Recommended Riemann Sum Method

[dsolva]

Hi, I have been using this integration a lot and I love it. It's great and simple and I'm even looking to add some of the light models I have.

Up util yesterday I had been using this only for lights and fixed consumptions determined by the state of my devices. So far, all numbers made sense. However, I started measuring my fridge energy with a plug and noticed my mistake on using the "trapezoid" default method due to the spikes (as documentation mentions) and started thinking about using the "left" method always unless it’s a plug with an uncertain polling of data and some other cases mentioned below. I'm open to being corrected if i make any mistake on the interpretation of how data and calculations work.

My line of thought is the following: If I'm measuring my fridge and for any reason remains at 0 since 8am until 8pm when its starts consuming 500W. HA's recorder will have two registries: 0 at 8am and 500W at 8pm and nothing in the middle because the value of 0 did not change until 8pm. The trapezoid method would draw a huge rectangle starting at 8am and touch 500W at 8pm which involves a huge overestimation. This is why the ideal method would be the "left" so it only draws a rectangle starting at 8pm until it drops.

I saw the setting "scan_interval" defaults to 10 minutes and i guesses that the trapezoid wouldn't be drawn far beyond the last scan (<10min) but it does. A real numerical situation is this: My fridge stays at 0W (constantly) for 30 minutes exactly and spiked to 447W. I see my energy increase from 1.71kWH to 1.82kWh instantly when the spike occurs. Making numbers it makes sense: (447W*30min/60) equals a square of .22kWh and thus a triangle (due to trapezoid method) of .11kWh which is exactly the jump I see in energy.

Now, that happened if i use my own power value from a switch (polled 1 per second) and I don't see energy jumps when I turn my lights on after a long time. Why is this? How does the calculation between calculated power and plug power to determine energy defer?

Because of this I'm thinking that the left method should (almost) always be used for lights and devices whose state might remain constant for long periods of time (turning on a fan, etc.) specially if their state is always determined by HA and the recorder will register it instantly (there wouldn't be a need for averaging with trapezoid since the left would be spot on). For example, if I have adaptive lighting and HA is changing brightness then HA is registering it on time and left is perfect. On the other hand, if the state of a device is used to determine the power and this value had to be polled or called by HA then maybe the trapezoid is ok to average for the unknown between states. This until I understand how the trapezoid works on power values from this integration...

What do you think?

[bramstroker]

@dsolva Thanks for your extensive writeup. I am in no means expert in how the riemann sensor exactly works and no mathematician guru ;-). Under the hood powercalc just creates the same riemann sensor which HA provides, so the logic is the same.
I think your best bet is to ask this question on community forum or discord (#energy channel) as there are probably people who know more about the exact implementation and can confirm your findings.

Your explanation sounds logic to me to make left the default integration method.
Also some users experienced issues with trapezoid in the past #837, #672 and they were also adviced to switch to left.

[dsolva]

I'm willing to write a more detailed explanation on use cases for both methods with visual examples so it can be added as a guide on how to choose one or the other. I know HA defaults to trapezoid because in real life it makes sense, not really in HA after understanding how records are managed. Once we have this guide I would suggest (if its convincing) to add in the documentation that people use left unless they know what they are doing and point to the guide if they are interested to know the details. I'm sure many people might suffer the miscalculations of trapezoid if they go for the defaults, since I see it should only apply for less than 10% of the time in plugs and never in measuring devices when HA defines their state.

[emufan]

I think as well, that left ist most of the time the best choice in HA. Trepzoid is in my point of view only good, if you have a real function and not a sum of discrete data point as you have in HA. I have already changes global settings from Powercalc in the past to left.

What I do not get as well, how my points are used by the Riemann sum within HA in cases with few points/state changes and esp. With a lot of data points/state changes.

[dsolva]

The longer a sensor stays without change, the bigger the "trapezoid". As you mention, its only worth if its a real function or maybe an energy sensor that is always changing in value and with consistent frequency, e.g. every 5,10,15 seconds there is a real change in value then drawing the trapezoid would be ok. But if there is 1hour of silence and then then bum many changes then that first change will produce huge error.

[rbolden]

I use LEFT for all of my HA energy sensors

[bramstroker]

Fyi I have changed default from trapezoid to left in version v0.26.0