2.1 基于模糊熵的直觉模糊VIKOR方法
2.1.1 问题描述
设某多属性决策问题有 m m m个方案 Y i ( i = 1 , 2 , ⋯ , m ) Y_{i}\left(i=1,2,\cdots,m\right) Yi(i=1,2,⋯,m),组成方案集 Y = ( Y 1 , Y 2 , ⋯ , Y m ) Y=\left(Y_{1},Y_{2},\cdots,Y_{m}\right) Y=(Y1,Y2,⋯,Ym),评价每个方案的属性(或指标)为 G j ( j = 1 , 2 , ⋯ , n ) G_{j}\left(j=1,2,\cdots,n\right) Gj(j=1,2,⋯,n),记属性集为 G = { G 1 , G 2 , ⋯ , G n } G=\left\{ G_{1},G_{2},\cdots,G_{n} \right\} G={G1,G2,⋯,Gn}。如果 F ~ i j = ⟨ μ i j , ν i j ⟩ ( i = 1 , 2 , ⋯ , m ; j = 1 , 2 , ⋯ , n ) \tilde{F}_{ij} = \left\langle \mu_{ij},\nu_{ij} \right\rangle \left(i=1,2,\cdots,m;j=1,2,\cdots,n\right) F~ij=⟨μij,νij⟩(i=1,2,⋯,m;j=1,2,⋯,n)为直觉模糊集,表示方案满足属性 G j ∈ G G_{j} \in G Gj∈G和不满足属性 G j ∈ G G_{j} \in G Gj∈G的程度,且 0 ≤ μ i j + ν i j ≤ 1 0 \leq \mu_{ij}+\nu_{ij} \leq 1 0≤μij+νij≤1,矩阵 F = ( ⟨ μ i j , ν i j ⟩ ) m × n F = \left(\left\langle \mu_{ij},\nu_{ij} \right\rangle\right)_{m×n} F=(⟨μij,νij⟩)m×n为该多属性决策问题的直觉模糊决策矩阵。现在的问题是依据直觉模糊决策矩阵 F F F,如何得到一个有效的决策分析方法来对所有方案进行优劣排序。
2.1.2 基于模糊熵的属性权重确定方法
模糊熵是Shannon信息熵在模糊数学领域的扩展,被用来解释模糊集所包含的信息量,信息量越多,则模糊性越低,它能够为决策者提供的信息也就越多。直觉模糊熵最早由Burillo等提出,后来很多学者对直觉模糊熵的定义及其计算公式进行了较深入的研究。本书借鉴文献关于模糊熵的公理化定义计算属性权重。
设 A ~ = { ⟨ x i , μ A ( x i ) , ν A ( x i ) ⟩ ∣ x i ∈ X , i = 1 , 2 , ⋯ , n } \tilde{A} = \left\{ \left\langle x_{i},\mu_{A}\left(x_{i}\right),\nu_{A}\left(x_{i}\right) \right\rangle | x_{i} \in X, i=1,2,\cdots,n\right\} A~={⟨xi,μA(xi),νA(xi)⟩∣xi∈X,i=1,2,⋯,n}为论域 X X X上的直觉模糊数,则
E ( A ~ ) = 1 n ∑ i = 1 n cos π ( μ A 2 ( x i ) − ν A 2 ( x i ) ) 2 (2.1) \color{red} { E\left(\tilde{A}\right) = \frac{1}{n}\sum_{i=1}^{n}\cos\frac{\pi\left(\mu_{A}^2\left(x_{i}\right) - \nu_{A}^2\left(x_{i}\right)\right)}{2} \tag{2.1} } E(A~)=n1i=1∑ncos2π(μA2(xi)−νA2(xi))(2.1)
是一个直觉模糊熵。
由于 π A ~ = 1 − ( μ A ~ ( x i ) − ν A ~ ( x i ) ) \pi_{\tilde{A}} = 1 - \left( \mu_{\tilde{A}}\left(x_{i}\right) - \nu_{\tilde{A}}\left(x_{i}\right) \right) πA~=1−(μA~(xi)−νA~(xi)),所以可以改写为
E ( A ~ ) = 1 n ∑ i = 1 n cos π ( μ A ( x i ) − ν A ( x i ) ) ( 1 − π A ( x i ) ) 2 (2.2) \color{red} { E\left(\tilde{A}\right) = \frac{1}{n}\sum_{i=1}^{n}\cos\frac{\pi\left(\mu_{A}\left(x_{i}\right) - \nu_{A}\left(x_{i}\right)\right)\left(1-\pi_{A}\left(x_{i}\right)\right)}{2} \tag{2.2} } E(A~)=n1i=1∑ncos2π(μA(xi)−νA(xi))(1−πA(xi))(2.2)
从式 ( 2.2 ) (2.2) (2.2)不难看出,直觉模糊熵 E ( A ~ ) E(\tilde{A}) E(A~)不仅考虑了隶属度与非隶属度的偏差 ( μ A ( x i ) − ν A ( x i ) (\mu_{A}\left(x_{i}\right) - \nu_{A}\left(x_{i}\right) (μA(xi)−νA(xi),而且考虑了犹豫度 π A ( x i ) \pi_{A}\left(x_{i}\right) πA(xi)的信息。
