Mismatch from mains to sum of all CTs

[wojo]

I'm seeing about 100W difference between my mains CTs and the sum of all my other CTs.

I have 42 circuits, most CTs are Micro-40s but 4x Micro-80s on larger loads. The mains are SPLIT-200As. Filled up 32 channels with 9 channels grouped up on the GEM.

In Grafana I've summed up the mains, subtracted the sum of the other CTs and that results in the following. I've double checked these values on the Dashbox and also again by my own queries on the Dashbox hourly data as well. Look over time, it never varies much off 100W except once in a while down to about 40-50W. I've tried correlating that with other loads and couldn't find anything that had a relationship to current draw.

From my own Influx data. Top is the missing load, bottom is just the consumption by channel much like the Dashbox shows:

Screen Shot 2020-07-26 at 23.51.19.png (325.08 KiB) Viewed 4869 times

From the Dashbox SQL database directly (in kWh). Matches up being around 100W per hour, so 0.1 kWh:

Screen Shot 2020-07-26 at 23.52.53.png (261.27 KiB) Viewed 4869 times

So overall I've tried to check everything I can think of:

• Ensuring all CT connections are solid on the GEM, manually testing each load
• That all grouped loads are set correctly in terms of polarity on the CTs (both for 240V balanced loads measured by one CT and also 240V unbalanced loads or other logically grouped loads e.g. two light circuits). Manually tested each as well to ensure no canceling out occurred.
• Channels are set to double properly for balanced 240V loads
• Trying to account for GFCI and AFCI/GFCI power consumption, which seems to be for my Eaton CH breakers between 0.75W and 1.3W per breaker. I have 28 non-dumb breakers, even at 1.3W each (the maximum for any of my breakers, the CHFAFGF115PN) that is only 42W. If it was just this accounting for the 100W, that'd be 3.5W per breaker. That doesn't seem right.
• I also shut off every breaker except the one powering the GEM/DB, and it resulted in 11W being consumed on both the mains and that single CT. Turning on two AFCI/GFCI combos resulted in an increase on the mains of around 3W... which makes sense if ~1.3W x 2 per Eaton is the power consumption of the breakers. I stopped there because all the rest of the circuits have _something_ on them, wanted to test with zero load.
• Shutting off all breakers (except that one) resulted in all channels showing 0W, so I guess I'm not getting any noise that seemingly requires the jumper

Breaker calucations:

11x AFCI/GFCI @ 1.3W = 14.3W
16x AFCI @ 0.75W = 12W
6x GFCIs 3W (unknown, assuming 3W on the high end. One breaker, rest outdoors/garage/mechanical in-wall) = 18W
= 44.3W

So 100W - 44.3 = 55.7W still unaccounted for.

This happens even with total house load at say, <1kW. The Micro-40s per the latest threads I could find are damn accurate it seems at 1% or so, so that'd only be <=10W difference (I forgot if that is under/overshoot). Either way the math doesn't add up.

So maybe I'm just obsessing, but 55.7W is still nothing to sneeze at 24/7.

My next steps I think are to calibrate off a known load, and go from there. Thoughts?

[ben]

The example at the top, is that 123 into ~9500W? If so, that would be around 1% and I'm unsure if you're going to be able to get it any closer. You can try the known load idea as you were thinking just in case some of your CTs are a tick or two off.

Also I would check the voltage vs the voltage at the same outlet to see if there's any adjustments that can be made there.

[wojo]

The example at the top, is that 123 into ~9500W? If so, that would be around 1% and I'm unsure if you're going to be able to get it any closer. You can try the known load idea as you were thinking just in case some of your CTs are a tick or two off.

Also I would check the voltage vs the voltage at the same outlet to see if there's any adjustments that can be made there.

The difference of 100W is pretty much constant from 1kW to 10kW for example, which is very odd.

I've checked the voltage with true RMS, I'l check again (haven't done it since install)... but wouldn't that impact all channels the same?

