use of correlation analysis and neural networks for the prediction of reactive power knowing the active power

Hi everyone, I'm an engineering trainee taking my first steps in the field of statistics.

What I have to do during my internship is, having the historical series of electrical energy of some cabins and determining through the use of neural networks if it is possible to predict the reactive power knowing the active power.

now, what I have here are the time series of the withdrawals of these substations (active power every hour and reactive power every hour for 2 years) and the time series of the MV production plants; what they asked me to do is to study the correlation between active and reactive power and then from the results create a neural network that allows the prediction of the reactive.

Not knowing the matter at all I ask you, how should I move? I have seen that for correlation analyzes there are a myriad of methods (Pearson index, spearman, autocorrelation) but I don't know what kind of data modeling I should do for these correlations to be correct and for them to be useful for building a neural network. I have read that the historical series must be analyzed by autocorrelation and subsequently the trend and periodicity must be eliminated, I am not fully understanding how to proceed and what I need to do precisely to treat my data correctly.

Can anyone help me? I hope I haven't been too confusing but at the moment I don't have very clear ideas.

[Giuseppe D'Amico]

heeeeelp