matlab模拟出现较大误差是什么原因,求助大神，使用BP神经网络预测数据，为什么误差很大？...

随机产生了25组数据，利用公式求得结果矩阵，现用20组数据训练BP网络，剩下的5组进行预测，产生的结果误差为什么很大，请大神指教，不甚感激

由于金币只有1.5个，所以没办法发求助帖

%测试BP神经网络

clear;

clc;

close all;

x = rand(25,4);

%%

%产生原始数据

indata = rand(25,4);

outdata = indata(:,1)*2+3*indata(:,2).^2 + indata(:,3) + indata(:,4)*0.25;

%设置训练数据

train_indata_innormal = indata(1:20,: )';

train_outdata_innormal = outdata(1:20,: )';

%设置测试数据

test_indata_innormal = indata(21:25,: )';

test_outdata_innormal = outdata(21:25)';

%训练数据归一化

[train_indata_normal,train_inpar] = mapminmax(train_indata_innormal);

[train_outdata_normal,train_outpar] = mapminmax(train_outdata_innormal);

%测试数据归一化

test_indata_normal = mapminmax('apply',test_indata_innormal,train_inpar);

%%

%BP神经网络

%创建网络训练函数

% for i=1:5000

net = newff(train_indata_normal,train_outdata_normal,[6 1],{ 'logsig' 'purelin' } , 'traingda');

%设置训练参数

net.trainparam.goal = 1e-5;

net.trainparam.epochs = 50000;

net.trainparam.min_grad = 1e-6;

%取消validation check功能

net.divideFcn = '';

[net,tr] = train(net,train_indata_normal,train_outdata_normal);

out_train = sim(net,train_indata_normal);

%计算训练误差

train_error = train_outdata_normal - out_train;

error_MSE = mse(train_error);

figure(1)

plot(train_outdata_normal,'-r');

hold on

plot(out_train,':g');

%仿真

out = sim(net,test_indata_normal);

%测试数据反归一化

test_outdata = mapminmax('reverse',out,train_outpar);

test_error = test_outdata - test_outdata_innormal;

figure(3)

plot(test_outdata,'-r');

hold on

plot(test_outdata_innormal,':g');