[net,tr,Y,E,Pf,Af]=train(net,P,T,Pi,Ai)

%% 清空环境变量
clear all
clc

%% 训练集/测试集产生
load spectra_data.mat
% 随机产生训练集和测试集
temp = randperm(size(NIR,1));
% 训练集——50个样本
P_train = NIR(temp(1:50),:)';
T_train = octane(temp(1:50),:)';
% 测试集——10个样本
P_test = NIR(temp(51:end),:)';
T_test = octane(temp(51:end),:)';
N = size(P_test,2);

%% BP神经网络创建、训练及仿真测试

% 创建网络
net = newff(P_train,T_train,9);
% 设置训练参数
net.trainParam.epochs = 1000;
net.trainParam.goal = 1e-3;
net.trainParam.lr = 0.01;
% 训练网络
net = train(net,P_train,T_train);
% 仿真测试
T_sim_bp = sim(net,P_test);

%% RBF神经网络创建及仿真测试

% 创建网络
net = newrbe(P_train,T_train,0.3);
% 仿真测试
T_sim_rbf = sim(net,P_test);

%% 性能评价

% 相对误差error
error_bp = abs(T_sim_bp - T_test)./T_test;
error_rbf = abs(T_sim_rbf - T_test)./T_test;
% 决定系数R^2
R2_bp = (N * sum(T_sim_bp .* T_test) - sum(T_sim_bp) * sum(T_test))^2 / ((N * sum((T_sim_bp).^2) - (sum(T_sim_bp))^2) * (N * sum((T_test).^2) - (sum(T_test))^2)); 
R2_rbf = (N * sum(T_sim_rbf .* T_test) - sum(T_sim_rbf) * sum(T_test))^2 / ((N * sum((T_sim_rbf).^2) - (sum(T_sim_rbf))^2) * (N * sum((T_test).^2) - (sum(T_test))^2));
% 结果对比
result_bp = [T_test' T_sim_bp' T_sim_rbf' error_bp' error_rbf']

%% 绘图
figure
plot(1:N,T_test,'b:*',1:N,T_sim_bp,'r-o',1:N,T_sim_rbf,'k-.^')
legend('真实值','BP预测值','RBF预测值')
xlabel('预测样本')
ylabel('辛烷值')
string = {'测试集辛烷值含量预测结果对比(BP vs RBF)';['R^2=' num2str(R2_bp) '(BP)' '  R^2=' num2str(R2_rbf) '(RBF)']};
title(string)