 Notice. New forum software under development. It's going to miss a few functions and look a bit ugly for a while, but I'm working on it full time now as the old forum was too unstable. Couple days, all good. If you notice any issues, please contact me. |
Forum Index : Microcontroller and PC projects : AI research project for this month
  Page 5 of 5 |
|||||
| Author | Message | ||||
| lizby Guru  Joined: 17/05/2016 Location: United StatesPosts: 3378 |
Aside from pure play with the PicoMite, etc., my focus has been on what I might be able to run in case of a grid power outage with my 10kWh of LiFePO4 batteries--8 100Ah 12V batteries, plus a 50Ah one. I have 2 of the batteries wired as 24V 100Ah and 4 as 24V 200Ah. I may rewire the 4 as 48V 100Ah. So my question is, what is my usage if I want to be able to run fridge, boiler, freezer, water pump (and various 12V/24V lights and fans which will have nominal current draw)? I haven't had a current monitor on the fridge (hard to get to the plug), but have had on boiler, freezer, and water pump. My current monitors are Sonoff Pow R2 16A switch/current sensors flashed with Tasmoto firmware and reporting to my data accumulator devices. Here is the graph using the program above for October, 2023:  Boiler usage obviously changes with winter temperature, but well pump and freezer usage are fairly consistent. And we are not in the house in the wintertime, so end-of-October boiler usage is probably good. So these three are generally going to use between 1 and 2 kilowatts per day. I wanted to know the total daily running time for each appliance, so I asked Gemini to produce a python program to run on a pi to generate an html file. Successful the first time:  OK. Well pump generally has a daily run time of less than an hour. Boiler has a maximum of over 2 hours. Freezer has a lot more--mostly between 3 and 5 hours a day. But how much power do they use per cycle? I asked Gemini to produce a distribution chart cycle times over the course of an entire year (2023). That ran the first time, but it turned out that most freezer cycles ran for longer than the range I had anticipated, so I asked for an extended range for it. For the well pump, the vast majority of times it runs between 50 and 99 seconds:  For the boiler, there are a lot of cycles between 100 and 120 seconds, and then some dozens in the range from 400 to 600 seconds.  And the freezer: 396 cycles under 20 minutes, but over 1,000 between 20 and 45.  The other important factor that can be derived from my data set is average power for each cycle (exclusive of the inrush current). I asked it to chart that for any specified day by taking the milliwatt hours of each cycle, multiplying it by 60, and dividing it by the duration of the cycle. This gives the wattage at which it is running to have used the watt-hours it reports. Here it is for the boiler on October 31, 2023 (the day of most usage for that month); maximum of a little over 180 watts:  And the freezer, with maximum of 111 watts:  And then, the hard lift--short duration, but power hungry--the water pump, using between 800 and 900 watts:  So my conclusion: with the batteries I have, I may be able in a power outage to get a few days of running with this $499 Canadian BLUETTI AC70 Generator Power Station which can be charged from solar, batteries, and grid. I intend to test this. The hard task will be the water pump inrush current. Supposedly the Bluetti can go up to 2kW for short periods (1kW continuous). I have designed an electrical manifold which can activate one appliance at a time (using Tasmotized Sonoff Pow R3 switch/current sensors). The freezer (in the basement) would be ok only running at night. I have never experienced a power outage of more than 10 hours, though my wife once went through one which lasted nearly a week. The process: for the first three programs, Gemini produced code that ran and gave the wished-for results except for one syntax error. Then it lost its mind and regressed to the flags program on the PicoMite and the LINE and BOX commands, producing programs I hadn't asked for. It took me maybe 6 iterations and 20 minutes to get it back on track. Overall, it released valuable information from my data set which confirms to my satisfaction that I have a testable path forward for powering needed appliances during a grid outage. PicoMite, Armmite F4, SensorKits, MMBasic Hardware, Games, etc. on fruitoftheshed |
||||
| lizby Guru Joined: 17/05/2016 Location: United StatesPosts: 3378 |
Looks like this will be a better option for me: https://www.amazon.ca/BLUETTI-Elite-100-V2-Generator/dp/B0F42TNLVH 4 1800W sockets with 3600W surge and ability to charge from grid, solar, and batteries. PicoMite, Armmite F4, SensorKits, MMBasic Hardware, Games, etc. on fruitoftheshed |
||||
| lizby Guru Joined: 17/05/2016 Location: United StatesPosts: 3378 |
If it works, "ChatGPT Agent ... drops the marginal cost of a knowledge-worker task to little more than the price of electricity and an API call." Exponential View PicoMite, Armmite F4, SensorKits, MMBasic Hardware, Games, etc. on fruitoftheshed |
||||
| lizby Guru Joined: 17/05/2016 Location: United StatesPosts: 3378 |