在多属性决策问题中,由于各方案 Y i ( i = 1 , 2 , ⋯ , m ) Y_{i}\left(i=1,2,\cdots,m\right) Yi(i=1,2,⋯,m)在属性 G j G_j Gj的评价值 F ~ i j = ⟨ μ i j , ν i j ⟩ \tilde{F}_{ij} = \left\langle \mu_{ij},\nu_{ij} \right\rangle F~ij=⟨μij,νij⟩为直觉模糊数,根据式 ( 2.2 ) (2.2) (2.2)可得属性 G j ( j = 1 , 2 , ⋯ , n ) G_j\left(j=1,2,\cdots,n\right) Gj(j=1,2,⋯,n)的直觉模糊熵 E j E_j Ej为
E j = 1 m ∑ i = 1 m cos π ( μ i j − ν i j ) ( 1 − π i j ) 2 , j = 1 , 2 , ⋯ , n (2.3) \color{red} { E_{j} = \frac{1}{m}\sum_{i=1}^{m}\cos\frac{\pi\left(\mu_{ij} - \nu_{ij}\right)\left(1-\pi_{ij}\right)}{2},j=1,2,\cdots,n \tag{2.3} } Ej=m1i=1∑mcos2π(μij−νij)(1−πij),j=1,2,⋯,n(2.3)
于是可得属性 G j ( j = 1 , 2 , ⋯ , n ) G_{j}\left(j=1,2,\cdots,n\right) Gj(j=1,2,⋯,n)的权重 ω j \omega_{j} ωj
ω j = 1 − E j n − ∑ j = 1 n E j , j = 1 , 2 , ⋯ , n (2.4) \omega_{j} = \frac {1- E_{j}} {n - \sum_{j=1}^{n}E_{j}},j=1,2,\cdots,n \tag{2.4} ωj=n−∑j=1nEj1−Ej,j=1,2,⋯,n(2.4)
2.1.3 基于模糊熵的直觉模糊VIKOR方法的决策步骤
S.1 确定多属性决策问题的方案集 Y = { Y 1 , Y 2 , ⋯ , Y m } Y=\{Y_1,Y_2,\cdots,Y_m\} Y={Y1,Y2,⋯,Ym}和属性集 G = { G 1 , G 2 , ⋯ , G n } G=\{G_1,G_2,\cdots,G_n\} G={G1,G2,⋯,Gn},获取多属性决问题中方案 Y i ∈ Y Y_{i} \in Y Yi∈Y关于属性 G j ∈ G G_{j} \in G Gj∈G的直觉模糊特征信息,构建直觉模糊决策矩阵 F F F。
S.2 根据直觉模糊多属性决策矩阵F,利用式 ( 2.3 ) (2.3) (2.3)和式 ( 2.4 ) (2.4) (2.4)计算属性 G j ( j = 1 , 2 , ⋯ , n ) G_{j}\left(j=1,2,\cdots,n\right) Gj(j=1,2,⋯,n)的模糊熵 E j E_{j} Ej和属性权重 ω j \omega_{j} ωj
S.3 根据直模糊多属性决策矩阵 F F F确定多属性解决问题的正理想解 Y + Y^{+} Y+和负理想解 Y − Y^{-} Y−:
Y + = ( Y 1 + , Y 2 + , ⋯ , Y n + ) (2.5) Y^{+} = \left( Y_{1}^{+},Y_{2}^{+},\cdots,Y_{n}^{+} \right) \tag{2.5} Y+=(Y1+,Y2+,⋯,Yn+)(2.5)
Y − = ( Y 1 − , Y 2 − , ⋯ , Y n − ) (2.6) Y^{-} = \left( Y_{1}^{-},Y_{2}^{-},\cdots,Y_{n}^{-} \right) \tag{2.6} Y−=(Y1−,Y2−,⋯,Yn−)(2.6)
式中, Y j + = ⟨ 1 , 0 ⟩ Y_{j}^{+}=\left\langle1,0\right\rangle Yj+=⟨1,0⟩, Y j − = ⟨ 1 , 0 ⟩ Y_{j}^{-}=\left\langle1,0\right\rangle Yj−=⟨1,0⟩。
S.4 计算的群体效益值 S i S_i Si、个体遗憾值 R i R_i Ri:
S i = ∑ j = 1 n ω j [ d ( Y j + , F ~ i j ) d ( Y j + , Y j − ) ] , i = 1 , 2 , ⋯ , m (2.7) S_{i} = \sum_{j=1}^{n}\omega_{j}\left[\frac{d\left(Y_{j}^{+},\tilde{F}_{ij}\right)}{d\left(Y_{j}^{+},Y_{j}^{-}\right)}\right],i=1,2,\cdots,m \tag{2.7} Si=j=1∑nωj⎣⎡d(Yj+,Yj−)d(Yj+,F~ij)⎦⎤,i=1,2,⋯,m(2.7)
R i = max j { ω j [ d ( Y j + , F ~ i j ) d ( Y j + , Y j − ) ] } , i = 1 , 2 , ⋯ , m (2.8) R_{i} = \max_{j} \left\{ \omega_{j}\left[ \frac{d\left(Y_{j}^{+},\tilde{F}_{ij}\right)}{d\left(Y_{j}^{+},Y_{j}^{-}\right)} \right] \right\},i=1,2,\cdots,m \tag{2.8} Ri=jmax⎩⎨⎧ωj⎣⎡d(Yj+,Yj−)d(Yj+,F~ij)⎦⎤⎭⎬⎫,i=1,2,⋯,m(2.8)
式中, d ( Y j + , F ~ i j ) d\left(Y_{j}^{+},\tilde{F}_{ij}\right) d(Yj+,F~ij), d ( Y j + , Y j − ) d\left(Y_{j}^{+},Y_{j}^{-}\right) d(Yj+,Yj−)为两个直觉模糊数的距离,用以下公式计算:
