人工神经网络及其应用实例—-毕业论文设计

则有:

????????????? p?? yi , ?bi????????a ji

? pqi?? x j

据此编写 MATLAB 程序如下(标准 BP 算法加动量法优化):

clear x0 x1 x2; beta = 0.1; miu = 0;

[n m] = size(x); for i = 1 : 1 : n

x0(1, i) = 1; x1(1, i) = x(i, 1); x2(1, i) = x(i, 2); y0(1, i) = 1; t(1, i) = x(i, 3); end

a01 = rand(); a11 = rand(); a21 = rand(); a02 = rand(); a12 = rand(); a22 = rand(); b0 = rand(); b1 = rand(); b2 = rand(); delta_a01 = 0; delta_a11 = 0; delta_a21 = 0; delta_a02 = 0; delta_a12 = 0; delta_a22 = 0; delta_b0 = 0; delta_b1 = 0; delta_b2 = 0; l = 1; while 1

u1 = a01 * x0 + a11 * x1 + a21 * x2; u2 = a02 * x0 + a12 * x1 + a22 * x2; y1 = 2 ./ (1 + exp(-u1)) - 1; y2 = 2 ./ (1 + exp(-u2)) - 1; v = b0 * y0 + b1 * y1 + b2 * y2; z = 2 ./ (1 + exp(-v)) - 1;

Page 17 of 25

error = 0;

for i = 1 : 1 : n

if (z(1, i) > 0 && t(1, i) == 1) || (z(1, i) < 0 && t(1, i) == -1) % else

error = error + 1; end end error

temp0 = -(t - z) .* exp(-v) ./ (1 + exp(-v)) .^ 2; temp1 = b1 .* exp(-u1) ./ (1 + exp(-u1)) .^ 2; temp2 = b2 .* exp(-u2) ./ (1 + exp(-u2)) .^ 2;

delta_b0 = miu * delta_b0 + (1 - miu) * beta * sum(-temp0 .* y0); delta_b1 = miu * delta_b1 + (1 - miu) * beta * sum(-temp0 .* y1); delta_b2 = miu * delta_b2 + (1 - miu) * beta * sum(-temp0 .* y2); delta_a01 = miu * delta_a01 + (1 - miu) * beta * sum(-temp0 .* temp1 .* x0);

delta_a11 = miu * delta_a11 + (1 - miu) * beta * sum(-temp0 .* temp1 .* x1);

delta_a21 = miu * delta_a21 + (1 - miu) * beta * sum(-temp0 .* temp1 .* x2);

delta_a02 = miu * delta_a02 + (1 - miu) * beta * sum(-temp0 .* temp2 .* x0);

delta_a12 = miu * delta_a12 + (1 - miu) * beta * sum(-temp0 .* temp2 .* x1);

delta_a22 = miu * delta_a22 + (1 - miu) * beta * sum(-temp0 .* temp2 .* x2);

b0 = b0 + delta_b0; b1 = b1 + delta_b1; b2 = b2 + delta_b2; a01 = a01 + delta_a01; a11 = a11 + delta_a11; a21 = a21 + delta_a21; a02 = a02 + delta_a02; a12 = a12 + delta_a12; a22 = a22 + delta_a22; l = l + 1; if l == 1000

break; end

Page 18 of 25

end j1 = 1; j2 = 1; k1 = 1; k2 = 1;

for i = 1 : 1 : n

if x(i, 3) == -1

if z(1, i) < 0

x11(j1, :) = x(i, :); j1 = j1 + 1; else

x12(j2, :) = x(i, :); j2 = j2 + 1; end else

if z(1, i) > 0

x21(k1, :) = x(i, :); k1 = k1 + 1; else

x22(k2, :) = x(i, :); k2 = k2 + 1; end end end hold on;

plot(x11(:, 1), x11(:, 2), 'g*'); plot(x12(:, 1), x12(:, 2), 'r+'); plot(x21(:, 1), x21(:, 2), 'bo'); plot(x22(:, 1), x22(:, 2), 'r+'); hold off;

axis([0 1 0 1]);

程序运行结果,经过训练,该神经网络对 200 个样本的分类正确 率达到了 96.5%,分类效果较好,具体分类情况如下图所示,其中被

错误分类的样本已用红色标示出,其它正确分类的样本仍用原类别对

应的颜色进行标示。

Page 19 of 25

1 0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0 0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1 之前的所有程序都是完整地按照多层前向神经网络的标准 BP 学 习算法过程进行编写的,程序较为复杂,而且如果出现差错,也不易

发现,因此应用有一定难度,下面介绍使用 MATLAB 的神经网络工具

箱进行人工神经网络的设计与训练的方法。

关于神经网络工具箱的帮助信息,可以在 Product Help 中搜索

“Neural Network Toolbox”获得,包含有较为详细的使用方法,可以

在需要时进行查阅。下面以一个函数拟合的问题为例,演示如何使用

神经网络工具箱。

首先介绍几个将要用到的函数:

premnmx:用于将网络的输入数据或输出数据进行归一化,归一

化后的数据将分布在 [?1,1]区间内。其语法格式为: Page 20 of 25

联系客服:779662525#qq.com(#替换为@)