You are on page 1of 8

Engineering Applications of Artificial Intelligence 91 (2020) 103556

Contents lists available at ScienceDirect

Engineering Applications of Artificial Intelligence


journal homepage: www.elsevier.com/locate/engappai

Alternatives selection for produced water management: A network-based


methodology✩
Shengzhong Mao a , Yong Deng a ,∗, Danilo Pelusi b
a Institute of Fundamental and Frontier Science, University of Electronic Science and Technology of China, Chengdu, 610054, China
b
Faculty of Communication Sciences, University of Teramo, Teramo, I-64100, Italy

ARTICLE INFO ABSTRACT


Keywords: Products and process strategies selection has attracted much attention, especially for produced water man-
Produced water management agement. In order to select best alternatives, it is necessary to develop a comprehensive decision-making
Visibility graph methodology applicable to complex and uncertain systems. In this paper, we propose a new network-based
Decision making
methodology of decision-making. Firstly, the discrete quantitative and qualitative input data are mapped into
Power average operator
networks based on ordered visibility graph. Then we take advantages of network topologies, which are node
Complex networks
degree and node distance, to establish the relationships among the input data. Finally, the data are aggregated
into a single value to make a decision. The case study indicates that our method is advantageous to make
decisions objectively while inheriting the characteristics of network structures.

1. Introduction taken into account to make decisions (Cano-Ruiz and McRae, 1998).
However, at the time the methodologies were mostly dealing with
Green products design (Liu et al., 2018) and renewable energy one object and less effective when handling multiple factors (Zhou
management (Zhang and Xu, 2019; Tsai, 2016) are two vital topics in et al., 2018). Based on life cycle assessment technology, Azapagic and
environment protection and ecological safety. Selecting and designing Clift studied a great many process selections to direct decision-making
proper strategies for green process has aroused great concerns and issues (Azapagic and Clift, 1998, 1999). Fu et al. also proposed a multi-
studies, especially in ecology assessment (Gandhi et al., 2018), risk objective framework under uncertainty to evaluate options in designing
analysis and perception (Danforth et al., 2019; Torres et al., 2017), process (Fu et al., 2000). Besides, the entropy function (Wen and Deng,
and produced water management (Hu et al., 2018; Zemlick et al., 2020b) and evidence theory (Li and Wei, 2019) are widely used to
2018). However, it is sometimes not easy to make reasonable decisions handle with uncertainty in decision-making. Furthermore, in order to
in the real world (Scholl et al., 2018). In a system, various factors solve the impact of large amounts of data when making decisions, many
interact with each other, increasing its complexity and uncertainty (Liu data aggregation methodologies were presented. The ordered weighted
and Zhang, 2019; Gao et al., 2019). In addition, since these different averaging (OWA) operator, which was proposed by Yager, is effective
elements often belong to different kinds of fields, how to deal with to fuse data under uncertainty (Yager, 1988; Yusoff et al., 2018).
their massive data is also an unsolved problem. For instance, there Afterwards, scientists studied the determination of weights for OWA
are four fundamental indexes in produced water management, which operator using maximum entropy methodology (Fullér and Majlender,
are feasible technique, cost index, damages to the environment, and 2001) and fuzzy theory (Sadiq and Tesfamariam, 2008). In addition
treatment effect (Sadiq et al., 2005). All the indicators are different to these traditional data processing methods, many methods of data
in their attributes and units, leading to a great deal of inconvenience aggregation have been proposed by researchers based on network
when building the connections among different indicators (Zhang et al., theories. For example, Jiang et al. conducted a research on power
2013). Therefore, under the influence of various indexes, how to make average (P-A) operator (Yager, 2001) to aggregate financial values
a decision for water management is still an open question (Bereskie by network structures (Jiang et al., 2016). Besides, Xu et al. utilized
et al., 2018). visibility graph method (Lacasa et al., 2012) for data aggregation and
Researchers have done a lot of work handling decision-making finally established a weighted network (Xu et al., 2018).
problems. In the early stage, Cano-Ruiz and McRae fully reviewed Although these methods helped a lot in decision-making and data
the technologies of process design, where environment factors were aggregation, they still have some limitations. For example, although

✩ No author associated with this paper has disclosed any potential or pertinent conflicts which may be perceived to have impending conflict with this work.
For full disclosure statements refer to https://doi.org/10.1016/j.engappai.2020.103556.
∗ Corresponding author.
E-mail address: dengentropy@uestc.edu.cn (Y. Deng).

https://doi.org/10.1016/j.engappai.2020.103556
Received 12 August 2019; Received in revised form 6 February 2020; Accepted 10 February 2020
Available online xxxx
0952-1976/© 2020 Elsevier Ltd. All rights reserved.
S. Mao, Y. Deng and D. Pelusi Engineering Applications of Artificial Intelligence 91 (2020) 103556

Fig. 1. Example of time series data and visibility graph: The time series data (7 data values) are mapped into graphs. The visibility graph contains the nodes corresponding to
time series data, and the edges between nodes.

Fig. 2. The ordered data are plotted in by vertical bars and transformed into OVG.

