Avoiding being overcharged by your smart meter by using AC line filters.

[Karel]

Does the OP have any specific evidence (other than weblinks) that support his hypothesis?

Have a look at the attachment.

[Fflint]

Wow, I'm away one day and there is almost 25 replies arguing over a proven fact rather than the key question :how to filter distorted waveform to avoid the metering error

Power factor has nothing to do with it. Modern meters measure the power factor correctly. The problem is most likely(in the order of importance) with the type of the measuring device (hall sensor, vs Rogowski coil vs CT), sampling frequency and the averaging method.

The studies I was talking about were both linked in this thread. First the university of Twente one, which no doubt had some issues including one that they used an analog meter for reference. Then the other (Polish) study that concluded the Twente guys results are "questionable", but if you read this study it actually supports the Twente guys results corrected by - 5% (the guy that did the Polish study did use a proper certified lab meter and he found out the old mechanical meters have an error of - 5% for his tested waveforms). He also used waveforms with THD "in accordance with the local standard" and found a smart meter error around 1%. Surprise, surprise, he used a lot less distorted waveforms and he found out the error is less pronounced!

The key point of the Twente guys study is that they obtained the horrible waveforms by using normal loads people have in their homes. Led lights, CFLs and dimmers. Then he went on to conclude if one has one or two led lights running a resistive load like a kettle, an oven, or incandescent lights it will correct the waveform so the error is practically zero. When he did write that in 2018, perhaps people did have LEDs, and incandescent lights on the same circuit as ovens and kettles. Now I already mentioned I have 75 led lights (not counting high power external floodlights) in the house and almost no other linear loads. So at best the total error found by Twente guys should be revised lower not debunked.

Both studies came out in 2017/2018. No one make any contrary findings I know about. So it is a fact. Now we can argue about it in next pages of this thread or we can focus on finding a solution.

uer166 as you design and certify meters I would be very interested in if you could tell us what sampling frequency meters typically use. There is a way of making an accurate meter that has low error even with distorted waveforms, but the amount of allowable distortion is specified in local standards and that's what meter designers design meters for. If we ask that they design them better how do we specify the limit of allowable distortion!? One could even say meters are fine, it is the people that use shitty non linear devices that emit conducted EMI in levels that far exceed the standards.

And this is, to some extent my point. One can't complain to a government regulator that the metering device is inaccurate if one feeds it a waveform outside its design parameters.

Therefore I'm looking for filters.

There are various kinds of filters, more or less expensive as well as using isolation transformers etc. I'm very interested in finding out about anything that could help even with just individual loads.

As mentioned previously, I have central heating pumps that run 24/7 in winter("smart" inverter driven) . Then there are LED lights(75 internal, and around 10 high power external) . I could put those (most offending) loads on separate circuits and build a filter. That's what I'll do once I have a way to measure the waveforms accurately. Until then I'm asking for additional ideas.

[Siwastaja]

Wow, I'm away one day and there is almost 25 replies arguing over a proven fact

No! There is no proven fact! You clearly don't understand what "proven fact" means. All that has been shown is that at certain point in time, certain un-named meters showed wrong values in presence of very spiky current with dimmers. Which is interesting, and the paper was interesting; I have a scientist background myself and have published papers. Specifically, the 107Hz indication was mind-blowing, and clear indication that particular meter is misbehaving. But still, you are very likely misinterpreting the situation. Because people care about money. If it was really that wide-spread, it would have been widely noticed and discussed already.

And it's exactly the right thing to discuss! Existence of problem is the key fact in any problem-solving case; playing your X-Y problem games isn't.

Because solving a problem which shouldn't exist, and very very likely does not exist, in the first place, is wasted effort. If the problem exists, it should be fixed where it is; working around it is more expensive.

If the problem does exist, then authorities should be involved and the meters fixed.

It's a real possibility that some meter series (the paper specifically shows those based on Rogowski coil; and I have to add to that, Rogowski coil itself does not have such limitation, it's an issue in the signal processing chain) have this problem with just that specific dimmer load profile. But does your meter behave that way, and even if it does, do you have such load profile?

I guess you are completely on wrong track, and you can throw a tantrum of us not being helpful, but that only throws you further on the wrong track. You need to start by proving the existence of the problem, otherwise there is nothing to design filter for.

I agree with you that arguing about it is waste of time. So start your work quickly, but start from the right premise: measuring the problem you are going to solve. Demonstrating the problem should be easy to do: just turn on all the LEDs and CFLs and other supposedly problematic loads, and look at the meter. Compare to the expected consumption based on the bulb ratings - they are not accurate ratings, but enough to give you the ballpark. Can you do that?

[uer166]

Wow, I'm away one day and there is almost 25 replies arguing over a proven fact rather than the key question :how to filter distorted waveform to avoid the metering error

Yeah I don't agree here at all. The papers looked at some very specific and most likely hard to reproduce edge cases (including seemingly literally broken meters at Siwastaja pointed out). Just because nobody wasted time to debunk, or at least put some context in front of those papers, doesn't mean they're fact. And even if they were fact, it doesn't mean that they're applicable to your situation, 99.99% chance they are not. Do you have a 2011-era smart meter? Do you have the exact same kind of LEDs and dimmers that the paper used? Do you literally not have a single other load except for those?