d ( α ~ 1 , α ~ 2 ) = 1 2 ( ∣ μ α ~ 1 − μ α ~ 2 ∣ + ∣ ν α ~ 1 − ν α ~ 2 ∣ + ∣ π α ~ 1 − π α ~ 2 ∣ ) d\left(\tilde{\alpha}_{1}, \tilde{\alpha}_{2}\right)=\frac{1}{2}\left(\left|\mu_{\tilde{\alpha}_{1}}-\mu_{\tilde{\alpha}_{2}}\right|+\left|\nu_{\tilde{\alpha}_{1}}-\nu_{\tilde{\alpha}_{2}}\right|+\left|\pi_{\tilde{\alpha}_{1}}-\pi_{\tilde{\alpha}_{2}}\right|\right) d(α~1,α~2)=21(∣μα~1−μα~2∣+∣να~1−να~2∣+∣πα~1−πα~2∣)
其中, α ~ 1 = ⟨ μ α ~ 1 , ν α ~ 1 ⟩ \tilde{\alpha}_{1} = \left\langle \mu_{\tilde{\alpha}_{1}},\nu_{\tilde{\alpha}_{1}} \right\rangle α~1=⟨μα~1,να~1⟩, α ~ 2 = ⟨ μ α ~ 2 , ν α ~ 2 ⟩ \tilde{\alpha}_{2} = \left\langle \mu_{\tilde{\alpha}_{2}},\nu_{\tilde{\alpha}_{2}} \right\rangle α~2=⟨μα~2,να~2⟩为两个任意的直觉模糊数。
S.5 计算各备选方案 Y i ∈ Y Y_{i} \in Y Yi∈Y的折中值 Q i Q_{i} Qi:
Q i = ν S i − S + S − − S + + ( 1 − ν ) R i − R + R − − R + , i = 1 , 2 , ⋯ , m (2.9) Q_{i} = \nu\frac{S_{i}-S^{+}}{S^{-}-S^{+}} + \left(1-\nu\right)\frac{R_{i}-R^{+}}{R^{-}-R^{+}},i=1,2,\cdots,m \tag{2.9} Qi=νS−−S+Si−S++(1−ν)R−−R+Ri−R+,i=1,2,⋯,m(2.9)
其中, S i S_{i} Si为最大群体效用,是 L 1 , j L_{1,j} L1,j测度; R i R_{i} Ri取小个体遗憾,是 L ∞ , j L_{\infty,j} L∞,j测度; ω j \omega_{j} ωj为各属性权重; S + = min i ( S i ) S^{+}=\min_{i}(S_{i}) S+=mini(Si), S − = max i ( S i ) S^{-}=\max_{i}(S_{i}) S−=maxi(Si), R + = min i ( R i ) R^{+}=\min_{i}(R_{i}) R+=mini(Ri), R − = max i ( R i ) R^{-}=\max_{i}(R_{i}) R−=maxi(Ri); ν \nu ν为决策机制系数, ν ∈ [ 0 , 1 ] \nu \in [0,1] ν∈[0,1]。当 ν > 0.5 \nu \gt 0.5 ν>0.5时,表示根据最大群体效用的决策机制进行决策;当 ν = 0.5 \nu = 0.5 ν=0.5时,表示依据决策者经过协商达成共识的决策机制进行决策;当 ν < 0.5 \nu \lt 0.5 ν<0.5时,表示根据最小个体遗憾的决策机制进行决策。
S.6 设按 S i S_i Si、 R i R_i Ri、 Q i Q_i Qi值分别从小到大排序,确定妥协解方案。按 Q i Q_i Qi值递增得到的排序为 Y ( 1 ) , Y ( 2 ) , ⋯ , Y ( J ) , ⋯ , Y ( m ) Y^{(1)},Y^{(2)},\cdots,Y^{(J)},\cdots, Y^{(m)} Y(1),Y(2),⋯,Y(J),⋯,Y(m),则依据排序条件 1 1 1和排序条件 2 2 2可确定备选方案的优劣排序和最优方案或折中方案。
案例分析
考虑突发事件应急预案评估问题。突发事件应急预案是针对各种突发事件类型而事先制订的一套能迅速、有效、有序地解决问题的行动计划或方案,为全面、客观地评判应急预案处置突发事件的能力,应从预案处置的快速性( G 1 G_1 G1)、预案内容的合理性( G 2 G_2 G2)、预案保障的充分性( G 3 G_3 G3)、预案消耗费用的合理性( G 4 G_4 G4)以及预案的广泛适用性( G 5 G_5 G5)等五个方面进行综合评价。假设现有五个应急预案 Y i ( i = 1 , 2 , 3 , 4 , 5 ) Y_i\left(i=1,2,3,4,5\right) Yi(i=1,2,3,4,5),决策者根据自己的知识、经验以及已有的统计数据确定出每个应急预案 Y i ( i = 1 , 2 , 3 , 4 , 5 ) Y_i\left(i=1,2,3,4,5\right) Yi(i=1,2,3,4,5)关于属性 G j ( j = 1 , 2 , 3 , 4 , 5 ) G_j\left(j=1,2,3,4,5\right) Gj(j=1,2,3,4,5)的直觉模糊评价信息,得到的直觉模糊决策矩阵 F = ( ⟨ μ i j , ν i j ⟩ ) 5 × 5 F=\left( \left\langle \mu_{ij},\nu_{ij} \right\rangle \right)_{5×5} F=(⟨μij,νij⟩)5×5,如下表所示。
G 1 G_1 G1 | G 2 G_2 G2 | G 3 G_3 G3 | G 4 G_4 G4 | G 5 G_5 G5 | |
---|---|---|---|---|---|
