With NILM (Non-Intrusive Load Monitoring) we face a number of challenges:
- Energy Patterns of different devices sometimes look alike. In this case, we can't differentiate between the different devices. (eg. A heating element of a kettle can have the same wattage as the heating element in a coffeemaker)
- Up until now, except for sub-metering with a Smappee Infinity we never really had a way to get a clean Energy Signature to use in our algorithms.
- We detect multiple energy patterns that are part of the same appliance (eg. Washing machine has a water pump, heating element and motor)
How does it work?
The Smappee starts with recognizing patterns.
Every time you turn an appliance On or Off, the used power will go up or down by a certain amount. If the Smappee detects enough of these Up's and Down's then it will be able to cluster this into what we call a "Find Me".
In general, the more you turn an Appliance On or Off, the faster the Smappee will detect it.
Once the Smappee has detected enough patterns, you can use one of the methods available in the App to start labelling the appliances.
We have different methods to label your appliances:
- Long-running program (Automatic)
- Learn with Switch (Semi-Automatic)
- Assisted learning (Manual)
- Auto labelling (Automatic)
- Event feedback
- The Appliance should be turned On and Off within a period of 6 hours.
- If it is turned on for longer than 6 hours such as an HVAC it will not be detected.
- The Appliance should draw the same amount of power every time it is turned on.
- Appliances that keep toggling to use more or less power, like heat pumps do not get detected
- An appliance with multiple modes, such as heater with mode 1,2,3,.. will result in a Find Me for each mode. The mode that is used the most will get detected first.
- 3 Phase appliances will have a pattern on 3 phases, this will result in 3 Find Me's
The appliance detection is based on the measurements and has some prerequisites to function optimally:
- The Phase mapping and measurements must be correct
- The production that is injected into the electrical installation must be measured with separate clamps
- For the Smappee Infinity, the correct type of load must be selected, we run the Appliance detection on the following type of loads
- Circuit - Power Outlet
- If you have selected any other type of load then this is subtracted from the Grid signal, meaning it gets skipped in the appliance detection.
- For all Smappee monitors, the Survey must be filled in (Otherwise NILM does not start!)
- For all Smappee monitors, the NILM only works when the installation is done as Residential
- For Smappee Plus/Pro you also have to activate the NILM on the different channels
- Never ever activate NILM on production channels
- When you have channels that are set as type "Submeter", the signal is subtracted from the Grid signal unless you activate NILM on that sub-metered channel.
The Survey will work as a basis for all the methods below and is mandatory to have correct auto labelling and correct long-running detection.
To fill in the Smappee survey you can click on the link on the dashboard page in the Smappee app or it can be found under Settings→ Your location(s) → Survey.
In the survey, we will ask you what kind of house you have, with how many people you live there and which type of appliances you have.
By using the feedback of the survey we know which Appliances we should and shouldn't detect. Which improves the accuracy of our Appliance detection and prevents wrongly detected devices.
The automatic methods will also not search more than the defined amount of appliances in the survey.
If you, for example, select 1 fridge in the survey, if the appliance list contains a fridge, then the automatic methods will stop searching for other fridges.
(For Infinity, sub-metered loads that have type Fridge also count as a Fridge)
Long-running program (Automatic)
The Long-Running program will serve as a basis to detect devices with potentially multiple parts (Find Me's) and devices of which we have a big database of signatures to match.
In the event summary, we show when they were turned on and how long they ran for.
Devices that can be detected with this technique: Dishwasher, Stove, Oven, Coffee Maker, Washing machine, Clothes dryer, Iron, Car charger.
Auto labelling (Automatic)
The Auto labelling algorithm searches for "Simple" signatures in your home that can be automatically labelled based on the survey. In the events summary, we show their On and Off event.
Devices that can be detected with this technique: Fridge, Freezer, Vacuum cleaner, Water Pump
Learn with Switch (Semi-Automatic)
The nice thing about our new Smart Plug (called the Smappee Switch) is that it can be used to measure a clean consumption and signature of the appliance that is connected to the Switch.
By turning on the "Learn with Switch function" in our new App you are essentially helping us in detecting the device that is connected to that Switch.
To do this detection the Switch compares it's measured signatures with that of the main consumption clamps. The Learn with Switch method found in the details of a Switch.
Devices that can be detected with this technique: Microwave, Kettle, Desktop computer, Laptop, Printer, Game console, Home Cinema, Media player, TV, Coffee Maker, Toaster, Steamer, Fryer, Vacuum Cleaner, Hairdryer, Water heater, Water pump, Heat pump, Tanning bed Bread oven
Assisted learning (Manual)
The assisted learning method, (which can be started by pressing the + symbol on the bottom right in the appliances section), is basically asking the user to turn on/off predefined devices which pinpoint the Smappee to the correct Find me's.
Devices that can be detected with this technique: This can be used for all the same devices like "Learn with Switch" with the addition of Fans, Garage doors and Blinds that are not connected to a power socket.
Please keep in my mind that you'll have to turn on/off a device 3 times within a 5-10 seconds period, it will be hard to detect for example computers using this method.
Finally, you can also Swipe on/off events to the left or to right to provide feedback to us whether they were good or bad. The feedback itself is analysed by our data-analysts and used to improve our algorithms.