# Understand the forecasting of hourly data with ARIMA

I'm trying to forecast the next 24 hours of electric consumption based on historical consumption, which constitute 24 records per day. I'm not sure if I understand the ARIMA model well, but if I want a good forecast from it for every hour of the next day, do I need to split my data into hours and forecast from that? Or is there any implemented feature in ARIMA that knows each hour's forecast should be based on that hour from previous days? I'm trying with a couple of options in R with seasonality period; something like:

timeSeries = ts(deviceData[,3], frequency=24)
timeSeries = msts(deviceData[,3], seasonal.periods=c(24,7*24))


but the result I got is very bad when I compare it with my historic data.

• You can try seasonal arima with seaon 24 – kjetil b halvorsen Mar 30 '17 at 6:10