程序实现：

上面的公式看似复杂，其实我们关心的就是最后的5个计算步骤，这里说明一下，有的书上以隶属度矩阵的某一范数小于一定值作为收敛的条件，这也可，不过计算量稍微要大一点了。

程序采用VB6.0编制，完全按照以上的步骤进行。

'程序实现功能：模糊聚类和硬聚类
'作    者： laviewpbt
'联系方式： laviewpbt@sina.com
'QQ:33184777
'版本：Version 2.3.1
'说明：复制请保留源作者信息，转载请说明，欢迎大家提出意见和建议

Private Declare Function GetTickCount Lib "kernel32" () As Long

Private Enum IniCenterMethod    '初始中心的方法
    CreateRandom                '随机的中心点
    CreateByHcm                 '由HCM创建的中心点
    CreateByRandomZadeh         '先随机创建隶属矩阵，然后计算得到的中心点

CreateByHand                '手工确定初始中心点

End Enum

Private Enum AntiFuzzyMethod    '反模糊的方法
    Max                         '最大隶属度法
    Middle                      '中位数法
    Mean                        '加权均值法
End Enum

Private Type FcmInfo
     Center() As Double         '聚类中心
     Degree() As Double         '隶属度，为Double类型
     Class() As Byte            '记录数据属于那一类
     TimeUse As Long            '所用时间
     Iterations  As Long        '迭带次数
     ErrMsg As String           '错误信息
End Type

Private Type HcmInfo
    Center() As Double          '聚类中心
    Class() As Byte             '记录数据属于那一类
    TimeUse As Long             '所用时间
    Iterations  As Long         '迭带次数
    ErrMsg As String            '错误信息
End Type

'*************************************************************************************
'*    作    者 ：    laviewpbt
'*    函 数 名 ：    Fcm
'*    参    数 ：    Data     -   待分类的样本，第一维的大小表示样本的个数，
'*                                第二维的大小表示样本的维数
'*                   Cluster  -   分类数
'*                   CreateIniCenter - 初始聚类中心的创建方法
'*                   AntiFuzzy -  反模糊化的方法
'*                   Exponent  -  一个控制聚类效果的参数，一般取2
'*                   Maxiterations  - 最大的迭代次数
'*                   MinImprovement - 最小的改进参数（两次迭代间聚类中心的距离）
'*    返回值 ：      FcmInfo结构，记录了相关的参数
'*    功能描述 ：    利用模糊理论的聚类方法把数据分类
'*    日    期 ：    2004-10-27 10.25.32
'*    修 改 人 ：    laviewpbt
'*    日    期 ：    2006-11-7 19.25.31
'*    版    本 ：    Version 2.3.1
'**************************************************************************************

Private Function Fcm(ByRef Data() As Double, ByVal Cluster As Long, Optional ByVal CreateIniCenter As IniCenterMethod = IniCenterMethod.CreateByHcm, Optional AntiFuzzy As AntiFuzzyMethod = Max, Optional Exponent As Byte = 2, Optional Maxiterations As Long = 400, Optional MinImprovement As Double = 0.01, Optional ByRef CenterByHandle As Variant) As FcmInfo
    If ArrayRange(Data) <> 2 Then
        Fcm.ErrMsg = "数据只能为二维数组"
        Exit Function
    End If
    Dim i As Long, j As Long, k As Long, l As Long, m As Long
    Dim DataNumber As Long, DataSize As Long
    Dim Temp As Double, Sum1 As Double, Sum2 As Double, Sum3 As Double, Index As Integer
    Dim OldCenter() As Double
    Fcm.TimeUse = GetTickCount
    DataNumber = UBound(Data, 1): DataSize = UBound(Data, 2)
    ReDim Fcm.center(1 To Cluster, 1 To DataSize) As Double
    ReDim Fcm.Degree(1 To Cluster, 1 To DataNumber) As Double
    ReDim Fcm.Class(1 To DataNumber) As Byte
    ReDim OldCenter(1 To Cluster, 1 To DataSize) As Double
    On Error GoTo ErrHandle:
    Randomize
    If CreateIniCenter = CreateRandom Then
        For i = 1 To Cluster
            For j = 1 To DataSize
                OldCenter(i, j) = Data(Rnd * DataNumber, j)    '产生随机初始中心点
            Next
        Next
    ElseIf CreateIniCenter = CreateByHcm Then
        Dim HcmCenter As HcmInfo
        HcmCenter = Hcm(Data, Cluster)
        For i = 1 To Cluster
            For j = 1 To DataSize
                OldCenter(i, j) = HcmCenter.center(i, j)   '产生HCM初始中心点
            Next
        Next
    ElseIf CreateIniCenter = CreateByRandomZadeh Then
        ReDim RndDegree(1 To Cluster, 1 To DataNumber) As Double
        Dim RndSum As Double
        For i = 1 To Cluster
            For j = 1 To DataNumber
                RndDegree(i, j) = Rnd           '创建随机的隶属度
            Next
        Next
        For j = 1 To DataNumber
            RndSum = 0
            For i = 1 To Cluster
                RndSum = RndSum + RndDegree(i, j)
            Next
            For i = 1 To Cluster
                RndDegree(i, j) = RndDegree(i, j) / RndSum   '隶属度矩阵每列之后必须为1
            Next
        Next
       
        For i = 1 To Cluster
            For j = 1 To DataSize
                Sum1 = 0: Sum2 = 0
                For k = 1 To DataNumber
                    Temp = Exp(Log(RndDegree(i, k)) * Exponent)  '其实就是RndDegree(i, k)^Exponent
                    Sum1 = Sum1 + Temp * Data(k, j)           '隶属度的平方乘以数值
                    Sum2 = Sum2 + Temp                        '隶属度的和
                Next
                OldCenter(i, j) = Sum1 / Sum2                 '得到聚类中心
            Next
        Next
    ElseIf CreateIniCenter = CreateByHand Then
        If IsMissing(CenterByHandle) Then
            Fcm.ErrMsg = "请提供初始聚类中心。."
            Exit Function
        ElseIf UBound(CenterByHandle, 1) <> Cluster Or UBound(CenterByHandle, 2) <> DataSize Then
            Fcm.ErrMsg = "手工提供的初始聚类中心维数有错误."
            Exit Function
        End If
        For i = 1 To Cluster
            For j = 1 To DataSize
                OldCenter(i, j) = CenterByHandle(i, j)
            Next
        Next
    End If

Do
        Fcm.Iterations = Fcm.Iterations + 1
        For i = 1 To Cluster

本文转自feisky博客园博客，原文链接：http://www.cnblogs.com/feisky/archive/2009/03/14/1586312.html，如需转载请自行联系原作者