Part II. S1. 基于直觉模糊混合平均算子的多属性决策方法

1.1 直觉模糊加权平均算子

---

  记论域X上所有直觉模糊数为$F\left(X\right)$

定义1.1

  设${\tilde{A}}_j=⟨{\mu }_j,{\nu }_j⟩\left(j=1,2,...,n\right)$是一组直觉模糊数，若$\mathrm{IFWA}$是一个映射：$F_n\to F$，使得：
$${\mathrm{IFWA}}_{\omega }\left({\tilde{A}}_1,{\tilde{A}}_2,...,{\tilde{A}}_n\right)={\omega }_1{\tilde{A}}_1\oplus {\omega }_2{\tilde{A}}_2\oplus ...\oplus {\omega }_n{\tilde{A}}_n\text{\hspace{1em}(1.1)}$$

  则称$\mathrm{IFWA}$为直觉模糊加权平均算子，其中$\mathbf{\omega }={\left({\omega }_1,{\omega }_2,...,{\omega }_n\right)}^T$为${\tilde{A}}_j=⟨{\mu }_j,{\nu }_j⟩\left(j=1,2,...,n\right)$的权重向量，${\omega }_j\in [0,1]\left(j=1,2,...,n\right),{\sum }_{j=1}^n{\omega }_j=1$

  特别地，若$\mathbf{\omega }={\left(\frac{1}{n},\frac{1}{n},...,\frac{1}{n}\right)}^T$，则$\mathrm{IFWA}$算子退化为直觉模糊平均(IFA)算子：
$${\mathrm{IFWA}}_{\omega }\left({\tilde{A}}_1,{\tilde{A}}_2,...,{\tilde{A}}_n\right)=\frac{1}{n}\left({\tilde{A}}_1\oplus {\tilde{A}}_2\oplus ...\oplus {\tilde{A}}_n\right)\text{\hspace{1em}(1.2)}$$

---

定理1.1
  设${\tilde{A}}_j=⟨{\mu }_j,{\nu }_j⟩\left(j=1,2,...,n\right)$是一组直觉模糊数，则有$\mathrm{IFWA}$算子运算得到的结果仍然是直觉模糊数，且

$$\underline{{\mathrm{IFWA}}_{\omega }\left({\tilde{A}}_1,{\tilde{A}}_2,...{\tilde{A}}_n\right)=⟨1-{\prod _{j=1}^n\left(1-{\mu }_j\right)}^{{\omega }_j},\prod _{j=1}^n{\left({\nu }_j\right)}^{{\omega }_j}⟩}\text{\hspace{1em}(1.3)}$$

推理过程：
  已知：（1）${\alpha }_1\oplus {\alpha }_2=\left({\mu }_{{\alpha }_1}+{\mu }_{{\alpha }_2}-{\mu }_{{\alpha }_1}{\mu }_{{\alpha }_2},{\nu }_{{\alpha }_1}{\nu }_{{\alpha }_2}\right)$；（2）$\lambda \alpha =\left(1-{\left(1-{\mu }_{\alpha }\right)}^{\lambda },{\nu }_{\alpha }^{\lambda }\right)$

  原式：${\mathrm{IFWA}}_{\omega }\left({\tilde{A}}_1,{\tilde{A}}_2,...,{\tilde{A}}_n\right)={\omega }_1{\tilde{A}}_1\oplus {\omega }_2{\tilde{A}}_2\oplus ...\oplus {\omega }_n{\tilde{A}}_n$

  S1.计算：${\omega }_1{\tilde{A}}_1,{\omega }_2{\tilde{A}}_2,...,{\omega }_n{\tilde{A}}_n$
  R1.可得：$T_1={\omega }_1{\tilde{A}}_1=\left(1-{\left(1-{\mu }_1\right)}^{{\omega }_1},{{\nu }_1}^{{\omega }_1}\right)$，$T_2={\omega }_2{\tilde{A}}_2=\left(1-{\left(1-{\mu }_2\right)}^{{\omega }_2},{{\nu }_2}^{{\omega }_2}\right)$，…，$T_n={\omega }_n{\tilde{A}}_n=\left(1-{\left(1-{\mu }_n\right)}^{{\omega }_n},{{\nu }_n}^{{\omega }_n}\right)$
  S2.计算：$T_1\oplus T_2$
  R2.可得：

