BP神经网络实例含源码(共16页).doc

精选优质文档-倾情为你奉上BP神经网络算法实现一：关于BP网络BP (Back Propagation)神经网络，即误差反传误差反向传播算法的学习过程，由信息的正向传播和误差的反向传播两个过程组成。输入层各神经元负责接收来自外界的输入信息，并传递给中间层各神经元；中间层是内部信息处理层，负责信息变换，根据信息变化能力的需求，中间层可以设计为单隐层或者多隐层结构；最后一个隐层传递到输出层各神经元的信息，经进一步处理后，完成一次学习的正向传播处理过程，由输出层向外界输出信息处理结果。当实际输出与期望输出不符时，进入误差的反向传播阶段。误差通过输出层，按误差梯度下降的方式修正各层权值，向隐层、输入层逐层反传。周而复始的信息正向传播和误差反向传播过程，是各层权值不断调整的过程，也是神经网络学习训练的过程，此过程一直进行到网络输出的误差减少到可以接受的程度，或者预先设定的学习次数为止。BP网络主要应用于以下方面：函数逼近、模式识别和分类、数据压缩。BP神经网络有较强的泛化性能，使网络平滑的逼近函数，能合理的响应被训练以外的输入。同时，BP网络又有自己的限制与不足，主要表现在：需要较长的训练时间、网络训练的结果可能使得权值逼近局部最优、训练数据范围外的数据泛化能力较差。为了避免训练陷入局部最优解，本程序采用改进的BP网络训练，既加入动量因子，使得网络在最优解附近有一定的震荡，跳出局部最优的范围。BP网络训练中学习速率与动量因子的选择很重要，在后面的内容中将进行详细的讨论。二：训练的函数程序中训练的函数为一个三输入一输出的非线性函数，如下所示：，网络结构为：351三：程序及相关界面（VB）1 主界面代码：Private Sub Command1_Click()form2.Visible = FalseForm3.Visible = TrueEnd SubPrivate Sub Command2_Click()form2.Visible = FalseForm1.Visible = TrueEnd SubPrivate Sub Command3_Click()form2.Visible = FalseForm4.Visible = TrueEnd SubPrivate Sub Command4_Click()form2.Visible = FalseForm5.Visible = TrueEnd SubPrivate Sub Command5_Click()EndEnd SubPrivate Sub Form_Load()Command3.Enabled = FalseCommand4.Enabled = FalseEnd Sub2 查看网络结构代码：Private Sub Command1_Click()Form3.Visible = Falseform2.Visible = TrueEnd Sub3 网络训练 代码：Private Sub Command1_Click()Form1.Visible = Falseform2.Visible = TrueEnd SubPrivate Sub Command2_Click()Label2.Caption = "样本训练中"Dim i As Integer, j As Integer, k As Integer, p As Integer, s As SingleDim Maxx(1 To 3) As Single, Minx(1 To 3) As Single, Meanx(1 To 3) As SingleDim x(1 To 3, 1 To 100) As Single, sumx(1 To 3) As Single, Temp As SingleDim Datex(1 To 3, 1 To 100) As Single, inputx(1 To 3) As Single, outputx(1 To 100) As SingleDim Ex(1 To 100) As SingleDim time(1 To 5000) As Integer, cishu(1 To 100) As IntegerDim Dv_1(1 To 5, 1 To 3) As Single, Dw_1(1 To 5) As SingleDim R As SingleDim Maxd As Single, Mind As SingleDim s1(1 To 5) As Single, y1(1 To 5, 1 To 100) As Single, s2 As Single, y2(1 To 100) As SingleDim deltW(1 To 100) As Single, deltV(1 To 5, 1 To 100) As SingleDim Dw(1 To 5) As Single, Dv(1 To 5, 1 To 3) As SingleDim MyIn(1 To 3) As SingleDim Errorx(1 To 5000) As SingleRandomizeFor i = 1 To 3 Maxx(i) = 0 Minx(i) = 0 Meanx(i) = 0Next iTemp = 0Maxd = 0Mind = 0For i = 1 To 5 For j = 1 To 3 Dv_1(i, j) = 0 v(i, j) = 2 * Rnd - 1 Next j Dw_1(i) = 0 w(i) = 2 * Rnd - 1Next iFor j = 1 To 3 For i = 1 To 100 x(j, i) = 4 * (2 * Rnd - 1) Next i sumx(j) = 0Next j'求最值For j = 1 To 3 For i = 1 To 100 If x(j, i) >= Maxx(j) Then Maxx(j) = x(j, i) End If If x(j, i) <= Minx(j) Then Minx(j) = x(j, i) Temp = Temp + x(j, i) End If Next i sumx(j) = Temp Temp = 0 Meanx(j) = sumx(j) / 100Next j'归一化For j = 1 To 3 For i = 1 To 100 If