Y 1 Y_1 Y1 | ⟨ 0.6 , 0.1 ⟩ \left\langle 0.6,0.1 \right\rangle ⟨0.6,0.1⟩ | ⟨ 0.3 , 0.1 ⟩ \left\langle 0.3,0.1 \right\rangle ⟨0.3,0.1⟩ | ⟨ 0.1 , 0.4 ⟩ \left\langle 0.1,0.4 \right\rangle ⟨0.1,0.4⟩ | ⟨ 0.7 , 0.1 ⟩ \left\langle 0.7,0.1 \right\rangle ⟨0.7,0.1⟩ | ⟨ 0.5 , 0.2 ⟩ \left\langle 0.5,0.2 \right\rangle ⟨0.5,0.2⟩ |
Y 2 Y_2 Y2 | ⟨ 0.4 , 0.2 ⟩ \left\langle 0.4,0.2 \right\rangle ⟨0.4,0.2⟩ | ⟨ 0.4 , 0.3 ⟩ \left\langle 0.4,0.3 \right\rangle ⟨0.4,0.3⟩ | ⟨ 0.3 , 0.4 ⟩ \left\langle 0.3,0.4 \right\rangle ⟨0.3,0.4⟩ | ⟨ 0.6 , 0.2 ⟩ \left\langle 0.6,0.2 \right\rangle ⟨0.6,0.2⟩ | ⟨ 0.8 , 0.1 ⟩ \left\langle 0.8,0.1 \right\rangle ⟨0.8,0.1⟩ |
Y 3 Y_3 Y3 | ⟨ 0.6 , 0.3 ⟩ \left\langle 0.6,0.3 \right\rangle ⟨0.6,0.3⟩ | ⟨ 0.7 , 0.2 ⟩ \left\langle 0.7,0.2 \right\rangle ⟨0.7,0.2⟩ | ⟨ 0.4 , 0.4 ⟩ \left\langle 0.4,0.4 \right\rangle ⟨0.4,0.4⟩ | ⟨ 0.4 , 0.1 ⟩ \left\langle 0.4,0.1 \right\rangle ⟨0.4,0.1⟩ | ⟨ 0.7 , 0.2 ⟩ \left\langle 0.7,0.2 \right\rangle ⟨0.7,0.2⟩ |
Y 4 Y_4 Y4 | ⟨ 0.3 , 0.5 ⟩ \left\langle 0.3,0.5 \right\rangle ⟨0.3,0.5⟩ | ⟨ 0.5 , 0.2 ⟩ \left\langle 0.5,0.2 \right\rangle ⟨0.5,0.2⟩ | ⟨ 0.6 , 0.3 ⟩ \left\langle 0.6,0.3 \right\rangle ⟨0.6,0.3⟩ | ⟨ 0.5 , 0.2 ⟩ \left\langle 0.5,0.2 \right\rangle ⟨0.5,0.2⟩ | ⟨ 0.9 , 0.0 ⟩ \left\langle 0.9,0.0 \right\rangle ⟨0.9,0.0⟩ |
Y 5 Y_5 Y5 | ⟨ 0.3 , 0.4 ⟩ \left\langle 0.3,0.4 \right\rangle ⟨0.3,0.4⟩ | ⟨ 0.2 , 0.2 ⟩ \left\langle 0.2,0.2 \right\rangle ⟨0.2,0.2⟩ | ⟨ 0.2 , 0.4 ⟩ \left\langle 0.2,0.4 \right\rangle ⟨0.2,0.4⟩ | ⟨ 0.3 , 0.5 ⟩ \left\langle 0.3,0.5 \right\rangle ⟨0.3,0.5⟩ | ⟨ 0.4 , 0.5 ⟩ \left\langle 0.4,0.5 \right\rangle ⟨0.4,0.5⟩ |
<代码待补充!>
2.2 属性权重信息未知情况下的直觉模糊VIKOR方法
2.2.1 问题描述
设某多属性决策问题有 m m m个方案 Y i ( i = 1 , 2 , ⋯ , m ) Y_{i}\left(i=1,2,\cdots,m\right) Yi(i=1,2,⋯,m),组成方案集 Y = ( Y 1 , Y 2 , ⋯ , Y m ) Y=\left(Y_{1},Y_{2},\cdots,Y_{m}\right) Y=(Y1,Y2,⋯,Ym),评价每个方案的属性(或指标)为 G j ( j = 1 , 2 , ⋯ , n ) G_{j}\left(j=1,2,\cdots,n\right) Gj(j=1,2,⋯,n),记属性集为 G = { G 1 , G 2 , ⋯ , G n } G=\left\{ G_{1},G_{2},\cdots,G_{n} \right\} G={G1,G2,⋯,Gn},假设属性权重向量 ω = ( ω 1 , ω 2 , ⋯ , ω n ) T \omega = \left(\omega_{1},\omega{2},\cdots,\omega_{n}\right)^{T} ω=(ω1,ω2,⋯,ωn)T。如果 F ~ i j = ⟨ μ i j , ν i j ⟩ ( i = 1 , 2 , ⋯ , m ; j = 1 , 2 , ⋯ , n ) \tilde{F}_{ij} = \left\langle \mu_{ij},\nu_{ij} \right\rangle \left(i=1,2,\cdots,m;j=1,2,\cdots,n\right) F~ij=⟨μij,νij⟩(i=1,2,⋯,m;j=1,2,⋯,n)为直觉模糊集,表示方案满足属性 G j ∈ G G_{j} \in G Gj∈G和不满足属性 G j ∈ G G_{j} \in G Gj∈G的程度,且 0 ≤ μ i j + ν i j ≤ 1 0 \leq \mu_{ij}+\nu_{ij} \leq 1 0≤μij+νij≤1,矩阵 F = ( ⟨ μ i j , ν i j ⟩ ) m × n F = \left(\left\langle \mu_{ij},\nu_{ij} \right\rangle\right)_{m×n} F=(⟨μij,νij⟩)m×n为该多属性决策问题的直觉模糊决策矩阵。现在的问题是依据直觉模糊决策矩阵 F F F,如何得到一个有效的决策分析方法来对所有方案进行优劣排序。
2.2.2 属性权重的确定方法
根据直觉模糊多属性决策矩阵 F F F确定多属性决策问题的正理想解 Y + Y^{+} Y+和负理想解 Y − Y^{-} Y−:
Y + = ( Y 1 + , Y 2 + , ⋯ , Y n + ) = ( ⟨ μ 1 + , ν 1 + ⟩ , ⟨ μ 2 + , ν 2 + ⟩ , ⋯ , ⟨ μ n + , ν n + ⟩ ) Y^{+} = \left( Y_{1}^{+},Y_{2}^{+},\cdots,Y_{n}^{+} \right) = \left(\left\langle\mu_{1}^{+},\nu_{1}^{+}\right\rangle,\left\langle\mu_{2}^{+},\nu_{2}^{+}\right\rangle,\cdots,\left\langle\mu_{n}^{+},\nu_{n}^{+}\right\rangle\right) Y+=(Y1+,Y2+,⋯,Yn+)=(⟨μ1+,ν1+⟩,⟨μ2+,ν2+⟩,⋯,⟨μn+,νn+⟩)
Y − = ( Y 1 − , Y 2 − , ⋯ , Y n − ) = ( ⟨ μ 1 − , ν 1 − ⟩ , ⟨ μ 2 − , ν 2 − ⟩ , ⋯ , ⟨ μ n − , ν n − ⟩ ) Y^{-} = \left( Y_{1}^{-},Y_{2}^{-},\cdots,Y_{n}^{-} \right) = \left(\left\langle\mu_{1}^{-},\nu_{1}^{-}\right\rangle,\left\langle\mu_{2}^{-},\nu_{2}^{-}\right\rangle,\cdots,\left\langle\mu_{n}^{-},\nu_{n}^{-}\right\rangle\right) Y−=(Y1−,Y2−,⋯,Yn−)=(⟨μ1−,ν1−⟩,⟨μ2−,ν2−⟩,⋯,⟨μn−,νn−⟩)
式中, ⟨ μ j + , ν j + ⟩ = ⟨ max i μ i j , min i ν i j ⟩ , ⟨ μ j − , ν j − ⟩ = ⟨ min i μ i j , max i ν i j ⟩ , j = 1 , 2 , ⋯ , n \left\langle\mu_{j}^{+},\nu_{j}^{+}\right\rangle = \left\langle \max_{i}\mu_{ij},\min_{i}\nu_{ij}\right\rangle, \left\langle\mu_{j}^{-},\nu_{j}^{-}\right\rangle = \left\langle \min_{i}\mu_{ij},\max_{i}\nu_{ij}\right\rangle, j=1,2,\cdots,n ⟨μj+,νj+⟩=⟨maxiμij,miniνij⟩,⟨μj−,νj−⟩=⟨miniμij,maxiνij⟩,j=1,2,⋯,n。
对于决策方案方案 Y i ( i = 1 , 2 , ⋯ , m ) Y_{i}\left(i=1,2,\cdots,m\right) Yi(i=1,2,⋯,m),用 d + ( ω ) d^{+}(\omega) d+(ω)和 d − ( ω ) d^{-}(\omega) d−(ω)分别表示方案 Y i ( i = 1 , 2 , ⋯ , m ) Y_{i}\left(i=1,2,\cdots,m\right) Yi(i=1,2,⋯,m)到正理想解 Y + Y^{+} Y+和负理想解 Y − Y^{-} Y−的加权偏差,定义 d + ( ω ) d^{+}(\omega) d+(ω)和 d − ( ω ) d^{-}(\omega) d−(ω)分别为:
d i + ( ω ) = 1 2 ∑ j = 1 n ω j [ ∣ μ i j − μ j + ∣ + ∣ ν i j − ν j + ∣ + ∣ π i j − π j + ∣ ] \color{blue} { d_{i}^{+}(\omega) = \frac{1}{2} \sum_{j=1}^{n}\omega_{j}\left[ \left|\mu_{ij}-\mu_{j}^{+}\right| + \left|\nu_{ij}-\nu_{j}^{+}\right| + \left|\pi_{ij}-\pi_{j}^{+}\right| \right] } di+(ω)=21j=1∑nωj[∣∣μij−μj+∣∣+∣∣νij−νj+∣∣+∣∣πij−πj+∣∣]
d i − ( ω ) = 1 2 ∑ j = 1 n ω j [ ∣ μ i j − μ j − ∣ + ∣ ν i j − ν j − ∣ + ∣ π i j − π j − ∣ ] \color{blue} { d_{i}^{-}(\omega) = \frac{1}{2} \sum_{j=1}^{n}\omega_{j}\left[\left|\mu_{ij}-\mu_{j}^{-}\right| + \left|\nu_{ij}-\nu_{j}^{-}\right| + \left|\pi_{ij}-\pi_{j}^{-}\right| \right] } di−(ω)=21j=1∑nωj[∣∣μij−μj−∣∣+∣∣νij−νj−∣∣+∣∣πij−πj−∣∣]
对于给定的权重向量 ω = ( ω 1 , ω 2 , ⋯ , ω n ) T \omega = \left(\omega_{1},\omega{2},\cdots,\omega_{n}\right)^{T} ω=(ω1,ω2,⋯,ωn)T, d + ( ω ) d^{+}(\omega) d+(ω)越小而 d − ( ω ) d^{-}(\omega) d−(ω)越大,则相应的方案 Y i ( i = 1 , 2 , ⋯ , m ) Y_{i}\left(i=1,2,\cdots,m\right) Yi(i=1,2,⋯,m)越优。
