Have AI and ML algorithms use new user entry caracteristics of Device Inventory for Current (Ampere) or Power (Watts).
One example: My On-Demand Hydronic Floor Water Heater has 4 heating elements using 5kW each (called stages). Each stage is turned ON/OFF successively. The device draws either 5kW, 10kW, 15kW or 20kW.
If I could tell AI and ML algorithms that my device uses 4 stages of 5kW each, then device detection would be eased.
Advantages would be to prevent false detections. For example, my drinkable water heater uses 4.5kW while my hydronic floor water heater uses 5.0kA stages. By knowing those caracteristics, AI and ML would not miss associate them.
@NorthMan, one of the tricky things about Sense is that it primarily looks at very short (maybe 1/2 second or less) ramps/transitions, not actual usage values. A lot of large usage devices, like EV chargers and some large heaters, have electronic controls that ramp the energy usage more slowly than the usual physics would dictate, so even though they are big users, their ramp eludes Sense’s main detector which is looking for a very short on and off transitions.
I use the words “main detector” because Sense has started looking for slower ramps, but those ramps are much trickier to categorize because they are all electronically controlled - they don’t have all the usual attributes of a physics-based ramp. I have two EV’s that charge at huge power levels (12kW and 20kW), but Sense is still working on figuring out those, and detections are lost every time the EV gets a change of firmware.
So the answer is yes, that info could be helpful, but right now, it’s not useful to the the main Sense types of models.
Thanks @kevin1 for detailled explaination on how Sense identify devices. I now understand why my 20kW floor water heater is not detected when a human eye could not mistake himself identifying it!
I look forward to see the new and to come algorithm that will look at slower ramps.
I will keep an eye on this feature to come. In the meantime, maybe I could order Flex sensors to plug onto my 20kW floor water heater. Would/could the data of my Flex sensors be used by Sense to train your algorithm? I might be part of the final solution. I’m ready to help
That’s exactly what the new Dedicated Circuit monitoring with Flex Sensors is for. It gives you immediate monitoring of the big usage circuits that pose challenges for current Sense capabilities, while also proving data to Sense so they can improve the machine learning. I would expect that the machine learning part is a long-term improvement.
