Multi-Criteria Group Decision-Making Method Based on an Improved Single-Valued Neutrosophic Hamacher Weighted Average Operator and Grey Relational Analysis ()
1. Introduction
Multi-criteria group decision-making (MCGDM) refers to a decision-making process in which several decision-makers (DMs) compare and select finite alternatives by using specific methods according to the evaluation information of multiple criteria [1] [2] . Due to the complex uncertainty of environmental conditions and the cognitive limitations of DMs, traditional quantitative methods usually cannot accurately describe decision-making problems. Many scholars have focused on fuzzy sets (FSs) [3] [4] [5] and their extensions (e.g., intuitionistic fuzzy sets (IFSs) [6] [7] [8] , hesitant fuzzy sets (HFSs) [9] [10] [11] , picture fuzzy sets (PFSs) [12] [13] [14] ), and fuzzy rough sets (FRSs) and so on [15] [16] ) to solve this issue. However, compared those extensions, neutrosophic sets (NSs) [17] [18] [19] characterized by truth-membership, indeterminacy-membership, and falsity-membership function can better describe the fuzzy uncertainty of decision-making provided by DMs. Wang et al. [20] and Ye [21] extended the scope of membership degrees from the non-standard unit interval ]0-, 1+[ to standard unit interval [0, 1] and defined single-valued neutrosophic sets (SNSs) to further promote the application of NSs.
The single-valued neutrosophic MCGDM and multi-criteria decision-making (MCDM) methods have been investigated mainly from three aspects: aggregation operators [22] [23] [24] , information measures [25] [26] [27] [28] , and outranking relations [29] [30] [31] . The aggregation operator, which is fundamental for the MCGDM and MCDM methods, can effectively present decision-making results in most cases. Ye [21] defined the single-valued neutrosophic weighted averaging (SNWA) operator and developed the corresponding MCDM method; meanwhile, Peng et al. [22] developed the single-valued neutrosophic weighted algebraic averaging (SNWAA) and single-valued neutrosophic weighted Einstein averaging (SNEWA) operators; Garg [24] defined the single-valued neutrosophic Frank weighted averaging (SNFWA) operator and applied it to handle MCDM problems; moreover, Liu et al. [32] developed an MCGDM method based on the single-valued neutrosophic Hamacher weighted averaging (SNHWA) operator. Finally, Kilic et al. [33] developed a single-valued neutrosophic leanness assessment methodology based on SNWAA operator and Mishra et al. [34] proposed a CRiteria Importance through Intercriteria Correlation (CRITIC) method using the SNWAA operator, while Pani et al. developed a multi-objective optimization based on ratio analysis with the full multiplicative form (MULTIMOORA) method based on the SNWAA operator.
However, existing single-valued neutrosophic weighed averaging operators and their corresponding MCGDM and MCDM methods have the following limitations. 1) The SNWA [21] , SNWAA [22] , SNEWA [22] , SNFWA [23] , and SNHWA operators [30] may produce unreasonable results in the aggregation process (i.e., indeterminate aggregated terms in some cases). 2) The corresponding decision-making methods, including the MCDM method using the SNWA operator [21] , the MCDM method using the SNWAA and SNEWA operators [22] , the MCDM method using the SNFWA operator [24] , and the MCGDM method using the SNHWA operator [32] can also produce unreasonable decision-making results, which do not allow to the real decision-making.
In this paper, we present an improved SNHWA (ISNHWA) operator that was developed to exceed the limitations of existing aggregation operators. An improved single-valued neutrosophic MCGDM method based on the ISNHWA operator and grey relational analysis (GRA) is also proposed to overcome the limitations of existing MCGDM and MCDM methods based on aggregation operators.
The structure of this paper is organized as follows. Section 2 reviews related concepts, including SNSs, single-valued neutrosophic numbers (SNNs), the comparison method of SNNs, some existing single-valued neutrosophic weighted averaging operators, and GRA. Section 3 analyzes the limitations of existing aggregation operators. Section 4 describes the proposed ISNHWA operator and discusses its properties. Section 5 introduces a single-valued neutrosophic MCGDM method based on the ISNHWA operator and GRA. Section 6 provides some application examples to verify the effectiveness and feasibility of the proposed method. Finally, conclusions are drawn in Section 7.
2. Preliminaries
2.1. Single-Valued Neutrosophic Averaging Operators
In this section, related concepts, including SNSs, SNNs, the comparison method of SNNs, and some existing single-valued neutrosophic averaging operators, are reviewed.
Definition 1 [19] . Let X be a space of points (objects) with a generic element in X, denoted by x. An SNS in X is characterized as
S={〈x,tS(x),fS(x),kS(x)〉|x∈X} , where
tS(x) ,
fS(x) , and
kS(x) are the truth-membership (satisfying
0≤tS(x)≤1 ), indeterminacy-membership (satisfying
0≤fS(x)≤1 ), and falsity-membership (satisfy
0≤kS(x)≤1 ) respectively. If X has only one element, then
S=〈tS(x),fS(x),kS(x)〉 is called an SNN and is denoted by
S=〈t,f,k〉 .
Definition 2 [22] . Assuming that
S1 and
S2 are two SNNs, then their comparison can be described as follows:
1) If
p(S1)>p(S2) , then
S1≻S2 ;
2) If
p(S1)=p(S2) and
q(S1)>q(S2) , then
S1≻S2 ;
3) If
p(S1)=p(S2) and
q(S1)=q(S2) , then
S1=S2 .
