GA与神经网络的结合的程序
GA与神经网络的结合的程序clear;
% Generate a set of points in the space
source_point = -10+20*rand(2,200);
% Classified boundary is y = sin(x)+2*cos^2(x)
points_c = zeros(1,200)
for i = 1:200
point_c(i) = source_point(2,i)>(sin(source_point(1,i))+2*(cos(source_point(1,i))^2));
end
save training_data
% Plot the result
figure();
hold on;
for i = 1:200
if point_c(i) == 1
plot(source_point(1,i),source_point(2,i),'*');
else
plot(source_point(1,i),source_point(2,i),'+');
end
end
%plot the boundary
x = -10:0.01:10;
y = sin(x)+2.*cos(x).^2;
plot(x,y,'r');
hold off;
% The training progress
% Defined the lower boundary
LB = zeros(1,25);
LB(1:18) = -1*ones(1,18);
LB(19:25) = -10*ones(1,7);
% Defined the upper boundary
UB = zeros(1,25);
UB(1:18) = 1*ones(1,18);
UB(19:25) = 10*ones(1,7);
% Set the training option
options=gaoptimset;
options.PopulationSize=20;
options.Generations=100;
options.StallGenLimit=Inf;
options.StallTimeLimit=Inf;
options = gaoptimset(options,'PlotFcns',{@gaplotbestf}')
= ga(@fitness_fuction,25,[],[],[],[],LB,UB,[],options);
% Generate 50 points to test the parameter
exam_point = -10+20*rand(2,50);
output = exam_function(par,exam_point);
% Plot the result
figure();
hold on;
for i = 1:50
if output(i) == 1
plot(exam_point(1,i),exam_point(2,i),'*');
else
plot(exam_point(1,i),exam_point(2,i),'+');
end
end
%plot the boundary
x = -10:0.01:10;
y = sin(x)+2.*cos(x).^2;
plot(x,y,'r');
hold off;
fitness function的源代码：
function fval = fitness_function(x);
load training_data;
h = zeros(6,200);
% Hidden layer output
h_out(1,:)= tanh(source_point(1,:)*x(1)+source_point(2,:)*x(2)+x(19));
h_out(2, :) = tanh(source_point(1,:)*x(3)+source_point(2,:)*x(4)+x(20));
h_out(3, :)= tanh(source_point(1,:)*x(5)+source_point(2,:)*x(6)+x(21));
h_out(4, :)= tanh(source_point(1,:)*x(7)+source_point(2,:)*x(8)+x(22));
h_out(5, :)= tanh(source_point(1,:)*x(9)+source_point(2,:)*x(10)+x(23));
h_out(6, :)= tanh(source_point(1,:)*x(11)+source_point(2,:)*x(12)+x(24));
% Output layer output
res = x(13:18)*h_out+x(25);
zm = zeros(1,200);
om = ones(1,200);
res(reszm) = om(res>zm);
fval = sum(abs(point_c-res));
检测程序的源代码：
function res = exam_function(x,data);
h = zeros(6,50);
% Hidden layer output
h_out(1, :)= tanh(data(1,:)*x(1)+data(2,:)*x(2)+x(19));
h_out(2, :)= tanh(data(1,:)*x(3)+data(2,:)*x(4)+x(20));
h_out(3, :)= tanh(data(1,:)*x(5)+data(2,:)*x(6)+x(21));
h_out(4, :)= tanh(data(1,:)*x(7)+data(2,:)*x(8)+x(22));
h_out(5, :)= tanh(data(1,:)*x(9)+data(2,:)*x(10)+x(23));
h_out(6, :)= tanh(data(1,:)*x(11)+data(2,:)*x(12)+x(24));
% Output layer output
res = x(13:18)*h_out+x(25);
zm = zeros(1,50);
om = ones(1,50);
res(reszm) = om(res>zm); 有点意思哦
页:
[1]