0+0+0+0+...=23W? Any way to improve small-load accuracy?

[jnh]

I installed my new GEM system just before the new year, and have been happy with it overall. So nice to be able to monitor individual circuits now! I kept the Split-200, Split-60, Micro-80s and Micro-50s in place from my 6-year-old ECM-1240, and added Micro-40's to all remaining circuits, spread across two panels. I'd originally ordered it with Micro-50's, hoping for the best accuracy, but these were unfortunately out of stock, so I agreed to substitute Micro-40's.

This is the first time I've used the Micro-40. Is this particular CT known to be less accurate with very small loads in the range of 1-2 watts or less? If so, is there any way to improve it, either by adjusting settings (preferred) or changing the way it's hooked up? Since most of the M-40's are on 15A and 20A branch circuits that don't need its full 40A range, I thought about putting in my own burden resistors of double what the GEM has between pins 2-3 of each input, connecting this and the CT in parallel to pins 1-2, then raising the CT Range value by one. Would this help? It'd be a bit of a wiring mess, since the GEM's push-in terminals don't look like they're meant for more than one wire.

Would it be worth considering an upgrade to Micro-50s, once those are available again?

What I'm seeing is a main-breaker measurement for my subpanel that's always anywhere from 13W to 25W higher than the total of all 17 individual breakers on that panel. Most breakers are reading 0W at any given time (Watt-seconds counters not incrementing at all), but turning these off one-by-one does reduce the residual load measured at the main by about 1W per breaker.

Most of these zero-reading breakers do have small phantom loads on them, such as GFCI receptacles, remote-controlled receptacles, appliances without mechanical switches, unloaded chargers, surge-protector strips with indicator lights, etc. I guess it's less than 1W on most circuits, but some must be slightly more, because there aren't enough breakers to fully account for the residual load sometimes hitting 23-25W.

Is the GEM capable of tracking fractional watt-seconds at all (using any variety of CT), such that, say, a 1/2 watt load would increment the W-s counter every other second? One of Teken's old posts mentioned a best-case precision of 0.27W, but I don't know how he figured that.

My GEM is mounted directly above the subpanel in question, with CT wires trimmed to less than 2' each. The subpanel main breaker has a pair of Micro-50's on it (leftover from ECM-1240, and formerly the only CTs in this panel), with Micro-40's on all individual breakers. This is my 120V-only, protected-loads panel that's kept alive from a hybrid solar+battery-based inverter during blackouts, and it has no load circuits greater than 20A. Both red & black "hot" legs coming into it come from the same inverter bus-bar.

In the main panel, a few feet away (12 mixed CT sets there are spliced into three Cat5 cables, about 10' each, run through attic & wall cavity to reach the GEM) I have a Split-60 on the single-pole 60A breaker feeding the inverter panel's AC-In, and indirectly the subpanel when it's operating from normal grid power. When no power is being supplied or drawn by the inverter panel, the Split-60 (feed to subpanel) and paralleled pair of Micro-50's (subpanel main) usually agree with one another to within a couple of watts. So, I'm inclined to trust those mains readings, and suspect individual circuits are reading a bit low rather than the main breakers a bit high. The GEM's PT and power supplies are plugged into a circuit from the subpanel (so that they'll keep running in "off grid" mode), which usually has no other active loads on it.

My inverter is a true-sinewave type (Outback GVFX-3524, 3.5kW) that's certified for feeding power back to the grid as well as operating standalone, so I don't think it's a factor here. Also, there's little or no difference in the residual (Main minus sum-of-loads) subpanel reading between when the inverter is active during sunny hours, vs. dormant at night and acting as only a pass-through (its transfer relay connecting AC-In straight through to AC-Out).

All inputs of the GEM are of equal accuracy and precision, right? It's not like the ECM-1240 where Ch1/Ch2 do a little better than Aux1-Aux5? I wired main panel loads starting to inputs 1, 2, 3... and subpanel loads counting down from from 32 (sub main), 31, 30, 29... so that leftover positions in the middle could be used to expand either panel while keeping contiguous groups.

Interestingly, the Main-minus-sum-of-circuits figure on my main panel is usually in the -5W to 5W range. It has mainly 240V loads, though, most measured by M-80s, with the few 120V circuits there (dishwasher, washing machine, outside lighting, charger for 12V battery system) on the opposite hot leg from the one going to my inverter and subpanel, for the sake of balance.

[ben]

Spec-wise the Micro-50 is superior on the low/high-end. In practice, we haven't seen a very big difference. I'm not sure when/if we'll have Micro-50s in stock again currently, we'll have to do more investigation to see if it's worth having both types stocked (how much of a difference does a Micro-50 add).