Fu et al.’s method succeeded in multi-attribute decision-making, how case, the sequence data are converted into network and we can take
to address massive amounts of data remained unsettled. Besides, the advantage of topology of networks to establish the relationships among
OWA and P-A operators are able to fuse data effectively but the weight different factors and deal with massive data.
determination method varies a lot and it is still open question (Song Inspired by previous work, this paper presents a new method of al-
and Deng, 2019). Each time an artificial parameter is necessary when ternatives selection for water management by network theory analysis.
data values are aggregated by OWA operator (Fullér and Majlender, Firstly, the discrete quantitative/qualitative input data values of each
2001), resulting in subjectivity and complexity. The network-based indicator are transformed into ordered visibility graphs. On the basis of
aggregation technology, like ordered visibility graph average (OVGA) considering node distance and node degree, a support function based
aggregation operator (Jiang et al., 2017), takes aggregation process on network structure is proposed to determine the weight distribution
only by node distance parameter in the network, which may cause among data. The case study shows the most appropriate alternative is
information loss and not take effect in other complex situations. There- down hole separation in the evaluation process of water management.
The proposed method is advantageous to make decisions objectively
fore, it is urgent to develop a comprehensive decision-making method
and retain more aggregation information.
applicable to different cases where a mass of data is involved.
We organize this paper as the following parts. Section 2 is the in-
Complex network has been studied for a long time, which has
troduction of visibility graph theory and support function. In Section 3,
the advantages of establishing relations in complex systems (Liang
the method proposed in this paper is introduced in detail. Section 4 is
et al., 2018; Wang et al., 2018b). Scientists studied network theories
the case study of water management and Section 5 is the conclusion of
in depth and applied them in spatial cooperation (Li et al., 2019), this paper.
resolving social dilemmas (Liu et al., 2019; Wang et al., 2018a), and
influential nodes identification (Wen and Deng, 2020a; Zhao et al., 2. Preliminaries
2020b), which are all received good performance. A novel topic, called
visibility graph, is a hot topic in recent years. The visibility graph 2.1. The visibility graph
(VG) was proposed by Lacasa et al. to map data values in time order
into graphs (Lacasa et al., 2008). Once this theory was put forward, it The visibility graph (VG) methodology, proposed by Lacasa et al.
was widely studied and applied to signal detection (Cai et al., 2019; (2008), is a network approach to transform time series data into
Gao et al., 2017), temporal irreversibility (Xiong et al., 2019b), and graphs (Li et al., 2016). The visibility graph preserves the information
time series analysis (Liu and Deng, 2019; Czechowski et al., 2016; of data in its network topology (Lacasa and Just, 2018) and its limited
Zhao et al., 2020a). It should point out that the order information and penetrability is also an crucial property (Wang et al., 2018c; Li et al.,
network structure are reserved in the transformation process. In this 2018). The visibility algorithm is defined as follows.

2
S. Mao, Y. Deng and D. Pelusi Engineering Applications of Artificial Intelligence 91 (2020) 103556

Table 1
14 best available technologies for produced water.
Technologies Alternatives
Physical separation technologies
Floatation 𝐴1
Sparging 𝐴2
Coalescence 𝐴3
Enhanced separation technologies
Hydrocyclones 𝐴4
PECT-F and/or Mare’s Tail 𝐴5
Centrifuges 𝐴6
Polishing technologies
MPPE 𝐴7
Adsorption 𝐴8
C-Tour 𝐴9
Membranes 𝐴10
Steam stripping 𝐴11
Biological 𝐴12
Produced water re-injection 𝐴13
Down hole separation 𝐴14

Definition 2.1. (𝑡𝑘 , 𝑥𝑘 ) is a set of time series data, which is located


between (𝑡𝑖 , 𝑥𝑖 ) and (𝑡𝑗 , 𝑥𝑗 ). The two sequential data are visible if (𝑡𝑘 , 𝑥𝑘 )
satisfies (Lacasa et al., 2008):
𝑥𝑖 − 𝑥𝑗 𝑥 − 𝑥𝑘
> 𝑖 . (1)
𝑡𝑖 − 𝑡𝑗 𝑡𝑖 − 𝑡𝑘
Fig. 1 describes the visibility algorithm more vividly. As can be seen,
a bar is linked if it can be seen by other bars in the histogram. The
right part of the diagram is the visibility graph, showing the nodes
corresponding to time series data, and the edges between nodes.
For example, (𝑡1 , 𝑥1 ) and (𝑡3 , 𝑥3 ) are visible, while (𝑡1 , 𝑥1 ) and (𝑡4 , 𝑥4 )
are invisible as they are blocked by (𝑡3 , 𝑥3 ). The visibility graph has the
following properties: Fig. 3. The evaluation indexes of each alternative.

(1) Connective: Adjacent nodes are directly connected.


(2) Non-directional: The links between two nodes are Non-
3. The proposed method
directional.
(3) Invariant: The relationships among nodes are invariant by scal-
In this section, inspired by the visibility graph and support function,
ing of the histogram.
we proposed a new method to determine the supports for data. Then
2.2. The support function an example is provided for illustrative purpose.

