Neural network with error BP

Hello

I am totally new to ANN, I have a simple problem which is I have to train a network with 2000 input output data set

The network consists of three inputs (x,y,z) and three outputs (t1, t2, t3), each row in the attached axcel files is input and its corresponding output with one hidden layer

this is a regression problem and I read some papers say that at hidden layer there should be sigmoid function and at output layer there sould be linear function x=y

But the problem is I got NAN when I use linear on output layer. Thank you for any idea or help

clear;
clc;
inpt = xlsread('inpt.xlsx'); 
outpt = xlsread('outpt.xlsx'); 
nh=50; % number of neurons in hidden layer
ni=3; % number of inputs
no=3; % number of outputs
wi = unifrnd(-1,1,[ni,nh]); % weights on inputs-hidden
wo = unifrnd(-1,1,[nh,no]); % weights on hidden-outputs
th_h = unifrnd(-1,1,[1,nh]); % bais on hidden layer
th_o = unifrnd(-1,1,[1,no]); % bais on output layer
a=0.1;
Error=30;
epoch = 1;
while abs(Error) >= 10  
for i=1:size(inpt,1)
    % estimate y
    for k=1:nh
       y(k) = sig(inpt(i,:)*wi(:,k)-th_h(k)) ;        
    end  
    % estimate thetas (output)
    for k=1:no       
       th(k) = (y*wo(:,k)-th_o(k)) ;  
    end 
    % estimate error
    e=outpt(i,:)-th;
    Error = sum((e).^2)
    % estimate delat at output layer
    do= th.*(1-th).*e;
    % error qradient at output layer
    for k=1:nh
        for j=1:no
        dwo(k,j)=a*y(k)*do(j);
        end
    end
    dth_o = a*(-1)*do;
    % error qradient at hidden layer 
    for k=1:nh
        D(k) = sum(wo(k,:).*do.*(y(k)*(1-y(k)))); 
    end
    for k=1:nh
        dwi(:,k)=a*inpt(i,:)'*D(k);
    end
    dth_h = a*(-1)*D;
%     update weights
    wi = wi+dwi;
    th_h= th_h+dth_h;
    wo = wo+dwo;    
    th_o = th_o+dth_o;     
epoch = epoch+1;
end
end
epoch