To satisfy your curiosity: modern Sigma-Delta simultaneous sampling ADCs would run at somewhere around 2~10kHz Fsamp, with antialiasing filter cutoff set to maybe half or less of Fsamp. You could definitely still have some aliasing since it's not a whole decade or two below, but the resultant error would be small in absolute terms.  Even then you'd need the sampling frequency to somehow match mains exactly for this to happen and not average out, which in real life would be impossible over the long term. Fsamp is not usually synched to mains (but RMS calculation windows are to avoid spectral leakage).

The much bigger and more annoying issue is making CT-based meters that can tolerate having ~60A DC current put through them, and still measure to 1% accuracy or whatever. Another one is bad power factor (linear only: inductive or capacitive), since it requires tight matching of phase delays in filters, or accurate calibration of those delays. But we're talking about +-1% here that is required per MID standard, not the ridiculous numbers you saw in the papers you posted.

[uer166]

If we ask that they design them better how do we specify the limit of allowable distortion!? One could even say meters are fine, it is the people that use shitty non linear devices that emit conducted EMI in levels that far exceed the standards.

Everything has an envelope, and everything will break eventually. If your microwave has a hole in it and it irradiates the meter, do you expect it to meter accurately or work at all? Modern meters are extremely robust and can handle great abuse, including surges, external magnetic fields, horrible mains waveforms, static discharges, etc etc etc.


I recommend getting some current clamps, and diff probes, and measuring your mains waveform consumption, just for funsies. I understand that you can't drop $30k on a reference meter (which is what the authors of the paper should have done), but it doesn't mean you have to be as janky as them.

[rstofer]

so what are the PF limits? the meanwell supply on my 3D printer is 0.5 PF according to some cheap power meter, it draws 350W

In industry, we pay a penalty if the power factor at the end of the billing period is less than 0.8.  But it's only about $1000 on a bill that is in the multiple 10s of thousands of dollars.  On a big system, you can't do anything to correct pf for $1000/month.  Just smile and wave!

[bdunham7]

In industry, we pay a penalty if the power factor at the end of the billing period is less than 0.8.  But it's only about $1000 on a bill that is in the multiple 10s of thousands of dollars.  On a big system, you can't do anything to correct pf for $1000/month.  Just smile and wave!

For inductive motor loads, fixed capacitor banks that parallel the load are simple and quite effective even with variable loading because the PF of the motor improves with load.  Sometimes they are worth it just to reduce the voltage drop in the wiring.  There are all sorts of products just for this purpose.  I don't know what the payback period would be, but in my experience the demand (peak) charges were way more than any PF levies. 

https://www.eaton.com/us/en-us/products/low-voltage-power-distribution-control-systems/power-factor-correction-capacitors.html

[ejeffrey]

The thing is power factor seems to cover a multitude of things the 3 main ones I can think of are

Chopped waveform (dimmers)
extreme peaks ( "smoothing" capacitors)
just plain shift

Either way I don't see how the generators cope, if there is a high demand for current that the generator cannot take but is trying to generate then it's going to be a problem. Why else do industrial loads get charged by power factor as well?

Generators cope the same way anything else does.  The voltage is set by the magnetic flux and the current is whatever the load draws.  Industrial loads get charged for poor power factor for two reasons: the increased transmission losses and because the transmission equipment is mostly limited by current and thus apparent power.

Think of it this way.  If you have a 16 amp / 230 V breaker in your house and you have a load that draws 10 amps but only consumes 1 kW.   If you want to plug a second one in it will exceed the circuit current limit.  You can't plug both in without upgrading the circuit or adding a second one.  If the loads were power factor corrected they would draw under 5A each and could operate on the same circuit.  To the power company, your excess apparent power draw prevents them from selling power to your neighbor unless they upgrade the transmission lines and transformers.

[Fflint]

Wow, I'm away one day and there is almost 25 replies arguing over a proven fact

No! There is no proven fact! You clearly don't understand what "proven fact" means. All that has been shown is that at certain point in time, certain un-named meters showed wrong values in presence of very spiky current with dimmers. Which is interesting, and the paper was interesting; I have a scientist background myself and have published papers. Specifically, the 107Hz indication was mind-blowing, and clear indication that particular meter is misbehaving. But still, you are very likely misinterpreting the situation. Because people care about money. If it was really that wide-spread, it would have been widely noticed and discussed already.

You assume that, but if the error adds up to let's say 10% of your usage that is extremely difficult to prove. Also as I stated before meters are certified for maximum allowable THD and devices used in our homes are not.

The Twente team used 8 different smart meters. 3 - 3 phase and 5 single phase. It is true they didn't give out models (I wonder why not), but if you look you'll find a high resolution photo of their setup here https://www.researchgate.net/profile/Bas-Ten-Have/publication/337229022/figure/fig1/AS:970012051378189@1608280303753/Ten-static-meters-placed-in-series.jpg to hopefully recognise some meters if you want to.

