abc.m

使用BP神经网络实现函数的逼近并附带着fig文件

% 产生训练样本与测试样本
%for  i = 1:20
%   X(1,i) = i;
%end
%for  j = 1:20;
%   Y(1,j) = j;
%end
%for i=1:20
%for j=1:20
%Z(i,j)=X(1,i)*log(X(1,i)^2+Y(1,j)^2);
%end
%meshc(X,Y,Z);
P=randint(2,200);
Value(1,200)=0;
for i=1:200
Value(1,i)=P(1,i)*log(P(1,i)^2+P(2,i)^2);
end
P1 = P(1,:);
P2 = P(1,:);
%column=200;
%初始化目标精度
%bp_goal=0.001;
%初始化隠层神经元数目
bp_hidenum=10;
net = newff([-100000 100000;-100000 100000],[bp_hidenum 1],{'logsig' 'logsig'});
net.IW{1,1}=rand(bp_hidenum,2);
%net.IW{2,1}=rand(1,handles.bp_hidenum);
net.trainFcn='traingda';
net.trainParam.goal=0.001;
net.trainParam.lr=0.02;
net.trainParam.epochs=6000;

net=train(net,P,Value);
S=sim(net,P);

subplot(121),plot3(P1,P2,Value);
subplot(122),plot3(P1,P2,S,':');
clear