How many devices and appliances can you detect?
There is no hard device limit, but the engine will have better performance when the total device number is less than 10.
What is the accuracy?
Accuracy, defined as whether a device is on or off at a given time, is called Activation Accuracy. Our activation accuracy of disaggregated devices is roughly 90% across our system.
Accuracy, defined as the power consumption of a specific, disaggregated end device, is called Reconstruction Accuracy. Our system is currently not publically benchmarked for Reconstruction Accuracy. We will post our expectations here when we have publishable results.
Is there anything you can't track?
Devices that consume < 10% of an overall circuit's energy (especially when they are on circuits with significantly larger devices) are difficult for the algorithm to identify.
Devices that always switch simultaneously with other devices (e.g. multiple lights or fans connected to the same switch or control signal) are impossible for the algorithm to separate.
Disaggregation may perform poorly on devices with highly variable activity states (e.g. plasma TVs where power draw depends heavily on media content).
Why don't the devices have names or labels to begin with?
Our algorithm for disaggregation is a blind signal separation engine. Disaggregation has a goal to separate and track devices over time. It will detect the signature of a device, but not the type, function or name of the device.
Naming a device is another class of algorithms called Classification. As of now, Verdigris offers a device labeling portal to help our customers identify and tag devices with labels. This labeled data is being collected, and in the near future, we will offer a device classifier to help guide or automate the labeling process.
Does a device need to be switched on or off to be tracked?
Yes, a device needs to switch at least 3 times per week to be tracked. Our algorithm needs to see the switching events to understand that devices contribution to the full power demand of the circuit. Once it sees the additive behavior of that one device, our algorithms will infer when the device is active in the full sum of power on a circuit.
Why do I see more devices than what I expect in the number of devices on a circuit?
Each unique activity state of a device may appear as a separate device. For example, a heater with low and high modes may appear as two separately tracked devices (low and high).
Devices are separated on each single phase of each breaker. 3-phase appliances (such as large motors) will show separate devices per phase, even though the three phases may work in concert to produce a singular device function.
Why don't I see devices on my circuit?
Disaggregation training requires a minimum of 2 weeks of time detect signatures. If your installation was within the last two weeks, you may not see devices populated in your dashboard.
If devices are too small to detect, or the device is always on, the disaggregation engine may lump your device as a "baseload". Baseload amounts are not show in our dashboard.
Why don't I see a specific device?
Device size (by power consumption) and device activity (on/off actions per week) are important to track devices. If you feel like a critical piece of equipment is not being tracked, please contact firstname.lastname@example.org for more information.