$$\begin{array}{rl}{T}_1\oplus T_2& =\left(1-{\left(1-{\mu }_1\right)}^{{\omega }_1}+1-{\left(1-{\mu }_2\right)}^{{\omega }_2}-\left(1-{\left(1-{\mu }_1\right)}^{{\omega }_1}\right)\left(1-{\left(1-{\mu }_1\right)}^{{\omega }_1}\right),{{\nu }_1}^{{\omega }_1}{{\nu }_2}^{{\omega }_2}\right)\\ & =\left(2-{\left(1-{\mu }_1\right)}^{{\omega }_1}-{\left(1-{\mu }_2\right)}^{{\omega }_2}-\left[1-{\left(1-{\mu }_2\right)}^{{\omega }_2}-{\left(1-{\mu }_1\right)}^{{\omega }_1}+{\left(1-{\mu }_1\right)}^{{\omega }_1}{\left(1-{\mu }_2\right)}^{{\omega }_2}\right],{{\nu }_1}^{{\omega }_1}{{\nu }_2}^{{\omega }_2}\right)\\ & =\left(1-{\left(1-{\mu }_1\right)}^{{\omega }_1}{\left(1-{\mu }_2\right)}^{{\omega }_2},{{\nu }_1}^{{\omega }_1}{{\nu }_2}^{{\omega }_2}\right)\end{array}$$
    同理可以计算得到：$T_1\oplus T_2\oplus T_3=\left(1-{\left(1-{\mu }_1\right)}^{{\omega }_1}{\left(1-{\mu }_2\right)}^{{\omega }_2}{\left(1-{\mu }_3\right)}^{{\omega }_3},{{\nu }_1}^{{\omega }_1}{{\nu }_2}^{{\omega }_2}{{\nu }_3}^{{\omega }_3}\right)$

    依次类推，可得到：$T_1\oplus T_2\oplus ...\oplus T_n=\left(1-{\left(1-{\mu }_1\right)}^{{\omega }_1}{\left(1-{\mu }_2\right)}^{{\omega }_2}...{\left(1-{\mu }_n\right)}^{{\omega }_n},{{\nu }_1}^{{\omega }_1}{{\nu }_2}^{{\omega }_2}...{{\nu }_n}^{{\omega }_n}\right)$

    合并结果为：$T_1\oplus T_2\oplus ...\oplus T_n=⟨1-{{\prod }_{j=1}^n\left(1-{\mu }_j\right)}^{{\omega }_j},{\prod }_{j=1}^n{\left({\nu }_j\right)}^{{\omega }_j}⟩$

  即：${\mathrm{IFWA}}_{\omega }\left({\tilde{A}}_1,{\tilde{A}}_2,...{\tilde{A}}_n\right)=⟨1-{{\prod }_{j=1}^n\left(1-{\mu }_j\right)}^{{\omega }_j},{\prod }_{j=1}^n{\left({\nu }_j\right)}^{{\omega }_j}⟩$

  $示例1.1$

  设${\tilde{A}}_1=⟨0.2,0.6⟩$，${\tilde{A}}_2=⟨0.4,0.5⟩$，${\tilde{A}}_3=⟨0.6,0.3⟩$，${\tilde{A}}_4=⟨0.3,0.7⟩$为四个直觉模糊数，$\omega ={\left(0.2,0.4,0.1,0.3\right)}^T$为其权重向量，则$\mathrm{IFWA}$结果是多少？

import numpy as np

IF_matrix = np.array([[0.2,0.4,0.6,0.3],[0.6,0.5,0.3,0.7]])
weight = np.array([0.2,0.4,0.1,0.3])

def calculate_IFWA(IF_matrix,weight):
    IFWA = [1,1]
    for i in range(IF_matrix.shape[1]):
        IFWA[0] *= np.power(1-IF_matrix[0][i],weight[i])
        IFWA[1] *= np.power(IF_matrix[1][i],weight[i])
    IFWA[0] = 1 - IFWA[0]
    return IFWA

print("IF矩阵：\n",IF_matrix)
print("权重向量：\n",weight)
print('IFWA：\n',calculate_IFWA(IF_matrix,weight))

  计算结果如下：

IF矩阵：
[[0.2 0.4 0.6 0.3]
[0.6 0.5 0.3 0.7]]
权重向量：
[0.2 0.4 0.1 0.3]
IFWA：
[0.36083364123865835, 0.5450794252184181]

---

1.2 直觉模糊有序加权平均算子

定义1.2
  设${\tilde{A}}_j=⟨{\mu }_j,{\nu }_j⟩\left(j=1,2,...,n\right)$是一组直觉模糊数，若$\mathrm{IFOWA}$是一个映射：$F_n\to F$，使得：
$${\mathrm{IFOWA}}_{\omega }\left({\tilde{A}}_1,{\tilde{A}}_2,...,{\tilde{A}}_n\right)={\omega }_1{\tilde{A}}_{\sigma (1)}\oplus {\omega }_2{\tilde{A}}_{\sigma (2)}\oplus ...\oplus {\omega }_n{\tilde{A}}_{\sigma (n)}\text{\hspace{1em}(1.4)}$$
  则称$\mathrm{IFOWA}$为直觉模糊有序加权平均算子，其中$\mathbf{\omega }={\left({\omega }_1,{\omega }_2,...,{\omega }_n\right)}^T$为与$\mathrm{IFOWA}$算子相关联的权重向量，${\omega }_j\in [0,1]\left(j=1,2,...,n\right),{\sum }_{j=1}^n{\omega }_j=1$，$\left(\sigma (1),\sigma (2),...,\sigma (n)\right)$为数组$\left(1,2,...,n\right)$的一个置换，使得对任意k，有${\tilde{A}}_{\sigma (k1)}\ge {\tilde{A}}_{\sigma (k)}$，即${\tilde{A}}_{\sigma (k)}$是直觉模糊数${\tilde{A}}_j=⟨{\mu }_j,{\nu }_j⟩\left(j=1,2,...,n\right)$按直觉模糊数的排序规则确定的第k个最大直觉模糊数。