The support function is used in the power average (P-A) operator 3.1. The ordered visibility graph
to determine the supports for data (Yager, 2001). Due to the efficiency
of P-A operator in data aggregation (Liu and Teng, 2018; Jiang et al., In Section 2.1, the VG method maps time series data into graphs,
2015), it has been widely used in decision-making (Song et al., 2017; which means the values are in time order. However, not all datasets
Liu and Liu, 2018; Xiong et al., 2019a). The power average operator are chronological in real life, like weight and height datasets. In order
and the support function are defined as follows. to eliminate the influence of time factor, the original data are sorted
first, and then they are mapped into an ordered visibility graph (OVG).
Definition 2.2. The P-A operator is applied in data aggregation, and
it is defined by Yager (2001) Definition 3.1. 𝑂 = {𝑜1 , 𝑜2 , … , 𝑜𝑛 } is a set of ordered data, where 𝑜𝑖
∑𝑛
(𝐹 (𝑥𝑖 ) + 1)𝑥𝑖 is the ith largest data in 𝑂. Similarly, the OVG algorithm is defined as
𝑃 − 𝐴(𝑥1 , 𝑥2 , … , 𝑥𝑛 ) = ∑𝑖=1
𝑛 , if (𝑘, 𝑜𝑘 ) is a set of ordered data, which is located between (𝑖, 𝑜𝑖 ) and
𝑖=1 (𝐹 (𝑥𝑖 ) + 1)
(2) (𝑗, 𝑜𝑗 ), then the two ordered data are visible if (𝑘, 𝑜𝑘 ) satisfies:
∑𝑛
𝐹 (𝑥𝑖 ) = 𝑆𝑢𝑝(𝑥𝑖 , 𝑥𝑗 ) 𝑜𝑖 − 𝑜𝑗 𝑜𝑖 − 𝑜𝑘
𝑗=1,𝑗≠𝑖 > . (3)
𝑖−𝑗 𝑖−𝑘
where 𝑆𝑢𝑝(𝑥𝑖 , 𝑥𝑗 ) means the support of 𝑥𝑖 from 𝑥𝑗 , and 𝐹 (𝑥𝑖 ) denotes
The OVG criteria is set by the order of data as well as numerical
all the argument values’ supports for 𝑥𝑖 .
values, which is supposed to inherit more information when mapping
As can be seen, the support function, i.e. 𝑆𝑢𝑝(𝑥𝑖 , 𝑥𝑗 ) is such a data into graphs. Likewise, the OVG has the following properties:
significant element when aggregating data are aggregated. However, connective, non-directional, and invariant.
how to determine the support function is still an open question. The
properties of support function are as follows: 3.2. Determination of support function
(1) 𝑆𝑢𝑝(𝑥𝑖 , 𝑥𝑗 ) ∈ [0, 1];
In Section 3.1, the ordered data are transformed into network. In
(2) 𝑆𝑢𝑝(𝑥𝑖 , 𝑥𝑗 ) = 𝑆𝑢𝑝(𝑥𝑗 , 𝑥𝑖 );
| | this section, the support function is determined based on the following
(3) 𝑆𝑢𝑝(𝑥𝑖 , 𝑥𝑗 ) ≥ 𝑆𝑢𝑝(𝑥𝑎 , 𝑥𝑏 ) if |𝑥𝑖 − 𝑥𝑗 | < ||𝑥𝑎 − 𝑥𝑏 ||.
| | two crucial characteristics of the network structure:

3
S. Mao, Y. Deng and D. Pelusi Engineering Applications of Artificial Intelligence 91 (2020) 103556

Table 2
The discrete quantitative/qualitative input data value of each indicator.
Environmental indicators(EnI) Technical feasibility(TFI)
Alternatives
RD GW AP CT SM EO EF ST CM
𝐴1 0.03 0.10 0.15 0.30 0.20 0.80 0.85 1.00 0.40
𝐴2 0.35 0.15 0.20 0.20 0.20 0.80 0.90 1.00 0.50
𝐴3 0.40 0.20 0.20 0.30 0.30 0.70 0.90 0.80 0.40
𝐴4 0.55 0.15 0.30 0.35 0.25 0.90 0.95 1.00 0.80
𝐴5 0.60 0.20 0.30 0.35 0.30 0.80 0.95 0.75 0.60
𝐴6 0.45 0.20 0.20 0.15 0.20 0.80 0.95 1.00 0.70
𝐴7 0.70 0.40 0.50 0.40 0.30 0.60 0.90 0.60 0.30
𝐴8 0.60 0.35 0.40 0.35 0.50 0.60 0.90 0.60 0.30
𝐴9 0.80 0.60 0.55 0.50 0.30 0.70 0.90 0.50 0.40
𝐴10 0.70 0.50 0.30 0.40 0.80 0.60 0.90 0.50 0.30
𝐴11 1.00 1.00 1.00 1.00 0.30 0.70 0.90 0.50 0.20
𝐴12 0.30 0.35 0.30 0.20 1.00 0.90 0.90 0.60 0.10
𝐴13 0.90 0.60 0.45 0.30 0.15 0.90 0.95 0.90 0.80
𝐴14 0.85 0.60 0.30 0.35 0.15 0.90 0.90 0.70 0.90
Treatment efficiency indicators(TI) Cost indicators(CI)
Alternatives
DO BTEX NPD PAH HM NORM WC OM CC
𝐴1 0.60 0.00 0.00 0.20 0.00 0.00 0.10 0.20 0.30
𝐴2 0.60 0.00 0.00 0.30 0.00 0.00 0.10 0.25 0.35
𝐴3 0.70 0.00 0.00 0.40 0.00 0.00 0.15 0.15 0.30
𝐴4 0.90 0.00 0.00 0.50 0.00 0.00 0.20 0.20 0.25
𝐴5 0.95 0.00 0.00 0.70 0.00 0.00 0.25 0.20 0.30
𝐴6 0.95 0.00 0.00 0.50 0.00 0.00 0.20 0.20 0.35
𝐴7 0.90 0.90 0.70 0.90 0.30 0.30 0.15 0.35 0.40
𝐴8 0.95 0.95 0.70 0.80 0.10 0.10 0.15 0.40 0.50
𝐴9 0.95 0.30 0.80 0.90 0.50 0.60 0.20 0.35 0.50
𝐴10 0.95 0.90 0.80 0.90 0.60 0.70 0.15 0.35 0.50
𝐴11 0.60 0.95 0.00 0.00 0.00 0.00 0.30 0.40 0.60
𝐴12 0.90 0.95 0.60 0.70 0.40 0.10 0.10 0.20 0.50
𝐴13 0.90 0.90 0.80 0.80 0.50 0.50 0.25 0.20 0.60
𝐴14 0.90 0.95 0.80 0.90 0.60 0.50 0.20 0.60 0.70

