time series predition with NARX

Dear all,

I am using Narx to do time series prediction. T is the target time series, and X is the external input. First of all I use open loop to train the model, then I transformed open loop to close loop. Finally, I make the prediction by specifying new external input X1 to the close loop model. The code I used is like the following, which works for my data sets.

net = narxnet(1:3,1:3,10); [x,xi,ai,t] = preparets(net,X,{},T); net = train(net,x,t,xi,ai); netc = closeloop(net); T1 = netc(X1);

The first question is that: I am wondering if it is by default that the final input states and layer states of the open-loop network automatically become the initial input states and layer state of the closed-loop network. As I just simply use closeloop(net) command without specifying the initial states.

Another question is I find the close loop didn’t yield results with reasonable accuracy, what should I do to improve this ?

Thanks in advance ! jun