 443 16 of 20

Table 10. Results of Borda ranking method.

Remediation Technology Votes Borda Rankings

Passive Collection Method (A1) 1 16
Shielding Method (A2) 0 18
Hydrodynamic and Control (A3) 8 7
Pump and Treat (A4) 15 3
Light Non-Aqueous Phase Liquid Recovery (A5) 17 1
In Situ Air Sparging (A6) 8 7
Enhanced Bioremediation (A7) 13 4
Permeable Reactive Barrier (A8) 4 14
In Situ Chemical Oxidation (A9) 13 5
Organic Clay (A10) 3 15
Electrochemical Dynamics (A11) 5 13
Groundwater Circulation Well (A12) 8 7
Monitoring Natural Attenuation (A13) 7 10
Soil Vapor Extraction and In Situ Aeration (A14) 1 16
Biological Aeration In Situ Aeration (A15) 6 12
Dual Phase Extraction Single Pump System (A16) 7 10
Dual Phase Extraction Double Pump System (A17) 10 6
Surfactant Enhanced Remediation (A18) 16 2

As shown in Figure 3, every method gives a priority order for each technology. When put into a
quadrangle graph, the more every quadrangle looks like a regular quadrangle, the more stable the four
methods are. The technologies with the highest and lowest ranking are more stable and have better
consistency among WP, SWA and TOPSIS. When it comes to CGT methods, there is a sharp change.
This is due to the fact that the application of CGT automatically removes all of the schemes with
minimum indicator values. Even if the other indicators are large (better), they will not be considered.
A simple linear weighting method cannot avoid the impact of the results of the correlation between
the indicators; a weighted product method will enlarge the characteristics of indicators, and may lead
to results being affected by individual indicators. TOPSIS will miss schemes containing very poor
indicators. The integration method avoids the shortcomings of the four programs to a certain extent.
The ranking of options based on a simple weighted addition method is Light Non-Aqueous Phase
Liquid Recovery, Surfactant Enhanced Remediation, Pump and Treat; the ranking of options based
on a weighted product method is Light Non-Aqueous Phase Liquid Recovery, Surfactant Enhanced
Remediation, Enhanced Bioremediation; the ranking of options based on cooperative game theory
is Pump and Treat, Dual Phase Extraction Double Pump System, and Dual Phase Extraction Single
Pump System; and the ranking of options based on TOPSIS is Surfactant Enhanced Remediation,
Light Non-Aqueous Phase Liquid Recovery, and Pump and Treat. Based on the four MCDA methods,
the 18 kinds of repair technology in the order of priority are as follows: Light Non-Aqueous Phase
Liquid Recovery (A5) > Surfactant Enhanced Remediation (A18) > Pump and Treat (A4) > [Enhanced
Bioremediation (A7), In Situ Chemical Oxidation (A9)] > Dual Phase Extraction Double Pump System
(A17) > [Hydrodynamic and Control (A3), In Situ Air Sparging (A6), Groundwater Circulation Well
(A12)] > [Monitoring Natural Attenuation (A13), Dual Phase Extraction Single Pump System (A16)]>
Biological Aeration In Situ Aeration (A15) > Electrochemical Dynamics (A11) > Permeable Reactive
Barrier (A8) > Organic Clay (A10) > [Passive Collection Method (A1), Soil Vapor Extraction and In Situ
Aeration (A14)] > Shielding Method (A2). Technologies in square brackets have the same priority.
 Water 2017, 9, x FOR PEER REVIEW 17 of 20

Phase Extraction Single Pump System (A16)]> Biological Aeration In Situ Aeration (A15) >
Electrochemical Dynamics (A11) > Permeable Reactive Barrier (A8) > Organic Clay (A10) > [Passive
Collection Method (A1), Soil Vapor Extraction and In Situ Aeration (A14)] > Shielding Method (A2).
Water 2017, 9, 443 in square brackets have the same priority.
Technologies 17 of 20

Figure 3. Ranking of four methods.

Figure 3. Ranking of four methods.