(1) Node degree: In the graph theory, node degree refers to the number Example 3.1. 𝑂 = {75, 65, 60, 60, 40, 35, 30} is a set of ordered data. The
of edges associated with the node. The more edges connected to a data are plotted in a histogram and transformed into OVG, as shown in
node, the more important that node is as it carries more network Fig. 2.
information. For example, in Fig. 1, there are two points connected As for node (1, 75), it is associated to (2, 65), (3, 60), and (4, 60),
to node (𝑡1 , 𝑥1 ) directly while five nodes are connected to node and the degrees of these three points are 3, 3, and 6, respectively. By
(𝑡3 , 𝑥3 ). Therefore, the degree of node (𝑡1 , 𝑥1 ) and node (𝑡3 , 𝑥3 ) are Eqs. (4) and (5), the total supports of node (1, 75) is calculated by
2 and 5, which implies node (𝑡3 , 𝑥3 ) is more important than node 𝑘2 𝑘 𝑘 3 3 6
(𝑡1 , 𝑥1 ) in this network. 𝐹 (𝑥1 ) = + 3 + 4 = + + = 6.5.
𝑑12 𝑑13 𝑑14 1 2 3
(2) Node distance: In the method of Yager (2001), the closer the
argument values are, the higher the support among them. Simi- Likewise, the total support of each node in set 𝑂 can be obtained.
larly, in Fig. 1, the closer the two connected vertical bars are, the Then the aggregated result of set 𝑂, denoted by Agg, can be obtained
higher they support each other. In the network structure, the node by Eq. (2).
distance is a crucial factor to reflect the interrelationship between 𝐴𝑔𝑔(75, 65, … , 30)
two nodes. In the visibility graph, the node distance is defined as 75 × (1 + 6.5) + 65 × (1 + 9) + ⋯ + 30 × (1 + 5)
𝑑 = |𝑡𝑛 − 𝑡𝑚 |. For instance, the distance between node (𝑡1 , 𝑥1 ) and = = 53.44.
(1 + 6.5 + 1 + 9 + ⋯ + 1 + 5)
node (𝑡3 , 𝑥3 ) is equal to 2.

It can be seen that in the network structure, the higher the degree
Algorithm 1 The proposed network methodology
of two nodes and the closer the distance between them, then the higher
support between them. Based on the mentioned considerations, the Input: The ordered dataset 𝑂
support function, i.e. 𝑆𝑢𝑝(𝑥𝑖 , 𝑥𝑗 ) is defined as follows and Algorithm Output: The aggregate value 𝐴𝑔𝑔(𝑂)
1 illustrates the proposed method in a more vivid way. 1: OVG algorithm: From ordered data to ordered visibility graph
2: for ∀ pair of connected nodes (𝑥𝑖 , 𝑥𝑗 ) do
Definition 3.2. In the ordered visibility graph, if 𝑥𝑖 is connected to 𝑥𝑗 , 3: Support function 𝑆𝑢𝑝(𝑥𝑖 , 𝑥𝑗 ) = 𝑘𝑖 /𝑑𝑖𝑗
4: Determine the supports of node 𝑥𝑖
the support of 𝑥𝑖 from 𝑥𝑗 , denoted by 𝑆𝑢𝑝(𝑥𝑖 , 𝑥𝑗 ), is defined by
5: end for
𝑘𝑗 6: Aggregated results 𝐴𝑔𝑔(𝑂)
𝑆𝑢𝑝(𝑥𝑖 , 𝑥𝑗 ) = . (4)
𝑑𝑖𝑗
where 𝑘𝑗 is the degree of node 𝑥𝑗 , and 𝑑𝑖𝑗 denotes the distance between
𝑥𝑖 and 𝑥𝑗 . 𝐹 (𝑥𝑖 ) denotes all the argument values’ supports for 𝑥𝑖 and it
is determined by 4. Case study


𝑛
𝐹 (𝑥𝑖 ) = 𝑆𝑢𝑝(𝑥𝑖 , 𝑥𝑗 ). (5) In this section, the proposed method is applied to select appropriate
𝑗=1,𝑗≠𝑖 alternatives for produced water management.

4
S. Mao, Y. Deng and D. Pelusi Engineering Applications of Artificial Intelligence 91 (2020) 103556

Fig. 4. The ordered visibility graph of 𝐴1 .

Fig. 5. Aggregated results obtained by different methods.

4.1. Application in water management resource depletion (RD), global warming (GW), air pollution (AP),
critical water mass (CT), and solid waste mass (SM). The indexes of ease
Green products, processes, and waste management policies are es- of operation (EO), efficiency (EF), status of technology (ST) and control
sential to environmental conservation as well as sustainable develop- measures (CM) are included in technical feasibility indicators (TFI). The
ment. There is a mass of data to be processed when choosing and
cost indicators (CI) are working capital (WC), operation & maintenance
designing green crafts and products (Sadiq et al., 2005). The proposed
(OM), and capital cost (CC). Treatment indicators (TI) contain dissolved
method is effective to aggregate varieties of influence factors to select
oil (DO), benzene, toluene, ethyl benzene, and xylene (BTEX), Poly-
appropriate alternatives under uncertainty.
Table 1 presents 14 best available technologies (BAT) for treating cyclic aromatic hydrocarbons (PAH), naphthalene, phenanthrene, and
produced water. Fig. 3 shows four primary indicators for evaluation dibenzothiophene (NPD), heavy metal (HM), and naturally occurring
of the technologies, which are environment, treatment, technical fea- radioactive material (NORM). Table 2 lists the input data value of each
sibility, and cost. Specifically, environmental indicators (EnI) include indicator (Sadiq et al., 2005).

