False power readings with solar diverter?

Applications
1

Just started looking at power monitoring, having had 12kW of solar panels since 2015. We’ve had an Apollo Gem solar diverter from the start connected to 3kW immersion water heater. As a quick and easy intro to monitoring I am trying a Shelly EM device, planning in time to move on to emon system once I have more time, as it seems somewhat more involved to set up than the Shelly app, albeit a lot more comprehensive. But I am seeing something very odd that I imagine may well have also have come up with emon users?

What I seem to observe is that when the Apollo diverter is routing full 3kW to the immersion, or zero, the readings for power used by the house are dead accurate, but when the solar diverter is routing any other amount of power to the immersion, worst case scenario half power, so 1.5kW to the immersion, the power consumed by the home appears to be over reading.

It seems to me that, roughly speaking, the power monitor thinks the diverter is delivering full power to the immersion, when it’s absolutely not. The net effect is that it reports the house is importing power from the grid, when in fact, as confirmed by import/export reading on the Apollo diverter, and also at the electricity meter itself, there is zero import/export, which of course is the intended function of the diverter.

I wonder if (a) anyone on here has come across similar behaviour affecting emon power readings with a solar diverter and (b) if anyone knows for sure what the waveform output of the Apollo Gem diverter actually is and how this may distort the power reading on the energy monitor? Monitoring is via standard CT clamps, one on the solar AC output (seems dead accurate at all times) and one on the supply to the house, after the Henley block to which the solar is connected i.e. it’s just the house consumption. This is the one that’s reading inaccurately when the diverter is in action. Also of course the Apollo has its own CT clamp on the grid/meter input.

Any emon users come across this before? (for one thing I’d like to know if the emon is immune from this error?. And can anyone throw light on waveform output of the Apollo? I can’t find it documented anywhere.

Thanks for looking.

Max

2

Hello @Max2 and welcome!
What hardware do you have? and when did you buy it?

It sounds like you may be running our earlier discreet sampling firmwares? These don’t work very well with PV diverters. The newer continuous monitoring firmware should give much better results.

3

No OEM hardware yet.

I suspect it’s your diverter. The Apollo was one of the early ones.
Winding down the power into a 3kW resistive heater is not a trivial problem.
It’s probably using a triac to turn the 240V on/off very quickly. This causes all sorts of issues, with harmonics induced into the usual 50Hz sine wave of the mains.
Have you or your neighbours noticed the lights flickering at all?
These harmonics are probably messing up the Shellys power monitoring. At the full 3kW, the immersion is “full on”, at 0kWh it’s completely off. In between all bets are off.

4

How are the house lead readings, when the diverter is off?
If these errors are still there, then it could be a power factor issue.

On my own system (no diverter) I am seeing up to 10% error between the measured loads, and GTI outpur. The difference being the import.
The CT driving the GTI does not sense voltage, and so is not going to compensate for power factor. Whilst he monitoring is an emonTX4, which does have voltage phase sensing, so I can rely on the readings more.
My base load is mostly capacitive from SMPSUs.
When I start adding resistive loads, the errors reduce. My assumption is, that this is watering down the power factor errors.

5

Might he a bit obvious, but where is your shelly device measuring the immersion power? It definitely needs to be on the input side of the diverter to have any chance of reading accurately, even then subject to harmonics re other responses.

I previously had a shelly power monitor on the input to my cheap and cheerful Solic diverter and it worked fine as does my emonpi now.

Rachel

6

Many thanks for replies. Yes Lee, you are correct as per my post, no Emon yet. I was just after dipping my toe in the water by a quick and easy solution with the Shelly EM prior to going down the Emon route. I thought that given the fact that just about anyone with solar PV will almost certainly have a solar diverter of some description I thought this must have come up many times, but alas a search of the group does not see this issue mentioned as far as I could see.

I have two CTs into the Shelly. One on solar PV AC out and one on the feed to the dist board after the Henley block that the solar connects to. In other words CTdist minus CTsolar = Grid export/import.