4.4.Conclusions
Conclusions
InInthis
thispaper,
paper,a fuzzy
a fuzzy multi-criteria
multi-criteriadecision
decisionanalysis
analysis(FMCDA)
(FMCDA)systemsystemwaswasestablished
establishedbased
based
ononthethe integration of a set of indicators, fuzzy sets theory, multi-criteria decision analysis(MCDA),
integration of a set of indicators, fuzzy sets theory, multi-criteria decision analysis (MCDA),
and
andfuzzy
fuzzyanalytical
analyticalhierarchy
hierarchyprocess
process(FAHP).
(FAHP).The Theinnovation
innovationofofthisthisresearch
researchmainly
mainlycomprised:
comprised:
(a)(a)ranking
ranking results were obtained according to various indicators of 18 technologies,reflecting
results were obtained according to various indicators of 18 technologies, reflecting
compromising
compromisingand andconflicting
conflictingfeatures
featuresamongamongthe thetechnologies
technologiesinintermstermsofofthe
theproposed
proposedsevenseven
dimensions
dimensions of evaluation indicators, and (b) uncertainties not only associated with weightsfor
of evaluation indicators, and (b) uncertainties not only associated with weights for
evaluation
evaluationindicators,
indicators,butbutalso
alsoforforthe
theevaluation
evaluationprocess
processofofMCDA
MCDAmethods methodswere
wereintroduced,
introduced,
improving
improving thethe
robustness
robustnessof the
of evaluation
the evaluationprocedure and theand
procedure results
the obtained. The proposed
results obtained. The FMCDA
proposed
includes three stages based on the evaluation scheme. The first stage is the fuzzy
FMCDA includes three stages based on the evaluation scheme. The first stage is the fuzzy impact impact transformation,
which includes two
transformation, steps:
which (a) the two
includes change of the
steps: descriptive
(a) the change of language variables
the descriptive into fuzzy
language sets, and
variables into
(b) the fuzzy sets were transformed into a single value. The results of this stage were
fuzzy sets, and (b) the fuzzy sets were transformed into a single value. The results of this stage were to generate
a new index matrix that contains only digital data. In the second stage, classical MCDA methods
were used to sort all kinds of decision alternatives. In the third stage, the results of MCDA methods
were integrated with different integration methods to get a more accurate result. The evaluation of
groundwater remediation technologies based on fuzzy multi-criteria decision analysis approaches
Water 2017, 9, 443 18 of 20

indicated that the method used in this research could reduce subjectivity and uncertainty, leading to
a more robust and defensible remedy selection, and as many remediation technologies are involved
as possible. This method can be applied to other areas for decision makers to select the best scheme
among different choices.

Acknowledgments: This study was supported by the National Basic Research Program of China (973 Program,
No. 2013CB430402), the National Natural Science Foundation of China (No. 51522901 and No. 51421065), and
the Fundamental Research Funds for the Central Universities (No. 2017EYT13). The authors also extend their
appreciation to the anonymous reviewers and editors for their constructive comments for improving the paper.
Author Contributions: Hao Wang and Yanpeng Cai analyzed the data; Yong Zeng and Hao Wang gathered and
analyzed the data; Yanpeng Cai reviewed the data results; Qian Tan and Yong Zeng contributed materials tools;
Hao Wang wrote the paper; Yanpeng Cai reviewed and edited the paper.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Lv, X.L.; Sun, J.C. Review on remediation technology for petrochemical hydrocarbon contamination of
groundwater. J. Anhui Agric. Sci. 2014, 42, 5567–5571.
2. Li, M.L.; Dong, W.H. Health risk assessment of typical petroleum contaminated groundwater and soil in
northeast China. China Rural Water Hydropower 2016, 7, 67–71.
3. Yu, C.; Yin, X.A.; Yang, Z.F.; Cai, Y.P.; Sun, T. A shorter time step for eco-friendly reservoir operation does not
always produce better water availability and ecosystem benefits. J. Hydrol. 2016, 540, 900–913. [CrossRef]
4. Tan, Q.; Huang, G.H.; Cai, Y.P.; Yang, Z.F. A non-probabilistic programming approach enabling risk-aversion
analysis for supporting sustainable watershed development. J. Clean. Prod. 2016, 112, 4771–4788. [CrossRef]
5. Zhang, B.Q.; An, D.; Wang, Y.; Liang, H.H.; Zhou, L.; Xi, B.D.; Tang, J. An optimization methodology of
groundwater remediation technologies of contaminated sites in the desert areas by multi-criteria decision
analysis. Chin. J. Environ. Eng. 2016, 10, 5521–5527.
6. Li, C.H.; Sun, L.; Jia, J.X.; Cai, Y.P.; Wang, X. Risk assessment of water pollution sources based on an
integrated k-means clustering and set pair analysis method in the region of Shiyan, China. Sci. Total Environ.
2016, 557–558, 307–316. [CrossRef] [PubMed]
7. Zhai, M.Z.; Zhang, K.L.; Dang, L.; Qin, W.; Ma, W.P. Study on American storage tank leakage prevention
technique system. Autom. Pet. Chem. Ind. 2016, 52, 14–17.
8. USEPA. UST Program Facts, Washington, DC, USA. [EB/OL]. 2013. Available online: http://www.epa.gov/
oust/pnhs/ustfacts.pdf (accessed on 10 September 2016).
9. Shang, J.T.; Wang, Z. Review of remediation technologies for oil-contaminated soil and effect of different soil
amendments on bioremediation efficacy. J. Environ. Sci. Manag. 2016, 41, 122–125.
10. Zhang, S.Y.; Su, X.X.; Lin, X.Y.; Zhang, Y.L.; Zhang, Y. Experimental study on the multi-media PRB reactor
for the remediation of petroleum-contaminated groundwater. Environ. Earth Sci. 2015, 73, 1–8. [CrossRef]
11. Saghi, H.; Lakzian, E. Optimization of the rectangular storage tanks for the sloshing phenomena based on
the entropy generation minimization. Energy 2017, 128, 564–574. [CrossRef]
12. Chen, H.L. Research progress in microbial remediation of petroleum-contaminated shallow aquifer.
Saf. Environ. Eng. 2015, 22, 66–72.
13. Zhao, L.; Zhang, Y.; Guo, J.S.; Zhang, D.; Deng, C.G. Investigation on pollution characteristics of petroleum
hydrocarbon in shallow groundwater around gas stations. Chin. J. Environ. Eng. 2016, 10, 131–136.
14. Cai, Y.P.; Yue, W.C.; Xu, L.Y.; Yang, Z.F.; Rong, Q.Q. Sustainable urban water resources management
considering life-cycle environmental impacts of water utilization under uncertainty. Resour. Conserv. Recyl.
2016, 108, 21–40. [CrossRef]
15. Hu, H.T. An enhanced remediation technology for contaminated groundwater. Appl. Mech. Mater. 2013,
253–255, 1089–1092. [CrossRef]
16. Han, H.L.; Chen, Z.; Yang, J.M.; Miao, C.C.; Zhang, K.; Jin, W.B.; Liu, Z. Field scale demonstration of
fungi-bacteria augmented remediation of petroleum-contaminated soil. Environ. Sci. 2008, 29, 454–461.
17. Li, P.; Tian, R.; Xue, C.; Wu, J. Progress, opportunities, and key fields for groundwater quality research under
the impacts of human activities in China with a special focus on western China. Environ. Sci. Pollut. Res. Int.
2017, 24, 13224–13234. [CrossRef] [PubMed]
Water 2017, 9, 443 19 of 20