Try messing with the CT range and dividing it on your end. CT range will half the signal as you increase it or double as you lower it. You might be able to use this to overcome the GEM 1-wattsecond accuracy.

All GEM Inputs are equal.

[jnh]

Cool, thanks for the suggestion on ranges! I guess this adjusts a pre-scaler in front of the ADC? Is there any risk of damage from accidentally setting a channels' range too low and driving it into overload/clipping, or just of bad data and possibly crosstalk?

I briefly changed one of the CT-40 channels that had a known small load on it from range 4 to 3 via terminal command, and its readings doubled as expected (until now I'd assumed ranges worked the other way around, given Split-200's need Range=2).

So, I just need to update my logger to allow for a per-channel divisor or shift-right count, and maybe dither post-division fractional Wh's across successive sample lines, to avoid losing text-column space to .5, .25, .75 readings. I mostly look at the realtime data in monospaced tabular form via ssh login, and of course .5+.5+.5+.5 vs. 1+0+1+0 won't matter for the cumulative reports.

[ben]

Cool, thanks for the suggestion on ranges! I guess this adjusts a pre-scaler in front of the ADC? Is there any risk of damage from accidentally setting a channels' range too low and driving it into overload/clipping, or just of bad data and possibly crosstalk?

I briefly changed one of the CT-40 channels that had a known small load on it from range 4 to 3 via terminal command, and its readings doubled as expected (until now I'd assumed ranges worked the other way around, given Split-200's need Range=2).

So, I just need to update my logger to allow for a per-channel divisor or shift-right count, and maybe dither post-division fractional Wh's across successive sample lines, to avoid losing text-column space to .5, .25, .75 readings. I mostly look at the realtime data in monospaced tabular form via ssh login, and of course .5+.5+.5+.5 vs. 1+0+1+0 won't matter for the cumulative reports.

I'm not sure how exactly it scales the value but there's no risk of damage. The worst that would happen is the counters may rollover more often if viewing heavy loads (1000A loads for example).

[jnh]

I staged the doubled range settings and post-readout halving+dithering of all 15/20A circuits to trigger just before my midnight log rotation, and it's been running in this configuration all day. This has helped in flushing out at least some of the hidden phantoms, while giving better resolution on other small loads (e.g. 9.5W LED now alternating between 9 and 10W on successive lines) and has reduced the residual subpanel load not accounted for in branch circuit readings by about 40%, now hovering around 11-15W most of the time. So, definitely a useful improvement.

I might experiment next with quadrupling and dividing by four. Since an M-80 reading up to 80A normally uses range=4 too, hopefully 20A*4 (range=2) on an M-40 wont overdrive the GEM's analog input scale. Good to know there's no risk of breaking anything.

[jnh]

Since midnight I've been running all M-40 channels (and one M-50) quadrupled, then divided by four while tracking fractional Wh and adding ++1 every four ticks. This has been good for revealing more small phantom loads, and further reduced subpanel residual wattage, now averaging around 10W, but I can't help still obsessing over where exactly those unaccounted ~10W are going

This is the what I'm seeing in realtime data-- that final column is subpanel-residual, the subpanel main (column 6) minus everything from the first curly brace to the last, and remains fairly constant regardless of which loads are running. Here the solar PV inverter is supplying power to both panels and backfeeding the grid, hence negative figures in columns 1-3, but it's nearly the same at night - no channels on the subpanel are configured for net-metering.

Code: Select all

11:01:55 125.3 -2662 -2672 -2774 102  0  9 0 2 0 0 {68 0 0 0 0}  9  3  (1 11 0) 1 0 0 2 {0 0} 0 0 1 -2 7
11:01:56 125.2 -2654 -2664 -2770 106  2  9 0 1 0 0 {68 0 0 0 0}  8  3  (1 11 0) 0 0 0 2 {0 0} 0 0 0 -2 13
11:01:57 125.2 -2653 -2662 -2771 109  1  8 0 3 0 0 {74 0 0 0 1}  9  3  (1  8 0) 0 0 0 1 {0 0} 0 0 1 -4 12
11:01:58 125.3 -2647 -2656 -2772 116  0  9 0 2 1 0 {82 0 0 0 0}  9  3  (1  9 0) 0 0 0 3 {0 0} 0 0 0 -3 9
11:01:59 125.3 -2641 -2650 -2765 115  1  9 0 2 0 0 {82 0 0 0 0}  8  3  (0 10 0) 0 0 0 1 {1 0} 0 0 1 -4 10
11:02:00 125.3 -2643 -2652 -2768 116  1 10 0 2 0 0 {83 0 0 0 0}  9  3  (1 11 0) 0 0 0 2 {0 1} 0 0 0 -4 6
11:02:01 125.4 -2642 -2652 -2766 114  1  9 0 1 0 0 {81 0 0 0 1}  8  2  (1 10 0) 0 0 0 2 {0 0} 0 0 0 -1 9
11:02:02 125.5 -2640 -2651 -2767 116  0  8 0 2 0 0 {82 0 0 0 0}  8  3  (1  8 0) 1 0 0 2 {0 0} 0 0 0 1 11
11:02:03 125.6 -2639 -2649 -2765 116  2  9 0 2 0 0 {82 0 0 0 0}  9  3  (1 10 0) 0 0 0 2 {0 0} 0 0 0 -3 9

