Methods for daily water consumption prediction



Data available:-
For a county in the USA
Following columns are there:-
MeterId, Consumer_Name, Address, Meter_Size
Note:- For each Consumer_Name there can be more than 1 meter and hence MeterId.
For each MeterId daily water consumption data transmitted by smart meters from May 2015 to May 2017.

Here is what i have tried:-
Analyse patterns in the data for meters belonging to a particular consumer_names like schools etc
Some of these meters show seasonality like day of the week,month etc
Also effect due to holidays are present in the meters belonging to school, banks etc

Apply auto.arima with external regressors day of the week and month of the year.
But results are not very accurate, I have data for June 2017 to verify.
Please suggest any other approach
P.S:- It isn’t possible to look at acf and pacf for each and every meter as number of meters are quite large.