Forecasting using a NARX network trained using catsampled training input

Question

Hugh Milligan 2021-4-15

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/803066-forecasting-using-a-narx-network-trained-using-catsampled-training-input

回答： Saurav Chaudhary 2021-4-20

Hi all,

I've been training different NARX networks using large volumes of training data, 236 sets, each containing a minimum of 100 inputs and outputs, and at most 1371.

I'm using input delays of 1:4 based on the significant lags I observed in the data. As I am using a time-series, and my data is discontinuous due to a large prescence of NAN in the data, I am using catsamples to concatonate the samples into one large discontinuous training set. This mean my inputs and targets are for the network are

1x1371 cell, each cell containing a 1x236 double arrays.

Once I've trained the net, I wish to forecast using new external inputs, however because of the formatting of catsamples in the training process, I am receiving a formatting error. Can anyone direct me as to how to prepare a small set of inputs and targets (7 hour forecast, so 7 inputs, 7 targets) to test to the network?

Thank you in advance.