e.g. 106 - (68+8+3+1+11+2) = 13W for line two.

Second-to-last is residual load for the main panel, which does correlate slightly to solar production, dipping negative a bit when the inverter is active, but that one rarely exceeds 5W either way. The voltage (measured from a subpanel circuit, downstream from inverter) does normally runs about 2V high when the inverter is sourcing significant power, since it has to exceed the grid voltage to push power back to the utility, but I've confirmed the accuracy of that PT reading with a good Fluke meter.

[ben]

Since midnight I've been running all M-40 channels (and one M-50) quadrupled, then divided by four while tracking fractional Wh and adding ++1 every four ticks. This has been good for revealing more small phantom loads, and further reduced subpanel residual wattage, now averaging around 10W, but I can't help still obsessing over where exactly those unaccounted ~10W are going

This is the what I'm seeing in realtime data-- that final column is subpanel-residual, the subpanel main (column 6) minus everything from the first curly brace to the last, and remains fairly constant regardless of which loads are running. Here the solar PV inverter is supplying power to both panels and backfeeding the grid, hence negative figures in columns 1-3, but it's nearly the same at night - no channels on the subpanel are configured for net-metering.

Code: Select all

11:01:55 125.3 -2662 -2672 -2774 102  0  9 0 2 0 0 {68 0 0 0 0}  9  3  (1 11 0) 1 0 0 2 {0 0} 0 0 1 -2 7
11:01:56 125.2 -2654 -2664 -2770 106  2  9 0 1 0 0 {68 0 0 0 0}  8  3  (1 11 0) 0 0 0 2 {0 0} 0 0 0 -2 13
11:01:57 125.2 -2653 -2662 -2771 109  1  8 0 3 0 0 {74 0 0 0 1}  9  3  (1  8 0) 0 0 0 1 {0 0} 0 0 1 -4 12
11:01:58 125.3 -2647 -2656 -2772 116  0  9 0 2 1 0 {82 0 0 0 0}  9  3  (1  9 0) 0 0 0 3 {0 0} 0 0 0 -3 9
11:01:59 125.3 -2641 -2650 -2765 115  1  9 0 2 0 0 {82 0 0 0 0}  8  3  (0 10 0) 0 0 0 1 {1 0} 0 0 1 -4 10
11:02:00 125.3 -2643 -2652 -2768 116  1 10 0 2 0 0 {83 0 0 0 0}  9  3  (1 11 0) 0 0 0 2 {0 1} 0 0 0 -4 6
11:02:01 125.4 -2642 -2652 -2766 114  1  9 0 1 0 0 {81 0 0 0 1}  8  2  (1 10 0) 0 0 0 2 {0 0} 0 0 0 -1 9
11:02:02 125.5 -2640 -2651 -2767 116  0  8 0 2 0 0 {82 0 0 0 0}  8  3  (1  8 0) 1 0 0 2 {0 0} 0 0 0 1 11
11:02:03 125.6 -2639 -2649 -2765 116  2  9 0 2 0 0 {82 0 0 0 0}  9  3  (1 10 0) 0 0 0 2 {0 0} 0 0 0 -3 9

e.g. 106 - (68+8+3+1+11+2) = 13W for line two.

Second-to-last is residual load for the main panel, which does correlate slightly to solar production, dipping negative a bit when the inverter is active, but that one rarely exceeds 5W either way. The voltage (measured from a subpanel circuit, downstream from inverter) does normally runs about 2V high when the inverter is sourcing significant power, since it has to exceed the grid voltage to push power back to the utility, but I've confirmed the accuracy of that PT reading with a good Fluke meter.

What CTs are you using to monitor the sub-panel? Try adding a jumper (just a piece of wire) between 3,4 of that channel if Micros. That'll help reduce any noise on the channel.