Here
p(Si)=ti+1−fi+1−ki3 and
q(Si)=ti−ki
(i=1,2) denote the score and accuracy functions, respectively.
Definition 3. It is hypothesized that
Sj=〈tj,fj,kj〉(j=1,2,⋯,n) is a group of SNNs and
ωj the corresponding weight of
Sj ,
ωj∈[0,1] and
n∑j=1ωj=1 . In this case, the following single-valued neutrosophic weighted average operators can be defined.
1) SNWA operator [21] :
SNWAω(S1,S2,⋯,Sn)=〈1−n∏j=1(1−tj)ωj,1−n∏j=1(1−fj)ωj,1−n∏j=1(1−kj)ωj〉 . (1)
2) SNWAA operator [22] :
SNWAAω(S1,S2,⋯,Sn)=〈1−n∏j=1(1−tj)ωj,n∏j=1fωjj,n∏j=1kωjj〉 . (2)
3) SNEWA operator [22] :
SNEWAω(S1,S2,⋯,Sn)=〈n∏j=1(1+tj)ωj−n∏j=1(1−tj)ωjn∏j=1(1+tj)ωj+n∏j=1(1−tj)ωj,2n∏j=1fωjjn∏j=1(2−fj)ωj+n∏j=1fωjj,2n∏j=1kωjjn∏j=1(2−kj)ωj+n∏j=1kωjj〉. (3)
4) SNFWA operator [24] :
SNFWAω(S1,S2,⋯,Sn)=〈1−logλ(1+n∏j=1(λ1−tj−1)ωj),logλ(1+n∏j=1(λfj−1)ωj), logλ(1+n∏j=1(λkj−1)ωj)〉(λ>1). (4)
5) SNHWA operator [32] :
SNHWAω(S1,S2,⋯,Sn)=〈n∏j=1(1+(γ−1)tj)ωj−n∏j=1(1−tj)ωjn∏j=1(1+(γ−1)tj)ωj+(γ−1)n∏j=1(1−tj)ωj, γn∏j=1fωjjn∏j=1(1+(γ−1)(1−fj))ωj+(γ−1)n∏j=1fωjj, γn∏j=1kωjjn∏j=1(1+(γ−1)(1−kj))ωj+(γ−1)n∏j=1kωjj〉. (5)
In Equation (5),
γ>0 . If
γ=1 , then the SNHWA operator would be reduced to the SNWAA operator in Equation (2); if
γ=2 , then the SNHWA operator would be reduced to the SNEWA operator in Equation (3).
2.2. Grey Relational Analysis
GRA is a method to quantify the grey and fuzzy information association relationship between sequences, which is suitable to solve the problem with uncertainty information or insufficient information. The comparison sequence and the reference sequence are determined firstly, and then the correlation degree between the sequences can be obtained based on the fitting degree of the geometry between different sequence curves. The more similar the geometry, the greater the degree of correlation between them. The mathematical expression for the grey correlation coefficients is presented as follows [35] :
τi(k)=minimink|ε0(k)−εi(k)|+ςmaximaxk|ε0(k)−εi(k)||ε0(k)−εi(k)|+ςmaximaxk|ε0(k)−εi(k)| . (6)
where
ς∈(0,1) denotes the distinguishable coefficient, and
|ε0(k)−εi(k)| denotes the absolute distance between the comparison values and the reference values. For the study of the specific value of distinguishable coefficient, most scholars have verified the fluctuation hours of the sequence data through examples, and
ς should take a larger value in the range (0, 1); otherwise, when the sequence data fluctuates greatly, it should take a smaller value.
ς=0.5 is determined in this paper.
3. Limitations of Existing Single-Valued Neutrosophic Weighted Average Operators
In this section, the limitations of existing single-valued neutrosophic weighted average operators (i.e., the SNWA [21] , SNWAA [22] , SNEWA [22] , SNFWA [24] , and SNHWA [32] operators) are discussed.
1) It is assumed that
S1=〈1,0,0〉 ,
S2=〈0,1,0〉 , and
S3=〈0,0,1〉 are three SNNs, and that their weight vector is
ω=(0,0,1)T . Based on the SNWA operator in Equation (1),
SNWAω(S1,S2,S3)=〈1−00×10×11,1−10×00×11,1−10×10×01〉 can be obtained. Since indeterminate value (00) can be found in the aggregated results, the SNWA operator [21] is unreasonable in some cases.
2) It is assumed that
S1=〈1,0,0.2〉 ,
S2=〈0.3,0.1,0〉 , and
S3=〈0.2,0.3,0.5〉 are three SNNs, and that their weight vector is
ω=(0,0,1)T . Based on the SNWAA operator in Equation (2),
SNWAω(S1,S2,S3)=〈1−00×0.30×0.21,00×0.10×0.31,0.20×00×0.51〉 can be obtained. The three terms in the aggregated results all include 00, which is an indeterminate value: the SNWAA operator [22] has limitations in some cases.
3) It is assumed that
S1=〈0.5,0.3,0.4〉 ,
S2=〈0.3,0,0.5〉 , and
S3=〈0.4,0.4,0〉 are three SNNs, and that their weight vector is
ω=(1,0,0)T . Based on the SNEWA operator in Equation (3), two terms
0.31×00×0.40 and
0.41×0.50×00 can be found in the aggregated results. Since they are indeterminate values, the SNEWA operator [22] is unreasonable in some cases.
