Different inputDelays and feedbackDelays for NARXNET network

2 次查看(过去 30 天)
Hello all the Matlab community,
It is the second day I've been looking for an answer in the forum, especially at Greg's answers, and I can confidently post this question since I still haven't found the answer...
I'm trying to define a NARXNET with different inut and feedback delays. For that I have generated a NARX script from the ntstool, and am modifying the generated script.
Typicaly, I want my delays to be:
inputDelays = 0:5;
feedbackDelays = 6:11;
hiddenLayerSize = 10;
net = narxnet(inputDelays,feedbackDelays,hiddenLayerSize,'open',trainFcn);
This is because in my situation, I use input data x(t) to x(t-5) and I can only know my output before another 5 timesteps have passed, therefore I can only look back at the previous output y(t-6) to y(t-11).
In the script, it is possible to inform those delays (like I have earlier), but I have a doubt concerning the fact that the network understands them and really trains the network the way I intent it to be trained. The reason why I have doubts is because when I prepare the timeseries with
[x,xi,ai,t] = preparets(net,X,{},T);
and I check x, then I see x{1,1) = X{12} and x{2,1} = T{12}, when I would have expected x{1,1) = X{1} and x{2,1} = T{7}. Of course there is the same wrong representation of delays on xi.
So, questions:
  • does my doubt come from a wrong understanding of how the timeseries are prepared for the narxnet, or is the narxnet indeed not prepared like would like?
  • can anyone confirm that the delays as I have defined them will be understood by the narxnet?
  • I wanted to try and investigate the net structure using analyseNetwork(net) in the hope I would see exactly which values are fed and how, but it doesn't work for narxnet network. Any alternative?
  • on top of having different delays for input and feedback, is it possible to have different spans? I am thinking: inputDelays = 0:5 and feedbackDelays = 6:20.
Thanks all!

回答(0 个)

类别

Help CenterFile Exchange 中查找有关 Sequence and Numeric Feature Data Workflows 的更多信息

产品


版本

R2019a

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by