AI for general research. This is becoming common. Almost daily now its "I wonder about x", and I ask an AI agent to make a chart or perform an analysis or summarize a document. I asked perplexity.ai about GDP (PPP) per capita for countries having a population greater than 30 million in 2025 and 1992 (post-USSR) in 2025 dollar terms, ranked by present GDP per capita. Only 4 of the 25 countries have moved more than 2 positions in ranking--Japan down 9 (and only country to have lower GDP per capita), Saudi Arabia up 8, China up 4 and the Philippines down 4. China grew by 820%, but Bangladesh and Nigeria grew more. Australia, Canada, France, Italy and the U.K. didn't change from positions 4-8. U.S. went from 2 to 1, replacing Japan, as Saudi Arabia moved from 11 to 3. Philippines and Russia were the only non "first world" countries to have less than 100% GDP per capita growth. Tremendous growth for the bottom 7 in 1992: 595-930%.  1992-2025.jpg) Edited 2025-07-22 22:44 by lizby PicoMite, Armmite F4, SensorKits, MMBasic Hardware, Games, etc. on fruitoftheshed |
||||
| lizby Guru Joined: 17/05/2016 Location: United StatesPosts: 3378 |
Well, a very unsatisfactory session with Gemini yesterday evening--unsatisfactory except that I ultimately got answers to the questions I was asking. There was frequent total disregard of the prompts I was posting, and reversions to programs (which I wasn't asking for) from weeks ago. At one point I asked it to repeat back to me its understanding of my prompt. It did pretty well. I prompted again with a very minor tweak, and asked it to execute the request, and it again reverted to a weeks-ago issue. This New York Times article on solar installations at remote off-grid homes on the Navaho Nation Reservation in New Mexico got me started: https://www.nytimes.com/2025/07/22/climate/new-mexico-navajo-power.html (shouldn't be paywalled). Looked like a great project. Very few details--the "panels ... can supply a maximum of 2,200 watt-hours each day" (a maximum--that implies maybe a 550W panel for a 4-hour sun day--but the ground mount racking they showed was for a lot more than a single panel). And "the battery provides enough energy for three full days without the sun". That would be a 6.6kWh battery--the nearest likely retail equivalent would be two 5kWh server rack batteries for 10kWh at less than $1,000 each. And "Each system, including labor, installation and five years of maintenance, is valued at $27,000 to $30,000." Well, "is valued at" doesn't necessarily mean "cost", but that seems extraordinarily high--maybe the cost of a roof-mounted 10kW system in the U.S. or Canada. Santan Solar in Phoenix, AX will sell 550W panels in a pallet of 31 for $185 each--enough to do 31 homes with one panel, or nearly 8 with a more reasonable 4-panel setup. Phoenix is a 6-hour drive from Ship Rock on the Navaho reservation. A single panel could directly charge a Bluetti Elite 200 V2 Power Station with 2kWh battery and 4 outlets capable of 2600W (total) with only racking and 10AWG wires needed between the panel and the power station. By personal experience, this could charge devices and run lights and fans on DC, and run an electric kettle to boil water for a cup of coffee (50W), fry 2 eggs and ham on an induction cooktop (60W), and cook a meal in a slow cooker. A bigger system could have 4 550W panels per homestead (or more), and 1 or 2 5kWh server rack batteries (which can directly charge the Bluetti). The first question I asked Gemini was about average daily insolation per month for Santa Fe, New Mexico (which would have similar insolation). It knew where to look, and after several of those mindless reversions, gave me this table for a 2.2kW system:  Well and good. On average, a 2.2kW system would produce between 8 and 13kWh per day. I then asked it to discover the production for the 4 worst consecutive days for insolation in 2023. After much more completely wrongheaded struggle, Gemini came up with this:  An average of 4.7kWh per day. If you used a quarter of the 10kWh battery each day, you'd have about 7.5kWh per day to spend. If you had a 1K gasoline generator, you could match your full average production. The good thing about the session, was that Gemini knew where to look for the local insolation data, the National Solar Radiation Database NRSDB https://nsrdb.nrel.gov/data-viewer, and it knew how to use the API to drill into that database. I trust that this serious issue of completely disregarding the prompt will be fixed in a later version. Meanwhile, I may subscribe to ChatGPT if Gemini persists with this cluelessness which you wouldn't tolerate in an employee. (Regarding the remote homestead installation, with 6 550W panels you could probably power a 9000 BTU or even a 12000 BTU mini-split heat pump with the Bluetti power station.) ~ Edited 2025-07-27 23:30 by lizby PicoMite, Armmite F4, SensorKits, MMBasic Hardware, Games, etc. on fruitoftheshed |
||||
| lizby Guru Joined: 17/05/2016 Location: United StatesPosts: 3378 |
"The Bitter Lesson — encoding human understanding into an AI tends to be worse than just letting the AI figure out how to solve the problem" The Bitter Lesson PicoMite, Armmite F4, SensorKits, MMBasic Hardware, Games, etc. on fruitoftheshed |
||||
| Page 5 of 5 |
|||||
 Print this page |
 |
The Back Shed's forum code is written, and hosted, in Australia. | © JAQ Software 2025 |
