Random solar drop-outs - real or a Sense error?

[erik]

I’ve had a whole bunch of random 1-2 minute dropouts of solar power reflected in the Sense app. Since SolarEdge monitoring is only provided in 15-minute increments, I don’t have any way (that I know of) to confirm that the solar power is actually dropping to zero, or if Sense is having a problem. I found a topic from well over a year ago from someone experiencing something similar, but it had no responses.

I’ve been bugging my solar provider to troubleshoot this from an electrical installation standpoint, but they haven’t been able to find a problem (my contact there has been responsive, but I’ve gotten no feedback from their tech support team). I don’t know if I should keep pressing them, or if I’m just going to look like a fool when they point out that Sense isn’t working right.

When this problem first occurred, the drops were consistently aligning with spikes in consumption as my clothes dryer cycled on (every solar drop was with a dryer spike, but there were also plenty of dryer spikes with no drop in solar). Since then, as I’ve periodically looked at the app, I’ve seen drop-outs that were not aligned with any consumption.

I can say that the problem always seems to occur during peak production times. Right now, that’s around 9kW, thanks I guess to the very high temperature outside. On a cooler day, it’s around 11.2kW.

If this is a known problem with Sense, does anyone know why it might be happening or what I might be able to do about it? If this could feasibly be caused by my actual hardware (Silfab panels feeding into a SolarEdge inverer), I’d love some ideas on how to isolate the problem to either Sense or the actual solar installation.

[ixu]

Guessing @kevin1 might have a solve for you but my general take is that any solar dropouts are going to be very interesting to the Sense team and contacting support should get you some quick action.

Solar dropouts = Emergency status!

A thought for the team: Where is the alert for @erik?

This is potentially something that should be bumped to the top of the support list.

Edit: Open circuit degradation is something you really need to stay on top of …

Solar cell degradation under open circuit condition in out-doors-in desert...

[kevin1]

I have had Sense dropouts in the past, but when it has happened, it has flatlined both my solar and total usage data. I’m betting that this is a legit measurement issue, not a monitor logjam / communications issue. Two thoughts:

• Watch the power meter for the next drop-off (I don’t think there is a way to set an alert for Solar). Run outside to your SolarEdge inverter. If it’s like mine, is has a tiny display that shows the actual instantaneous production (DC and AC voltage, current and power). Might be tough to catch since it seems to only last (suspiciously) for 1 minute.

• If you don’t have many clouds, you might be able to chart 15 min production vs. theoretical max for that 15 min period to see if you can spot 1-2 minute losses.

[erik]

@kevin1
The power meter is a net meter that displays an integer number of kWh used. It goes up when I’ve used 1kWh more than I’ve generated, and it goes down when I’ve generated 1kWh more than I’ve used. Unfortunately, that means that it’s not perceptive to minute-long dropouts in solar production. And as you said, it would be very difficult to catch. If it was perceptive to short drops, I’d be tempted to set up a camera to monitor it continuously, so that I could go back and check the footage at the time of a dropout.

The analytical approach, analyzing SolarEdge’s 15-min production data to find 1-ish minute drops would be challenging, because the natural fluctuation in the curve is routinely in the 0.1kW range. If I could catch the problem on a cooler day, when the production is saturated (11,288-11,289W, per Sense) for at least an hour (I don’t think that’s physically possible with my configuration. I’ve seen saturation in the Sense graph, but never in SolarEdge’s 15-min resolution daily graph. I just looked back through mid-May to confirm), and has drop-outs, I’d have more confidence in my analysis. Right now, though, I don’t think I can distinguish natural variation from a few minutes of dropped measurements. I will try some comparisons in Excel, though, to see if the Sense and SolarEdge measurement differences change during periods with dropouts. The first step will be to figure out if the SolarEdge measurement for 12:30 is the 15 minutes ending at, centered on, or beginning at 12:30. There’s enough of a delay in the SolarEdge chart that I can’t be sure which it is.

To @ixu 's point, Sense does not alert me when this happens. I only see it when I happen to scroll through the timeline and see it. I don’t know how often this happens. I can see that it happened once today, once yesterday, and not at all on Monday.

