MATLAB: Different Result of Analytical and Trained ANN Recurrent Neural Network

Question

Hi All,

I have question regarding to my analytical equation vs trained ANN (Recurrent Neural Network) with 5 input and 1 output. From my trained ANN, I obtain IW, LW, and b to create analytical ANN equation. The function of this equation is shown below :

============================================================================================================

function answer = ANN(u,t) ;

global netc

[IW] = netc.IW{1,1} ; [LW1] = netc.LW{1,1} ; [LW2] = netc.LW{2,1} ; [b1] = netc.b{1,1} ; [b2] = netc.b{2,1} ;

a0 = zeros(20,1) ; a1(:,1) = tanh(IW*u + LW1*a0 + b1) ; a2(1) = LW2*a1(:,1) + b2 ;

for t = 1:(t-1) ; a1(:,t+1) = tanh(IW*u + LW1*a1(:,t) + b1) ; a2(t) = LW2*a1(:,t) + b2 ; end

answer = a2(end,end) ; ============================================================================================================

However, the curve produce by this analytical method is different even though the trend looks similar. The comparison between this two curves shown below :

NOTE : Blue = Trained ANN / Red = Analytical Equation

I want to ask are there something wrong with my analytical equation? Please advise.

Regards,

HKT

Answer 1

You forgot that the MATLAB variables are automatically normalized before training and denormalized after training.

Try searching in both the NEWSGROUP and ANSWERS using

 greg analytic

Hope this helps

Thank you for formally accepting my answer

Greg