4) It is assumed that
S1=〈1,0,0.5〉 and
S2=〈0.5,0.3,0.4〉 are two SNNs, and that their weight vector is
ω=(0,1)T . Based on the SNFWA operator in Equation (4), two indeterminate terms, i.e.,
00×(λ0.5−1) and
00×0.31 , can be found in the aggregated results: the SNFWA operator [24] is unreasonable in some cases.
5) It is assumed that
S1=〈1,0.2,0.3〉 ,
S2=〈0.5,0,0.5〉 , and
S3=〈0.5,0.6,0〉 , and that weight vector is
ω=(1,0,0)T . Based on the SNHWA operator in Equation (5), two indeterminate terms (
0.21×00×0.60 and
0.31×0.50×00 ) can be obtained in the aggregated results: the SNHWA operator [32] is unreasonable in some cases.
4. The Improved Single-Valued Neutrosophic Weighted Average Operator
In this section, we defined an ISNHWA operator that can overcome the limitations of the SNWA, SNWAA, SNEWA, SNFWA, and SNHWA operators.
Definition 4. Assuming that
Sj=(tj,fj,kj)(j=1,2,⋯,n) is a group of SNNs and
ωj the weight of
Sj ,
ωj∈[0,1] and
n∑j=1ωj=1 , then the ISNHWA operator can be defined as a function
Sn→S :
ISNHWAω(S1,S2,⋯,Sn)=〈n∏j=1(1+(γ−1)tj)ωj−(1−1ρ(1−n∏j=1(1−ρtj)ωj))n∏j=1(1+(γ−1)tj)ωj+(γ−1)(1−1ρ(1−n∏j=1(1−ρtj)ωj)) , γ(1−1ρ(1−n∏j=1(1−ρ(1−fj))ωj))n∏j=1(1+(γ−1)fj)ωj+(γ−1)(1−1ρ(1−n∏j=1(1−ρ(1−fj))ωj)) , γ(1−1ρ(1−n∏j=1(1−ρ(1−kj))ωj))n∏j=1(1+(γ−1)(1−kj))ωj+(γ−1)(1−1ρ(1−n∏j=1(1−ρ(1−kj))ωj))〉. (7)
In the above equation,
γ>0 and
0<ρ<1 .
1) If
γ=1 , then the ISNHWA operator can be reduced to the improved single-valued neutrosophic weighted average (ISNWA) operator:
ISNWAω(S1,S2,⋯,Sn)=〈1ρ(1−n∏j=1(1−ρtj)ωj), 1−1ρ(1−n∏j=1(1−ρ(1−fj))ωj), 1−1ρ(1−n∏j=1(1−ρ(1−kj))ωj)〉. (8)
2) If
γ=2 , then the ISNHWA operator can be reduced to the improved single-valued neutrosophic Einstein weighted average (ISNEWA) operator:
ISNEWAω(S1,S2,⋯,Sn)=〈n∏j=1(1+tj)ωj−1+1ρ(1−n∏j=1(1−ρtj)ωj)n∏j=1(1+tj)ωj+1−1ρ(1−n∏j=1(1−ρtj)ωj) , 2(1−1ρ(1−n∏j=1(1−ρ(1−fj))ωj))n∏j=1(2−fj)ωj+1−1ρ(1−n∏j=1(1−ρ(1−fj))ωj) , 2(1−1ρ(1−n∏j=1(1−ρ(1−kj))ωj))n∏j=1(2−kj)ωj+1−1ρ(1−n∏j=1(1−ρ(1−kj))ωj) 〉. (9)
If
ρ=1 , then the ISNHWA operator in Equation (9) can be reduced to the SNHWA operator in Equation (5). For convenience, in this study we considered
ρ=0.98 and
γ=3 . Apparently, the larger the value of
ρ , the higher will be the similarity degree between the aggregated results using the ISNHWA operator and those using the SNHWA operator.
Here, we will utilize the proposed ISNHWA operator to handle the problems presented in Section 3, overcoming the limitations of the SNWA [21] , SNWAA [22] , SNEWA [22] , SNFWA [24] , and SNHWA [32] operators. The corresponding results are summarized below.
1) We assumed that
S1=〈1,0,0〉 ,
S2=〈0,1,0〉 , and
S3=〈0,0,1〉 were three SNNs, and that their weight vector was
ω=(0,0,1)T . In this case, using the ISNHWA operator, we obtained
ISNHWAω(S1,S2,S3)=〈0,0,1〉 , which is in accord with the actual decision-making process: the ISNHWA operator can overcome the limitations of the SNWA operator [21] .
2) We assumed that
S1=〈1,0,0.2〉 ,
S2=〈0.3,0.1,0〉 , and
S3=〈0.2,0.3,0.5〉 were three SNNs, and that their weight vector was
ω=(0,0,1)T . In this case, using the ISNHWA operator, we obtained
ISNHWAω(S1,S2,S3)=〈0.2,0.3,0.5〉 , which is in accord with the actual decision-making process: the ISNHWA operator overcomes the shortcomings of the SNWAA operator [22] .
3) We assumed that
S1=〈0.5,0.3,0.4〉 ,
S2=〈0.3,0,0.5〉 , and
S3=〈0.4,0.4,0〉 were three SNNs, and that their weight vector was
ω=(1,0,0)T . In this case, using the ISNHWA operator, we obtained
ISNHWAω(S1,S2,S3)=〈0.5,0.3,0.4〉 , which is also in accord with the actual decision-making process: the ISNHWA operator overcomes the limitations of the SNEWA operator [22] .
