Why didn't I think of this before?!?

[Inor]

I only have about 6 weeks of data for the well, our freezers and the refrigerator. So far, I have partially trained 2 separate models, one for the well and one for the freezers. (I will save messing with the refrigerator for later.) This morning, I turned on the two semi-trained models for the freezers and the well. They are not actually adjusting the settings on anything yet. They are only logging the decisions they "would have made" if they were fully active. I want to watch how they react and see what they would do in normal circumstances and try to get an estimate for how much, if anything, they would save if they were fully active. I also want to see how much improvement they make over the course of a week or two as the models continue to train on the new data coming in over the next several days.

[Jester-ND]

[Inor]

My models have been live and actually adjusting the settings on the freezers, refrigerator and well pump for a few months now, so I thought I should do an update to this post with the results...

For about the last 4-6 weeks, the software seems to be saving me a little more than 7% and a little less than 8% on my electric bill. It is hard to quantify it precisely because I set up a stock tank in the pasture behind our fence line for the cattle for our rancher neighbor. So we are burning through quite a bit more water than we have in previous years. But, based on having the well pump run a little more intelligently only running at night when the electricity rate is cheaper (except when the water level in our tank gets too low), it is definitely having a positive impact.

I have the freezers keeping the temp below 8 degrees during the day and below 0 at night and that seems to be working perfectly. The model for the freezers is also smart enough to take into account the outside air temperature when making a decision on whether to cool the freezers down more during the day as well. That part is actually working better than I thought it would. We had a couple weeks when the daytime highs were above 100 and I wanted a fallback to cool the freezers below 0 on those days just in case the power went out. (The power did not go out, but I wanted to make sure we were safe just in case.)

The models are still re-training once per week so they are continuing to improve. But it seems unlikely at this point that they will get too much better. I figure the best I can hope for is maybe 1/10 of a percent or so improvement over the next year.

So putting everything in terms of dollars and cents, I figure this project is saving me somewhere between $10 and $15 per month on my electric bill. To follow that up with the questions: "Was it worth the effort to do this project?" My answer would be a qualified "yes". In total, I have somewhere between 40 and 60 hours of development time invested in coding the software and getting everything set up on my network. If I were to bill that at my normal billable rate for contract programming the project would never pay for itself. But since I did not have to pay for any hardware and my programming time was basically free... Yeah, it was worth doing. It was certainly a fun project and there is a tangible and measurable benefit to doing it.

"Would I recommend that you build something similar for your situation?" Maybe. If you enjoy writing code and learning new stuff AND you have an unused computer laying around not doing anything... Sure, go for it. If your unused computer is some kind of Intel/AMD based computer, it is probably not worth the effort since the power consumption for the computer is high enough that it will be unlikely to save much overall. But if you have a cheap ARM-based computer (like a Raspberry Pi or similar), yeah, you will get some benefit.

So overall, I am calling this project a "qualified success".

[red442joe]

Yay!
Thanks for the update and sharing!

Joe

[Inor]

Another update on my Jetson project...

It has been live for several months now and it has retrained several times with new observation data as it has been running. The freezer and well modules have continued to get better and now appear to be saving me a touch over 8% per month on my total electric bill. I also wrote a new modules for the kitchen refrigerator and the furnace. They are have only been running for a couple months, so their contribution to the savings has yet to be determined. But given that my first models were only saving about 7.5% on my electric bill and the savings is now over 8%, they are adding something to the mix.

I have also been watching the resource utilization on the Jetson to see how much computing horsepower is being used. It is so little that it is barely measurable. So I began thinking: "What else can I do with it?"

I decided to pick up a couple cheap cameras to see if I could make a simple security system with them. The first camera and chip came this week. It is called an ESP32-CAM. It is a little bigger than a postage stamp and including the camera, they go for about $8.50. (The chip is about $5 and the camera module is about $3.50.)



Getting it working was a serious pain in the neck because the technical documentation for it described different pin-outs than what was actually printed on the board. But once I got that figured out I was able to get it running. For an $8 "computer" it works surprisingly well. The image quality, while not great, is good enough for what I want it for. It has built-in WIFI and Bluetooth. The built-in WIFI antenna leaves something to be desired in terms of range, but I figured out how I can extend that with an external antenna. Getting the external antenna working will require one change to the circuitry on the board which does not excite me because I am the world's worst solderer. But it is only one connection and if I screw it up, it only costs me a couple dollars.

Loading up a sample firmware application from the chip developers to demonstrate the capabilities, it does seem to support most everything I can think that I would need to build some simple game/security cameras.



They do make some other cameras that I can try with it as well. (The one I am currently testing is the cheapest, crappiest model they offer). Using it for security, I am probably going to want a camera that does not filter out infrared light. I am sure on cameras as cheap as these, the infrared filter must just be a piece of plastic glued to the lens. I have not yet taken it apart to know for sure, but I am pretty sure they would not spend the money for infrared filtering glass for the lens.

The other thing that looks pretty good with it is the power requirement. When it is fully active and everything is turned on, it burns roughly 110 milliamps @ 5V per hour. I can also put it in "deep sleep" mode and it burns less than 6 milliamps. So I can easily run it with a small solar panel and a couple 18650 lithium cells for backup at night and cloudy days.

Just doing some back-of-the-envelope math to get a sense for how expensive a complete camera unit might cost, I should be able to construct a complete camera unit with motion sensors, motorized pan and tilt, solar power with lithium battery backup power, and a 4GB local image storage, plus a plastic housing for the whole mess for less than $50 per unit. I figure I can construct several camera units, place them around our property and have them communicate to the Jetson to allow us to monitor and control them from a single location.

I also ordered another different camera board that is supposed to be a little better quality. It has a much better camera and a more powerful chip. But it does require more power (still under 200 milliamps when fully in use but I think it is somewhere around 20 milliamps when sleeping), and it is bigger (about an inch by 2 inches). The power does not bother me but the larger size might be an issue. If it is, I might see if I can get the camera for the new board to work with the ESP32 chip.

Whichever way I decide to go, writing the software to run the whole thing looks like it is going to be easy once get the electronics hardware part figured out.

So phase II of this project should give me something to tinker with inside this winter.

[Box of frogs]

My boy Inor is wicked smart !!

[Prepared One]

I am lazy, I bought ring cameras with solar panels so I have cameras that cover the largest part of the cleared property plus the gates. Easy peasy. I am going to put a couple cellular game cameras back in the woods and down by the creek as well. I'll know who's coming at least.

[TJC44]

Are you using an Arduino to run it, or something else? Looks like the price on it (like everything else) has gone up. Amazon has them for about $10 now. I was toying with the idea of getting the development kit just to have something to play with to keep me out of trouble. Having a camera up by the road (there is a turnout there that people use to turn around) would be nice. Hmmm...

[T-Man 1066]

Definitely a cool project that I will continue to follow your progress on.

What kind of camera range from the house do you expect to get?

[Dwight55]

Nice to know I'm not the only one falling into the "learn something new" pit.

Caught the laser engraver bug . . . and the drone bug . . . so now I've got a Phantom 4 drone decorated like a Vietnam era Phantom F-4 . . . including USS Coral Sea designation as her home carrier . . .

And as of last week I now have two laser engravers . . . a 3000 mw and a 10 watt . . .

Learning how to do all of this at 79 is a bit of a challenge . . . but it beats the heck out of the View and the Price is Right.

Nowadays when my wife asks me if there is anything on TV I just tell her dust . . .

But anyway . . . lets all learn something new and have fun doing it.

May God bless,
Dwight