And it's exactly the right thing to discuss! Existence of problem is the key fact in any problem-solving case; playing your X-Y problem games isn't.

Because solving a problem which shouldn't exist, and very very likely does not exist, in the first place, is wasted effort. If the problem exists, it should be fixed where it is; working around it is more expensive.

If the problem does exist, then authorities should be involved and the meters fixed.

But, as I said before, if the usage error is a result of your loads resulting in higher THD than the norm allows then whose problem is it really? The authorities that are supposed to ensure the meter is able to accurately measure waveforms that comply with a standard or people's that use devices that cause the waveforms that exceed the distortion specified?

It's a real possibility that some meter series (the paper specifically shows those based on Rogowski coil; and I have to add to that, Rogowski coil itself does not have such limitation, it's an issue in the signal processing chain) have this problem with just that specific dimmer load profile. But does your meter behave that way, and even if it does, do you have such load profile?

I guess you are completely on wrong track, and you can throw a tantrum of us not being helpful, but that only throws you further on the wrong track. You need to start by proving the existence of the problem, otherwise there is nothing to design filter for.

I can only believe what I read in a published scientific study. You are correct there are certain issues with the Twente study. Here you have a link to a much better Slovak study of the same thing done in 2021. https://link-springer-com.libproxy.viko.lt/article/10.1007/s00202-021-01365-8?error=cookies_not_supported&code=1af7722f-9d29-4015-9500-ccedc905263e titled "Contemporary electric energy meters testing under simulated nonsinusoidal field conditions". They did find errors as high as 37% in I quote "state of the art smart meters".

I agree with you that arguing about it is waste of time. So start your work quickly, but start from the right premise: measuring the problem you are going to solve. Demonstrating the problem should be easy to do: just turn on all the LEDs and CFLs and other supposedly problematic loads, and look at the meter. Compare to the expected consumption based on the bulb ratings - they are not accurate ratings, but enough to give you the ballpark. Can you do that?

Yes I can, and I did with my central heating pumps. The datasheet for the pump has a chart that shows power consumption at each setting (there are 6 settings, 3 constant pressure, and 3 constant flow). I set the pumps to the lowest setting. I switched off all the circuits in the breaker panel other than the pumps overnight and I took readings before I went to bed. Then in the morning. Instead of few tens of watts the load registered was close to 200W.

This is just one example.

There was time when I tracked this quite carefully, but eventually it required more effort than the money spent on the excess electricity was worth and I gave up. Then recently I found out all this research about meter errors with non linear loads. I'm also building a PV setup and electricity goes up in price by 50% next year in my country. So I became more actively interested in not being overcharged.

If we ask that they design them better how do we specify the limit of allowable distortion!? One could even say meters are fine, it is the people that use shitty non linear devices that emit conducted EMI in levels that far exceed the standards.

Everything has an envelope, and everything will break eventually. If your microwave has a hole in it and it irradiates the meter, do you expect it to meter accurately or work at all? Modern meters are extremely robust and can handle great abuse, including surges, external magnetic fields, horrible mains waveforms, static discharges, etc etc etc.


I recommend getting some current clamps, and diff probes, and measuring your mains waveform consumption, just for funsies. I understand that you can't drop $30k on a reference meter (which is what the authors of the paper should have done), but it doesn't mean you have to be as janky as them.

I don't even blame the meters (that much) if the loads truly generate such chopped waveforms.

When the CTs I ordered arrive beginning of Feb I'll record loads waveforms and I'll find out if what I'm seeing is due to distortion, or error at the lowest measurable current or something else.

For now I wanted to see if others are interested in the same thing. If others found problems and solutions I could make use of with smart meters.

Wow, I'm away one day and there is almost 25 replies arguing over a proven fact rather than the key question :how to filter distorted waveform to avoid the metering error

Yeah I don't agree here at all. The papers looked at some very specific and most likely hard to reproduce edge cases (including seemingly literally broken meters at Siwastaja pointed out). Just because nobody wasted time to debunk, or at least put some context in front of those papers, doesn't mean they're fact. And even if they were fact, it doesn't mean that they're applicable to your situation, 99.99% chance they are not. Do you have a 2011-era smart meter? Do you have the exact same kind of LEDs and dimmers that the paper used? Do you literally not have a single other load except for those?

OK, fine (about old meters). How about the 2021 study I linked above?

As for my smart meter, is it a 2011 era meter? Yes, actually it has been about 10 years since I bought the land and I had power installed. So the meter quite literally is a 2010/2011 model(Apator 16ec3r). There are lots of people with such "old" and much older meters. When I was renting a flat in the city (5~6 years ago in UK) it had a mechanical meter that looked like it could've been 30 years old. People don't upgrade their meters unless there is a reason for it. However, my meter will be upgraded once I finish my PV setup as the power company requires a different model for bi-directional billing (despite my model supporting that).

As for my loads I did describe them pretty well in this thread. I literally have two linear loads in my house. One is a well water pump,the other is a flow water heater. They see 20 min of usage per day if that much. Everything else is literally LEDs (most of the time 1 of the 3 phases has only LEDs, the other 2 have a mix of LEDs and switched power supplies). In near future I'll be adding a transformerless PV inverter too.