4) We assumed that
S1=〈1,0,0.5〉 and
S2=〈0.5,0.3,0.4〉 were two SNNs, and that their weight vector was
ω=(0,1)T . In this case, using the ISNHWA operator, we obtained
ISNHWAω(S1,S2,S3)=〈0.5,0.3,0.4〉 , which is in accord with the actual decision-making process: the ISNHWA operator overcomes the limitations of the SNFWA operator [24] .
5) We assumed that
S1=〈1,0.2,0.3〉 ,
S2=〈0.5,0,0.5〉 , and
S3=〈0.5,0.6,0〉 are three SNNs, and that their weight vector was
ω=(1,0,0)T . In this case, using the ISNHWA operator, we obtained
ISNHWAω(S1,S2,S3)=〈1,0.2,0.3〉 , which is in accord with the actual decision-making process: the ISNHWA operator overcomes the shortcomings of the SNHWA operator [32] .
To sum up, the ISNHWA operator overcomes the shortcomings of the SNWA [21] , SNWAA [22] , SNEWA [22] , SNFWA [24] , and SNHWA [32] operators.
Based on Definition 4, some properties of the ISNHWA operator will be discussed below.
Theorem 1 (Idempotency). Let
Sj(j=1,2,⋯,n) be a group of SNNs and
ωj be the weight of
Sj (
ωj∈[0,1] and
n∑j=1ωj=1 ). If
S1=S2=⋯=Sn=S=〈t,f,k〉 , then
ISNHWAω(S1,S2,⋯,Sn)=S .
Proof. Since
S1=S2=⋯=Sn=S=〈t,f,k〉 :
ISNHWAω(S1,S2,⋯,Sn)=〈(1+(γ−1)t)n∑i=1ωi−(1−1ρ(1−(1−ρt)n∑i=1ωi))(1+(γ−1)t)n∑i=1ωi+(γ−1)(1−1ρ(1−(1−ρt)n∑i=1ωi)), γ(1−1ρ(1−(1−ρ(1−f))n∑i=1ωi))(1+(γ−1)(1−f))n∑i=1ωi+(γ−1)(1−1ρ(1−(1−ρ(1−f))n∑i=1ωi)), γ(1−1ρ(1−(1−ρ(1−k))n∑i=1ωi))(1+(γ−1)(1−k))n∑i=1ωi+(γ−1)(1−1ρ(1−(1−ρ(1−k))n∑i=1ωi))〉=〈1+(γ−1)t−(1−t)1+(γ−1)t+(γ−1)(1−t),γ(1−(1−f))1+(γ−1)(1−f)+(γ−1)f,γ(1−(1−k))1+(γ−1)(1−k)+(γ−1)k〉=〈t,f,k〉.
Theorem 2 (Boundedness). Let
Sj(j=1,2,⋯,n) be a group of SNNs. If
S−=〈mintj,maxfj,maxkj〉(j=1,2,⋯,n) and
S+=〈maxtj,minfj,minkj〉(j=1,2,⋯,n) , then
S−≤ISNHWAω(S1,S2,⋯,Sn)≤S+ .
Proof. Assuming that
x=n∏j=1(1+(γ−1)tj)ωj ,
y=1−1ρ(1−n∏j=1(1−ρtj)ωj) , and
ISNHWAω(S1,S2,⋯,Sn)=〈p(tj),p(fj),p(kj)〉 , then
p(tj)=x−yx+(γ−1)y=1−γx/y+(γ−1) .
Since
xy=n∏j=1(1+(γ−1)tj)ωj1−1ρ(1−n∏j=1(1−ρtj)ωj) and
xy is a monotonous increasing function of
tj , also
p(tj) is a monotonous increasing function of
tj . Thus,
p(tj)≥n∏j=1(1+(γ−1)mintj)ωj−(1−1ρ(1−n∏j=1(1−ρmintj)ωj))n∏j=1(1+(γ−1)mintj)ωj−(γ−1)(1−1ρ(1−n∏j=1(1−ρmintj)ωj))=mintj
and
p(tj)≤n∏j=1(1+(γ−1)maxtj)ωj−(1−1ρ(1−n∏j=1(1−ρmaxtj)ωj))n∏j=1(1+(γ−1)maxtj)ωj−(γ−1)(1−1ρ(1−n∏j=1(1−ρmaxtj)ωj))=maxtj.
Similarly, we can prove that
minfj≤p(fj)≤maxfj and
minkj≤p(kj)≤maxkj . Hence,
mintj+1−maxfj+1−maxkj≤ISNHWAω(S1,S2,⋯,Sn)≤maxtj+1−minfj
+1−minkj , i.e.,
p(S−)≤p(ISNHWAω(S1,S2,⋯,Sn))≤p(S+) . So
S−≤ISNHWAω(S1,S2,⋯,Sn)≤S+ is true.
Theorem 3 (Monotonicity). Let
Sj(j=1,2,⋯,n) and
ˉSj(j=1,2,⋯,n) be two groups of SNNs. If
Sj≤ˉSj ,
tj≤ˉtj ,
fj≥ˉfj , and
kj≥ˉkj(j=1,2,⋯,n) , then
ISNHWAω(S1,S2,⋯,Sn)≤ISNHWAω(ˉS1,ˉS2,⋯,ˉSn) .
Proof. Since
tj≤ˉtj ,
fj≥ˉfj and
kj≥ˉkj(j=1,2,⋯,n) , based on Theorem 2, we can obtain
p(ISNHWAω(S1,S2,⋯,Sn))≤p(ISNHWAω(ˉS1,ˉS2,⋯,ˉSn)) . Therefore,
ISNHWAω(S1,S2,⋯,Sn)≤ISNHWAω(ˉS1,ˉS2,⋯,ˉSn) .