The readings are very accurate with the diverter either fully off or fully on, but anywhere in between causes large errors. TrystanLea I am borrowing a 'scope so in next couple of days I will be able to see the exact output of the Apollo unit. I’m certainly led to think by the behaviour of the readings that, as you say, it’s very simply rapidly turning the AC on/off with differing mark/space ratio. My guess is the Shelly takes short samples then uses the highest sample to get the power reading. So from what you said about the latest Emon using continuous sampling then yes, it would seem that this would make Emon largely immune to this issue? Perhaps that’s why nothing much seen on here about it.

It’s not a small error. With solar delivering excess power, and diverter delivering about 2000W (from the Apollo display), to hold import/export at Zero (confirmed by revolving disk stationary on DNO mechanical meter) the Shelly says Grid is importing 770W, so a 770W over-reading of grid import.

Anyway, for those interested once I can get a scope reading and see the actual waveform out of the Apollo (and into it… my guess it will be very similar) then I will post it here.

But I would still be interested in anyone’s comments on whether this could be an issue with solar diverters generally, although it seems there are quite a few different methods employed to achieve the same result. Perhaps the more modern units don’t tend to use this “burst” mode (if that is indeed what it turns out to be…but seems likely). Perhaps they chop the AC more like a large light dimmer so it largely preserves a more continuous AC waveform? I guess this begs another question. Would large numbers of overhead spots with light dimmers in a system potentially cause inaccuracy in power readings?

But overall I think TrystanLea may well have answered my question to a large degree in the issue being sample-length related. I think I read somewhere that Shelley uses 1 second updated instantaneous samples so that could well be the issue is my guess?

Thanks again for all inputs.

7

What is its rated power?

8

I can’t speak for emon, which is not my thing, but here suggest a possible, for emon people to comment on. Suppose that your diverter ‘chops’ power into the 3kW immersion heater, or uses a big triac dimmer switch, or something like that. Are there ways for the chop timings to drift in and out of synch with emon sampling times ? For example, a worst possible, and we have no indication that it is this bad, suppose that your diverter buzzed at 600Hz when half-on, with no consideration for when in the mains cycle it should start every chop. Sometimes emon would sample inside an ‘on’ interval, and sometimes it would sample inside a ‘chopped’ millisecond and see ‘OFF’. If this were happening, you might hear a screech from the diverter box and from the immersion heater, and you would see :
sometimes 3kW
sometimes 0kW
sometimes in between
some minutes in which it rapidly hops amongst those
some seconds in which it seems settled for a bit.

To get from instantaneous readings, which are necessarily so with mains AC, to ‘1 second samples’ would need two differences :
many measurement samples inside 20 milliseconds mains cycle
averaging of those for 49 whole mains cycles; at least 950 measurement samples averaged for one reporting sample. With a 600Hz chop, even that would struggle and would see a little bit of unexplained variation.

A possible solution, which is probably not practical, “buy a better solar diverter”, is to replace rectify and chop to immersion heater, with rectify then chop to immersion heater via a generously sized inductor coil and with freewheel diodes inside the diverter box. I expect that everyone else on this site knows what a buck DC downconverter is and the acceptable types which work with emon.

9

Was this ever solved? I have exactly the same problem also with an Apollo Gem diverter. I too think it must be down to the sampling rate of the Shelley.

I am told by a reasonably reliable source that all well designed PV diverters don’t actually truncate the waveform, because that would cause unacceptable harmonics. What I am told they do is to switch on/off at the zero crossing points, adjusting the number of cycles they ‘allow through’ to match the load to available PV on average. This means that they do in fact alternately import and export more or less the difference between the full 3kW and the power available from the PV. The averaging is apparently done over <1 second, which mechanical meters ignore because of inertia, and smart meters are designed to ‘smooth out’ (rather than recording the import and export separately) as part of their spec.

