The Sweatpants Effect
By Alex Panchula, Director of Product Management, Data Science at Oracle Utilities –
Like many Americans, I’m working from home alongside my working spouse, kids and pet fish. Truthfully, I never believed last year’s entreatments: “We promise we will take care of the fish – you will never need to clean the tank.”
When the stay-at-home orders were first issued due to the health crisis, there was some curiosity within Oracle Utilities around how the expected energy shift would impact individuals and utilities. There are some assumptions we can safely make: office energy usage declines and home energy usage increases.
One of my favorite changes is what we’ve termed “The Sweatpants Effect.” Our deep learning detection model on weekend dryer usage shows there is basically no change before and after shelter-in-place orders. Laundry peaks around 9 a.m. and slowly drops over the course of the day.
Looking at weekday dryer usage we learned that, before shelter-in-place orders, dryer usage slowly rose throughout the day and peaked around 8 p.m. when people got back from work. But during the shelter-in-place order the weekdays now look very similar to the weekend. My conclusion is that sitting at home in sweatpants doing laundry during the day is no longer a weekend activity. And if it seems like the weekdays and weekends are blurring together, you’re right – this supported by our deep learning analytics of usage patterns.
Since Oracle Utilities offers personalized home energy management communication to millions of utility customers using machine learning models, the data science team was also asked, “Do the models still work with such a large shift in energy usage?”
In fact, the models weren’t specifically trained to distinguish between a work day and a weekend day when most people were home, and therefore the team was confident that the models would disaggregate stay-at-home usage without any issue. I’m happy to report that this is indeed the case; our deep-learning models remain incredibly accurate. An example of this accuracy is shown in graph on a single-family home.
Right now, my youngest is making a pitch for a pet hamster and assures me that I’ll never have to clean out the cage. I don’t need a machine learning model to know how that’s going to go.