10 minutes moving average of voltage - how?

[mmmmm009]

Hi,

I use Esp32 and a powermeter. Every 1sec I get a new reading of voltage (let's assume 1 phase).
I'd like to calculate a moving average of Voltage in a 10 minute window.
I have no clue on how to approach this problem.

1. Are arrays implemented in Berry?
2. I'd like to create some kind of circular buffer with 600 readings of my voltage (10min => 60x10)
3. 600 floating point values...is it 2400bytes?
4. If I'd like to get a new/refreshed calculation every 1 sec... would it overload somehow Esp32? (in the mean time I need to do some other stuff in a fast_loop)

Thx for any help/suggestions.

[sfromis]

In Berry, you can use list objects as indexed arrays. I'd expect the list to occupy more than 2400 bytes, but with only 600 values, you should not get into trouble.

Simple calculations on at small array of 600 elements should be no problem several times a second, but obviously does depend on specifics. Any loop code you have will be blocking, meaning that fast loop cannot get in between. If you monitor timing to see it becoming "too slow", you could split list processing into segments of 100 (or whatever fits), and use tasmota.set_timer(0, ...) to get to next 50-ms tick of the main loop.

List processing with index access in a loop for i:0..599 ... end can be sped significantly up by not using the convenient for / range construct, but instead "old school" loops where you count up the index, and check if the max is reached.

Of course, you also have the basic for value:mylist .... end but usage of that depends on your circumstances.

[sfromis]

For a simple moving average, you do not even need to loop through the array, you can push/pop values at the end, and maintain a total, which you then can divide by the array size. Obviously depending on details of what you mean by "moving average".

[joba-1]

There is an alternative average algorithm that uses nearly no resources. Most of the time more than good enough.

avg = (newValue + avg * 600) / 601

hope it helps

[s-hadinger]

Well not quite. If you have less than 600 readings it's very bad.

See https://en.m.wikiversity.org/wiki/Moving_Average/Cumulative_Moving_Average

[s-hadinger]

Hint: the formula is

So you can use this until t is 600 and then use the formula above from @joba-1

[mmmmm009]

Thx guys.
I can wait first 10min before getting correct results.
so @joba-1 solution seems to be the best for me.
I’ll try.

[joba-1]

you can make it much better during start if instead of 600 you use min(collectedTimes, 600).

edit: and of course use that min + 1 for the division.

[mmmmm009]

off topic...

Is it possible to display my calculations here:

[sfromis]

If you create your Berry code as a driver, you do indeed have features for providing values to appear on the web UI front page, and as part of the regular sensors MQTT messages.

This is included in this example (if you ignore the I2C-related parts):
https://tasmota.github.io/docs/Berry-Cookbook/#creating-an-i2c-driver
Step 6 and 7 is where it happens, as long as you have added your class instance as a driver.

[mmmmm009]

so here is my calculations:

var avg1=0
var avg1_old=0
var avg_i=0

var avg2=0
var avg2_old=0

var avg3=0
var avg3_old=0

def every_second()

	if (avg_i>600)
	avg_i=600
	end

	avg1=(Volt1+avg1_old*avg_i)/(avg_i+1)
	avg1_old=avg1

	avg2=(Volt2+avg2_old*avg_i)/(avg_i+1)
	avg2_old=avg2

	avg3=(Volt3+avg3_old*avg_i)/(avg_i+1)
	avg3_old=avg3

avg_i=avg_i+1

        mqtt.publish(m1111, str(avg1))
	mqtt.publish(m2222, str(avg2))
	mqtt.publish(m3333, str(avg3))

end

In other loop (also every 1 second) I send all Voltages to Home Asssistant.
In HA I applied a filter: Moving Average (time based) 10minutes to all 3 Voltages.
(despite I send all 3 Voltages to mqtt broker every 1 second, HA for some reason receives those Voltages every 3 seconds).
(so HA does all the calculations on samples that comes every 3 seconds, and Tasmota does the same with 1 second intervals).
Here are the results (3 phases, blue lines are calculated by Tasmota and yellow by HA).
(so In fact HA and Tasmota they both are using not exactly the same samples).