当属性权重信息不完全时,若权重向量满足: ω ‾ j ≤ ω j ≤ ω ‾ j ∑ j = 1 n ω j 2 = 1 , ω j ≥ 0 ( j = 1 , 2 , ⋯ , n ) {\underline{\omega}_{j} \leq \omega_{j} \leq {\overline{\omega}_{j}}} \sum_{j=1}^{n}\omega_{j}^{2} = 1,\omega_{j} \geq 0 \left(j=1,2,\cdots,n\right) ωj≤ωj≤ωj∑j=1nωj2=1,ωj≥0(j=1,2,⋯,n),可建立多目标优化模型:
{ min d i + ( ω ) = 1 2 ∑ j = 1 n ω j [ ∣ μ i j − μ j + ∣ + ∣ ν i j − ν j + ∣ + ∣ π i j − π j + ∣ ] , i = 1 , 2 , ⋯ , m max d i − ( ω ) = 1 2 ∑ j = 1 n ω j [ ∣ μ i j − μ j − ∣ + ∣ ν i j − ν j − ∣ + ∣ π i j − π j − ∣ ] , i = 1 , 2 , ⋯ , m s.t. ω ‾ j ⩽ ω j ⩽ ω ˉ j , ∑ j = 1 n ω j = 1 , ω j ⩾ 0 , j = 1 , 2 , ⋯ , n (2.10) \left\{ \begin{array}{l} \min d_{i}^{+}(\omega)=\frac{1}{2} \sum_{j=1}^{n} \omega_{j}\left[\left|\mu_{i j}-\mu_{j}^{+}\right|+\left|\nu_{i j}-\nu_{j}^{+}\right|+\left|\pi_{i j}-\pi_{j}^{+}\right|\right], i=1,2, \cdots, m \\ \max d_{i}^{-}(\omega)=\frac{1}{2} \sum_{j=1}^{n} \omega_{j}\left[\left|\mu_{i j}-\mu_{j}^{-}\right|+\left|\nu_{i j}-\nu_{j}^{-}\right|+\left|\pi_{i j}-\pi_{j}^{-}\right|\right], i=1,2, \cdots, m \\ \text { s.t. } \underline{\omega}_{j} \leqslant \omega_{j} \leqslant \bar{\omega}_{j}, \sum_{j=1}^{n} \omega_{j}=1, \omega_{j} \geqslant 0, j=1,2, \cdots, n\end{array} \right. \tag{2.10} ⎩⎨⎧mindi+(ω)=21∑j=1nωj[∣∣μij−μj+∣∣+∣∣νij−νj+∣∣+∣∣πij−πj+∣∣],i=1,2,⋯,mmaxdi−(ω)=21∑j=1nωj[∣∣μij−μj−∣∣+∣∣νij−νj−∣∣+∣∣πij−πj−∣∣],i=1,2,⋯,m s.t. ωj⩽ωj⩽ωˉj,∑j=1nωj=1,ωj⩾0,j=1,2,⋯,n(2.10)
令
d + ( ω ) = ∑ i = 1 m d i + ( ω ) = 1 2 ∑ j = 1 n ∑ i = 1 m ω j [ ∣ μ i j − μ j + ∣ + ∣ ν i j − ν j + ∣ + ∣ π i j − π j + ∣ ] \begin{aligned} d^{+}(\omega) &= \sum_{i=1}^{m} d_{i}^{+}(\omega) \\ &= \frac{1}{2} \sum_{j=1}^{n} \sum_{i=1}^{m} \omega_{j}\left[\left|\mu_{i j}-\mu_{j}^{+}\right|+\left|\nu_{i j}-\nu_{j}^{+}\right|+\left|\pi_{i j}-\pi_{j}^{+}\right|\right] \end{aligned} d+(ω)=i=1∑mdi+(ω)=21j=1∑ni=1∑mωj[∣∣μij−μj+∣∣+∣∣νij−νj+∣∣+∣∣πij−πj+∣∣]
d − ( ω ) = ∑ i = 1 m d i − ( ω ) = 1 2 ∑ j = 1 n ∑ i = 1 m ω j [ ∣ μ i j − μ j − ∣ + ∣ ν i j − ν j − ∣ + ∣ π i j − π j − ∣ ] \begin{aligned} d^{-}(\omega) &= \sum_{i=1}^{m} d_{i}^{-}(\omega) \\ &= \frac{1}{2} \sum_{j=1}^{n} \sum_{i=1}^{m} \omega_{j}\left[\left|\mu_{i j}-\mu_{j}^{-}\right|+\left|\nu_{i j}-\nu_{j}^{-}\right|+\left|\pi_{i j}-\pi_{j}^{-}\right|\right] \end{aligned} d−(ω)=i=1∑mdi−(ω)=21j=1∑ni=1∑mωj[∣∣μij−μj−∣∣+∣∣νij−νj−∣∣+∣∣πij−πj−∣∣]
则 d + ( ω ) d^{+}(\omega) d+(ω)、 d − ( ω ) d^{-}(\omega) d−(ω)分别表示所有方案 Y i ( i = 1 , 2 , ⋯ , m ) Y_{i}\left(i=1,2,\cdots,m\right) Yi(i=1,2,⋯,m)到正理想解 Y + Y^{+} Y+和负理想解 Y − Y^{-} Y−的加权偏差总和。每个方案都是公平竞争的,不存在任何偏好关系,因此模型 ( 2.10 ) (2.10) (2.10)等权集结为如下单目标最优化模型:
{ min d + ( ω ) − d − ( ω ) = 1 2 ∑ j = 1 n ∑ i = 1 m ω j [ ∣ μ i j − μ j + ∣ + ∣ ν i j − ν j + ∣ + ∣ π i j − π j + ∣ − ∣ μ i j − μ j − ∣ − ∣ ν i j − ν j ∣ − ∣ π i j − π j − ∣ ] s.t. ω ‾ j ⩽ ω j ⩽ ω ˉ j , ∑ j = 1 n ω j = 1 , ω j ⩾ 0 , j = 1 , 2 , ⋯ , n (2.11) \left\{ \begin{array}{c} \min d^{+}(\omega)-d^{-}(\omega)=\frac{1}{2} \sum_{j=1}^{n} \sum_{i=1}^{m} \omega_{j}\left[\left|\mu_{i j}-\mu_{j}^{+}\right|+\left|\nu_{i j}-\nu_{j}^{+}\right|+\left|\pi_{i j}-\pi_{j}^{+}\right|\right. \\ \left.\quad-\left|\mu_{i j}-\mu_{j}^{-}\right|-\left|\nu_{i j}-\nu_{j}\right|-\left|\pi_{i j}-\pi_{j}^{-}\right|\right] \\ \text {s.t. } \underline{\omega}_{j} \leqslant \omega_{j} \leqslant \bar{\omega}_{j}, \sum_{j=1}^{n} \omega_{j}=1, \omega_{j} \geqslant 0, j=1,2, \cdots, n \end{array} \right. \tag{2.11} ⎩⎨⎧mind+(ω)−d−(ω)=21∑j=1n∑i=1mωj[∣∣μij−μj+∣∣+∣∣νij−νj+∣∣+∣∣πij−πj+∣∣−∣∣μij−μj−∣∣−∣νij−νj∣−∣∣πij−πj−∣∣]s.t. ωj⩽ωj⩽ωˉj,∑j=1nωj=1,ωj⩾0,j=1,2,⋯,n(2.11)
解最优化模型 ( 2.11 ) (2.11) (2.11),可得属性权重向量 ω = ( ω 1 , ω 2 , ⋯ , ω n ) T \omega = \left(\omega_{1},\omega{2},\cdots,\omega_{n}\right)^{T} ω=(ω1,ω2,⋯,ωn)T。
如果属性权重信息完全未知,则可构建以 d + ( ω ) d^{+}(\omega) d+(ω)最小化或 d − ( ω ) d^{-}(\omega) d−(ω)最大化为目标的最优化模型式 ( 2.12 ) (2.12) (2.12)和式 ( 2.13 ) (2.13) (2.13):
{ min d + ( ω ) = 1 2 ∑ j = 1 n ∑ i = 1 m ω j [ ∣ μ i j − μ j + ∣ + ∣ ν i j − ν j + ∣ + ∣ π i j − π j + ∣ ] s.t. ∑ j = 1 n ω j 2 = 1 , ω j ⩾ 0 , j = 1 , 2 , ⋯ , n (2.12) \left\{ \begin{array}{l} \min d^{+}(\omega)=\frac{1}{2} \sum_{j=1}^{n} \sum_{i=1}^{m} \omega_{j}\left[\left|\mu_{i j}-\mu_{j}^{+}\right|+\left|\nu_{i j}-\nu_{j}^{+}\right|+\left|\pi_{i j}-\pi_{j}^{+}\right|\right] \\ \text {s.t. } \sum_{j=1}^{n} \omega_{j}^{2}=1, \omega_{j} \geqslant 0, j=1,2, \cdots, n \end{array} \right. \tag{2.12} {mind+(ω)=21∑j=1n∑i=1mωj[∣∣μij−μj+∣∣+∣∣νij−νj+∣∣+∣∣πij−πj+∣∣]s.t. ∑j=1nωj2=1,ωj⩾0,j=1,2,⋯,n(2.12)
{ max d + ( ω ) = 1 2 ∑ j = 1 n ∑ i = 1 m ω j [ ∣ μ i j − μ j − ∣ + ∣ ν i j − ν j − ∣ + ∣ π i j − π j − ∣ ] s.t. ∑ j = 1 n ω j 2 = 1 , ω j ⩾ 0 , j = 1 , 2 , ⋯ , n (2.13) \left\{ \begin{array}{l} \max d^{+}(\omega)=\frac{1}{2} \sum_{j=1}^{n} \sum_{i=1}^{m} \omega_{j}\left[\left|\mu_{i j}-\mu_{j}^{-}\right|+\left|\nu_{i j}-\nu_{j}^{-}\right|+\left|\pi_{i j}-\pi_{j}^{-}\right|\right] \\ \text {s.t. } \sum_{j=1}^{n} \omega_{j}^{2}=1, \omega_{j} \geqslant 0, j=1,2, \cdots, n \end{array} \right. \tag{2.13} {maxd+(ω)=21∑j=1n∑i=1mωj[∣∣μij−μj−∣∣+∣∣νij−νj−∣∣+∣∣πij−πj−∣∣]s.t. ∑j=1nωj2=1,ωj⩾0,j=1,2,⋯,n(2.13)
通过构造拉格朗日函数,利用极值理论求解最优化模型式 ( 2.12 ) (2.12) (2.12)或式 ( 2.13 ) (2.13) (2.13),可得其最优解分别为