Theorem 4 (Commutativity). Let
Sj(j=1,2,⋯,n) be a group of SNNs. If
˜Sj(j=1,2,⋯,n) is an arbitrary permutation of
Sj , then
ISNHWAω(S1,S2,⋯,Sn)=ISNHWAω(˜S1,˜S2,⋯,˜Sn) .
The process of proof is omitted here.
5. Improved Single-Valued Neutrosophic MCGDM Method Based on the ISNHWA Operator and GRA
Let
S={S1,S2,⋯,Sn} be a group of alternatives,
C={C1,C2,⋯,Cm} a group of criteria, and
D={D1,D2,⋯,Dz} a group of DMs. The weight vectors of the criteria and DMs will be
w=(w1,w2,⋯,wm)T and
λ=(λ1,λ2,⋯,λz)T , respectively, and satisfy the following conditions:
wj,λl∈[0,1] ,
m∑j=1ωj=1 , and
z∑l=1λl=1 . It is assumed that
Rlij=〈tlij,flij,klij〉
(i=1,2,⋯,n;j=1,2,⋯,m;l=1,2,⋯,z) is the criteria value provided by the DM
Dl for the alternative
Si under criterion
Cj . The steps involved in the application of the new MCGDM method, which is based on such promises, are described in the following subsections. Notably, if there is only one DM, then Step 2 can be omitted: the single-valued neutrosophic MCDM method is a special case of the MCGDM method. Moreover, if the weight information of criteria is completely known, then Step 3 can be omitted here; otherwise, if the weight information of criteria is completely unknown, the weight of criteria will be determined using the grey correlation analysis method. The steps for selecting the alternative(s) are provided in the following and the flowchart of the proposed method is shown in Figure 1 and Algorithm 1 respectively.
Step 1. Normalization of the decision matrix.
Since the criteria may be of the cost or benefit type, the criteria normalization method can be determined as:
ˉRlij={Rlij, for the benefit criteria Cj(Rlij)c, for the cost criteria Cj ,
(i=1,2,⋯,n;j=1,2,⋯,m) , (10)
where
(Rlij)c=(klij,flij,tlij) denotes the complement of
Slij .
Step 2: Calculation of the comprehensive evaluation values of all DMs.
From the ISNHWA operator, let
ρ=0.98 and
γ=3 . In this case, the aggregated comprehensive evaluation values
δij(i=1,2,⋯,n;j=1,2,⋯,m) of all
Figure 1. Flowchart of the proposed method.
Algorithm 1. The score function and the accuracy function values of each alternative.
Input: The single-valued neutrosophic decision matrix , the alternatives is , and the distinguish coefficient is . |
Output: The score function and the accuracy function values of each alternative. |
function the score function and the accuracy function values of each alternative for each do function the general decision matrix |
calculate the aggregated decision matrix in Equation (11) end for |
return |
end function |
function the weight of criteria |
for each criterion do |
calculate the arithmetic mean value of the alternative under criteria |
end for |
calculate the in Equation (12) |
return |
for each alternative under criterion , determine the grey relation coefficient do |
Calculate the in Equation (13), calculate the grey correlation degree in Equation (14), calculate the weight of criteria in Equation (15) |
end for |
return |
end function |
for each alternative do |
calculated the aggregated overall values of each alternative in Equation (16), calculated the score function and accuracy function values of each alternative and in Equation (17) |
end for |
return and |
end function |
翻译:
DMs are the following:
δij=〈tij,fij,kij〉=ISNHWAλ(R1ij,R2ij,⋯,Rzij) =〈z∏l=1(1+2tlij)λl−(1−10.98(1−z∏l=1(1−0.98tlij)λl))z∏l=1(1+2tlij)λl+2(1−10.98(1−z∏l=1(1−0.98tlij)λl)) , 3(1−10.98(1−z∏l=1(1−0.98(1−flij))λl))z∏l=1(1+(γ−1)flij)λl+2(1−10.98(1−z∏l=1(1−0.98(1−flij))λl)), 3(1−10.98(1−z∏l=1(1−0.98(1−klij))λl))z∏l=1(1+2(1−klij))λl+2(1−10.98(1−z∏l=1(1−0.98(1−klij))λl))〉. (11)
Step 3: Determination of the weight of criteria
Grey correlation analysis describes the proximity between comparison sequence (CS) and reference sequence (RS), which can be measured by the grey correlation coefficient [36] [37] . First, we will use the corresponding column of the comprehensive decision-making matrix provided by all DMs as the CS, i.e.,
δj=(δ1j,δ2j,⋯,δnj)(j=1,2,⋯,m) . Second, the RS is the key factor to determine the weight of criteria. According to the maximum entropy principle, the weight of all criteria is equal without any prior information. Then the RS can be considered as the arithmetic mean value of the alternative under all criteria and denoted as:
ˉδ=(ˉδ1,ˉδ2,⋯,ˉδn) , where
ˉδi=ISNHWA1/n(δi1,δi2,⋯,δim) =〈m∏j=1(1+2tij)1/n−(1−10.98(1−m∏j=1(1−0.98tij)1/n))m∏j=1(1+2tij)1/n+2(1−10.98(1−m∏j=1(1−0.98tij)1/n)), 3(1−10.98(1−m∏j=1(1−0.98(1−fij))1/n))m∏j=1(1+(γ−1)fij)1/n+2(1−10.98(1−m∏j=1(1−0.98(1−fij))1/n)), 3(1−10.98(1−m∏j=1(1−0.98(1−kij))1/n))m∏j=1(1+2(1−kij))1/n+2(1−10.98(1−m∏j=1(1−0.98(1−kij))1/n))〉. (12)
Then the corresponding grey correlation coefficient between
ˉδi and
δij can be obtained as:
τ(ˉδi,δij)=minjminid(ˉδi,δij)+ςmaxjmaxid(ˉδi,δij)d(ˉδi,δij)+ςmaxjmaxid(ˉδi,δij) . (13)
where
d(ˉδi,δij)=13(|tˉδi−tδij|+|fˉδi−fδij|+|kˉδi−kδij|) .