18. Lv, H. Research on Biodegradation of Petroleum Hydrocarbon Contaminants and Enhanced In-Situ
Remediation in Groundwater. Ph.D. Thesis, Jilin University, Changchun, China, 2014.
19. Zhao, Y.S. Risk management and screening of remediation technologies for contaminated groundwater site.
J. Jilin Univ. 2012, 42, 1426–1433.
20. Li, W.; Wang, M.Y.; Han, Z.T.; Zhang, M.; Liu, L.Y. Screening process of brownfield site groundwater remedial
technologies: A case study of an abandoned chemical factory contaminated site. Hydrogeol. Eng. Geol. 2016,
3, 131–140.
21. Bai, L.P.; Luo, Y.; Liu, L.; Zhou, Y.Y.; Yan, Z.G.; Li, F.S. Research on the screening method of soil remediation
technology at contaminated sites and its application. Environ. Sci. 2015, 36, 4218–4224.
22. Bossel, H. Indicators of Sustainable Development: Theory, Method, Applications; International Institute of
Sustainable Development: Winnipeg, MB, Canada, 1999; pp. 20–55.
23. Bossel, H. The human actor in ecological-economic models: Policy assessment and simulation of actor
orientation for sustainable development. Ecol. Econ. 2000, 34, 337–355. [CrossRef]
24. Loucks, D.P. Sustainability Criteria for Water Resources System; Cambridge University Press: Cambridge, UK,
1999; pp. 33–90.
25. Loucks, D.P. Sustainable water resources management. Water Int. 2000, 25, 3–10. [CrossRef]
26. Faures, J.M. Indicators for Sustainable Water Resources Development; Land and Water Development Division,
FAO: Rome, Italy, 1998; pp. 1–10.
27. Meng, B.; Chi, G. Evaluation index system of green industry based on maximum information content.
Singap. Econ. Rev. 2016. [CrossRef]
28. Li, H.; Zhang, H.; Wang, F.; Wang, Z. Social risk assessment index system by composite catastrophe models:
A case study in contemporary china. In Proceedings of the Tenth International Conference on Management
Science and Engineering Management; Springer: Singapore, 2017.
29. Chen, S.J.J.; Hwang, C.L.; Beckmann, M.J.; Krelle, W. Fuzzy multiple attribute decision making: Methods
and applications. Lect. Notes Econ. Math. Syst. 2011, 375, 1–531.
30. Baas, S.M.; Kwakernaak, H. Rating and ranking of multiple-aspect alternatives using fuzzy sets. Automatica
1977, 13, 47–58. [CrossRef]
31. Bonissone, P.P. A fuzzy sets based linguistic approach: Theory and applications. In Proceedings of the
Conference on Winter Simulation, Orlando, FL, USA, 3–5 December 1980; IEEE Press: Piscataway, NJ, USA,
1980; pp. 99–111.
32. Chen, S.M. A new approach to handling fuzzy decision-making problems. IEEE. Trans. Syst. Man Cybern.
1988, 18, 1012–1016. [CrossRef]
33. Efstathiou, J.; Rajkovic, V. Multi-attribute decision making using a fuzzy heuristic approach. IEEE. Trans.
Syst. Man. Cybern. 1979, 9, 326–333. [CrossRef]
34. Efstathiou, J.; Tong, R. Ranking fuzzy sets: A dession theoretic approach. IEEE. Trans. Syst. Man Cybern.
1982, 2, 655–659.
35. Kerre, E.E. The use of fuzzy set theory in electrocardiological diagnostics. In Approximate Reasoning in
Decision Analysis; Gupta, M.M., Sanchez, E., Eds.; North-Holland Publishing: Amsterdam, The Netherlands,
1982; pp. 277–278.
36. Wenstop, F. Fuzzy set simulation models in a systems dynamic perspective. Kybernetes 1976, 5, 209–218.
[CrossRef]
37. Wang, P.; Zhu, Z.; Wang, Y. A novel hybrid MCDM model combining the saw, TOPSIS and GRA methods
based on experimental design. Inf. Sci. 2016, 345, 27–45. [CrossRef]
38. Xue, Y.X.; You, J.X.; Zhao, X.; Liu, H.C. An integrated linguistic MCDM approach for robot evaluation and
selection with incomplete weight information. Int. J. Prod. Res. 2016, 54, 5452–5467. [CrossRef]
39. Mardani, A.; Jusoh, A.; Zavadskas, E.; Zakwan, N.; Valipour, A.; Kazemilari, M. Proposing new hierarchical
framework to evaluating of quality management practices: A new combined fuzzy hybrid mcdm approach.
J. Bus. Econ. Manag. 2016, 17. [CrossRef]
40. Karsak, E.E.; Dursun, M. An integrated fuzzy MCDM approach for supplier evaluation and selection.
Comput. Ind. Eng. 2015, 82, 82–93. [CrossRef]
41. Hipel, K.W. Fuzzy set methodologies in multi-criteria modeling. In Fuzzy Information and Decision Processes;
Gupta, M.M., Sanchez, E., Eds.; North-Holland Publishing: Amsterdam, The Netherlands, 1982; pp. 279–287.
Water 2017, 9, 443 20 of 20