Maxx(j) - x(j, i) >= x(j, i) - Minx(j) Then R = Maxx(j) - x(j, i) Else R = x(j, i) - Minx(j) End If Datex(j, i) = (x(j, i) - Meanx(j) / R Next iNext j'期望输出For i = 1 To 100 For j = 1 To 3 inputx(j) = Datex(j, i) Next j outputx(i) = 2 * (inputx(1) + Sin(inputx(2) + Exp(inputx(3)Next i'输出归一化For i = 1 To 100 If Maxd <= outputx(i) Then Maxd = outputx(i) End If If Mind >= outputx(i) Then Mind = outputx(i) End IfNext iFor i = 1 To 100 Ex(i) = (outputx(i) - Mind) / (Maxd - Mind)Next i'训练For s = 1 To 5000 Step 1 time(s) = s For p = 1 To 100 cishu(p) = p For i = 1 To 3 MyIn(i) = Datex(i, p) Next i For i = 1 To 5 For j = 1 To 3 Temp = Temp + v(i, j) * MyIn(j) Next j s1(i) = Temp Temp = 0 Next i For i = 1 To 5 y1(i, p) = 1 / (1 + Exp(-s1(i) Next i For i = 1 To 3 Temp = y1(i, p) * w(i) + Temp Next i s2 = Temp Temp = 0 y2(p) = 1 / (1 + Exp(-s2) deltW(p) = (Ex(p) - y2(p) * y2(p) * (1 - y2(p) For i = 1 To 5 deltV(i, p) = deltW(p) * w(i) * y1(i, p) * (1 - y1(i, p) Next i Next p'误差 For i = 1 To 100 Temp = Temp + (Ex(i) - y2(i) 2 Next i Errorx(s) = Temp Temp = 0'调整权值 For i = 1 To 5 Dw_1(i) = Dw(i) Next i For i = 1 To 5 For j = 1 To 100 Temp = Temp + deltW(j) * y1(i, j) Next j Dw(i) = Temp Temp = 0 Next i For i = 1 To 5 For j = 1 To 3 Dv_1(i, j) = Dv(i, j) Next j Next i For i = 1 To 5 For j = 1 To 3 For k = 1 To 100 Temp = Temp + deltV(i, k) * Datex(j, k) Next k Dv(i, j) = Temp Temp = 0 Next j Next i For i = 1 To 5 w(i) = 0.2 * Dw(i) + 0.2 * Dw_1(i) + w(i) Next i For i = 1 To 3 For j = 1 To 5 v(j, i) = 0.2 * Dv(j, i) + 0.2 * Dv_1(j, i) + v(j, i) Next j Next i'画图 Picture1.Cls Picture1.ScaleTop = 1.5 Picture1.ScaleHeight = -2 Picture1.ScaleLeft = -10 Picture1.ScaleWidth = 120 Picture1.Line (-9, 0)-(110, 0) Picture1.Line (0, 0)-(0, 1.5) For i = 1 To 100 Picture1.PSet (cishu(i), Ex(i), RGB(128, 128, 0) Picture1.PSet (cishu(i), y2(i), RGB(128, 0, 0) Next i For i = 1 To 99 Picture1.Line (cishu(i), Ex(i)-(cishu(i + 1), Ex(i + 1), RGB(128, 128, 0) Picture1.Line (cishu(i), y2(i)-(cishu(i + 1), y2(i + 1), RGB(128, 0, 0) Next i'延时 For j = 1 To 1000 For k = 1 To 50 Next k Next j Picture2.Cls Picture2.Print s DoEventsNext sLabel2.Caption = ""form2.Command3.Enabled = Trueform2.Command4.Enabled = True'泛化Dim test(1 To 3, 1 To 20) As Single, sumE(1 To 3) As SingleDim MaxE(1 To 3) As Single, MinE(1 To 3) As Single, MeanE(1 To 3) As SingleDim MaxxE As Single, MinxE As SingleDim des(1 To 3) As Single, outE(1 To 20) As SingleDim MIn(1 To 3) As Single, s11(1 To 5) As Single, y11(1 To 5, 1 To 20) As Single, s22 As SingleDim DateE(1 To 3, 1 To 20) As SingleFor i = 1 To 20 For j = 1 To 3 test(j, i) = 4 * (2 * Rnd - 1) Next jNext iFor j = 1 To 3 For i = 1 To 20 If test(j, i) >= MaxE(j) Then