  特别地，若$\mathbf{\omega }={\left(\frac{1}{n},\frac{1}{n},...,\frac{1}{n}\right)}^T$，则$\mathrm{IFOWA}$算子退化为直觉模平均(IFA)算子：
$${\mathrm{IFOWA}}_{\omega }\left({\tilde{A}}_1,{\tilde{A}}_2,...,{\tilde{A}}_n\right)=\frac{1}{n}\left({\tilde{A}}_1\oplus {\tilde{A}}_2\oplus ...\oplus {\tilde{A}}_n\right)$$

定理1.2
  设${\tilde{A}}_j=⟨{\mu }_j,{\nu }_j⟩\left(j=1,2,...,n\right)$是一组直觉模糊数，${\tilde{A}}_{\sigma (k)}$是直觉模糊数${\tilde{A}}_j=⟨{\mu }_j,{\nu }_j⟩\left(j=1,2,...,n\right)$按直觉模糊数的排序规则确定的第k个最大直觉模糊数，则有$\mathrm{IFOWA}$算子运算得到的结果仍然是直觉模糊数，且

$$\underline{{\mathrm{IFOWA}}_{\omega }\left({\tilde{A}}_1,{\tilde{A}}_2,...{\tilde{A}}_n\right)=⟨1-{\prod _{j=1}^n\left(1-{\mu }_{\sigma (j)}\right)}^{{\omega }_j},\prod _{j=1}^n{\left({\nu }_{\sigma (j)}\right)}^{{\omega }_j}⟩}\text{\hspace{1em}(1.5)}$$

  $示例1.2$

  设${\tilde{A}}_1=⟨0.1,0.2⟩$，${\tilde{A}}_2=⟨0.2,0.4⟩$，${\tilde{A}}_3=⟨0.6,0.2⟩$，${\tilde{A}}_4=⟨0.4,0.2⟩$为四个直觉模糊数，$\omega ={\left(0.2,0.3,0.3,0.2\right)}^T$为其权重向量，则$\mathrm{IFOWA}$结果是多少？

import numpy as np

def IF_rank(IF_matrix):
    # 计算每个直觉模糊数的得分值score=μ-v
    score = IF_matrix[0] - IF_matrix[1]
    # 按照得分值从大到小进行重新排序
    index = np.argsort(-score)
    IF_sequence_matrix = np.zeros(IF_matrix.shape)
    for i in range(IF_matrix.shape[1]):
        IF_sequence_matrix[0][i] = IF_matrix[0][index[i]]
        IF_sequence_matrix[1][i] = IF_matrix[1][index[i]]
    return IF_sequence_matrix

def calculate_IFOWA(IF_sequence_matrix,weight):
    IFOWA = [1,1]
    for i in range(IF_sequence_matrix.shape[1]):
        IFOWA[0] *= np.power(1-IF_sequence_matrix[0][i],weight[i])
        IFOWA[1] *= np.power(IF_sequence_matrix[1][i],weight[i])
    IFOWA[0] = 1 - IFOWA[0]
    return IFOWA

IF_matrix = np.array([[0.1,0.2,0.6,0.4],[0.2,0.4,0.2,0.2]])
weight = [0.2,0.3,0.3,0.2]

# 首先对IF_matrix进行排序
print('原始模糊直觉为：\n',IF_matrix)
IF_sequence_matrix = IF_rank(IF_matrix)
print('有序模糊直觉为：\n',IF_sequence_matrix)
# 然后计算IFOWA
IFOWA = calculate_IFOWA(IF_sequence_matrix,weight)
print('IFOWA：\n',IFOWA)

  计算结果如下：

原始模糊直觉为：
[[0.1 0.2 0.6 0.4]
[0.2 0.4 0.2 0.2]]
有序模糊直觉为：
[[0.6 0.4 0.1 0.2]
[0.2 0.2 0.2 0.4]]
IFOWA：
[0.33816744854675806, 0.22973967099940704]

---

1.3 直觉模糊混合平均算子

定义1.3
  设${\tilde{A}}_j=⟨{\mu }_j,{\nu }_j⟩\left(j=1,2,...,n\right)$是一组直觉模糊数，若$\mathrm{IFHA}$是一个映射：$F_n\to F$，使得：
$${\mathrm{IFHA}}_w\left({\tilde{A}}_1,{\tilde{A}}_2,...,{\tilde{A}}_n\right)=w_1{\widetilde{A^{{}^{\prime }}}}_{\sigma (1)}\oplus w_2{\widetilde{A^{{}^{\prime }}}}_{\sigma (2)}\oplus ...\oplus w_n{\widetilde{A^{{}^{\prime }}}}_{\sigma (n)}\text{\hspace{1em}(1.6)}$$