ς(0<ς<1) represents
the distinguished degree. The smaller the parameter value is, the higher the distinguished degree between the correlation coefficients is. In general, if the sequence data fluctuation is small, then the parameter should take a larger value in the interval of (0, 1); otherwise, when the sequence data fluctuation is large, the smaller value should be determined. Based on the empirical value,
ρ=0.5 will be used in this paper.
Then the grey correlation degree between the RS and the CS can be determined:
π(ˉδ,δj)=1nn∑i=1τ(ˉδi,δij) . (14)
The RS is the aggregation information of each alternative that the weight of all criteria is equal, which can be regarded as a virtual criterion. Similar to the consensus problem in group decision-making, the higher the correlation between the DM and the consensus opinion of the group, the higher the weight should be given to the corresponding DM. Therefore, the weight of criteria can be determined as:
wj=π(ˉδ,δj)m∑j=1π(ˉδ,δj) . (15)
Step 4: Calculation of the overall values for each alternative.
From Step 2, the aggregated overall values
δi(i=1,2,⋯,n) of each alternative can be determined using the ISNHWA operator as follows:
δi=〈ti,fi,ki〉=ISNHWAw(Ri1,Ri2,⋯,Rim) . (16)
Step 5: Calculation of the score function and of the accuracy function.
The score function
p(δi) and accuracy function
q(δi)(i=1,2,⋯,n) can be determined from Definition 2.
Step 6. Ranking of the alternatives.
The larger the value of
p(δi) , the better is the alternative
Si . Based on Step 4, the final ranking can be obtained.
6. Application Examples
With the rapid development of urbanization and industrialization in China, the environmental problems are becoming more and more serious. Under the background of rapid global climate change, the surface wind speed in most regions has decreased, and the air quality has touched the red line many times, which has seriously damaged the health of the people. Especially in autumn and winter, the degree of air pollution and the number of air pollution days increase more, so it is very important to predict and estimate the air quality. The 19th Asian Games 2022 will be held in Hangzhou during September 10-25. In order to forecast the air quality during the 19th Asian Games, the local environmental department intends to present a comprehensive evaluation of the air quality in Hangzhou during September 2018, 2019, 2020, and 2021. To this aim, three air-quality monitoring stations (i.e., three DMs) can be considered:
D={D1,D2,D3} , and the corresponding weight of three stations is
λ=(0.31,0.4,0.29)T . The monthly air quality during September 2018, 2019, 2020, and 2021 can be denoted as alternatives:
S={S1,S2,S3,S4}= {September 2017, September 2018, September 2019, September 2020}. From the Implementing Details for Urban Environmental Comprehensive Treatment and Quantitative Examination (General Office of State Environmental Protection Administration (GOSEPA, 2006), the important criteria for evaluation can be denoted instead as
C={C1,C2,C3,C4,C5}={NO2,PM2.5,PM10,SO2,CO,O3} .
For the data in experiment, all members of the monitoring station department should be provided the evaluation information in linguistic terms based on the actual data for attributes of air quality and DMs’ knowledges respectively. The transformation rules between the linguistic terms and SNNs are presented in Table 1. For example, one DM in
D1 provides the language evaluation information “higher” for the CO2 emissions in 2018, then the corresponding SNN is
〈0.8,0.1,0.2〉 . Second, according to the evaluation information for all DMs in
D1 , the average values for three parameters can be obtained as
〈0.5,0.3,0.2〉 . Finally, since the criterion for CO2 emissions is cost-type, then the normalized value of evaluation information is
〈0.2,0.3,0.5〉 based on Equation (10).
In order to rank the four alternatives, the measured values can be converted to SNNs based on the monitoring stations:
Rl=(Rlij)4×3(l=1,2,3) .
R1=(〈0.2,0.3,0.5〉〈0.5,0.3,0.5〉〈0.4,0.4,0.3〉〈0.2,0.1,0.3〉〈0.5,0.1,0.2〉〈0.3,0.1,0.3〉〈0.2,0.4,0.4〉〈0.2,0.4,0.4〉〈0.4,0.3,0.3〉〈0.3,0.1,0.2〉〈0.4,0.2,0.1〉〈0.5,0.2,0.2〉〈0.5,0.4,0.3〉〈0.4,0.3,0.3〉〈0.5,0.3,0.2〉〈0.4,0.2,0.3〉〈0.4,0.2,0.3〉〈0.4,0.2,0.3〉〈0.3,0.5,0.2〉〈0.4,0.3,0.3〉〈0.5,0.2,0.3〉〈0.1,0.4,0.3〉〈0.3,0.3,0.3〉〈0.3,0,0.3〉) ;
Linguistic term | SNN |
Very high | (0.9, 0, 0.1) |
High | (0.8, 0.1, 0.2) |
Medium high | (0.7, 0, 0.1) |
Medium | (0.5, 0.1, 0.1) |
Medium low | (0.4, 0, 0.1) |
Low | (0.3, 0.1, 0.2) |
Very low | (0.2, 0, 0.1) |
翻译:
Table 1. The transformation rules used for the evaluation of air quality.
