Virtual Submetering

The mirubee platform identifies the individual consumption of some appliances using a single general meter (virtual submetering). Just as a speech recognition system identifies words, mirubee identifies household appliances. To be able to identify home appliances by software, virtual submetering technology, the meter reads and sends one sample per second of active, reactive power and voltage. This results at the end of the day in a graph of 86,000 points on which we apply the pattern recognition algorithm. The analysis routine is launched every night to analyze the previous 24 hours, it is not a real-time process, so the disaggregated information is obtained the next day.

The algorithm looks for steps of power used (corresponds to the turning on or off of devices) and other particular features of each domestic appliance such as for example:

 

fridge

washing machine

dishwasher

microwave

 

The refrigerator usually has a peak of consumption every time the compressor starts, the washing machine makes very characteristic 'ups-and-downs' each time it starts and stops the motor that turns the drum, the oven and the vitroceramic hob make some characteristic curves as they take temperature and then make automatic on / off cycles (the vitro with a frequency higher than the oven), etc.

 

Virtual Submetering

 

Therefore what is measured is REAL, it is not estimated nor statistical. However, it must also be said that it will never be as precise as a physical measurement with an individual meter in each appliance (real submetering). The error in the allocation of energy is around 20%. In any case, we believe that it is a good orientation and that the savings in hardware costs and inconvenience compensates.

The algorithm automatically detects some appliances, but for others it needs help. One way to help the algorithm to make the detections is the confirmation of several questions that the system launches automatically. That is, the system analyzes the consumption graph and when it suspects that a pattern resembles something known, it sends a notification to the mobile app (or email) of the style: 'Yesterday at 12 o'clock you used the dishwasher?' If the answer is affirmative, the system saves that pattern as the specific one of that dishwasher.

Another way to help the algorithm is the 'manual trainer'. The trainer is like a pattern recorder and is basically designed to indicate to the algorithm what the step of power of simple devices, ON / OFF type, such as an electric heater, a toaster or an iron is. The virtual submetering works with appliances that have characteristic patterns (normally associated with operating cycles).

The devices that can currently be detected with the algorithm are the following:

 

Automatic detection

Refrigerator, washing machine, dryer, microwave, hot water boiler.

 
 

Training assisted by questions

Dishwasher, oven, resistance ceramic hob.

 
 

Manual training

Electric heater, toaster, iron, kettle, etc.

 

For now, the induction hob, inverter-type air conditioning, lighting, televisions, computers and electronic devices in general are beyond the scope of the algorithm.