%% II. 声明全局变量
global P_train % 训练集输入数据
global T_train % 训练集输出数据
global R % 输入神经元个数
global S2 % 输出神经元个数
global S1 % 隐层神经元个数
global S % 编码长度
S1 =7;
%% 训练集/测试集产生
input=xlsread('input.xlsx');
output=xlsread('output.xlsx');
% 随机产生训练集和测试集
temp = randperm(size(input,1));
% 训练集——25个样本
P_train = input(temp(1:28),:)';
T_train = output(temp(1:28),:)';
% 测试集——5个样本
P_test =input((28:end),:)';
T_test = output((28:end),:)';
N = size(P_test,2);
%% III. 数据归一化
[p_train, ps_input] = mapminmax(P_train,0.2,0.8);
p_test = mapminmax('apply',P_test,ps_input);
[t_train, ps_output] = mapminmax(T_train,0.2,0.8);
%% IV. BP神经网络创建、训练及仿真测试
%%
% 1. 创建网络
net = newff(p_train,t_train,[S1],{'logsig','purelin'},'trainlm');
%%
% 2. 设置训练参数
net.trainParam.epochs = 50;
net.trainParam.goal =0.1;
net.trainParam.lr = 0.06;
%%
% 3. 训练网络
net = train(net,p_train,t_train);
%%
% 4. 仿真测试
t_sim = sim(net,p_test);
%
T_sim = mapminmax('reverse',t_sim,ps_output);
error = abs(T_sim - T_test)./T_test;
%% V. GA-BP神经网络
R = size(p_train,1);
S2 = size(t_train,1);
S = R*S1 + S1*S2 + S1 + S2;
aa = ones(S,1)*[-1,1];
%% VI. 遗传算法优化
%%
% 1. 初始化种群
popu = 50; % 种群规模
initPpp = initializega(popu,aa,'gabpEval',[],[1e-6 1]); % 初始化种群
%%
% 2. 迭代优化
gen = 100; % 遗传代数
% 调用GAOT工具箱,其中目标函数定义为gabpEval
[x,endPop,bPop,trace] = gaot(aa,'gabpEval',[],initPpp,[1e-6 1 1],'maxGenTerm',gen,...
'normGeomSelect',0.09,'arithXover',2,'nonUnifMutation',[2 gen 3]);
%%
% 3. 绘均方误差变化曲线
figure(1)
plot(trace(:,1),1./trace(:,3),'r-');
title( 'GA优化BP神经网络,绘制均方误差变化曲线—Jason niu')
hold on
plot(trace(:,1),1./trace(:,2),'b-');
xlabel('Generation');
ylabel('Sum-Squared Error');
%%
% 4. 绘制适应度函数变化
figure(2)
plot(trace(:,1),trace(:,3),'r-');
title( 'GA优化BP神经网络,绘制适应度函数变化曲线—Jason niu')
hold on
plot(trace(:,1),trace(:,2),'b-');
xlabel('Generation');
ylabel('Fittness');
%% VII. 解码最优解并赋值
%%
% 1. 解码最优解
[W1,B1,W2,B2,val] = gadecod(x);
inputnum=size(P_train,1); % 输入层神经元个数
outputnum=size(T_train,1); % 输出层神经元个数
w1num=R*S1; % 输入层到隐层的权值个数
w2num=S2*S1;% 隐层到输出层的权值个数
N=w1num+S1+w2num+outputnum; %待优化的变量的个数
%%
% 2. 赋值给神经网络
w1=x(1:R*S1);
B1=x(R*S1+1:R*S1+S1);
w2=x(R*S1+S1+1:R*S1+S1+S1*S2);
B2=x(R*S1+S1+S1*S2+1:R*S1+S1+S1*S2+S2);
net.iw{1,1}=reshape(w1,S1,R);
net.lw{2,1}=reshape(w2,S2,S1);
net.b{1}=reshape(B1,S1,1);
net.b{2}=B2;
%% VIII. 利用新的权值和阈值进行训练
net = train(net,p_train,t_train);
%% IX. 仿真测试
s_ga = sim(net,P_test) %遗传优化后的仿真结果
T_sim = mapminmax('reverse',s_ga,ps_output);%%
% 1. 相对误差error
error = abs(T_sim - T_test)./T_test;
%%
% 2. 决定系数R^2
R2 = (N * sum(T_sim .* T_test) - sum(T_sim) * sum(T_test))^2 / ((N * sum((T_sim).^2) - (sum(T_sim))^2) * (N * sum((T_test).^2) - (sum(T_test))^2));
%%
% 3. 结果对比
result = [T_test' T_sim' error'];
2020-3-27 15:15 上传
2020-3-27 15:15 上传