I have been looking at making a "monitor" node - The idea is it will be looking at data/events in all the nodes/devices and determine if something "strange" is happening. E.g. detect a water leak (slow or major), detect circuits being on when not normally on, determine changes in power consumption - etc. - I had my roof heat tape for snow removal running in the summer due to a faulty switch, have had small water leaks when people not turning a water hose completely off and so on.

The plan would be to use machine learning as part of the node so it over time learns the "normal" behavior. This will include home/away, time of year (maybe climate data) etc.

The current idea is to have the node send an email and maybe raise a warning flag - The node will not be able to control any devices - only set flags (to be used in programs) or send messages

The configuration will be complex (through configuration tab) - need to define what the different variables that need to be monitored represent (monitor event). It is possible an AI will be able to help generate the configuration over time

Is there interest in such functionality - I am looking for additional monitor events to plan for if there is interest. If not I'll just keep it as a private node (as I have a personal need/desire)

I do that kind of thing through regular programming. It's like an expansion of leak detection, alarm intrusion, etc. I look for weak batteries in my zwave locks, etc. Not sure how a distinct, separate node could really integrate a diverse set of unrelated conditions. At some point I might group these programs together in a distinct folder, that's about it.

Maybe AI could make suggestions in this regard, based on the monitoring functions of the devices you have.

i get your point, but you can only do what you can predict (think of) - The idea here is to identify items that behave different than normal - it may be ok, but in some cases it is not

e.g. getting a message that power consumption is 20% higher than normal for a circuit (this time of year), that water usage is higher than normal etc.

I expect most information is in them generated message - not the flag

What you seem to be saying is that AI could be used to learn what's normal, like monitoring energy use, run times, water use, etc and possibly correlate it to influencing factors, like outside temperatures. Then you could ask it to watch for anything it considers unusual, and alert you to it. Interesting idea. There are certainly lots of sensors now available that could make this possible.

@Panda88 Absolutely interested. It would make programming these tasks easier. Thanks for bringing it up.

Edited September 11Sep 11 by BobM99

@Panda88 ,

I love the idea. The only question I have is the integration with AI. Does the plugin require its own AI interface to LLMs or do you expect to reuse the eisy-ai ?

With kind regards,
Michel

Michel
I am planning to use internal ML (low CPU requirement) - it is not AI per say - the issue becomes setting up the configuration - determine what the different node variables represent, and which ones are the important ones to monitor / track. This is where I believe the AI can get involved. However, I see this as secondary - I need to get the core working first, and once this is working I'll look at adding AI to generate the configuration - I have not had a chance to play with eisy-ai yet, but I would imagine it can do the job well assuming it can generate configuration files (or json etc is that is easier) - I would imagine it would not have the right to access my node's database

This is interesting. I'm sure it wouldn't be hard to do, Today FLUME collects all my water usage and tells me when something is running when it shouldn't be its about 95% correct. I have been collecting my power meter with a Rainforest zigbee device collecting the stats from power every 2 seconds ( I like what they did, they allow you to add a POST url to the system. So when they collect from the meter, they POST to your web server) . I haven't looked but I know there are logs in eISY, but if eISY would allow me to see all the events on/off/motion/... I could write them to my DB. Then using all 3 DB's it would have a "general" view of my house and could do like flume and say, you left the bathroom light on, Or the yard lights are not drawing ### watts which it normally does, its over or under. And so much more. Its getting all that HOUSE data in one place or accessable in one place. The AI could then as you said generate hi/low points or start time/end time events. If would make it easy and give you a pretty good starting points for your alerts.

@Panda88 , awesome. Let me know if I can be of any assistance.

@awysocki , you can already do that with REST APIs. We collect everything in a database.

With kind regards,
Michel

This is a very interesting idea. Thanks for bringing it up. I look forward to seeing it progress.

Lots of AI models do this for various other things. It sounds like a potentially useful item, but I don't know how to do it without AI. Of course it would take a couple years to really figure out what things are predictable and what things are chaotic. And then there is the part about things that actually matter. If 99% of the time you only use a light switch in the evening, and then one day you use it in the morning, do you really want a notifciation?