  则称$\mathrm{IFHA}$为直觉模糊混合平均算子，其中$\mathbf{w}={\left(w_1,w_2,...,w_n\right)}^T$为与$\mathrm{IFHA}$算子相关联的权重向量，$w_j\in [0,1]\left(j=1,2,...,n\right),{\sum }_{j=1}^n{w}_j=1$，$\left({\widetilde{A^{{}^{\prime }}}}_{\sigma (1)},{\widetilde{A^{{}^{\prime }}}}_{\sigma (2)},...,{\widetilde{A^{{}^{\prime }}}}_{\sigma (n)}\right)$为加权的直觉模糊数数组$\left({\widetilde{A^{{}^{\prime }}}}_1,{\widetilde{A^{{}^{\prime }}}}_2,...,{\widetilde{A^{{}^{\prime }}}}_n\right)$的一个置换，使得对任意k，有${\widetilde{A^{{}^{\prime }}}}_{\sigma (k-1)}\ge {\widetilde{A^{{}^{\prime }}}}_{\sigma (k)}$，即${\widetilde{A^{{}^{\prime }}}}_{\sigma (k)}$是直觉模糊数${\widetilde{A^{{}^{\prime }}}}_j=⟨{\mu }_j^{{}^{\prime }},{\nu }_j^{{}^{\prime }}⟩\left(j=1,2,...,n\right)$按直觉模糊数的排序规则确定的第k个最大直觉模糊数。$\mathbf{\omega }={\left({\omega }_1,{\omega }_2,...,{\omega }_n\right)}^T$为${\tilde{A}}_j=⟨{\mu }_j,{\nu }_j⟩\left(j=1,2,...,n\right)$的权重向量，${\omega }_j\in [0,1]\left(j=1,2,...,n\right),{\sum }_{j=1}^n{\omega }_j=1$，$n$为平衡系数。

  特别地，若$\mathbf{w}={\left(\frac{1}{n},\frac{1}{n},...,\frac{1}{n}\right)}^T$，则$\mathrm{IFHA}$算子退化为直觉模糊加权平均算子$\mathrm{IFWA}$；若$\mathbf{\omega }={\left(\frac{1}{n},\frac{1}{n},...,\frac{1}{n}\right)}^T$，则$\mathrm{IFHA}$算子退化为直觉模糊有序加权平均算子$\mathrm{IFOWA}$。
  直觉模糊混合平均算子$\mathrm{IFHA}$是直觉模糊加权平均算子$\mathrm{IFWA}$和直觉模糊有序加权平均算子$\mathrm{IFOWA}$的拓展，它充分考虑了待集结直觉模糊集自身重要性及其所在位置重要程度两方面的信息。

定理1.3
  设${\tilde{A}}_j=⟨{\mu }_j,{\nu }_j⟩\left(j=1,2,...,n\right)$是一组直觉模糊数，令${\widetilde{A^{{}^{\prime }}}}_j=n{\omega }_j{\tilde{A}}_j=⟨{\mu }_j^{{}^{\prime }},{\nu }_j^{{}^{\prime }}⟩\left(j=1,2,...,n\right)$，${\widetilde{A^{{}^{\prime }}}}_{\sigma (k)}$是直觉模糊数${\widetilde{A^{{}^{\prime }}}}_j=⟨{\mu }_j^{{}^{\prime }},{\nu }_j^{{}^{\prime }}⟩\left(j=1,2,...,n\right)$中按直觉模糊数的排序规则确定的第k个最大直觉模糊数，则有$\mathrm{IFHA}$算子运算得到的结果仍然是直觉模糊数，且

$$\underline{{\mathrm{IFHA}}_{\omega ,w}\left({\tilde{A}}_1,{\tilde{A}}_2,...{\tilde{A}}_n\right)=⟨1-{\prod _{j=1}^n\left(1-{\mu }_{\sigma (j)}^{{}^{\prime }}\right)}^{{\omega }_j},\prod _{j=1}^n{\left({\nu }_{\sigma (j)}^{{}^{\prime }}\right)}^{{\omega }_j}⟩}\text{\hspace{1em}(1.7)}$$

  其中，$\mathbf{w}={\left(w_1,w_2,...,w_n\right)}^T$为与$\mathrm{IFOWA}$算子相关联的权重向量， $w_j\in [0,1]\left(j=1,2,...,n\right),{\sum }_{j=1}^n{w}_j=1$；
$\mathbf{\omega }={\left({\omega }_1,{\omega }_2,...,{\omega }_n\right)}^T$为${\tilde{A}}_j=⟨{\mu }_j,{\nu }_j⟩\left(j=1,2,...,n\right)$的权重向量，${\omega }_j\in [0,1]\left(j=1,2,...,n\right),{\sum }_{j=1}^n{\omega }_j=1$，$n$为平衡系数。