FWIW, here’s a drop yesterday (strangely, along with an actual data drop, just a few minutes earlier). In this case, the drop does correspond to a step change in power consumption:

However, when I first discovered the issue in May, it was happening much more frequently. Here’s the first screenshot (from my tablet) that I sent to my solar provider. You can see that all of the drops except for the last one coincide with a large step increase in consumption (A/C compressor and clothes dryer cycle):

[kevin1]

@erik,

I guess you don’t have a “meter display” on the front of your inverter like I do. It gives second-by-second solar production DC Voltage (371.3V) and output production AC voltage (245.5V) and power (3.1kW)

.

[ixu]

@erik

I’d be tempted to set up a camera to monitor it continuously, so that I could go back and check the footage at the time of a dropout.

I love it when one thinks in that direction. Sometimes it’s the easiest method! I’ve done that with a webcam on timelapse in many situations: water leaks; temperature; contractors mixing concrete.

I tend to always want redundancy in systems … dual-Senses would help here. As would built-in solar alerts.

Feels like support needs to see your data.

[erik]

@kevin1

No such luck. I’d go snap a photo, but the sky just opened up and we’re under a flash flood warning. So instead, here’s part of the photo the installer asked me to take for their marketing (before I bought the Sense). The inverter is just a plain beige box, and it goes through not one, but two junction boxes (not sure why) before connecting to the main power.

[erik]

Ah darn it… So much for doing some comparative analysis in Excel. While Sense displays data in 1-second increments, it’ll only EXPORT data in day or hour increments. So I’d be looking for 1-ish minute signals in 15-minute and 60-minute samples.

[kevin1]

I tried a quick experiment to see whether my aggregated 15min SolarEdge outputs, for the past few days, aggregated into hours would be close enough to the Sense data to spot a 1-2 min dropout.

First off, I needed to offset my SolarEdge data forward by 15min (and divide by 4) before aggregating into hours, to match up with Sense. That means the solar output is “named” by the end of the 15 minute interval, instead of the start time of the interval in SolarEdge-land. Green is an hour by hour comparison after the 15 min offset, orange is without.

Here’s just the data with a 15 min offset on the SolarEdgedata aggregated into hours

Here’s the same data with 3 midday readings randomly reduced by a 1.5 minute dropout. Can you spot them ?

[erik]

I can spot them (it appears that you reduced three SolarEdge values near the top), but only because I have the before-edit and after-edit graphs to compare to (it becomes a game of “spot the difference” instead of “spot the outlier”). I might be able to see two of them without the comparison, but I see similar differences in the 1.5kW region, too.

I’ll try something similar in a bit, and see if anything jumps out at me.

[kevin1]

I did 5 runs on my solar data, randomly negatively altering 3 midday points by 1.5 minutes, then applying a linear model to that altered data, without any comparison to the original data. When I look at the 3 biggest residuals, 2 times out of 5, they are all 3 the same as the altered points. 3 times out of 5, 2 out of 3 of the largest residuals were the altered points. Here’s a case where all three of the largest residuals were the dropouts.

So the real SolarEdge inverter power dropouts are likely findable by looking at the largest residuals.

[erik]

I’m finding just the opposite about my SolarEdge reporting. I have to assume that the time starts with the timestamp to get better correlation! Below, I’ve reported both versions, where the “12:00” Sense data is compared with SolarEdge’s 12:00, 12:15, 12:30, and 12:45, and where the “12:00” Sense data is compared with SolarEdge’s 12:15, 12:30, 12:45, and 1:00. You’ll see that the former gives me better correlation, whereas you seemed to indicate the latter gives you better correlation. It’s very strange that SolarEdge wouldn’t stick with consistent labeling across its customer base!

My 5/31 data is VERY badly correlated (I added an X=Y line in red to show where the trend line “should” be), regardless of which SolarEdge hour grouping I use:

However, when I do the same with my data from today, I get better correlation with an almost-perfect X=Y trendline, but I can NOT see the dropout, which either suggests that my SolarEdge-Sense disagreement is larger than the influence of a 1-ish minute Sense outage, or that the outage is real:

You’ll note that in this 7/22 data, there’s one point near the top of the graph that is noticeably further to the left… unfortunately, that is from the 1PM hour, and the drop was during the 12PM hour (12:21PM). So the biggest outlier is not the one I’m looking for.