R2=(〈0.3,0.5,0.1〉〈0.1,0.5,0.4〉〈0.2,0.4,0.4〉〈0.2,0.4,0.2〉〈0.5,0.1,0.1〉〈0.3,0.1,0.2〉〈0.2,0.6,0.2〉〈0.4,0.3,0.3〉〈0.3,0.4,0.3〉〈0.4,0.1,0.1〉〈0.6,0,0.2〉〈0.4,0.1,0〉〈0.5,0.5,0.2〉〈0.5,0.4,0.3〉〈0.5,0.6,0.1〉〈0.3,0.2,0.2〉〈0.4,0,0.1〉〈0.4,0.2,0.2〉〈0.4,0.5,0.1〉〈0.4,0.4,0.3〉〈0.4,0.3,0.3〉〈0.2,0.3,0.2〉〈0.5,0.1,0.2〉〈0.5,0.1,0.2〉) ;
R3=(〈0.3,0.5,0.2〉〈0.5,0.4,0.3〉〈0.4,0.2,0.3〉〈0.2,0.1,0.3〉〈0.4,0.1,0〉〈0.5,0.5,0〉〈0.1,0.6,0.3〉〈0.5,0.4,0.4〉〈0.6,0.2,0.2〉〈0.3,0.2,0.4〉〈0.5,0.1,0.1〉〈0.4,0.2,0.1〉〈0.5,0.3,0.1〉〈0.5,0.2,0.3〉〈0.5,0.6,0.2〉〈0.4,0.1,0.3〉〈0.5,0.2,0.2〉〈0.4,0,0.1〉〈0.6,0.5,0.1〉〈0.5,0.3,0.3〉〈0.4,0.5,0.2〉〈0.5,0,0.2〉〈0.6,0.1,0.1〉〈0.4,0.1,0〉) .
Step 1. Normalization of the decision matrix.
Since these criteria are of the cost type, the normalized decision matrix can be obtained from Equation (9):
ˉRl=Rl .
Step 2. Calculation of the comprehensive evaluation values of all DMs.
Based on the MCGDM method using the ISNHWA operator, and considering
ρ=0.98 and
γ=3 (see Equation (10)), the comprehensive values of all DMs can be determined as:
R=(〈0.269,0.4712,0.4452〉〈0.3442,0.4601,0.4032〉〈0.3209,0.4226,0.3291〉〈0.1706,0.5081,0.5346〉〈0.3695,0.4403,0.3577〉〈0.4236,0.4067,0.301〉〈0.5,0.4612,0.4074〉〈0.4698,0.4067,0.3〉〈0.5,0.4911,0.4878〉〈0.4323,0.5,0.5023〉〈0.4297,0.4298,0.3368〉〈0.4318,0.41,0.309〉 〈0.2,0.3023,0.1788〉〈0.4718,0.2143,0.1026〉〈0.36,0.2856,0.1722〉〈0.3404,0.246,0.1261〉〈0.5125,0.1203,0.0513〉〈0.4318,0.2824,0.1571〉〈0.3604,0.296,0.1652〉〈0.4297,0.1434,0.0655〉〈0.4,0.1865,0.0912〉〈0.2579,0.2378,0.1369〉〈0.4723,0.268,0.1443〉〈0.4109,0.1194,0.0516〉).
Step 3. Determination of the weight of criteria
Form Equation (11)-Equation (12), the reference sequence is determined as:
ˉf1=〈0.3286,0.4327,0.2479〉 ,
ˉf2=〈0.3766,0.4052,0.2122〉 ,
ˉf3=〈0.4442,0.404,0.2114〉 and
ˉf4=〈0.4068,0.4014,0.2078〉 . The grey relational coefficient can be obtained:
τ(ˉfi,Rij)=(0.67880.807110.64420.50420.72790.43800.82390.91580.69440.46210.75590.66400.95890.56440.74900.56300.59450.56480.83630.92090.59330.71750.5472) .
Based on Equation (13)-Equation (14), the grey correlation degree between the RS and the CS can be obtained as:
π(ˉδ,δ1)=0.5864 ,
π(ˉδ,δ2)=0.8565 ,
π(ˉδ,δ3)=0.8503 ,
π(ˉδ,δ4)=0.6702 ,
π(ˉδ,δ5)=0.5617 , and
π(ˉδ,δ6)=0.6563 . Then the weight of criteria can be determined as:
w1=0.1402 ,
w2=0.2048 ,
w3=0.2033 ,
w4=0.1603 ,
w5=0.1343 , and
w6=0.1571 .
Step 4. Calculation of the overall values for each alternative.
Based on the ISNHWA operator and Equation (15), the overall aggregated values for each alternative can be obtained:
δ1=〈0.3262,0.4381,0.2593〉 ,
δ2=〈0.3785,0.4137,0.2235〉 ,
δ3=〈0.4472,0.4130,0.2271〉 , and
δ4=〈0.4066,0.4063,0.2150〉 .
Step 5. Determination of the score and accuracy functions.
From the comparison method in Definition 2, the score value of each alternative can be obtained:
p(δ1)=0.5430 ,
p(δ2)=0.5804 ,
p(δ3)=0.6024 , and
p(δ4)=0.5951 .