To satisfy your curiosity: modern Sigma-Delta simultaneous sampling ADCs would run at somewhere around 2~10kHz Fsamp, with antialiasing filter cutoff set to maybe half or less of Fsamp. You could definitely still have some aliasing since it's not a whole decade or two below, but the resultant error would be small in absolute terms.  Even then you'd need the sampling frequency to somehow match mains exactly for this to happen and not average out, which in real life would be impossible over the long term. Fsamp is not usually synched to mains (but RMS calculation windows are to avoid spectral leakage).

The much bigger and more annoying issue is making CT-based meters that can tolerate having ~60A DC current put through them, and still measure to 1% accuracy or whatever. Another one is bad power factor (linear only: inductive or capacitive), since it requires tight matching of phase delays in filters, or accurate calibration of those delays. But we're talking about +-1% here that is required per MID standard, not the ridiculous numbers you saw in the papers you posted.

It is interesting. I definitely would prefer a meter with Fsamp closer to 10kHz than 2kHz. Unfortunately, one has no choice in the matter. If it measures pure sinusoidal waveform with a resistive load correctly it meets the norm and that's it. You call the 500% results "ridiculous". No one is expecting errors of this magnitude in real life. 10% is a lot if it is across a year of use. The other issue you mentioned might very well compound with the first. If I'm reading the documentation of my meter correctly the lowest current it can measure is 250mA. That's 60W load at 240V. What if I had 3 30W loads each on its own phase? (lets assume linear for simplicity's sake) Will it register them as 3*60W continuous? Or will it miss them? Either way it is quite an error over multiple days and months. If I had a 3 phase PV setup and in winter during overcast sky it would produce 150W split across 3 phases would my meter show 0? Perhaps. And I haven't even started talking about non linear loads of the kind discussed in the research papers linked.

[Siwastaja]

It is true they didn't give out models (I wonder why not),

Because they do not trust their own paper, and/or their higher-ups in the university management do not trust it, because they fear lawsuits they can't win, and they don't also truly believe the matter is important. On the other hand, publishing a paper is a way to get funding for the department, so why take the risk? This is how research really works: come up with something interesting, don't waste too much effort on it (your personal goal might be publishing 10-20 papers every year), make it good enough so it passes the journal standards.

The fear against publishing product names is a tell-tale sign of mediocre science. It's everywhere. I have seen lot worse, but also lot better.

Another sign is the "root cause" section which doesn't talk about root causes at all, just briefly mentions the existence of Rogowski coil. Root cause analysis is hard and tedious. Likely the team doesn't have the suitable experts. If this was a plane crash killing 300 people, root cause analysis would look quite different.

Third is, they don't have a reference at all. The "reference" is just an arbitrary type of commercial meter, not proven to give correct results at all. This is understandable, because a real reference unit would be expensive, and proving its reliability yet more work.

And don't get me wrong, I'm not saying this is bullshit or fake science. I appreciate the work, but you have to understand the context. This is just everyday reality how science works. The sole idea of the publication is to trig interest in the subject, to make someone with more experienced scientists, much more resources, and better trust in their own science, to repeat the measurements - and provide the brands and models of the tested products.

It's common to hide the product names, but it's not some kind of standard that is always followed. I have seen papers regarding commercial off-the-shelf li-ion cells which identify the exact brands and products.

I remember a scandal here maybe a decade ago when a university published a study that all Vitamin D supplement brands except one have significantly too little Vitamin D in them. In the end, it turned out their measurement was just completely flawed, and the products were fine - except the one which had too much.

It's considered proven fact only after it has been independently repeated, verified and discussed.

[Siwastaja]

Yes I can, and I did with my central heating pumps. The datasheet for the pump has a chart that shows power consumption at each setting (there are 6 settings, 3 constant pressure, and 3 constant flow). I set the pumps to the lowest setting. I switched off all the circuits in the breaker panel other than the pumps overnight and I took readings before I went to bed. Then in the morning. Instead of few tens of watts the load registered was close to 200W.

OK, this is indeed interesting.

1) You are 100% sure there are no other loads? Did you let your refrigerator and freezer warm up during night? Can you repeat the same with the circulation pump turned off as well, to verify zero reading? And I mean, turned off at the pump, not at the breaker - sometimes house wiring contains surprises like undocumented branches.

2) What does it mean to have "load registered to 200W"? Is this load number reported by the meter in Watts? Or did you calculate it by (morning_kwh - evening_kwh) / hours_between_readings? The latter should give you the correct result.

[Simon]

The thing is power factor seems to cover a multitude of things the 3 main ones I can think of are

Chopped waveform (dimmers)
extreme peaks ( "smoothing" capacitors)
just plain shift

Either way I don't see how the generators cope, if there is a high demand for current that the generator cannot take but is trying to generate then it's going to be a problem. Why else do industrial loads get charged by power factor as well?

Generators cope the same way anything else does.  The voltage is set by the magnetic flux and the current is whatever the load draws.  Industrial loads get charged for poor power factor for two reasons: the increased transmission losses and because the transmission equipment is mostly limited by current and thus apparent power.

