Dear Kangujam

It is totally normal. no problem if you got different results. but in case of optimization you should choose the best fit. I mean chose the best architecture, which is the one that MSE of training set is close to validation and test set. At the same time validation set must have the lowest value. look at the file attached here.

Thank you @Maryam kh . But I am getting the best fit at mse =60.The mse value is coming out to be large.Is it correct?Is there any allowable range for the mse value in load forecasting?

R-square should be large not MSE, MSE is known as error. the best fit is the one that has lowest MSE and the greater Rsquare.

Hi Kangujam,

You need to fix the random number generator to get the reproducibility. It is because neural networks use some random initialisation of weights.

In Matlab you can use rng to fix the random seed.

Hope this helps.

Thank you @ Nam Le.Can you please elaborate on where and how to use rng in nftool?

If you are using nftool, i think this link is more helpful :)

https://uk.mathworks.com/help/nnet/gs/fit-data-with-a-neural-network.html