5
S. Mao, Y. Deng and D. Pelusi Engineering Applications of Artificial Intelligence 91 (2020) 103556

Table 3
The weights of each indicator for different alternatives.
𝑤1 𝑤2 𝑤3 𝑤4 𝑤5 𝑤6 𝑤7 𝑤8 𝑤9 𝑤10 𝑤11 𝑤12 𝑤13 𝑤14 𝑤15 𝑤16 𝑤17 𝑤18
𝐴1 0.0320 0.0524 0.0720 0.0764 0.0712 0.0596 0.0560 0.0359 0.0510 0.0643 0.0569 0.0560 0.0665 0.0555 0.0455 0.0523 0.0564 0.0402
𝐴2 0.0362 0.0625 0.0749 0.0718 0.0661 0.0364 0.0525 0.0623 0.0656 0.0543 0.0705 0.0738 0.0585 0.0537 0.0306 0.0384 0.0525 0.0394
𝐴3 0.0454 0.0659 0.0666 0.0652 0.0629 0.0536 0.0542 0.0370 0.0558 0.0628 0.0518 0.0642 0.0670 0.0669 0.0481 0.0462 0.0499 0.0366
𝐴4 0.0261 0.0354 0.0457 0.0650 0.0787 0.0679 0.0697 0.0580 0.0597 0.0566 0.0718 0.0697 0.0745 0.0604 0.0325 0.0369 0.0517 0.0397
𝐴5 0.0334 0.0457 0.0604 0.0607 0.0601 0.0669 0.0739 0.0768 0.0576 0.0570 0.0616 0.0635 0.0628 0.0620 0.0483 0.0390 0.0409 0.0295
𝐴6 0.0236 0.0397 0.0701 0.0796 0.0858 0.0598 0.0748 0.0699 0.0397 0.0565 0.0616 0.0647 0.0666 0.0552 0.0298 0.0347 0.0491 0.0387
𝐴7 0.0096 0.0128 0.0478 0.0528 0.0638 0.0600 0.0670 0.0743 0.0626 0.0383 0.0629 0.0819 0.0749 0.0628 0.0706 0.0763 0.0530 0.0287
𝐴8 0.0119 0.0437 0.0663 0.0856 0.0669 0.0557 0.0712 0.0708 0.0596 0.0631 0.0557 0.0596 0.0593 0.0658 0.0567 0.0418 0.0378 0.0285
𝐴9 0.0201 0.0302 0.0428 0.0570 0.0710 0.0714 0.0637 0.0740 0.0715 0.0503 0.0644 0.0800 0.0744 0.0626 0.0553 0.0514 0.0397 0.0201
𝐴10 0.0274 0.0311 0.0423 0.0512 0.0578 0.0599 0.0578 0.0592 0.0534 0.0634 0.0489 0.0719 0.0807 0.0786 0.0712 0.0667 0.0515 0.0267
𝐴11 0.0136 0.0181 0.0271 0.0452 0.0633 0.0671 0.0769 0.0678 0.0746 0.0720 0.0588 0.0618 0.0793 0.0708 0.0407 0.0475 0.0659 0.0496
𝐴12 0.0319 0.0412 0.0316 0.0585 0.0628 0.0633 0.0589 0.0731 0.0791 0.0604 0.0719 0.0719 0.0695 0.0518 0.0519 0.0403 0.0437 0.0384
𝐴13 0.0400 0.0496 0.0469 0.0478 0.0565 0.0597 0.0496 0.0689 0.0709 0.0662 0.0634 0.0625 0.0758 0.0629 0.0551 0.0496 0.0441 0.0303
𝐴14 0.0364 0.0452 0.0427 0.0435 0.0514 0.0602 0.0661 0.0672 0.0602 0.0619 0.0502 0.0703 0.0747 0.0703 0.0502 0.0527 0.0535 0.0434

Table 4
The comparison of aggregation results by different methods.
Fullér and Majlender (2001) Wang et al. (2015) Proposed method
𝛼 = 0.1 Order 𝛼 = 0.4 Order 𝛼 = 0.5 Order 𝛼 = 0.6 Order 𝛼 = 0.9 Order Results Order Results Order
𝐴1 0.0161 14 0.2035 14 0.2906 14 0.3885 14 0.7893 13 0.3215 14 0.2958 14
𝐴2 0.0261 13 0.238 13 0.3278 12 0.4263 12 0.8122 12 0.3769 11 0.3486 12
𝐴3 0.0303 11 0.2438 12 0.3278 13 0.4173 13 0.7447 14 0.3762 12 0.3441 13
𝐴4 0.0269 12 0.2582 11 0.3556 11 0.4607 11 0.8357 11 0.3606 13 0.4027 11
𝐴5 0.039 9 0.3027 9 0.4028 9 0.507 9 0.8483 10 0.4294 9 0.4161 9
𝐴6 0.0315 10 0.2774 10 0.3806 10 0.492 10 0.8843 6 0.4042 10 0.4058 10
𝐴7 0.2557 4 0.4579 5 0.5333 5 0.6118 6 0.8579 9 0.5165 6 0.4959 7
𝐴8 0.1722 6 0.4285 6 0.5139 7 0.6004 8 0.8756 7 0.5346 5 0.5277 5
𝐴9 0.2952 1 0.5048 4 0.575 4 0.6465 4 0.875 8 0.5817 4 0.5695 4
𝐴10 0.2788 2 0.5254 3 0.6028 3 0.6789 3 0.8927 5 0.6014 3 0.581 3
𝐴11 0.0509 8 0.4023 8 0.525 6 0.6462 5 0.9672 1 0.4743 8 0.4798 8
𝐴12 0.1415 7 0.4059 7 0.5056 8 0.6078 7 0.9175 2 0.512 7 0.4963 6
𝐴13 0.2499 5 0.5459 2 0.6333 2 0.7152 2 0.9082 4 0.6581 2 0.6416 2
𝐴14 0.2712 3 0.5725 1 0.6556 1 0.7322 1 0.9095 3 0.6588 1 0.6515 1