Think of it this way.  If you have a 16 amp / 230 V breaker in your house and you have a load that draws 10 amps but only consumes 1 kW.   If you want to plug a second one in it will exceed the circuit current limit.  You can't plug both in without upgrading the circuit or adding a second one.  If the loads were power factor corrected they would draw under 5A each and could operate on the same circuit.  To the power company, your excess apparent power draw prevents them from selling power to your neighbor unless they upgrade the transmission lines and transformers.

It's worse than that. Take for example a load that draws power  50% of the time. To achieve the same average power draw the wiring sees twice the load, this is because you now pull twice the current 50% of the time. now I assume that so you had I^2 losses, now you have (2I)^2 losses for half the time. So lets say that the average power is conducted un 1/3 cycle pulses, now you have 3 times the loss. Now think of things running off rectifiers, when that smoothing capacitor charges, how long does it charge for? less than 1/10 of the cycle? whoops

I don't know what this does to a circuit breaker, I don't know if their operation is thermal but obviously like the rest of the system they take a pasting.

I assume such extraordinary peaking on the line will drag the voltage down at the peak.

[T3sl4co1l]

Well yes, the circuit breaker just sees whatever RMS current is flowing.

If the load draws power at 50% duty, the RMS is sqrt(1/50%) ~= 1.4 times higher than the same power draw at PF=1.  So the PF is 0.7 in this case.

Mind, PF is usually averaged over some cycles, not a long period, so we wouldn't normally claim that a resistive heater has a terrible power factor!  If we allow such a definition for a moment, however, this does indeed work out the same way: the heater might be doing say 750W average, but it's doing it in gulps of 1.5kW or whatever, and so has a PF of 71%.

Anyway, when this occurs within a cycle, as with SCR phase controllers or cap-input rectifiers, the traditional measure of PF will indeed show the same difference.  Mind that it's weighted by the power available at any given instant: PF suffers very little indeed if you remove the flanks near zero crossing, but suffers greatly if you chop off some of the peak.

And yes, most power is a bit flat-topped for this reason.


Digression:

A breaker takes some time to respond thermally (minutes to seconds, depending on overload).  That response time actually serves as the averaging period over which we measure power factor, for purposes of breaker rating.  We could have a 30A heater doing an every-other-cycle modulation pattern, or even on and off every second or so, and still operate apparently safely into a breaker nominally rated for 20A.

This isn't problematic for safety purposes, AFAIK; in-wall wiring heats up in exactly the same way as the breaker does.  Albeit at different rates.  It could be that some connections heat up much faster (say those shove-insert / spring blade connections some receptacles have -- varies by locality!), and thus reach higher peak temperatures, experience more oxidation and cycling, and eventually open up and melt and arc.  So, that conclusion does depend on nothing being more sensitive, over shorter time scales, than the breaker.  Which maybe isn't a great bet, but should be an acceptable assumption with approved wiring.

AFAIK, instruments measure PF per cycle, or over a window of some cycles.  You'd get a better representation of behavior like this, from the peak/average consumption, or crest factor or the like, on a PQA.

Tim

[uer166]

AFAIK, instruments measure PF per cycle, or over a window of some cycles.

Yeah it's actually the exact same window that RMS values are calculated from, since PF = Power/(RMS(I)*RMS(V)). The minimum required is one full AC cycle, any less and the RMS value for that signal does not exist. I've seen it done over a window of one second (50 or 60 cycles) that is synched to mains.

The energy is of course accumulated at the ADC sampling rate, and within a single AC cycle you can have both positive and negative energy flow into the accumulator when you have an inductive or capacitive load. This presents interesting challenges when meters implement separate "imported" and "exported" energy, since there is again a fundamental minimum of one AC cycle to determine the net energy flow direction. If you have a single "net" accumulator then no special handling required.

Other cool thing is some meters operate entirely in the frequency domain: you take the FFT of all voltage/current harmonics over some window, and calculate power with some math kung-fu by multiplying voltage and current harmonics. I really don't like this since it's not intuitive compared to a simple integral of I*V samples, but it is cool. Anyway, this has nothing to do with OP's problem but I find this stuff exciting so 

[uer166]

Okay so re: some notes for that second longer paper (Contemporary electric energy meters testing under simulated nonsinusoidal field conditions)
 

The results presented in [17] suggest that digital meters are less subject to measurement errors as long as the signals are still compatible with the sampling process. However, the induction-type electricity metering devices are highly sensitive to the distortions in the supply voltage.

I can totally believe the above, in fact, I don't see how modern digital meters can be any less accurate than the old ones, in general (especially with nonlinear loads).

And:

Two meters
(in a total of six devices) presented errors greater than their respective accuracy classes. However, given that under nonsinusoidal conditions the MUTs’ error margins can be extended (according to the standards, as shown in Sect. 4.1), only one sample demonstrated the errors, which exceed its tolerance limits. All the deviations from the readings of the reference device were positive, the meters over registered.

The absolute worst error above was ~+8%, when the vast majority of meters was much less than that. The Class 2 meters are allowed up to 6% error in those conditions, so TBH it's not too far off, and not worth losing your sleep over that extra 2%.