[ben]

The example at the top, is that 123 into ~9500W? If so, that would be around 1% and I'm unsure if you're going to be able to get it any closer. You can try the known load idea as you were thinking just in case some of your CTs are a tick or two off.

Also I would check the voltage vs the voltage at the same outlet to see if there's any adjustments that can be made there.

The difference of 100W is pretty much constant from 1kW to 10kW for example, which is very odd.

I've checked the voltage with true RMS, I'l check again (haven't done it since install)... but wouldn't that impact all channels the same?

It'll affect channels differently based on power factor. Your Main Panel for example is probably around 80% whereas a lighting breaker should be 100%.

[Teken]

Having done this more than eight years ago here are my thoughts and insight regarding obtaining the best accuracy & performance.

- Line Voltage: Use a high quality True RMS meter to measure the outlet and adjust the GEM accordingly.

- Burden Resistor: It costs you nothing but time to insert a jumper to see what happens. I added a few jumpers to several channels to null out very low level power readings.

- CT's: As you and Ben noted depending upon CT's they all have a tolerance and variance.

- Reference Load: Depending upon the CT type and tolerance validate your resistive loads. Again using a high quality meter and (OHMS Law) that can offer voltage and current. You'll need at least three loads from low, mid, high to offer the best insight about variance. You will find that the variance may be at the lower end vs the high end depending upon each CT in place or load. Again, I can't stress enough the reference load being used must be known and accurate otherwise adjusting the CT values will just give you out of band values.

I used several incandescent light bulbs in various wattage's which I tracked in my projects thread. Along with a hair dryer, space heater, and infra red heating lamp. All of the loads were measured with a calibrated clamp and several other high quality DMM's. You would be surprised to see that a standard incandescent bulb of any wattage isn't reflective of the stated wattage.

You'll need to monitor the line voltage and do OHM's Law while you're measuring to give you perspective and reference. Lastly, if you extended the CT's at all this is something that will impact the system.

[wojo]

Thanks Ben and Teken!

Checked the PT today and all good there per my True RMS DMM.

Will start checking some of the loads with a known reference and try the jumper, but I'm mostly happy now that it's stable and I can at least track deviations from that (e.g. wire fell out, phases got screwed up on a bundle, etc).

[wojo]

This is interesting.

At the top is voltage, middle is the missing watts (around 100W not accounted for), and the last one is the per channel dwh value. I chose a few of the top channels, notice that there's a striking correlation to each dip in the missing watts to a coordinated decrease in load across ALL channels proportional to load. Conversely, when all channels spike higher, it results in a higher amount of missing watts in the overall calculation.

What's really weird is to see all the channels move together like this. The missing watts doesn't change when a load turns on/off at all, small or large (e.g. that bottom green load ch18, nor big ones like the A/C.

... however, it does change +/- 40-50% based on these spikes up and down across all channels. This seems odd?

That first blue line marked dip is 21%/52W decrease 245W->193W on ch8 (red). The rest of the channels following proportionally. ch4 (green) is 123W->97W is a 21%/26W decrease, and same story for ch7 (orange) at 74W->57W for a delta of just over 22%/17W.

Meanwhile at the same time the missing watts changes from 90W to 60W or 35.45% decrease.

Screen Shot 2020-07-28 at 8.20.49 AM.png (229.41 KiB) Viewed 4850 times

(for my reference this image was generated from this timecode: http://nas:3000/d/iCJKZUnMk/power-detai ... 5930407377)

[ben]

What's the packet send interval you're using? Looks like it makes up for the loss with the next packet, could be slow to process at times.

[wojo]

What's the packet send interval you're using? Looks like it makes up for the loss with the next packet, could be slow to process at times.

Interesting, it's 15 seconds currently. Sometimes it makes it, sometimes it doesn't (which seems like it wouldn't be a timing/framing issue).

I'll play with some send intervals regardless and see what it does.