  $示例1.3$

  设${\tilde{A}}_1=⟨0.2,0.5⟩$，${\tilde{A}}_2=⟨0.3,0.4⟩$，${\tilde{A}}_3=⟨0.5,0.1⟩$，${\tilde{A}}_4=⟨0.7,0.2⟩$，${\tilde{A}}_4=⟨0.6,0.3⟩$为五个直觉模糊数，$\omega ={\left(0.25,0.15,0.20,0.18,0.22\right)}^T$为${\tilde{A}}_j\left(j=1,2,3,4,5\right)$的权重向量，$w={\left(0.112,0.236,0.304,0.236,0.112\right)}^T$为与$\mathrm{IFHA}$算子相关联的权重向量， 则$\mathrm{IFHA}$结果是多少？

import numpy as np

def calculate_weight_IF(IF_matrix,weight_IF):
    weight_IF *= IF_matrix.shape[1] # 计算<A_j_'> = n×<w_j>×<A_j>
    weighted_IF_matrix = np.zeros(IF_matrix.shape) # 初始化矩阵用于存放加权后的结果
    weighted_IF_matrix[0] = 1 - np.power(1-IF_matrix[0],weight_IF)  # 计算加权后的μ_i = 1-(1-<u_i>)^<w_i>
    weighted_IF_matrix[1] = np.power(IF_matrix[1],weight_IF) # 计算加权后的v_i = (<v_i>)^<w_i>
    return weighted_IF_matrix

def IF_rank(IF_matrix):
    # 计算每个直觉模糊数的得分值score=μ-v
    score = IF_matrix[0] - IF_matrix[1]
    # 按照得分值从大到小进行重新排序
    index = np.argsort(-score)  # 获取从大到小排序索引
    IF_sequence_matrix = np.zeros(IF_matrix.shape)  # 初始化结果矩阵
    for i in range(IF_matrix.shape[1]):  # 更新结果矩阵
        IF_sequence_matrix[0][i] = IF_matrix[0][index[i]]
        IF_sequence_matrix[1][i] = IF_matrix[1][index[i]]
    return IF_sequence_matrix

def calculate_IFHA(IF_sequence_matrix,weight):
    IFOWA = [1,1]
    for i in range(IF_sequence_matrix.shape[1]):
        IFOWA[0] *= np.power(1-IF_sequence_matrix[0][i],weight[i])
        IFOWA[1] *= np.power(IF_sequence_matrix[1][i],weight[i])
    IFOWA[0] = 1 - IFOWA[0]
    return IFOWA

# 输入
IF_matrix = np.array([[0.2,0.3,0.5,0.7,0.6],[0.5,0.4,0.1,0.2,0.3]])
weight_IF = np.array([0.25,0.15,0.20,0.18,0.22])
weight_IFHA = np.array([0.112,0.236,0.304,0.236,0.112])

# S1: 计算加权直觉模糊数A_j_'
weighted_IF_matrix = calculate_weight_IF(IF_matrix,weight_IF)
print('Weighted_IF_matrix：\n',weighted_IF_matrix)
# S2：计算A_j_'的得分值
# S3：排序
ranked_IF = IF_rank(weighted_IF_matrix)
print('Ranked_IF：\n',ranked_IF)
# S4：计算IFHA
IFHA = calculate_IFHA(ranked_IF,weight_IFHA)
print('IFHA：\n',IFHA)

  计算结果如下：

Weighted_IF_matrix：
[[0.24340671 0.23471442 0.5 0.66161654 0.63502259]
[0.42044821 0.50297337 0.1 0.23492379 0.26597045]]
Ranked_IF：
[[0.66161654 0.5 0.63502259 0.24340671 0.23471442]
[0.23492379 0.1 0.26597045 0.42044821 0.50297337]]
IFHA：
[0.49698679510742627, 0.2491508434303773]

---

1.4 基于直觉模糊混合平均算子的多属性决策步骤与实例分析

  直觉模糊多属性决策问题的描述

  设某多属性决策问题有m个方案$Y_i\left(i=1,2,...,m\right)$组成方案集 Y = { Y 1 , Y 2 , . . . , Y m } Y=\left\{ Y_{1},Y_{2},...,Y_{m} \right\} Y={Y1​,Y2​,...,Ym​}，评价每个方案的属性（或指标）为$G_j\left(j=1,2,...,n\right)$，记属性集为 G = { G 1 , G 2 , . . . , G n } G=\left\{ G_{1},G_{2},...,G_{n} \right\} G={G1​,G2​,...,Gn​}。假设方案$Y_i\in Y$关于属性$G_j\in G$的评价值可以表示为直觉模糊集 F i j ~ = { ⟨ G j , Y i ⟩ , μ i j , ν i j } ( i = 1 , 2 , . . . , m ; j = 1 , 2 , . . . , n ) \tilde{F_{ij}}=\left\{ \left\langle G_{j},Y_{i}\right\rangle,\mu_{ij},\nu_{ij} \right\}\left(i=1,2,...,m;j=1,2,...,n\right) Fij​~​={⟨Gj​,Yi​⟩,μij​,νij​}(i=1,2,...,m;j=1,2,...,n)，其中${\mu }_{ij}\in [0,1]$和${\nu }_{ij}\in [0,1]$分别表示方案$Y_i\in Y$满足属性$G_j\in G$和不满足属性$G_j\in G$的程度，且$0\le {\mu }_{ij}+{\nu }_{ij}\le 1$。