Fig. 6. Aggregated results obtained by Fuller & Majlender’s method. When 𝛼 = 0.4, 𝛼 = 0.5, and 𝛼 = 0.6, the trends of change are similar. However, the results vary a lot when 𝛼
takes extreme values like 0.1 and 0.9.

Take 𝐴1 as an example, we can get its OVG, shown in Fig. 4 by Firstly, we compare our method with Fuller & Majlender’s method.
Eq. (3). Then we calculate the support of each indicator as well as As shown in Fig. 6, the final results obtained by Fuller & Majlender’s
their weight distribution of 𝐴1 by Eqs. (4) and (5). Finally, we get the method are different when 𝛼 changes. The curves vary a lot, especially
aggregated result of 𝐴1 , which is calculated by Eq. (2). By taking the
when the value of 𝛼 is taken extreme values like 0.1 or 0.9. It is because
same operation, we can get all the OVGs of other alternatives, and their
weight distributions are shown in Table 3. 𝛼 is a parameter, which is set artificially according to the experience
of experts. Therefore, the aggregate results are subjective, leading to
4.2. Comparison and analysis the loss of objectivity when making decisions. Besides, whenever we
conduct an experiment, we need to set the parameters artificially,
Table 4 lists the aggregation results obtained by the proposed which adds levels of complexity. More importantly, it is hard to find out
method, together with other two methods proposed by Fullér and
the most appropriate parameter values under different experimental
Majlender (2001) and Wang et al. (2015). As can be seen, the most
appropriate alternative is 𝐴14 , i.e. down hole separation, in the evalu- conditions. So it is necessary to develop an aggregation method that
ation as it ranks 1st. The results of aggregation by three methods are removes subjective factors. In terms of this view, the proposed method
plotted in Fig. 5. is more objective as it aims to build the relationships among data by

6
S. Mao, Y. Deng and D. Pelusi Engineering Applications of Artificial Intelligence 91 (2020) 103556

Fig. 7. The details of the aggregated results obtained by three methods (𝛼 = 0.5). The curve of the proposed method (red curve) is between Fuller & Majlender’s method (𝛼 = 0.5)
and Wang et al.’s method (see in the square boxes). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

their supports. Moreover, it is more efficient because it is fully data- Acknowledgment