Other random points I gathered were:

• They injected 10kHz current waveform as a load. This is well beyond the BW of any 60Hz meter, so 
• The quoted >60% error was for reactive power, not real power. Residential users don't get billed for reactive power. Moreover, in certain condition, you cannot calculate reactive power accurately when using CTs for current measurement. This is a fundamental math problem (CTs can't measure even harmonic currents), not a accuracy/resolution/whatever problem.

All in all, that second paper, which is substantially more thorough and believable, would lead us to believe that this is a non-issue, even with special "engineered" waveforms designed to defeat the meters. They still use a "meh" reference meter, but it is a massive step-up from a mechanical one.

[Fflint]

Siwastaja, I don't disagree with you regarding the Twente study. However it served its purpose it highlighted the problem and other teams looked at the issue too.

Yes I can, and I did with my central heating pumps. The datasheet for the pump has a chart that shows power consumption at each setting (there are 6 settings, 3 constant pressure, and 3 constant flow). I set the pumps to the lowest setting. I switched off all the circuits in the breaker panel other than the pumps overnight and I took readings before I went to bed. Then in the morning. Instead of few tens of watts the load registered was close to 200W.

OK, this is indeed interesting.

1) You are 100% sure there are no other loads? Did you let your refrigerator and freezer warm up during night? Can you repeat the same with the circulation pump turned off as well, to verify zero reading? And I mean, turned off at the pump, not at the breaker - sometimes house wiring contains surprises like undocumented branches.

I'm as sure as one can be. I guess 99%?. My fridge/freezer is fine without electricity if it isn't opened for 7h or so. It is a new build house I was involved with building every step of the way. I didn't watch the electrician put in the wiring, but I have the schematic. I will be 100% sure when I have the CTs arrive and I measure the waveform right in front of the meter and compare with the readings.

2) What does it mean to have "load registered to 200W"? Is this load number reported by the meter in Watts? Or did you calculate it by (morning_kwh - evening_kwh) / hours_between_readings? The latter should give you the correct result.

I compared the readings before I went to bed, and then first thing in the morning (I even took pictures back then). It shows usage to 1/10th of a kWh.

Also with this meter it is possible to "read" current usage by measuring the time between the "ticks". It blinks the led every so many watt-hours. I can't currently remember how often (this was over a year ago). I spent lots of time doing that, switching off/on various citcuits, remeasuring etc.

Okay so re: some notes for that second longer paper (Contemporary electric energy meters testing under simulated nonsinusoidal field conditions)
 

The results presented in [17] suggest that digital meters are less subject to measurement errors as long as the signals are still compatible with the sampling process. However, the induction-type electricity metering devices are highly sensitive to the distortions in the supply voltage.

I can totally believe the above, in fact, I don't see how modern digital meters can be any less accurate than the old ones, in general (especially with nonlinear loads).

And:

Two meters
(in a total of six devices) presented errors greater than their respective accuracy classes. However, given that under nonsinusoidal conditions the MUTs’ error margins can be extended (according to the standards, as shown in Sect. 4.1), only one sample demonstrated the errors, which exceed its tolerance limits. All the deviations from the readings of the reference device were positive, the meters over registered.

The absolute worst error above was ~+8%, when the vast majority of meters was much less than that. The Class 2 meters are allowed up to 6% error in those conditions, so TBH it's not too far off, and not worth losing your sleep over that extra 2%.

Other random points I gathered were:

• They injected 10kHz current waveform as a load. This is well beyond the BW of any 60Hz meter, so 
• The quoted >60% error was for reactive power, not real power. Residential users don't get billed for reactive power. Moreover, in certain condition, you cannot calculate reactive power accurately when using CTs for current measurement. This is a fundamental math problem (CTs can't measure even harmonic currents), not a accuracy/resolution/whatever problem.

All in all, that second paper, which is substantially more thorough and believable, would lead us to believe that this is a non-issue, even with special "engineered" waveforms designed to defeat the meters. They still use a "meh" reference meter, but it is a massive step-up from a mechanical one.

The point is not that modern meters are less precise than old ones, but that with high use of non linear loads creating choppy wavorms there is a significant error. You mentioned the highest error was 8% while 6% is allowable. Firstly, I didn't realise 6% error is allowable. If it is, that is less than ideal.

However, the purpose of the study wasn't to quantify actual error possible in real life with realloads. The goal in my opinion was to show that with non sinusoidal waveforms the meters show not insignificant errors.

Now another team should measure waveform that happen with actual use of dozens of LEDs and lots of other "modern" loads. Then see how meters fare with those waveforms.

Regarding your critique that injecting non sinusoidal waveforms with 10kHz frequencies into the ac line is unfair to a smart meter than usually handles 60Hz (50Hz here). In my opinion it is not a matter of fairness, but wmthe studies are bringing us closer to understanding if we are being charged correctly for our usage and generation. I can think of many loads that generate harmonics in tens of KHz in current as well as voltage (not at the same time of course, it is usually one or the other depending on what kind of load is it).

Also If my meter had an error of 8% (overcharging on usage and again under measuring on PV generation) I could loose 16% total if my PVs cover 100% of my usage. That in monetary terms is enough to think about what one can do to fix that on one's end.

