Forgive my lack of statistical terms, but what I'm looking at is trying to determine is on a specific day and a specific hour, a certain house has a change in their electricity usage.
The dataset has columns: house, datetime, and electricity usage (kwH) and spans 1 year. What I'm trying to figure out is if a house is told not to use electricity on a day at a certain hour, if the house actually did it or not.
Sample of the dataset
2017-11-24 06:30:00 0.110 52
2017-11-20 20:00:00 0.110 3
2017-10-29 10:30:00 0.080 7
2017-04-11 20:30:00 0.981 29
2017-07-10 17:30:00 0.530 55
The nature of the data is such that everyday there is a pattern of high electricity usage in the morning and in the evening, when people wake up and get ready for work, and when they come home.
So if the house was told to reduce electricity usage between 7pm and 9pm on 2017-11-25, is it possible to determine if they did, knowing the past household history? If there were more sampling times, like more days on which they were told to reduce usage, would methodology change?
My attempts have been to look at decomposing the seasonality and trend, and then looking at the residuals. So I would have a distribution of residuals for that 2 hour period for the house for the past year, and then do a t-test. Any further insight? Should I try separating the dataset and then using a flavour of ARIMA to predict, and then compare?