driven without setting parameters every time, preserving the objectivity
of the original data. The work is partially supported by National Natural Science Foun-
Besides, we compare our method with Wang et al.’s method, which dation of China (Grant No. 61973332).
is also based on network theories. As shown in Fig. 7, the curve of the
proposed method (red curve) is between Fuller & Majlender’s method References
(𝛼 = 0.5) and Wang et al.’s method (the square boxes in Fig. 7). The
top three proper alternatives (𝐴14 , 𝐴13 , and 𝐴10 ) are the same by the Azapagic, A., Clift, R., 1998. Linear programming as a tool in life cycle assessment.
three methods. In addition, the trend of three curves is consistent, indi- Int. J. Life Cycle Assess. 3 (6), 305–316.
Azapagic, A., Clift, R., 1999. Life cycle assessment as a tool for improving process
cating our method is feasible to aggregate results and make the correct
performance: a case study on boron products. Int. J. Life Cycle Assess. 4 (3), 133.
assessment when selecting alternatives. It should be pointed out that Bereskie, T., Delpla, I., Rodriguez, M.J., Sadiq, R., 2018. Drinking-water management
although Wang et al. also adopted network method to make decisions, in Canadian provinces and territories: a review and comparison of management
they only consider the degree of nodes in the network to determine the approaches for ensuring safe drinking water. Water Policy 20 (3), 565–596.
weights of data, which may cause information loss when aggregating Cai, Q., Gao, Z.-K., Yang, Y.-X., Dang, W.-D., Grebogi, C., 2019. Multiplex limited
penetrable horizontal visibility graph from EEG signals for driver fatigue detection.
values. Our approach not only considers the degree of nodes, but also
Int. J. Neural Syst. 29 (5), 1850057.
takes the effect of node distance into account. Therefore, the proposed Cano-Ruiz, J., McRae, G., 1998. Environmentally conscious chemical process design.
method not only can get accurate results, but also effectively retains Ann. Rev. Energy Environ. 23 (1), 499–536.
more data and network structure features. Czechowski, Z., Lovallo, M., Telesca, L., 2016. Multifractal analysis of visibility
graph-based Ito-related connectivity time series. Chaos 26 (2), 023118.
5. Conclusion Danforth, C., McPartland, J., Blotevogel, J., Coleman, N., Devlin, D., Olsgard, M.,
Parkerton, T., Saunders, N., 2019. Alternative management of oil and gas produced
water requires more research on its hazards and risks. In: Integrated Environmental
In this paper, we proposed a novel method to select the most Assessment and Management. Wiley Online Library.
appropriate alternatives based on network structures. The method first Fu, Y., Diwekar, U.M., Young, D., Cabezas, H., 2000. Process design for the environ-
sorts the input data and then map them into ordered visibility graphs. ment: A multi-objective framework under uncertainty. Clean Prod. Process. 2 (2),
By taking the node distance and node degree into consideration, we 92–107.
Fullér, R., Majlender, P., 2001. An analytic approach for obtaining maximal entropy
proposed a network-based methodology to determine the support func-
OWA operator weights. Fuzzy Sets and Systems 124 (1), 53–57.
tion. In the case study, 𝐴14 , i.e. down hole separation is selected as the Gandhi, H., Sadiq, R., Hu, G., Hewage, K., 2018. Ecological risk assessment of accidental
most appropriate alternative for produced water management. release of flowback water: A conceptual framework. Hum. Ecol. Risk. Assess. 24
The advantages of the proposed method are as follows. On one (2), 398–426.
hand, the method explores the relationships among input data to Gao, Z.-K., Cai, Q., Yang, Y.-X., Dong, N., Zhang, S.-S., 2017. Visibility graph
from adaptive optimal kernel time-frequency representation for classification of
determine the support function, so it is fully data-driven and no ar-
epileptiform EEG. Int. J. Neural Syst. 27 (04), 1750005.
tificial parameters are required for decision-making, which can avoid Gao, J., Xu, Z., Liang, Z., Liao, H., 2019. Expected consistency-based emergency
subjectivity and randomness. On the other hand, it considers the factors decision making with incomplete probabilistic linguistic preference relations.
of node degree and node distance in the network to obtain reasonable Knowl.-Based Syst. 176, 15–28.
results as well as retaining more original information and network Hu, Y., Mackay, E., Vazquez, O., Ishkov, O., et al., 2018. Streamline simulation of
barium sulfate precipitation occurring within the reservoir coupled with analyses
features, which is conducive to further researches in decision-making.
of observed produced-water-chemistry data to aid scale management. SPE Prod.
In summary, the proposed network operator is a general method for Oper. 33 (01), 85–101.
data aggregation because it removes subjectivity as well as conserving Jiang, L., Liu, H., Cai, J., 2015. The power average operator for unbalanced linguistic
more network structures. In the future, we will continue to improve our term sets. Inf. Fusion 22, 85–94.
method in handling larger amounts of data and explore its theoretical Jiang, W., Wei, B., Tang, Y., Zhou, D., 2017. Ordered visibility graph average
aggregation operator: An application in produced water management. Chaos 27
and practical applications in data aggregation for complex systems such
(2), 023117.
as finance, health and industry. Jiang, W., Wei, B., Zhan, J., Xie, C., Zhou, D., 2016. A visibility graph power averaging
aggregation operator: A methodology based on network analysis. Comput. Ind. Eng.
CRediT authorship contribution statement 101, 260–268.
Lacasa, L., Just, W., 2018. Visibility graphs and symbolic dynamics. Physica D 374–375,
35–44.
Shengzhong Mao: Conceptualization, Methodology, Software, In-
Lacasa, L., Luque, B., Ballesteros, F., Luque, J., Nuño, J.C., 2008. From time series to
vestigation, Writing - original draft. Yong Deng: Resources, Writing complex networks: The visibility graph. Proc. Natl. Acad. Sci. 105 (13), 4972–4975.
- review & editing, Supervision. Danilo Pelusi: Resources, Writing - Lacasa, L., Nunez, A., Roldán, É., Parrondo, J.M., Luque, B., 2012. Time series
review & editing. irreversibility: a visibility graph approach. Eur. Phys. J. B 85 (6), 217.