I only use the 8% as an example. Currently what the true error is, is anyone's guess.

I think, as the cost of electricity becomes even more bonkers considerations like this will become more and more popular. Perhaps I should ping this thread in 2 years time. I bet there will be more interest in filtering etc. Perhaps there will be more research by then. Also, as time goes by more and more people will have houses that contain only non linear loads that try to limit their power use by pulling power in very short time steps (possibly at or even below the current detection limit). Then they will get readings that make no sense.

This is just an anecdote, but imagine my surprise when there were 2 months during which I had to use an electric resistive heater. I set up a simple power measuring device (plug with WiFi) not to be surprised by the bill. I used up around 30% more electricity during those two months. Then once the bill arrived I paid less than usual for the time period(my bills are bi-monthly) Not by much, just few percent, but still this was bonkers. (The readings are not estimated, a guy from the electric company comes every month and physically reads the meter). Also I work from home, my usage is very consistent and predictable beyond the expected seasonal changes.

I'm mentioning the above not as proof, but just to demonstrate the kind of inconsistencies that do happen.

[Siwastaja]

OK, you might be into something.

This would make an interesting class action lawsuit. IMHO, meter manufacturers cannot hide behind the claim that some types of loads are "not within specifications", if these loads pass the regulations i.e. are legal, as the intended purpose of an electricity meter is to measure electricity used by such legal loads.

If I had such case, I would just add an illegal outlet before the meter. Yes, stealing, yes, illegal; exactly like installing a meter that meters incorrectly. If you made sure you "steal" exactly the same amount the meter miscalculates (by installing a good reference meter), it would make a pretty interesting lawsuit because you could logically argue you are not stealing!

A more serious note, just file a complaint and request a new meter to be installed. See what they have to say. It's possible this is a known issue with a specific brand of meter and they just replace them whenever someone complains.

[Kleinstein]

The load they used in the first study was actually beyound the limits for a legal load (too much power for phase for phase angle modulation).
AFAIR some of the meters that had the most issues were old designs, no longer in use with some known issues with extreme loads.

The newer, more detailed study showed lower errors and only 1 meter 2 % higher than the spec limit (6% allowed vs actually 8%). I am nut sure if this includes to possible error of there reference meter. With most tested meters reading high, there is a chance that there refence meter was reading low. So part of the error, if not most could be from there ref. meter. This same problem applies to the old study. It could very well be that the old analog meter was reading low - AFAIK the limits there are much wider.

The relatively large error was for a pure high distortion load, which is not very common. The regulation are so that the larger loads (like the PC, TV, electric kettle, microwave, and even LED lamps) need to be more well behaved.The classic incandecent lamp with phase angle control in the first study is hardly used any more.

So you may get up too 6% over charged on a small part (maybe 5% usually less) of the consumption, but not on the main part.
It is also that this it the worst case error. Some meters will also charge to little.

[thm_w]

The reference meter ENA330 from Elcom: https://fei1.vsb.cz/kat410/studium/studijni_materialy/pmd/Spolecne/Ena330_manual.pdf

ENA330 was chosen as a reference device for measuring the values of powers (i.e., active, reactive, apparent), electric energies, and network parameters. It is a modular PQ measurement instrument with capability for measurement and analysis of electrical power network parameters and quality according to valid international standards IEC 61000-4-7, IEC 61000-4-15, EN 50,160, and IEC61000-4-30. The ENA analyzer fully complies with the IEC 61000-4-30 standard, class A. The measurement accuracy for both voltage and current inputs is ± 0.1%, while the current clamps i200 can drop the accuracy of current estimation by ≤ 0.5%.

They claim the i200 is accurate within 0.5% when the actual spec is 1% + 0.5A: https://www.fluke.com/en-ca/product/accessories/current-clamps/fluke-i200
i200 has bandwidth of 10kHz. But ENA330 bandwidth is only 45Hz to 2.5kHz right? Is it valid to compare 10kHz injected signals?

Attached is another paper where they use the ENA330 and measured a few percent low compared to their reference meter (Schrack LZQJ-XC).

[uer166]

Is it valid to compare 10kHz injected signals?

I don't think so. Unless the reference and DUT meters have the exact same frequency response, any signals that are outside of either one of the meters' BW are not valid for comparison. I really don't get the point of this experiment since real energy at that high of a frequency is minute anyway. And of course, as a customer you're actually ahead here since you get it for free.

There's been a lot of talk with very little data, it would be cool to get to the bottom of the issue and see who is wrong and how. Papers are cool and all but they have absolutely nothing to do with OP's situation.

[Fflint]

OK, you might be into something.

This would make an interesting class action lawsuit. IMHO, meter manufacturers cannot hide behind the claim that some types of loads are "not within specifications", if these loads pass the regulations i.e. are legal, as the intended purpose of an electricity meter is to measure electricity used by such legal loads.

If I had such case, I would just add an illegal outlet before the meter. Yes, stealing, yes, illegal; exactly like installing a meter that meters incorrectly. If you made sure you "steal" exactly the same amount the meter miscalculates (by installing a good reference meter), it would make a pretty interesting lawsuit because you could logically argue you are not stealing!

