I would know what are the best predictive model to forecast consumption data with R or Python, I have got data from one month (each day and each 15 minutes) and i have to forecast data for tomorrow for each 15 minutes (00h00, 00h15,...,23h45).

Thank You

Edit :

X1 : Names of region

X2 : Date (01/01/2017,...,Today)

X3 : Hour (00h00, 00h15, 00h30,...,23h45)

X4 : Power consumption (5000, 6000,...,5500)

X5 : Production of Power - 1 (400,200,...)

X6 : Production of Power - 2 (10000,12000,...)

X7 : Production of Power - 3 (400,200,...)

X8 : Production of Power - 4 (100,200,...)

X9 : Production of Power - 5 (0,2,...)

X10 : Production of Power - 6 (4000,2000,...)

X11 : Production of Power - 7 (0,-2000,-10,...)

X12 : Difference between consumption and production (-8000,-8000,...)

I have to predict X4, that's my TARGET, I heard about ARIMA, time series is it the best model in this case ?

  • $\begingroup$ what is your data and what do you want to forecast ? state your variables. $\endgroup$ – Subhash C. Davar May 21 '17 at 22:52
  • $\begingroup$ I edited my post $\endgroup$ – John_Trump May 21 '17 at 23:32

Warm welcome to data mining!
You will need to do some serious learning.
The best model for a given problem can't be identified theoretically. It's more a trial and error exercise based on experience. You will also need a lot of data with as much as possible helper research as possible. Like energy consumption may be influenced by:

  • Day is a weekday or a weekend
  • Or maybe better: day of week
  • Length of the night vs. length of daylight
  • Industrial activity in the area
  • Trading activity if available
  • ...make some research for more ideas

This is feature engineering or 'hacking', take a look into it.

Since you are planning to forecast something the choices are a dense DNN or a multi layer RNN. A dense DNN is more simple but it can be used for prediction only as a benchmark. Any proper RNN models must be superior to a dense DNN.
Secondly you need more data. Forecasting a day based on a month of data may be too ambitious.

I would like to suggest that you go in the direction with learning:

  • Data normalization
  • Deep neural networks (Dense)
  • Recurring neural networks (RNN), like LSTM or GRU. Ultimately you will need these.

I started with these sources, but you may prefer other points to start:
Artificial Intelligence for Humans

Application of deep neural networks

P.S: I'm not sure that X5 through X12 adds any value for the prediction. I believe that would need more industry specific knowledge from me.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.