  为方便起见，通常将 F i j ~ = { ⟨ G j , Y i ⟩ , μ i j , ν i j } ( i = 1 , 2 , . . . , m ; j = 1 , 2 , . . . , n ) \tilde{F_{ij}}=\left\{ \left\langle G_{j},Y_{i}\right\rangle,\mu_{ij},\nu_{ij} \right\}\left(i=1,2,...,m;j=1,2,...,n\right) Fij​~​={⟨Gj​,Yi​⟩,μij​,νij​}(i=1,2,...,m;j=1,2,...,n)简记为$\widetilde{F_{ij}}=⟨{\mu }_{ij},{\nu }_{ij}⟩\left(i=1,2,...,m;j=1,2,...,n\right)$，即${\mu }_{ij}$表示方案$Y_i\in Y$满足属性$G_j\in G$的程度，即${\nu }_{ij}$表示方案$Y_i\in Y$不满足属性$G_j\in G$的程度。于是$Y_i\in Y\left(i=1,2,...,m\right)$的所有$n$个属性评价值可以简单地记为向量${\tilde{F}}_i={\left({\tilde{F}}_{i1},{\tilde{F}}_{i2},...,{\tilde{F}}_{in}\right)}^T={\left(⟨{\mu }_{i1},{\nu }_{i1}⟩,⟨{\mu }_{i2},{\nu }_{i2}⟩,...,⟨{\mu }_{in},{\nu }_{in}⟩\right)}^T$。此时，所有方案$Y_i\left(i=1,2,...,m\right)$关于属性$G_j\left(j=1,2,...,n\right)$的直觉模糊评价信息可以用一个直觉模糊决策矩阵$\mathbf{F}={\left(⟨{\mu }_{ij},{\nu }_{ij}⟩\right)}_{m\times n}$表示。

表2-1 直觉模糊决策矩阵$F$

	$G_1$	$G_2$	⋯	$G_n$
$Y_1$	$⟨{\mu }_{11},{\nu }_{11}⟩$	$⟨{\mu }_{12},{\nu }_{12}⟩$	⋯	$⟨{\mu }_{1n},{\nu }_{1n}⟩$
$Y_2$	$⟨{\mu }_{21},{\nu }_{21}⟩$	$⟨{\mu }_{22},{\nu }_{22}⟩$	⋯	$⟨{\mu }_{2n},{\nu }_{2n}⟩$
…	…	…	…	…
$Y_m$	$⟨{\mu }_{m1},{\nu }_{m1}⟩$	$⟨{\mu }_{m2},{\nu }_{m2}⟩$	⋯	$⟨{\mu }_{mn},{\nu }_{mn}⟩$

---

1.5 基于直觉模糊混合平均算子的多属性决策步骤

  基于直觉模糊混合平均算子的多属性决策步骤
  步骤 1 确定多属性决策问题的方案集 Y = { Y 1 , Y 2 , . . . , Y m } Y=\left\{Y_{1},Y_{2},...,Y_{m}\right\} Y={Y1​,Y2​,...,Ym​}和属性集 G = { G 1 , G 2 , . . . , G n } G=\left\{G_{1},G_{2},...,G_{n}\right\} G={G1​,G2​,...,Gn​}。
  步骤 2 获取多属性决策问题中各方案$Y_i\in Y$关于属性$G_j\in G$的直觉模糊特征信息，构建直觉模糊决策矩阵$F$。
  步骤 3 确定多属性决策问题各属性的权重，得到属性权重向量$\mathbf{w}={\left(w_1,w_2,...,w_n\right)}^T$。
  步骤 4 利用正态分布赋权法等确定与$\mathrm{IFHA}$算子相关联的权重向量（或位置向量）$\mathbf{\omega }={\left({\omega }_1,{\omega }_2,...,{\omega }_n\right)}^T$。
  步骤 4 利用${\mathrm{IFHA}}_{\omega ,w}\left({\tilde{A}}_1,{\tilde{A}}_2,...{\tilde{A}}_n\right)=⟨1-{{\prod }_{j=1}^n\left(1-{\mu }_{\sigma (j)}^{{}^{\prime }}\right)}^{{\omega }_j},{\prod }_{j=1}^n{\left({\nu }_{\sigma (j)}^{{}^{\prime }}\right)}^{{\omega }_j}⟩$计算方案$Y_i$的综合属性值${\tilde{d}}_i={\mathrm{IFHA}}_{w,\omega }\left(\widetilde{(}F{)}_{i1},\widetilde{(}F{)}_{i2},...,\widetilde{(}F{)}_{in},\right)$。首先利用属性权重$\mathbf{w}={\left(w_1,w_2,...,w_n\right)}^T$和平衡系数$n$计算加权的直觉模糊数${\widetilde{F^{{}^{\prime }}}}_{ij}=n{w}_j{\tilde{F}}_{ij}$，然后利用直觉模糊数的排序规则对${\widetilde{F^{{}^{\prime }}}}_{ij}\left(j=1,2,...,n\right)$进行排序，得到直觉模糊数组${\widetilde{F^{{}^{\prime }}}}_{\sigma (1)},{\widetilde{F^{{}^{\prime }}}}_{\sigma (2)},...,{\widetilde{F^{{}^{\prime }}}}_{\sigma (n)}$；最后根据与$\mathrm{IFHA}$算子相关的权重向量$\mathbf{\omega }={\left({\omega }_1,{\omega }_2,...,{\omega }_n\right)}^T$计算方案$Y_i$的综合属性值${\tilde{d}}_i={IFHA}_{w,\omega }\left({\tilde{F}}_{i1},{\tilde{F}}_{i2},...,{\tilde{F}}_{in}\right)$

