Data Intelligence, Business Analytics
I have a timeseries data for last 5 years of electricity consumption for 52K users. I have to provide 3 month forecast for future consumption level for the all 52 K users for the pupose advance billing . The data available has granulairy of daily level. and forecast is required for monthly.
I have previously worked on time series data for forecast but the peculiarity of this situation is - Here we have multiple users and we cannot do forecast for all of them.
Please suggest me the best way to carry out such exercise.
Does the classic methods of forecast work here of I have to adopt a hybrid method (may be a combination of cluster and forecast) to overcome this .
Thanks in advance !!
Why do you say you can't forecast all 52k??? Of course, you can.
Yes, classic forecasting methods work here. I would recommend 3+ years of daily data and then roll up to a monthly level as day of the week and week of the year or month of the year fixed effects are important to model. Include holiday dummy variables, hdd, wind and future expected values of these.
Visit us at autobox.com for more on this type of work.
Many thanks for the reply , Tom !!
Please suggest, if the following approach is good and practical way ,
Step 1 - Shoud I pick just one customer , roll up his daily numbers to monthly , include other factors such as holiday, hdd, wind etc. and then forecast the numbers.
Step 2- Apply the same model to other customers.
Assuming models is what has the world in a heap of trouble. Most software systems pick 30 models from a list and that is what you get. If you got a Rx for your glasses from a list of 30, would you be happy?
We have a tool that can do this automatically. I can run a demo on your data and show you what we can do. Send an email to firstname.lastname@example.org.
I was just thinking about this problem of yours. I've never myself come across forecast ever for more than one group/cluster/cohort/individual!
I would presume you might have to aggregate data to some level before going about forecasting - unless you want to forecast as Panel Data?!
@ Tom- Thanks again, Tom , will send you the mail at mentioned id right away.
@ Arun - Thanks Arun. This is for a project on operation analytics for a electricity distribution company. You'll find similar pain areas in water or gas distribution comapanies since they generate the monthly/quarterly future estimates of usage and send advance bill.