The panels are the part people photograph. The rack, the neat rows, the glint of glass — that’s the money shot. But the decisions that actually make or break a solar build often happen in the unglamorous space between the array and the inverter, where a calculation you get wrong can cook a $3,000 inverter on a cold morning. This is the post about that space: how I sized the strings for three EG4 18kPV inverters, why bifacial panels made me redo the current math, and the couple of mistakes I’m working around as I wire it up.
The number that Can Kill Your Inverter: cold Voc
Here’s the counterintuitive thing about solar panels: the moment you have to design around isn’t a blazing July afternoon. It’s the coldest morning of the year, at first light, before the array is even making usable power.
A panel’s open-circuit voltage — Voc, the voltage with nothing connected — goes up as the cells get colder. My panels drop about 0.29–0.31% of their Voc for every degree C above 25, which means they gain that much for every degree below. String a bunch in series and those volts stack up pretty quick. Push the total past the inverter’s ceiling on a cold morning and you don’t get a warning; you get damage. When a manufacturer says don’t exceed 600 volts they mean it. You don’t have extra leeway built into that number, the leeway is already taken up. You’re probably going to find that at least one of the MPPTs is dead.
Most modern hybrid or off-grid inverters assume you’re going to mostly connect your panels in series. When you do that you can get a lot of power without having to run heavy wiring. A series string at 500 volts and ten amps is 5000 watts and you can connect that to your inverter with 10 gauge wire, or even 12 gauge. A parallel string at 50 volts for 5000 watts is 100 amps. For anything more than a few feet you’ll be spending more on wire that looks like it belongs on a welder than you did on panels.
But that high voltage will rise when it gets cold out. So the number of panels you can connect in series comes down to: how cold does it actually get here, and what’s the inverter’s limit?
In my case, the EG4 18kPV has an absolute maximum PV input of 600 V, with an over-voltage protection threshold at 550 V, and it wants the operating voltage to live in a 140–500 V window. For the cold, I went looking for Hood River’s record low: −33 °C, set in 1919. That’s far colder than code actually requires — the NEC method uses an ASHRAE design minimum that’s more like −15 °C here — but I’d rather design to the record and never think about it again.
The math, per panel, at −33 °C: my 415 W Canadian Solar panels have a 47.8 V Voc at 41 degrees C, and the temperature coefficient is 0.29% per degree C so at −33 °C that becomes 47.8 × (1 + 0.0029 × 58) ≈ 55.8 V. Now the maximum number of panels is just the number of panels times 55.8V:
- 8 in series → 447 V. Comfortable.
- 9 in series → 503 V. Still under the 550 V protection point.
- 10 in series → 558 V. Over the protection threshold.
- 11 in series → 614 V. Over the 600 V damage limit.
I’d briefly convinced myself 11 was fine — it pencils out under 600 at milder temperatures. But at the real record cold, 11 series would put 614 V into an inverter rated for 600. That’s the whole lesson in one line: design to the coldest morning, not the average one. I settled on strings of 8, which also happens to divide my 32-panel rack cleanly into four. You don’t want to be over the protection limit either — the MPPTs just dump the excess.
Bifacial changed the current, not the voltage
Both my arrays are bifacial — glass on both faces, making power off the back as well as the front. A year ago, I painted the south side and top of my shipping container white so the older tilting-rack panels would have something bright bounce light to the back of them. For the new rack I’m going to lay down orchard cloth–white on the top, black on the bottom to control weeds and give me some bifacial gain. And I’m going to place white marble offcuts to hold the cloth down–I have tons of it stacked on pallets by my back fence. My shop used to be a granite and marble cutting business
The trap is assuming bifacial only helps and never complicates. It complicates the current. Voltage barely notices the extra rear light — Voc rises with irradiance only logarithmically, a fraction of a percent. But current scales almost linearly with total light, so the rear gain shows up there. On a good day my panels pull noticeably more amps than their front-only rating.
That matters because each MPPT input has a current limit — 25 A on the big one, 15 A on the two smaller ones. And here’s the part that bites: if you parallel two strings onto one MPPT and the combined, rear-boosted current climbs past that input’s rating, the inverter simply clips the top off. You’d be throwing away exactly the bifacial gain you built a white apron and painted a container to get.
The fix is simple: one string per MPPT wherever possible, so nothing parallels and nothing clips. String length doesn’t change this — paralleling two strings doubles the current no matter how long they are; only the number of paralleled strings does.
Three inverters, one puzzle
Sixty-one panels, three 18kPVs, and the panels don’t divide evenly. The two new inverters each take 16 of the new bifacial panels as two single strings of 8 — one per MPPT, nothing paralleled, the third MPPT left open for future expansion. Clean, and every amp of rear gain gets through.