If this explanation of how PV diverters work is correct, and it seems very plausible, then a mismatch between the Shelley sampling period and the cycle period for the diverter would cause exactly what we observe. It also means that ‘buying a better diverter’ may well not solve the problem (you may need a poorer one!)

10

I can only answer for the sort of PV diverters that I’ve read about, analysed or helped with the design of, but of course I can comment about the OEM sampling software.

Basically, there are three modes of operation for a PV diverter. All involve chopping the wave up, the main difference is in the rate at which this happens. These are all covered under “PV Diversion” in the Docs section, so I don’t need to repeat any of that here.

When it comes to measuring both the diverted power and the nett grid power, then the way the power is sampled is most important. As has been mentioned, emonLib samples for approximately 200 ms every 10 s. This is fine for loads that are switched on and off at intervals of minutes or hours, not good when the switching intervals get less than a few minutes - because clearly it must assume the load remains in the same state for up to 10 s, which may well be wrong. emonLib was the default software used in the emonTx V2 and in particular circumstances in the emonTx V3.
Generally, the emonTx V3 used emonLibCM - the ‘CM’ standing for Continuous Monitoring - and this samples every voltage and current channel ‘continuously’ - one set of samples comprising the voltage and 4 currents (in the case of the emonTx V3.4 with 4 channels in use) at a rate slightly faster than than 1900 sample sets per second. It calculates rms voltage, rms current, real & apparent power & power factor and reports those values averaged over the 10 s reporting period. So each quantity (V or I) is sampled 38 times during every mains cycle (ignoring those users in N. America, where it’s less due to their 60 Hz supply). EmonLibDB has significantly more work to do, and its sample rate is lower: at better than 1480 sample sets per second – using 3 + 12 single-phase channels of an emonTx4 @ 50 Hz, which equates to 29 or 24 samples per mains cycle at 50 Hz or 60 Hz respectively. And note, neither emonLibDCM or emonLibDB’s sampling system is synchronised to the supply frequency, though the reporting system is.

The result is, unless there’s little or no filtering on the diverter, emonLibCM, emonLibDB and the emonPi3 software in development won’t have a problem with accurately measuring both the diverted power and the nett grid power, both of which of course will have rapidly varying powers whilst the diverter is operating. (And if there is no filtering on the diverter, the network operation may well take an interest if any complaints are received, because the are rules relating to how much harmonic current can be injected into the public electricity supply network.)

It might be worth mentioning that some decades ago, when coal was mined in the UK, I worked for a firm that made large electrical machines. The projects I worked on were winding engines for coal mines, one particular one was a 2957 kW (4000 HP) d.c. motor. The word was, everyone in the village next to the mine could sit at home in the evening and count every load that came up the shaft because the lights dipped every time the winder accelerated, and brightened as it regenerated back into the supply to decelerate. From memory, the current to accelerate to top speed was about 7.5 kA d.c., which translated to about 200 A per phase at 11 kV.

11

Given the ADE ICs used in the Shellys I doubt it’s a sampling rate issue - those things typically sample at 8000 samples/sec.

I think that’s likely to be the cause of your discrepancy. AFAIK there is no spec on how quickly an energy meter should respond to a power reversal. From a revenue point of view I suspect energy providers would say “the faster the better”. I think the diverter folk refer to this as their “energy bucket” - basically how much slop the meter gives them before it starts counting in its Import or Export register. It’s possible your Shelly has a much smaller bucket than your revenue meter. When your Shelly reports that you’ve imported energy during diversion, does it also show an identical export of energy?

There some more detail here on what’s going on inside any meter (Shelly or revenue) trying to deal with this. That detail refers to the ADE7816 which is a very close cousin to the ADE ICs used inside the Shelly. Scroll down a few posts to see a scope trace of how quickly the IC detects a power reversal.

12

Interesting.

I am principally monitoring power not energy and of course I don’t get 8000 data points per second. It may be that the Shelley reports an instantaneous power periodically, but integrates the samples to get energy. I will take a look at the energy data and the references to see if I can figure that out.

13

There’s some more discussion on this thread