The 7/22 data weakly suggests the drops are real, but I don’t think it’s conclusive. The 5/31 data, on the other hand, more strongly suggests the Sense dropouts are reading errors on Sense’s part. The dots are on both sides of the trendline, but if you look at the red X=Y line, you’ll note that it fits better with the low-end points, and the high-end points are either on the line or to the left of it - which means Sense was consistently providing LOWER values than SolarEdge. And the worst outlier is the 1PM dot, which is where the repeated outages in the screen capture a few messages above were (6 solar reading drops in about 20 minutes)

[kevin1]

Good analysis. Three thoughts.

• My SolarEdge data has come via the Solar City website, then via the Tesla app, so perhaps it’s offset from the direct SolarEdge data. I discovered that I had a time offset between Sense and Solar City a while back when my straight line became a loop.
  How Accurate is Sense vs. Utility Metering? - #12 by kevin1
• From my experience, and that of others on the forum, inverter readings run about 3-5% higher than Sense. But the combined Sense net calculation ends up much closer to my hourly utility net readings vs. combining Sense Total Utilization with inverter solar data. So treat the Sense solar reading as more accurate than the inverter data. Take a look at my fit lines - slope is 1.03 (inverter is 3% higher).
• It might be easier to pick out the errant points if you charted, analyzed and fit multiple days worth of data in one shot, to give regression a broader statistical sample.

[erik]

@kevin1

It might be easier to pick out the errant points if you charted, analyzed and fit multiple days worth of data in one shot, to give regression a broader statistical sample.

I considered that, but since the hour with a drop-out in it looked “good” and a similar hour with no drop-out looked like an outlier (in the right direction to look like the signal I’m seeking) in the 7/22 data, I figured the extra work was unlikely to pay off. While I’d need a lot of examples to prove the idea that these drops will fall away from the trend in a detectable manner, I only really needed one to disprove it, and the first one I looked at did it.

The 5/31 data is weird enough that I might go back and check out some more days from when I first found this problem if I have free time this weekend. I might also try replicating those early drops by timing my laundry to run the dryer mid-day again (I’ve been avoiding that to prevent losing valuable solar power), and seeing how that affects the comparison now.

It’s really interesting that you’re finding Sense to more closely match your utility meter than the inverter. I wonder if that suggests there are 3-5% power losses between your inverter and the utility’s meter. That seems like an awful lot of loss for such a short run, though. Unfortunately, my utility (Dominion VA Power) doesn’t “properly” report meter values on the bill. It only reports increases, making it hard to do a comparison without compulsively running outside to read the meter and manually record values to trend. My bill’s “previous” and “current” meter values are stuck at “10541” (kWh) until they next get a reading that’s above that value, even though it’s currently displaying a value below 10,000. I guess Dominion doesn’t explicitly track my production credits, they just don’t bill me for generation/transmission/fuel until they see a higher number (I assume they have some buffer defined for when customers roll back from 00000 to 99999, and don’t send out a bankrupting power bill, and since net credits are supposedly only good for a year, I’m also guessing they’d reset the value on my bill after 12 consecutive months of no net increase)

[pswired]

@erik, you might consider buying a clamp-on ammeter with a min/max reporting function. You could then clamp this on the inverter output in the middle of a sunny, cloud-free day (definitely not this week here in the DC area) and checking later in the day to see if the minimum current value indicates a dropout as seen by Sense. If so, you could then put it between your panels and the inverter to isolate whether the inverter or the string is dropping out.

This would do the trick for AC inverter output, but not the DC bus:

Klein Tools 400A AC Auto-Ranging Digital Clamp Meter CL330 - The Home Depot

Or if it’s something you’ll have a use for in the future, you might consider a more expensive Fluke meter with DC current measurement capability.

[kevin1]

My inverter is about 8-10 feet away from my Sense, so I doubt much of the difference stems from resistive wiring loss. I haven’t been able to ascertain where inside the inverter, the SolarEdge data comes from, but I suspect it comes from the processor controlling MPPT and the DC to AC conversion. Newer SolarEdge’s than mine (2013) have the option for revenue grade metering but that seems to be a separate module. Given the regularity of the forum comments about this difference, I’m guessing that some of the difference might be attributable to power lost to the output filter on the inverter.

Good luck sorting out the dropout events. You might want to check in with support@sense.com.