The original inverter is the messy one, and I’ll be honest: I built that side over a year ago and I didn’t actually remember the string configuration I used. I had to climb up on the container and look at the wiring. Yikes — one 12-panel string and one 10-panel. When I searched the web for panel specs I was guessing at what I had, and from the calculations it looked like last year’s mild winter let me dodge a bullet. But I was sure I had run the calculations and was safe.
Turns out the panels are a different panel than I’d been assuming — the DNA-120 is a 120-half-cell module, which runs a lower Voc than the 144-cell I thought I had, and that materially changes the over-voltage picture — 41.4 Voc, not 48.2 V (−0.28 %/°C). Redoing the math with the real number: cold Voc per panel at the −33 °C record = 41.4 × (1 + 0.0028 × 58) = 48.1 V.
The 12-series never threatened the 600 V damage limit — it wouldn’t get there until about −49 °C, colder than Hood River has ever been. It only crosses the 550 V protection threshold below roughly −13 °C. The worst that string does on a genuinely cold morning is trip over-voltage protection and sit there doing nothing until the sun warms the panels a few degrees — a little lost harvest on the coldest mornings, not a fried inverter. I wasn’t rolling the dice on damage; at most I was leaving a few kWh on the table a handful of mornings a year.
Still, I need to do a bit of tidy-up work up on the container, and my knee has recovered enough to climb the ladder, so I’m going to reconfigure the wiring of the tilting rack from 12 + 10 to 11 + 11:
- MPPT-1 (25 A): experiment 7 — leave it (337 V cold)
- MPPT-2 (15 A): tilting 11 (529 V cold)
- MPPT-3 (15 A): tilting 11 (529 V cold)
That’s moving one panel from the long string to the short one — no re-terminating three strings, no parallel pair, no clipping (each 11-string pulls ~12–13 A even with the container’s bifacial boost, comfortably under the 15 A inputs). Two clean single strings of 11 is the answer.
So yay, it all fits, with a little headroom, but it’s still the least tidy corner of the whole system — and a good reminder that past-me doesn’t tend to leave great documentation for future me–hence these posts. And you thought I was doing this for altruistic reasons.
The unglamorous mistakes
I never get a big project done without a few screwups, so here my latest:
I’m running with a dead disconnect right now. My PV disconnect switch failed, and until I rewire it, the tilting array has no proper isolation. I don’t recommend this — with solar, the panels are live whenever the sun is up, and the only way to break the circuit without a disconnect is pulling connectors under load, which makes exactly the arc you don’t want to make. It’s first on my list to restore before I go elbow-deep in anything else. Do as I say, not as I’m currently doing.
For the new array, I bought a disconnect that’s one string too small. I picked up a three-string PV disconnect, figuring three would be plenty. Two things conspired against that. First, I’d planned on 30 panels, but I’d bought the “large module” version of the rack, which assumes substantially wider panels than mine — so it swallowed 32. Second, and more fundamentally, even the original 30 panels can’t be wired in three strings: the cold-Voc limit caps me near 8 per string, and three strings of nine is only 27 panels. Thirty-two forces four strings of eight, the math is nice, eight panels is right in the sweet spot of the MPPTs. A four-string switch was a minimal upcharge at the time, but I bought the three string switch–dumbass. The lesson is annoyingly general: the string count is set by physics. Do the math then buy the disconnect. But hey, I need a new disconnect switch for the tilting rack — and it needs to be a three string. So I ordered a new four string disconnect for the Chiko rack. Sometimes I’m stupid but sometimes I’m lucky.
And now I’m counting wire. I’ve got two reels of solar cable, one red and one black, and I thought what was left on the reels must be plenty when I was thinking three strings. Four strings means four red home runs and four black ones from the rack to the inverters in the container — a full extra run per color, about a third more cable than I planned for. I measured the run, and did the arithmetic — because the one thing worse than coming up short is splicing a home run in the middle of a field. Turns out that I’ve got plenty of wire. I’ll have a few hundred feet to spare. And of course I checked the voltage drop over the distance. It’s tiny. Lucked out again.
Lessons Learned
The rack and panel install rewards patience, a good string line, and a bit of precision–which I have in short supply. Next time I’d spend a bit more time fabricating some tools that would make the Trident mount installation more precise. My finished product is a bit wobbly because I didn’t get the bases perfect, but it’s strong as heck and looks OK. The DC design rewards a specific kind of paranoia. Design to the record cold, not the comfortable average. Realize that bifacial gain affects the current, so don’t lose it because the MPPT clipped the excess. Buy the disconnect that matches the math. Measure your wire twice. And write down what you did, because in a year I’ll be standing in front of my inverter trying to remember what past-Bill was thinking. That’s what this post is all about — looking out for Future Bill.