02-28-2017 01:42 PM
Are you not a reader of my superb blog Neural Market Trends? http://www.neuralmarkettrends.com/tagged/raspberry-pi
So yes, Studio can be installed and runs on the newer version of the Pi.
03-01-2017 07:23 AM
I put Studio v6 something on a Pi2 and it works, but took 5 minutes to load up. The newer versions of the Pi with more onboard RAM should be faster.
My ultimate goal was to put Studio on a Pi and send it up on a weather ballon in space and then have RapidMiner in space. lol
04-19-2017 12:35 PM
Tom Ott wisely recommended that I look at his blog as he has previously tackled this exact question. I actually am pretty close but am hitting the memory max issue on the Pi. There is a way to use a partition of the SD card as a virtual memory store via wmswap but all the online forums say not to do this...something with SD cards not liking all the r/w that a vmswap would do.
The current use case is purely an IoT hobbyist one. Here in VT I heat my home with a stove that burns wood "pellets". The stove is currently controlled by a custom circuitboard which does a pretty terrible job dealing with a variety of issues unique to the stove. I'd like to replace the custom circuitboard with the Pi3 so I can collect data via the sensors and create a ML model that will decide how to run the stove to maximize net energy output (BTU heat out minus electrical energy + pellet in).
Thoughts / ideas welcome!
04-24-2017 11:38 AM
Oh that sounds like a great project as we're getting ready to install a pellet stove insert into our new house.
My initial thought is that it would a time series problem similar to electrical usage prediction that was presented at RCOMM 2010. I have the paper somewhere but it was about predicting demand based on current usage. So for your problem I could see temperature trend and forecast being an input, possibly the time and day (i.e.during weekends you might be skiing), and of course the existing ambient temperature of your house. We get lots of sun so we have an passive solar heating going on, which depends on the calendar day (the lower the sun, the less passive heating) and my location.
04-24-2017 02:45 PM
Excellent. I like your idea about the time series and inside/outside temps as variables. I would like the Pi to control all aspects of the stove, not just on/off, so I was going to hard-wire the pi to
1. replace the crappy control board with the Pi:
- auger on/off
- combustion fan on/off
- convection fan on/off
- ignitor on/off
2. replace my programmable thermostat with Pi inside sensor
3. some new sensors that are needed:
- burn pot temp
- burn pot pellet height (mine tends to overflow)
- outside temp (as you said)
I guess I was thinking to get the Pi to control the stove first - not in a ML sense but just a basic system flow. Once that is up and running, collect other day to begin optimization via ML time series(?).
04-25-2017 10:13 AM
Yeah, I think you want to measure and capture the sensor data first and then look to optimize it.
Another option is to save all the sensor data as a CSV and upload it to a RM Server. Set a trigger and then run some processes to do the ML and predictions out there. You can then tweak the controller manually.