Narx model GIVES POOR PERFORMANCE

I am trying to train a NN using NEWNARXSP. When I execute the following commands:

current={ [3.38] [3.37706] [3.37412] [3.37118] [3.36824] [3.3653] [3.36236] [3.35942] [3.35648] [3.35354] [3.3506] [3.34766] [3.34472] [3.34178] [3.33884] [3.3359] [3.33296] [3.33002] [3.32441] [3.32147] [3.32] [3.31804] [3.31706] [3.31559] [3.31412] [3.31118] [3.30922] [3.30771] [3.3053] [3.3] [3.29804] [3.29706] [3.29412] [3.29265] [3.29118] [3.2884] [3.28294] [3.28] [3.27608] [3.27412] [3.27118] [3.26824] [3.267505] [3.26677] [3.26660] [3.26660] [3.26660] [3.2702] [3.27412] [3.27559] [3.27608] [3.27706] [3.27853] [3.28] [3.28147] [3.28294] [3.28588] [3.2898] [3.29559] [3.3] [3.30294] [3.30588] [3.31265] [3.31559] [3.32] [3.3249] [3.32724] [3.33314] [3.33608] [3.34] [3.34294] [3.34735] [3.35078] [3.3547] [3.36] [3.36441] [3.36882] [3.37176] [3.37323] [3.37412] [3.37706] [3.38] [3.38294] [3.38]};
resistance={[6.02] [6.02] [6.02] [6.02] [6.02] [6.02] [6.02] [6.02] [6.02] [6.02] [6.02] [6.02] [6.02] [6.02] [6.02] [6.02] [6.02] [6.02] [6.04] [6.06] [6.10] [6.12] [6.14] [6.16] [6.18] [6.22] [6.24] [6.28] [6.31] [6.40] [6.44] [6.48] [6.52] [6.56] [6.60] [6.68] [6.78] [6.86] [7] [7.07] [7.23] [7.42] [7.5] [7.58] [7.66] [7.74] [7.9] [8] [8] [8] [7.9] [7.82] [7.74] [7.66] [7.58] [7.5] [7.43] [7.31] [7.16] [7.07] [7] [6.93] [6.83] [6.79] [7.72] [6.65] [6.60] [6.54] [6.5] [6.46] [6.42] [6.38] [6.34] [6.30] [6.26] [6.22] [6.18] [6.15] [6.14] [6.12] [6.11] [6.1] [6.06] [6.02]};

a=cell2mat(current);
b=cell2mat(resistance);
d1=[1 2];
 d2=[1 2];
 sus=newnarxsp({[3.26660 3.38],[6.02 8]},d1,d2,[5 1],{'tansig','purelin'});
 p=[a;b];
 T=b;
 sus.trainFcn='trainlm';
 sus.trainparam.show=100;
 sus.trainparam.epochs=1000;
 sus=train(sus,p,T);

IT GIVES POOR PERFORMANCE. Actually I want to measure the value of resistance by using the current. here current is input and resistance is target . what can i do now?

[Walter Roberson]

Note: all those [] are unnecessary. Especially as you cell2mat() anyhow, suggesting that you would be better off writing

current = [3.38 3.3776 3,37412 <etc>];

Answer 1

[nick]

Hi c.m.f.s.,

Here are a few suggestions to improve performance of your NARX neural network:

Data Normalization: You can normalize your input and target data. Models trained on normalized data tend to have better generalization capabilities, resulting in more accurate predictions on unseen data.

Network Architecture: You can experiment with different network architectures with different numbers of neurons in the hidden layer or different activation functions.