  $示例1.4$

  考虑公共投资项目绩效评估问题。公共投资项目绩效评估要同时考虑公共投资的经济性、效率性、有效性和社会公平性，可以从管理绩效（$G_1$）、经济绩效（$G_2$）、社会绩效（$G_3$）、生态绩效（$G_4$）和可持续发展绩效（$G_5$）五个方面进行绩效评估。设利用层次分析法得到属性$G_j(j=1,2,3,4,5)$的权重向量为$\omega ={(0.052,0.141,0.269,0.269,0.269)}^T$；通过现场调查和专家咨询可以得到五个公共投资项目$Y_i(i=1,2,3,4,5)$关于属性$G_j(j=1,2,3,4,5)$的直觉模糊评价结果如下表所示。

	$G_1$	$G_2$	$G_3$	$G_4$	$G_5$
$Y_1$	$⟨0.7,0.3⟩$	$⟨0.7,0.2⟩$	$⟨0.4,0.5⟩$	$⟨0.7,0.3⟩$	$⟨0.6,0.2⟩$
$Y_2$	$⟨0.6,0.2⟩$	$⟨0.4,0.3⟩$	$⟨0.5,0.3⟩$	$⟨0.7,0.1⟩$	$⟨0.5,0.2⟩$
$Y_3$	$⟨0.8,0.1⟩$	$⟨0.6,0.2⟩$	$⟨0.4,0.3⟩$	$⟨0.2,0.6⟩$	$⟨0.3,0.4⟩$
$Y_4$	$⟨0.5,0.4⟩$	$⟨0.3,0.6⟩$	$⟨0.6,0.3⟩$	$⟨0.3,0.4⟩$	$⟨0.5,0.3⟩$
$Y_5$	$⟨0.6,0.4⟩$	$⟨0.5,0.4⟩$	$⟨0.6,0.2⟩$	$⟨0.5,0.3⟩$	$⟨0.7,0.2⟩$

  代码如下：

import numpy as np

def calculate_weight_IF(IF_matrix,weight_IF):
    weight = weight_IF*IF_matrix.shape[1] # 计算<A_j_'> = n×<w_j>×<A_j>
    weighted_IF_matrix = np.zeros(IF_matrix.shape) # 初始化矩阵用于存放加权后的结果
    weighted_IF_matrix[0] = 1 - np.power(1-IF_matrix[0],weight)  # 计算加权后的μ_i = 1-(1-<u_i>)^<w_i>
    weighted_IF_matrix[1] = np.power(IF_matrix[1],weight) # 计算加权后的v_i = (<v_i>)^<w_i>
    return weighted_IF_matrix

def IF_rank(IF_matrix):
    # 计算每个直觉模糊数的得分值score=μ-v
    score = IF_matrix[0] - IF_matrix[1]
    # 按照得分值从大到小进行重新排序
    index = np.argsort(-score)  # 获取从大到小排序索引
    IF_sequence_matrix = np.zeros(IF_matrix.shape)  # 初始化结果矩阵
    for i in range(IF_matrix.shape[1]):  # 更新结果矩阵
        IF_sequence_matrix[0][i] = IF_matrix[0][index[i]]
        IF_sequence_matrix[1][i] = IF_matrix[1][index[i]]
    return IF_sequence_matrix,index

def calculate_IFHA(IF_sequence_matrix,weight):
    IFHA = [1,1]
    for i in range(IF_sequence_matrix.shape[1]):
        IFHA[0] *= np.power(1-IF_sequence_matrix[0][i],weight[i])
        IFHA[1] *= np.power(IF_sequence_matrix[1][i],weight[i])
    IFHA[0] = 1 - IFHA[0]
    return IFHA