42. Cheng, K.S. Development of a Fuzzy Multi-Criteria Decision Support System for Municipal Solid Waste
Management. Master’s Thesis, University of Regina, Regina, SK, Canada, 2000.
43. Jain, R. A procedure for multi-aspect decision making using fuzzy sets. Int. J. Syst. Sci. 1977, 8, 1–7.
[CrossRef]
44. Chen, S.H. Ranking fuzzy numbers with maximizing set and minimizing set. Fuzzy Sets Syst. 1985, 17,
113–129. [CrossRef]
45. Chang, S.H.; Wen, C.G.; Cheng, Y.L. A fuzzy multi-objective programming approach for optimal management
of the reservoir watershed. Eur. J. Oper. Res. 1997, 99, 289–302. [CrossRef]
46. Hong, I.; Vogel, D.R. Data and model management in a generalized mcdm-dss. Decis. Sci. 1991, 22, 1–25.
[CrossRef]
47. Maniezzo, V.; Mendes, I.; Paruccini, M. Decision support for siting problems. Decis. Support Syst. 1998, 23,
273–284. [CrossRef]
48. Haastrup, P.; Maniezzo, V.; Mattarelli, M.; Rinaldi, F.M.; Mendes, I.; Paruccini, M. A decision support system
for urban waste management. Eur. J. Oper. Res. 1998, 109, 330–341. [CrossRef]
49. Gershon, M. The role of weights and scales in the application of multi-objective decision making. Eur. J.
Oper. Res. 1984, 15, 244–250. [CrossRef]
50. Lau, P. An Architectural Framework for Construction of a Flexible Expert System for Engineering Selection.
Master’s Thesis, University of Regina, Regina, SK, Canada, 1996.
51. Ruan, Y.F.; Liu, Y.F.; Wang, D.; Mou, J. Research on application of fuzzy consistent matrix theory in
optimization of foundation design schemes. Ind. Constr. 2008, 38, 636–640.

© 2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).