Step 6. Ranking of the alternatives.
Since
p(δ3)>p(δ4)>p(δ2)>p(δ1) ,
S3≻S4≻S2≻S1 . The best and worst alternatives would be
S3 and
S1 .
In order to further explore the effectiveness of the proposed method, the sensitivity analysis of different parameters will be carried out below to explore the influence of different parameter value on the final decision-making results.
If
ρ=0.98 ,
ς=0.5 and
0<γ<10 , then the corresponding weight and the final rankings are shown in Figure 2 & Figure 3. Obviously, the change of parameter
γ have less impact on
w4 and more effects on other five weights. However, according to the results in Figure 3, as the parameter value
γ become larger, the final ranking of four alternatives gradually tends to stabilize, i.e.,
S3≻S4≻S2≻S1 . The best and worst alternatives would be
S3 and
S1 . If
ρ=0.98 ,
0<ς<1 , and
γ=3 , then the corresponding weight and the final rankings are shown in Figure 4 & Figure 5. From Figure 4, the change of parameter
ς have more impacts on
w2 and
w3 , and less effects on other four weights. Moreover, the different values of the parameter
ς do not affect the ordering of four alternatives, and the final ranking is always
S3≻S4≻S2≻S1 . If
0<ρ<1 ,
ς=0.5 , and
γ=3 , then the corresponding weight and the final rankings are shown in Figure 6 & Figure 7. According to Figure 5, different values of the parameter
ρ have less effect on the weights, and the final rankings of four alternatives are not affected by the parameter
ρ .
Therefore, from the analysis presented above, although the parameter changes
Figure 2. Weight with
ρ=0.98 ,
ς=0.5 and
0<γ<10 .
Figure 3. Final rankings with
ρ=0.98 ,
ς=0.5 and
0<γ<10 .
Figure 4. Weights with
ρ=0.98 ,
0<ς<1 , and
γ=3 .
Figure 5. Final rankings with
ρ=0.98 ,
0<ς<1 , and
γ=3 .
Figure 6. Weights with
0<ρ<1 ,
ς=0.5 , and
γ=3 .
Figure 7. Final rankings with
0<ρ<1 ,
ς=0.5 , and
γ=3 .
affect the value of the corresponding weights, it did not influence the final rankings of four alternatives, ensuring the stability of the proposed decision method.
At the same time, DMs can choose different parameter values according to their own preferences, providing more choices for different decision-making problems.
Moreover, since the existing MCDM methods based on those operators discussed earlier cannot handle decision-making problems where the weight information is completely unknown. Then if Example 6.1 is solved by applying the SNWA-operator based MCDM method [21] , the SNWAA- and SNEWA-operator based MCDM methods [22] , the SNFWA-operator based MCDM method [24] , and the SNHWA-operator based MCGDM method [32] , the weight of criteria in Step 2 is used:
ω=(0.2942,0.3367,0.3691)T . The corresponding compared results shown in Table 2 can be obtained. Notably, the aggregation operators in the compared MCDM methods can be used twice to handle MCGDM problems.
MCDM methods | Ranks |
SNWA-operator based MCDM method [21] | |
SNWAA-operator based MCDM method [22] | |
SNEWA-operator based MCDM method [22] | |
SNFWA-operator based MCDM method ( ) [24] | |
SNHWA-operator based MCGDM method ( ) [32] | |
ISNHWA-operator based MCGDM method ( ) | |
翻译:
Table 2. Compared results.
From Table 1, it can be seen that the ranking obtained by using the proposed single-valued neutrosophic MCGDM method is the same as that obtained by using the SNWAA-operator based MCDM method [22] and SNHWA-operator based MCGDM method (
γ=3 ) [31] :
S3≻S4≻S1≻S2 and the best alternative is
S3 . However, the ranking obtained by applying the proposed MCGDM method based on the ISNHWA operator is different from that obtained by applying the SNWA-operator based MCDM method [19] , the SNEWA-operator based MCDM method [22] , and the SNFWA-operator based MCDM method (
λ=2 ) [25] , and the best alternative would be
S4 . Moreover, the SNWA-operator based MCDM method [21] , the SNEWA-operator based MCDM method [22] , and the SNFWA-operator based MCDM method [24] have limitations, as discussed in Examples 5.1 and 5.2. Thus, the ISNHWA-operator based MCGDM method, the SNWAA-operator based MCDM method [22] , and the SNHWA-operator can produce more reasonable results in this case.
In the following, another example with known weight information is provided to make a comparative analysis to further verify the effectiveness and superiority of the proposed method.
7. Conclusion
A novel single-valued neutrosophic MCGDM method based on the ISNHWA operator is proposed in this paper. The proposed ISNHWA operator avoids the limitations of existing aggregation operators (i.e., the SNWA [21] , SNWAA [22] , SNEWA [22] , SNFWA [24] , and SNHWA [31] operators). Additionally, the novel single-valued neutrosophic MCGDM method overcomes the limitations of the SNWA-operator based MCDM method [21] , the SNWAA- and SNEWA-operator based MCDM methods [22] , the SNFWA-operator based MCDM method [24] , and the SNHWA-operator based MCGDM method [31] . Several application examples have been provided to demonstrate that the ISNHWA operator and the single-valued neutrosophic MCGDM method are effective and feasible for solving MCDM and MCGDM problems. Future research should focus on the improvement of other single-valued neutrosophic aggregation operators (i.e., the Bonferroni mean operator and Heronian mean operators), and of the corresponding MCDM and MCGDM methods.