How to find best combination of delay size and neurons for narx network?

I'm using a time series prediction narx neural network for my dataset. The input & feedback delay, number of Hidden neurons are chosen randomly to train the network. However i dont know what is the maximum accuracy my model can reach. So, What is the best way to find good combination of delay size and number of neurons for the network. Does sweeping the parameter and iterating over helps?

[yanqi liu]

yes，sir，it is a optimization problem，may be check use GA、ABC and so on algorithm

Answer 1

[Vatsal]

Hi,

I understand that you want to find the optimal delay size and number of neurons for your network to achieve good accuracy. I recommend referring to the following MathWorks documentation link, which provides a detailed explanation on how to determine the optimal input delay and hidden nodes for 'NARXNET':

https://in.mathworks.com/matlabcentral/answers/296646-finding-optimal-id-fd-and-hidden-nodes-for-narxnet?s_tid=srchtitle

I hope this helps!