7
S. Mao, Y. Deng and D. Pelusi Engineering Applications of Artificial Intelligence 91 (2020) 103556

Li, Z., Jia, D., Guo, H., Geng, Y., Shen, C., Wang, Z., Li, X., 2019. The effect of Wang, Z., Jusup, M., Shi, L., Lee, J.-H., Iwasa, Y., Boccaletti, S., 2018a. Exploiting
multigame on cooperation in spatial network. Appl. Math. Comput. 351, 162–167. a cognitive bias promotes cooperation in social dilemma experiments. Nature
Li, X., Sun, M., Gao, C., Han, D., Wang, M., 2018. The parametric modified limited Commun. 9 (1), 2954.
penetrable visibility graph for constructing complex networks from time series. Wang, J., Li, C., Xia, C., 2018b. Improved centrality indicators to characterize the nodal
Physica A 492, 1097–1106. spreading capability in complex networks. Appl. Math. Comput. 334, 388–400.
Li, R., Wang, J., Yu, H., Deng, B., Wei, X., Chen, Y., 2016. Fractal analysis of the short Wang, H., Mo, H., Sadiq, R., Hu, Y., Deng, Y., 2015. Ordered visibility graph
time series in a visibility graph method. Physica A 450, 531–540. weighted averaging aggregation operator: A methodology based on network
Li, P., Wei, C., 2019. An emergency decision-making method based on DS evidence analysis. Comput. Ind. Eng. 88, 181–190.
theory for probabilistic linguistic term sets. Int. J. Disaster Risk Reduct. 37, 101178. Wang, M., Vilela, A.L., Du, R., Zhao, L., Dong, G., Tian, L., Stanley, H.E., 2018c. Exact
Liang, K., Dai, M., Shen, H., Wang, J., Wang, Z., Chen, B., 2018. L2-L synchronization results of the limited penetrable horizontal visibility graph associated to random
for singularly perturbed complex networks with semi-Markov jump topology. Appl. time series and its application. Sci. Rep. 8 (1), 5130.
Math. Comput. 321, 450–462. Wen, T., Deng, Y., 2020a. Identification of influencers in complex networks by local
Liu, F., Deng, Y., 2019. A fast algorithm for network forecasting time series. IEEE information dimensionality. Inform. Sci. 512, 549–562.
Access 7 (1), 102554–102560. Wen, T., Deng, Y., 2020b. The vulnerability of communities in complex networks: An
Liu, C., Guo, H., Li, Z., Gao, X., Li, S., 2019. Coevolution of multi-game resolves social entropy approach. Reliab. Eng. Syst. Saf. 196, 106782.
dilemma in network population. Appl. Math. Comput. 341, 402–407. Xiong, S.-H., Chen, Z.-S., Chang, J.-P., Chin, K.-S., 2019a. On extended power average
Liu, P., Liu, X., 2018. The neutrosophic number generalized weighted power averaging operators for decision-making: A case study in emergency response plan selection
operator and its application in multiple attribute group decision making. Int. J. of civil aviation. Comput. Ind. Eng. 130, 258–271.
Mach. Learn. Cybern. 9 (2), 347–358. Xiong, H., Shang, P., Hou, F., Ma, Y., 2019b. Visibility graph analysis of temporal
Liu, A., Liu, H., Xiao, Y., Tsai, S.-B., Lu, H., 2018. An empirical study on design partner irreversibility in sleep electroencephalograms. Nonlinear Dynam. 96 (1), 1–11.
selection in green product collaboration design. Sustainability 10 (1), 133. Xu, P., Zhang, R., Deng, Y., 2018. A novel visibility graph transformation of time series
Liu, P., Teng, F., 2018. Multiple attribute decision making method based on normal into weighted networks. Chaos Solitons Fractals 117, 201–208.
neutrosophic generalized weighted power averaging operator. Int. J. Mach. Learn. Yager, R.R., 1988. On ordered weighted averaging aggregation operators in multicriteria
Cybern. 9 (2), 281–293. decisionmaking. IEEE Trans. Syst. Man Cybern. 18 (1), 183–190.
Liu, P., Zhang, X., 2019. Some intuitionistic uncertain linguistic Bonferroni mean Yager, R.R., 2001. The power average operator. IEEE Trans. Syst. Man Cybern. A 31
operators and their application to group decision making. Soft Comput. 23 (11), (6), 724–731.
3869–3886. Yusoff, B., Merigó, J.M., Ceballos, D., Peláez, J.I., 2018. Weighted-selective aggregated
Sadiq, R., Khan, F.I., Veitch, B., 2005. Evaluating offshore technologies for produced majority-OWA operator and its application in linguistic group decision making. Int.
water management using greenpro-ia risk-based life cycle analysis for green and J. Intell. Syst. 33 (9), 1929–1948.
clean process selection and design. Comput. Chem. Eng. 29 (5), 1023–1039. Zemlick, K., Kalhor, E., Thomson, B.M., Chermak, J.M., Graham, E.J.S., Tidwell, V.C.,
Sadiq, R., Tesfamariam, S., 2008. Developing environmental indices using fuzzy 2018. Mapping the energy footprint of produced water management in New
numbers ordered weighted averaging (FN-OWA) operators. Stoch. Environ. Res. Mexico. Environ. Res. Lett. 13 (2), 024008.
Risk Assess. 22 (4), 495–505. Zhang, Y., Xu, Z., 2019. Efficiency evaluation of sustainable water management using
Scholl, I., LaRussa, A., Hahlweg, P., Kobrin, S., Elwyn, G., 2018. Organizational-and the HF-TODIM method. Int. Trans. Oper. Res. 26 (2), 747–764.
system-level characteristics that influence implementation of shared decision- Zhang, K., Zargar, A., Achari, G., Islam, M.S., Sadiq, R., 2013. Application of decision
making and strategies to address them–a scoping review. Implement. Sci. 13 (1), support systems in water management. Environ. Rev. 22 (3), 189–205.
40. Zhao, J., Mo, H., Deng, Y., 2020a. An efficient network method for time series
Song, Y., Deng, Y., 2019. A new soft likelihood function based on power ordered forecasting based on the DC algorithm and visibility relation. IEEE Access 8,
weighted average operator. Int. J. Intell. Syst. 34 (11), 2988–2999. 7598–7608.
Song, M., Jiang, W., Xie, C., Zhou, D., 2017. A new interval numbers power average Zhao, J., Wang, Y., Deng, Y., 2020b. Identifying influential nodes in complex networks
operator in multiple attribute decision making. Int. J. Intell. Syst. 32 (6), 631–644. from global perspective. Chaos Solitons Fractals.
Torres, L., Yadav, O.P., Khan, E., 2017. Perceived risks of produced water management Zhou, L., Wu, X., Xu, Z., Fujita, H., 2018. Emergency decision making for natural
and naturally occurring radioactive material content in North Dakota. J. Environ. disasters: An overview. Int. J. Disaster Risk Reduct. 27, 567–576.
Manag. 196, 56–62.
Tsai, S.-B., 2016. Using grey models for forecasting china’s growth trends in renewable
energy consumption. Clean Technol. Environ. Policy 18 (2), 563–571.

You might also like