Using different input dimension test data in Neural Network

Hello !

I am working on neural network based time series forcasting. I have an input dataset of (153*8640) where 150rows represent my original data and the 3 rest rows are additional features (which are add to my data set to improve prediction performance). the target data is (150 * 8640), T = X(1:150, 2:end).

I have divided my data into train 70%, val 15% and test 15% datasets, using "dividrand". but the problem is about testing dataset which must be unique and do not contain additional features (the 3 additional rows) as new input data! so :

1) how to separate test data ?

2) is it right that test dataset musn't contain additional features as new input to the same trained net1 ?

3) does testing data need to have a target dataset to or we need just to predict ytest from: ytest = net1(Xtest) ? or we must train 'net1' again on this test dataset (Xtest) :

[net2, trtest] = train(net1, xtest, test, xitest, aitest) ? if yes, how to determine "xitest" and "aitest" ?

I have trayed to predict ytest directly from "ytest = net1(Xtest)" using 15% of input data where I took only (150 row) but, for sure, the net1 do not accept different input dimension (diffrent number of input rows 150 =/= 153)!!

4) is it right to test our network using an "input test data" composed of all rows containing additional features and 15% of the original data columns ?

Please! help me to get a solution