IF_matrix_4 = np.array([[[0.7,0.7,0.4,0.7,0.6],[0.3,0.2,0.5,0.3,0.2]],
                      [[0.6,0.4,0.5,0.7,0.5],[0.2,0.3,0.3,0.1,0.2]],
                      [[0.8,0.6,0.4,0.2,0.3],[0.1,0.2,0.3,0.6,0.4]],
                      [[0.5,0.3,0.6,0.3,0.5],[0.4,0.6,0.3,0.4,0.3]],
                      [[0.6,0.5,0.6,0.5,0.7],[0.4,0.4,0.2,0.3,0.2]]])
weight_attr_4 = np.array([0.052,0.141,0.269,0.269,0.269])
weight_IFHA_4 = np.array([0.112,0.236,0.304,0.236,0.112]) # 正态分布赋权法
# S1: 计算加权直觉模糊数A_j_'
weighted_IF_matrix_4 = []
for item in IF_matrix_4:
    print(weight_attr_4)
    weighted_IF_matrix_4.append(calculate_weight_IF(item,weight_attr_4))
print('\nWeight_IF_Matrix:\n',weighted_IF_matrix_4)

# S2：计算A_j_'的得分值
# S3：排序
ranked_IF_4 = []
for item in weighted_IF_matrix_4:
    tmp_rlt = IF_rank(item)
    ranked_IF_4.append(tmp_rlt[0])
print('\nRanked_IF：\n',ranked_IF_4)
# S4：计算IFHA
IFHA_4 = []
for item in ranked_IF_4:
    IFHA_4.append(calculate_IFHA(item,weight_IFHA_4))
print('\nIFHA：\n',IFHA_4)
# 再次排序
IFHA_4 = np.array(IFHA_4)
tmp_matrix = np.zeros((2,IFHA_4.shape[0]))
for i in range(tmp_matrix.shape[1]):
    tmp_matrix[0][i] = IFHA_4[i][0]
    tmp_matrix[1][i] = IFHA_4[i][1]

final_result,index = IF_rank(tmp_matrix)
print('\nFinal Result: \n',final_result)
print('Final Index: \n',index+1)

  计算结果如下：

[0.052 0.141 0.269 0.269 0.269]
[0.052 0.141 0.269 0.269 0.269]
[0.052 0.141 0.269 0.269 0.269]
[0.052 0.141 0.269 0.269 0.269]
[0.052 0.141 0.269 0.269 0.269]
Weight_IF_Matrix:
[array([[0.26877417, 0.57207222, 0.49694736, 0.80197132, 0.7084113 ],
[0.73122583, 0.32153344, 0.39365399, 0.19802868, 0.11478505]]), array([[0.21198298, 0.30241581, 0.60634601, 0.80197132, 0.60634601],
[0.65806349, 0.42792778, 0.19802868, 0.04518559, 0.11478505]]), array([[0.34193651, 0.47585398, 0.49694736, 0.25927665, 0.38104815],
[0.54954087, 0.32153344, 0.19802868, 0.50305264, 0.2915887 ]]), array([[0.16491208, 0.2223322 , 0.7084113 , 0.38104815, 0.60634601],
[0.78801702, 0.69758419, 0.19802868, 0.2915887 , 0.19802868]]), array([[0.21198298, 0.38655751, 0.7084113 , 0.60634601, 0.80197132],
[0.78801702, 0.52414602, 0.11478505, 0.19802868, 0.11478505]])]
Ranked_IF：
[array([[0.80197132, 0.7084113 , 0.57207222, 0.49694736, 0.26877417],
[0.19802868, 0.11478505, 0.32153344, 0.39365399, 0.73122583]]), array([[0.80197132, 0.60634601, 0.60634601, 0.30241581, 0.21198298],
[0.04518559, 0.11478505, 0.19802868, 0.42792778, 0.65806349]]), array([[0.49694736, 0.47585398, 0.38104815, 0.34193651, 0.25927665],
[0.19802868, 0.32153344, 0.2915887 , 0.54954087, 0.50305264]]), array([[0.7084113 , 0.60634601, 0.38104815, 0.2223322 , 0.16491208],
[0.19802868, 0.19802868, 0.2915887 , 0.69758419, 0.78801702]]), array([[0.80197132, 0.7084113 , 0.60634601, 0.38655751, 0.21198298],
[0.11478505, 0.11478505, 0.19802868, 0.52414602, 0.78801702]])]
IFHA：
[[0.6044416137848779, 0.2746528218446278], [0.5490833706472749, 0.20246395100727907], [0.39806747813910426, 0.3527361010122677], [0.442009916381061, 0.34998743727745046], [0.5924687656619477, 0.24057371055311083]]
Final Result:
[[0.59246877 0.54908337 0.60444161 0.44200992 0.39806748]
[0.24057371 0.20246395 0.27465282 0.34998744 0.3527361 ]]
Final Index:
[5 2 1 4 3]

---

特别说明：本专栏主要参考郭子雪等所著《直觉模糊多属性决策理论方法及应用研究》书籍，部分代码计算结果与书中有所出入，请仔细甄别！