RatLabGuy
You look like a monkey and smell like one too
- Joined
- May 18, 2005
- Location
- Churchville, MD
The Duke Energy bill thread has got me thinking again about monitoring sources of power use around the house.
In the past when I looked at these things, the cost to buy a complete COTS system seemed to outweigh or greatly extend the payoff. However their getting cheaper and the cost of energy is going up.
Is anybody using something? Whats been your experience on the level of detail available?
It seems for residential use, they fall into 2 categories. The classical way is a bunch of CT sensors that clip onto individual circuits in your box, then you track the consumption according to circuit. A complete setup w/ a bunch of sensors is several hundred $$, like TED Pro.
The new way is via a single sensor or two on the main line that samples specific details of the inductive load and figures out the actual devices based on their distortion profile, which takes some time for a machine learning algorithm to figure out. Things like Sense, Smappee, etc. I actually understand how this works, I know guys who do exactly that kind of signal processing with device-caused electrical distortion and we use this type of SVM adaptive learning in our lab. The big downside to these though is that you are reliant on the algorithm to "find" devices, and I know from our experience that they aren't great and need a lot of data.
I'm tempted to try out a Smappee, but I don't know if the lack of specific accuracy of circuits will be frustrating. I haven't found a vendor that combines both yet? I' ma nerd, and would be happy just building my own thing w/ a Pi, Arduino and bunch of CT sensors and write a data logger but having access to the pre-made learning algorithm API would be nice....
In the past when I looked at these things, the cost to buy a complete COTS system seemed to outweigh or greatly extend the payoff. However their getting cheaper and the cost of energy is going up.
Is anybody using something? Whats been your experience on the level of detail available?
It seems for residential use, they fall into 2 categories. The classical way is a bunch of CT sensors that clip onto individual circuits in your box, then you track the consumption according to circuit. A complete setup w/ a bunch of sensors is several hundred $$, like TED Pro.
The new way is via a single sensor or two on the main line that samples specific details of the inductive load and figures out the actual devices based on their distortion profile, which takes some time for a machine learning algorithm to figure out. Things like Sense, Smappee, etc. I actually understand how this works, I know guys who do exactly that kind of signal processing with device-caused electrical distortion and we use this type of SVM adaptive learning in our lab. The big downside to these though is that you are reliant on the algorithm to "find" devices, and I know from our experience that they aren't great and need a lot of data.
I'm tempted to try out a Smappee, but I don't know if the lack of specific accuracy of circuits will be frustrating. I haven't found a vendor that combines both yet? I' ma nerd, and would be happy just building my own thing w/ a Pi, Arduino and bunch of CT sensors and write a data logger but having access to the pre-made learning algorithm API would be nice....