A more serious note, just file a complaint and request a new meter to be installed. See what they have to say. It's possible this is a known issue with a specific brand of meter and they just replace them whenever someone complains.

They have a complaint procedure that involves taking the meter away and recertifying (with a nice resistive load on a pure sinusoidal waveform Im told). If they find issues. They don't tell you the magnitude of the error. They just correct it or they install a new meter and that's it. You may have been overpaying for years, but you'll get no money back. However, if they find the meter is "fine" they charge you for the labour of recertification. So it makes much more sense to get to the bottom of it on one's own first

I'm in a middle of building a PV setup and as a part of registering it they will swap the meter for a different model so if this is indeed a dodgy unit it will help. However, if this is distortion etc. the new meter may be better or worse.

My CTs should arrive in few weeks. It is very likely I'll have plenty of time before they swap the old meter to make measurements.

Is it valid to compare 10kHz injected signals?

I don't think so. Unless the reference and DUT meters have the exact same frequency response, any signals that are outside of either one of the meters' BW are not valid for comparison. I really don't get the point of this experiment since real energy at that high of a frequency is minute anyway. And of course, as a customer you're actually ahead here since you get it for free.

Do you know a youtube that has a channel titled "Mr Carlson's Lab"? He has just uploaded a video of repairing a coffee machine with a SMPS. That coffee machine is broken, but even before it broke it emitted huge amount of (conducted) EMI in the range of hundreds of kHz. The owner noticed it, because he happens to repair old radios. He also found out it is not a simple case of a circuit emitting EMI into surrounding space, but that it uses the AC feed as a huge antenna (he checked this by plugging it via an isolation transformer with an rf shunt).

So here you have an example of one device emitting signals in hundreds of Khz range into AC.

All I'm saying is that the 10kHz test wasn't supposed to be an approximation of typical usage, but all sorts of scenarios happen. Sometimes for months on end.

There's been a lot of talk with very little data, it would be cool to get to the bottom of the issue and see who is wrong and how. Papers are cool and all but they have absolutely nothing to do with OP's situation.

Well, there is no data, because the title of the thread isn't "I think my power meter lies". I was attempting to make this not about my particular situation, but a whole class of problems and solutions. You're not convinced that it is indeed a problem. That's fine. Maybe that changes in future when more studies come out, or data is captured by amateurs.

Either way the subject of filtering AC distortion is an interesting one.

[richard.cs]

CTs can't measure even harmonic currents

CTs can't measure DC, so a half-wave load (which contains both DC and even harmonics) would measure incorrectly, but that does not mean the even harmonics cannot be measured. 100 Hz, 200 Hz, 300 Hz, etc. (120, 240, 360 for the 60 Hz world) pass through a CT just fine.

[Kleinstein]

The new smart meter circuits in principle have the ability to check the power quality and detect cases with excessive crap on the voltage (e.g. bad inverter in the neighborhood) or extreme distorted current  (e.g. poor consumers or bad PV installation) or brown outs with dropping supply voltage (e.g. overloaded transformer, thin lines). It is a shame that they not at least show an indication of the power quality.

I don't think the chance of getting significantly overcharged is high. The tests with the low load with highly distored wavefrom may be off by a significant percentage, but this is still only low power and thus little money. In the sum the current waveform is usually still pretty good and the performance in the extreme cases is not that relevant.
The extreme case would mainly apply with a broken, non compliant PV installation, possibly with a neighbor. So a shame the meter would not warn about this. Still this would be a rate case.

The lower the discrepancy found, the less attentions the studies get - more like the worst one get the most publicity.
There are cases when a new smart meter reads higher than an old style mechanical, but this is likely with an error with the old one.

[T3sl4co1l]

CTs can't measure even harmonic currents

CTs can't measure DC, so a half-wave load (which contains both DC and even harmonics) would measure incorrectly, but that does not mean the even harmonics cannot be measured. 100 Hz, 200 Hz, 300 Hz, etc. (120, 240, 360 for the 60 Hz world) pass through a CT just fine.

Not completely.  If the current goes to zero inbetween, and the L/R time constant of the CT + DCR + burden resistor is adequate, it will read just fine.  This normally requires a diode though (to get the "off" direction L/R much shorter than "on"), so is probably a special case.

Other than that, some DC can be tolerated, in which case the AC waveform and Vpp will be correct, but the baseline obviously won't.  Not sure offhand what saturation current mode mains-frequency CTs have -- it could be as low as some 10s of mA, making this quite an important discrepancy, given most loads are much more than that!

Tim

[Kleinstein]

The current transformers should be resonable tolerant to a DC part. With some loads a DC component can happen. A relatively common example are hair dryers on the lower power setting that use a diode to reduce the power to half. AFAIK for hand held use this is allowed in many regions.

I would not expect the CT to saturate under some 100mA, as few materials get saturation under 1 A/m and 10 cm magentic path is still relative small. So it would be hard to make the CT saturate so early. Chances are they would prepare for some DC and intentionally use a material with current needed for saturation / leaving the good range.