It would be great if the Home Assistant had a power consumption learning mode. Having only one central wattmeter, for appliances with a constant consumption (light bulbs) in a mode where all non-permanent consumers are switched off, one could note in HA enter the total consumption when the light is on, then enter the total consumption when the light is off. And thus for HA to determine the approximate consumption of the light bulb. And then connect artificial intelligence to this system, connect only one central wattmeter to HA, and since HA knows the moment when the light is on and how much consumption has changed, so that artificial intelligence makes the characteristics of the devices over time. And it should learn conditions, what voltage is now, how bright is light. After long period of learning, HA will know models of light. And after what microphones for example can be used to learn consumption of vacuum cleaner and washing machines. And HA should show icons on devices, that they are matmatical ready models
I think Power Calc does a lot of what you’re referring to (minus ‘AI’ somehow determining the unmeasurable loads if every single power consuming device in a house is connected to HA).
Um… light bulbs usually have the wattage written on the bottom. 
It is very strange that they work bad, because after a month of learning result must be nice. I believe that HA knows about devices more, than such equipment. Maybe HA should have a list of products, for example, list of bulb lamps, and before install new lamp, HA should scan barcode. And use croud data for same devices.
It is impossible to map a house with only one wattmeter.
Undimmable lightbulbs are easy, but many devices today are impossible.
A fridge and freezer change characteristic based on outside temperature and I side temperature due to openings.
A water heater change characteristic based in water consumption.
But worst of all is modern chargers that can change characteristic in so many ways based on the device that is connected and it commands. These will obscure any wattmeter with a collective measurement.
Powercalc does this.
This means that Home Assistant must develop its own wattmeter, which makes reports in nanoseconds, adds up the data in batches, and then transmits to HA every 5 seconds. If artificial intelligence does not work well, then the reports are not accurate, then you need to make super accurate reports. After all, a Tesla car can even determine that the wiring is not suitable for charging a car. If you increase the accuracy of the readings, there will be success
One of my purposes with HA is to save energy, you seems to have ideas of making it an energy-consuming system, so what’s your point of having this power-consuming “feature”, is it just for the fun of it, knowing how much your 10W lightbulps consuming ( just count the min/hours ), have you thought about how much energy you HA-device consume with an i.e 20% increase in CPU, IO(disk-read/Write) ?
And i don’t mention lifespan of SSD, or SD for that matter(those who use this)
My Home Assistant runs on a Raspberry Pi - cost about £50. How much does a Tesla go for nowadays? 
And come to think of it, HA can do it too… Pretty impressive, eh?
if the calculations are at the ASIC level, it will not spend anything
I sound like you would also demand a 100% precise weather forecast for the next month.
When you AI can handle that then HA can probably also handle all the possible variables for power consumption.
only on GPS coordinates level, using AI to predict wind-directions, and upstream windcasts around his block 
Are you familiar with these models Wally? I am interested in exploring this type of analysis but would love to hear from others that have attempted this before. I am personally only interested in bigger consumption players like fridges, electric heaters, water heaters, etc. I would be curious to know the required sampling rates that devices like Sense require. You can apparently get away with 5 sec sampling rate but I am not sure how well this works…
Energy Disaggregation is using the characteristic “noise” a device create when it use the energy from the mains.
This work well for devices that have a constant energy usage pattern, but it becomes trickier with the more modern smart devices that change the usage pattern.
A non-smart bulb will use the same energy all the time it is turned on, but a smart will adapt the energy to what it needs in order to show the correct brightness, colors, warmth and so on, which creates a huge amount of different patterns for that bulb.
A non-smart electric heater might have 5 settings, so it will have 5 patterns, which is possible to handle.
A smart electric heater might have 100 settings that the control circuit can set to hit the right temperature.
The devices you wanted to track might be possible, because they are often not that smart.
The problematic devices are the computer controlled ones, like smart bulbs, media devices and computers and smart chargers.
Such analyzers work without HA, they will work better with it. Because the data can be enriched with the time of turning on the lights, TVs. States are already in the Home assistant. Now you just need to combine the data of the smart wattmeter and the Home assistant and the GPT model
Please let the user draw at least in the UI, in the picture, that if 400 watts is a vacuum cleaner, if 5 watts is a lamp. This crutch will help.
Power calc can already manage all the fixed value devices and it also have a lot of mapped devices with variable values.
The issue is mapping several variable devices simultaneously.
See Energy Disaggregation or Non-Intrusive Load Monitoring (NILM) blueprint for a blueprint that approximates power consumption of some devices. I’ve used it successfully for dryer, oven, stove, microwave, rice cooker, and EVSE.
Blueprint it is nice, but native integration can have more capabilities