MaxE(j) = test(j, i) End If If test(j, i) <= MinE(j) Then MinE(j) = test(j, i) Temp = Temp + test(j, i) End If Next i sumE(j) = Temp Temp = 0 MeanE(j) = sumE(j) / 100Next j'归一化For j = 1 To 3 For i = 1 To 20 If MaxE(j) - test(j, i) >= test(j, i) - MinE(j) Then R = MaxE(j) - test(j, i) Else R = test(j, i) - MinE(j) End If DateE(j, i) = (test(j, i) - MeanE(j) / R Next iNext j'求输出 For p = 1 To 20 Ti(p) = p For i = 1 To 3 MIn(i) = DateE(i, p) Next i For i = 1 To 5 For j = 1 To 3 Temp = Temp + v(i, j) * MIn(j) Next j s11(i) = Temp Temp = 0 Next i For i = 1 To 5 y11(i, p) = 1 / (1 + Exp(-s11(i) Next i For i = 1 To 3 Temp = y11(i, p) * w(i) + Temp Next i s22 = Temp Temp = 0 y22(p) = 1 / (1 + Exp(-s22) Next p'输出及归一化For j = 1 To 20 For i = 1 To 3 des(i) = DateE(i, j) Next i outE(j) = 2 * (des(1) + Sin(des(2) + Exp(des(3)Next j'输出归一化For i = 1 To 20 If MaxxE <= outE(i) Then MaxxE = outE(i) End If If MinxE >= outE(i) Then MinxE = outE(i) End IfNext iFor i = 1 To 20 outD(i) = (outE(i) - MinxE) / (MaxxE - MinxE)Next iEnd Sub4 查看训练结果代码：Private Sub Command1_Click()Form5.Visible = Falseform2.Visible = TrueEnd SubPrivate Sub Command2_Click()Picture1.ClsPicture2.ClsDim i As Integer, j As IntegerFor i = 1 To 5 For j = 1 To 3 Picture2.Print v(i, j); Spc(4); Next j Picture2.Print Picture2.Print Picture1.Print w(i);Next iEnd Sub5 泛化代码：Private Sub Command1_Click()Form4.Visible = Falseform2.Visible = TrueEnd SubPrivate Sub Command2_Click()For s = 1 To 20 'Picture1.Cls Picture1.ScaleTop = 1.5 Picture1.ScaleHeight = -2 Picture1.ScaleLeft = -5 Picture1.ScaleWidth = 30 Picture1.Line (-5, 0)-(25, 0) Picture1.Line (0, -0.5)-(0, 1.5) For i = 1 To 20 Picture1.PSet (Ti(i), outD(i), RGB(128, 128, 0) Picture1.PSet (Ti(i), y22(i), RGB(128, 0, 0) Next i For i = 1 To 19 Picture1.Line (Ti(i), outD(i)-(Ti(i + 1), outD(i + 1), RGB(128, 128, 0) Picture1.Line (Ti(i), y22(i)-(Ti(i + 1), y22(i + 1), RGB(128, 0, 0) Next iNext sEnd Sub6 全局模块Public w(1 To 5) As Single, v(1 To 5, 1 To 3) As SinglePublic Ti(1 To 20) As Single, y22(1 To 20) As Single, outD(1 To 20) As Single四：相关分析及讨论以上编程实现了对一个三输入、一输出非线性函数的逼近，在模型训练中采用改进的BP网络动量因子法，输入是随机产生的100组数据，输出是通过已知函数得到的相应期望输出，通过BP网络的5000代训练可以与期望输出拟合的很好，泛化也较理想，训练误差和泛化误差都在可接受范围内。前已提及，在BP网络训练中，学习速率和动量因子的选择是很重要的，一下分别对两者进行研究：1 关于学习速率对于前述函数，分别取学习速率为：0.01, 0. 2, 0.3, 0.5, 0.9可得训练结果如下：从对网络的训练及相应的曲线可以看出，学习速率小的话训练稳定，但较慢，由于BP网络采用的是梯度下降法，训练的好坏很大程度上决定于开始的若干步，如果学习速率过小，开始训练过于缓慢，则有可能使整个训练最终无法达到要求。学习速率越大，训练越快，但稳定性变差，当大于一定数值后甚至最终无法收敛。基于此，引进动量项，以改进训练性能。2 关于学习速率一下分别取动量因子为：0.01, 0.1, 0.2, 0.5, 0.8, （学习速率为0.5）结果如下：有训练过程及相应曲线可已看出，当引进动量项以后即使在较大的学习速率下，训练过程依然较平稳，而且由于在每次修改权值的时候考虑了前一次的修改作用，使得权值的修正更具合理性，但是，如果动量项过大的话亦会使得训练过程震荡，无法寻得最优值。综上，在BP神经网络的训练中综合考虑学习系数和动量因子，可以使得训练过程即迅速而且平稳，能在较快的时间内找到最佳的权值组合。专心-专注-专业