Hi everyone,

I am trying to build an artificial neural network for some load predictions. So, as one of the inputs, I have selected the actual load at the (t-1) time to predict the load at time 't'. By doing that results seems great, which lead me to the following questions.

My question is, is there any theoretical problem in this kind of approach? Is this similar to the moving window approach ? if it's possible, could anyone point me to a paper where this kind of approach (target at t-1 is used as an input) is used?

I really appreciate your help regarding this.

Thanks in advance.