Sliding average dummy device

[gibbon_]

Hi,

for measuring values over a certain time and averaging them, I would find it useful to have a dummy device that takes the last X values one sends to it and represents the average over these values.
Best case:

• X is configurable (i am thinking about 2-16 values)
  Only set values are taken into account (and unset values are ignored)

Do any such efforts exist? Did anyone hack this already? I was trying to use custom events and the values of several dummy devices, but did not manage to get to a usable solution yet.

I know that normally this kind of job would be done by a central home automation system, but I am trying to avoid this as much as possible. I am trying to use this for building an room activity meter (using several PIR sensors) that sanitizes the values measured in the ESP and output a calculated result value.

[jsilver]

This sort of thing might certainly find uses. Even if you have a central machine doing processing of raw data, it's often useful to use local horsepower to do simple smoothing, deadband processing, deduping, exception filtering, rate of change calculations, hysteresis or other operations on the raw data before forwarding it.

[gibbon_]

Agreed.

In this case it would be to do thresholding and filtering of unreliable input. As you said: We have the cycles, why not put them to good use.

[lyndondr]

I'm also looking for a moving or sliding average function also. Is this possible in the Rules and Custom Variables possibly?

[toffel969]

I'm also looking for a moving or sliding average function also. Is this possible in the Rules and Custom Variables possibly?

Ya it sure is, you can use rules and dummy devices to do that.

[lyndondr]

I'm also looking for a moving or sliding average function also. Is this possible in the Rules and Custom Variables possibly?

Ya it sure is, you can use rules and dummy devices to do that.

Do you have any examples you can post?

[Domosapiens]

Not a real moving average (FIFO), but an approximation.
Simplified (understandable) version of my hourly and daily temperature averaging.
(with thanks to my '80 teacher Digital Filtering, Prof. Dr. Hermann, of the University Twente)

Assumption:
1 sample per 10 min

Hour filter length=6
Define a Dummy task <x>, AV_hour
Val1: published Average
Val2: calculated Average
Val3: newest sample

Rule for hour filter:
On newest_sample Do
Taskvalueset,x,3,%eventvalue% //store newest 1/6th hour sample
Taskvalueset,x,2,(([AV_hour#Val3]*0.166)+([AV_hour#Val1]*0.833)) //take 1/6th of the new value plus 5/6th of the published hour value
Taskvalueset,x,1,[AV_hour#Val2] //promote the calculated Avarage to the published Average
EndOn

Day filter length=144
Define a Dummy task <y>, AV_day
Val1: published Average
Val2: calculated Average
Val3: newest sample

Rule for day filter:
On newest_sample Do
Taskvalueset,x,3,%eventvalue% //store newest 1/6th hour sample
Taskvalueset,x,2,(([AV_day#Val3]*0.0.006)+([AV_day#Val1]*0.0.993)) //take 1/6*1/24th of the new value plus 143/144*th of the published day value
Taskvalueset,x,1,[AV_day#Val2] //promote the calculated Avarage to the published Average
EndOn

For faster filter run-in give the published Average (Val1) a reasonable start value in the mid of the expected range.
For outside temperature .. say 10 degrees C.

On System#Boot Do
"initiate values"
EndOn


With the latest possibilities, you could use (I think):

On Time#Initialized Do
TaskRun <sensortask> //force first reading
"store first reading as published Average"
EndOn