ω j ∗ = ∑ i = 1 m [ ∣ μ i j − μ j + ∣ + ∣ ν i j − ν i j + ∣ + ∣ π i j − π j + ∣ ] ∑ j = 1 n [ ∑ i = 1 m [ ∣ μ i j − μ j + ∣ + ∣ ν i j − ν j + ∣ + ∣ π i j − π j + ∣ ] ] 2 , j = 1 , 2 , ⋯ , n (2.14) \omega_{j}^{*}=\frac{\sum_{i=1}^{m}\left[\left|\mu_{i j}-\mu_{j}^{+}\right|+\left|\nu_{i j}-\nu_{i j}^{+}\right|+\left|\pi_{i j}-\pi_{j}^{+}\right|\right]}{\sqrt{\sum_{j=1}^{n}\left[\sum_{i=1}^{m}\left[\left|\mu_{i j}-\mu_{j}^{+}\right|+\left|\nu_{i j}-\nu_{j}^{+}\right|+\left|\pi_{i j}-\pi_{j}^{+}\right|\right]\right]^{2}}}, \quad j=1,2, \cdots, n \tag{2.14} ωj∗=∑j=1n[∑i=1m[∣∣μij−μj+∣∣+∣∣νij−νj+∣∣+∣∣πij−πj+∣∣]]2∑i=1m[∣∣μij−μj+∣∣+∣∣νij−νij+∣∣+∣∣πij−πj+∣∣],j=1,2,⋯,n(2.14)
或
ω j ∗ = ∑ i = 1 m [ ∣ μ i j − μ j − ∣ + ∣ ν i j − ν i j − ∣ + ∣ π i j − π j − ∣ ] ∑ j = 1 n [ ∑ i = 1 m [ ∣ μ i j − μ j − ∣ + ∣ ν i j − ν j − ∣ + ∣ π i j − π j − ∣ ] ] 2 , j = 1 , 2 , ⋯ , n (2.15) \omega_{j}^{*}=\frac{\sum_{i=1}^{m}\left[\left|\mu_{i j}-\mu_{j}^{-}\right|+\left|\nu_{i j}-\nu_{i j}^{-}\right|+\left|\pi_{i j}-\pi_{j}^{-}\right|\right]}{\sqrt{\sum_{j=1}^{n}\left[\sum_{i=1}^{m}\left[\left|\mu_{i j}-\mu_{j}^{-}\right|+\left|\nu_{i j}-\nu_{j}^{-}\right|+\left|\pi_{i j}-\pi_{j}^{-}\right|\right]\right]^{2}}}, \quad j=1,2, \cdots, n \tag{2.15} ωj∗=∑j=1n[∑i=1m[∣∣μij−μj−∣∣+∣∣νij−νj−∣∣+∣∣πij−πj−∣∣]]2∑i=1m[∣∣μij−μj−∣∣+∣∣νij−νij−∣∣+∣∣πij−πj−∣∣],j=1,2,⋯,n(2.15)
对 ω j ∗ \omega_{j}^{*} ωj∗进行归一化处理可得属性 G j ( j = 1 , 2 , ⋯ , n ) G_{j}\left(j=1,2,\cdots,n\right) Gj(j=1,2,⋯,n)的权重为:
ω j = ∑ i = 1 m [ ∣ μ i j − μ j + ∣ + ∣ ν i j − ν i j + ∣ + ∣ π i j − π j + ∣ ] ∑ j = 1 n ∑ i = 1 m [ ∣ μ i j − μ j + ∣ + ∣ ν i j − ν j + ∣ + ∣ π i j − π j + ∣ ] , j = 1 , 2 , ⋯ , n (2.16) \omega_{j} = \frac {\sum_{i=1}^{m}\left[\left|\mu_{i j}-\mu_{j}^{+}\right|+\left|\nu_{i j}-\nu_{i j}^{+}\right|+\left|\pi_{i j}-\pi_{j}^{+}\right|\right]} {\sum_{j=1}^{n}\sum_{i=1}^{m}\left[\left|\mu_{i j}-\mu_{j}^{+}\right|+\left|\nu_{i j}-\nu_{j}^{+}\right|+\left|\pi_{i j}-\pi_{j}^{+}\right|\right]}, \quad j=1,2, \cdots, n \tag{2.16} ωj=∑j=1n∑i=1m[∣∣μij−μj+∣∣+∣∣νij−νj+∣∣+∣∣πij−πj+∣∣]∑i=1m[∣∣μij−μj+∣∣+∣∣νij−νij+∣∣+∣∣πij−πj+∣∣],j=1,2,⋯,n(2.16)
或
ω j = ∑ i = 1 m [ ∣ μ i j − μ j − ∣ + ∣ ν i j − ν i j − ∣ + ∣ π i j − π j − ∣ ] ∑ j = 1 n ∑ i = 1 m [ ∣ μ i j − μ j − ∣ + ∣ ν i j − ν j − ∣ + ∣ π i j − π j − ∣ ] , j = 1 , 2 , ⋯ , n (2.17) \omega_{j} = \frac {\sum_{i=1}^{m}\left[\left|\mu_{i j}-\mu_{j}^{-}\right|+\left|\nu_{i j}-\nu_{i j}^{-}\right|+\left|\pi_{i j}-\pi_{j}^{-}\right|\right]} {\sum_{j=1}^{n}\sum_{i=1}^{m} {\left[\left|\mu_{i j}-\mu_{j}^{-}\right|+\left|\nu_{i j}-\nu_{j}^{-}\right|+\left|\pi_{i j}-\pi_{j}^{-}\right|\right]}}, \quad j=1,2, \cdots, n \tag{2.17} ωj=∑j=1n∑i=1m[∣∣μij−μj−∣∣+∣∣νij−νj−∣∣+∣∣πij−πj−∣∣]∑i=1m[∣∣μij−μj−∣∣+∣∣νij−νij−∣∣+∣∣πij−πj−∣∣],j=1,2,⋯,n(2.17)
<案例分析待补充!>