Project note
Nineteen Days Before the Snow: What a Rush Off-Grid Build Taught Me About Sizing, Winter Storage, and First Impressions
6:40 p.m., late October 2024
Nasir called from Gilgit. He manages a small medical clinic up there — eight beds, a vaccine fridge, an autoclave, and a diesel generator that had just thrown a rod through the block. The clinic already had panels on the roof, 4.2 kWp of them, installed two summers earlier. What it didn't have was a battery.
Normal lead time on a system like that is three to four weeks. From that call to first snow on the Karakoram Highway, we had nineteen days. Missing that window didn't mean a late delivery. It meant the equipment sitting in a shed at the bottom of the valley until April, and a clinic running on a borrowed generator all winter.
I said, “Let me see what's possible.” He heard, “We'll make it.” That distinction matters more than it should, and it comes back later.
Sizing before shopping
My first instinct was to start pricing hardware. That's the wrong instinct, and I've been burned by it enough times that I now force myself through the load math first — even on a rush job. Especially on a rush job.
Nasir sent me two weeks of readings off the old inverter's display. Night load averaged about 5.8 kWh — lights, the vaccine fridge cycling, the autoclave running during evening clinic hours. Day load he could mostly cover off the array directly, so the bank only had to carry the dark hours.
Battery energy storage system sizing looks simple on paper: usable kWh divided by depth of discharge, then divided again by efficiency losses. In practice, the assumptions are where people get themselves into trouble.
- Autonomy target: one day, not three. He had a generator, and a new one was already ordered. Sizing for three days of autonomy would have roughly tripled the battery line item and put us past the deadline for no real gain.
- Depth of discharge: 75%. The LiFePO4 packs we were quoting could run to 80%, but I sized to 75% because winter nights up there are long and I'd rather leave margin than explain a low-voltage trip in January.
- Round-trip efficiency: 0.85. Combined inverter and charge path. Conservative, which is what I want to be when I can't send a tech back in three days.
So: 5.8 ÷ 0.75 ÷ 0.85 ≈ 9.1 kWh. We specced a 10.2 kWh bank — 48V nominal, 200Ah, LiFePO4. Slightly oversized against the math, deliberately.
Here's the detail I almost missed. The clinic had a separate monitoring cabinet — router, data logger, two security cameras — and it lived on the generator circuit. If the main bank ever tripped on low voltage, that cabinet went dark, and the clinic lost remote monitoring at exactly the moment someone would want it. We added a 48v 20ah lifepo4 battery as a dedicated buffer for that cabinet. Roughly a kilowatt-hour, enough to ride out a trip and keep the logger writing. Cheap insurance, and it turned out to be the component the client asked about most, six months later.
Two controllers, not one
With 4.2 kWp on the roof and a 48V nominal bank, a single charge controller gets uncomfortable fast. At 57V absorption you're looking at somewhere around 70–80A at full irradiance, and Gilgit-Baltistan is cold and high — panels up there routinely beat their nameplate in winter, which is the opposite of what most people expect.
One 50A unit would have clipped hard. Not a catastrophe, but you'd be throwing away generation from about 11 a.m. to 2 p.m. every clear day, and on a clinic running an autoclave that's real money.
So we went with two epever 50a mppt charge controllers, one per 2.1 kWp array half. I went back and forth on this for the better part of a day, honestly. One larger controller is tidier — one fewer failure point, one fewer set of DC breakers, one fewer thing to document. But the split gave us something I couldn't buy otherwise: shading tolerance. There's a poplar on the east end of that roof that Nasir would not cut down, and it shades roughly a third of the array until about 9:30 a.m. Split arrays with independent MPPTs handle partial shade far better than one tracker reading an averaged input.
The redundancy was a bonus I didn't pay extra for. If a controller fails in February, the clinic drops to half generation instead of zero, and I can put a replacement on the next truck north.
The single-controller route would have saved maybe $200. That's a trade I've made wrong before, and I didn't want to make it again on this one.
Where the wiring spec went sideways
This is the part I have to be honest about, because it cost us two days.
I wrote the string plan up and sent it to the local electrician with a note reading “50A per controller.” He read 50A and sized the DC run off the roof as a single 6mm² home run feeding one unit. Reasonable reading, wrong conclusion — I meant the controller's output rating, he took it as the array's service rating. We were using the same words and meaning different things.
I only caught it because he sent a photo of the conduit on the Tuesday and something about one home run disappearing into a 50A box didn't sit right with me. Reworking it meant a second pull, new gland entries, and an extra day on site. On a nineteen-day clock, a day is a lot.
The fix wasn't technical. It was a one-page diagram with the array halves drawn separately, every circuit labelled in both amps and mm², and a callback before anyone pulls wire. We've used that format on every multi-controller job since, and it hasn't failed yet.
The 5kW inverter price question
Nasir asked the question every client asks, and my answer was less satisfying than he wanted.
If you're searching for epever 5kw inverter price in pakistan, here's the honest shape of it: the number quoted online is almost never the number you pay. What we saw in late 2024, across three distributors — one in Lahore, one in Islamabad, one in Gilgit — was roughly PKR 250,000 to PKR 400,000 for a 48V, 5kW pure sine unit. That is a range, not a quote, and it will be different by the time you read this. The rupee moves, duties change, and anyone giving you a firm figure over WhatsApp without checking landed stock is guessing.
The spread had almost nothing to do with the hardware. It came down to four things:
- Stock position. A distributor holding inventory quotes one thing. A distributor who has to import to order quotes another, plus three to five weeks of waiting you didn't plan for.
- Inverter-only vs. inverter/charger. The charger adds a transfer switch and a lot of convenience if you're integrating a generator. Nasir was, so we needed it — that moved the number up meaningfully.
- Freight and duty. Getting a 40kg box to Gilgit-Baltistan isn't free, and nobody puts the last 600km in their headline price.
- Warranty handling. The cheapest quote is often a seller who will tell you to ship the unit back to the factory yourself. That's a decision you make once, at a bad moment, from a cold clinic.
I'd rather pay mid-range to someone who picks up the phone in February than save 12% with a seller who won't.
The mistake that surfaced three weeks later
We didn't have a formal receiving check for battery state of charge. That's the whole gap. Not a technical failure — a missing twenty-minute procedure.
The packs shipped at roughly 30% SOC, which is fine for shipping; nobody did anything wrong. But they then sat in a Karachi warehouse for three weeks and on a truck north for four days. By the time they landed at the clinic, two of the four modules were lower than I'd like going into commissioning.
Not damaged. Just low, and low for longer than I was comfortable with. We brought them up slowly off the array before the first full load test and everything checked out. But I spent that week half-waiting for a BMS fault that never came, and I should have prevented the waiting entirely.
The third time something like this happens, you write a checklist. I finally wrote it after the second time, which is later than it should have been: ship at 50% SOC, log the BMS reading on arrival, log it again at commissioning, flag anything under 25%. One page. It took twenty minutes to write and should have existed in 2022.
What the client actually judged
Here's the thing I keep relearning. Nasir will never see the MPPT tracking algorithm. He'll never know whether those controllers run at 97% or 99% efficiency. Those numbers are invisible to him, and honestly they're invisible to most of us.
What he saw was the enclosure door.
Labelled breakers, every one of them. A printed single-line diagram in a plastic sleeve bonded inside the door. Cable glands torqued, DC runs in conduit with a drip loop at the wall entry. No wire nuts anywhere. Battery modules on a rack with 100mm clearance on all sides and a temperature sensor positioned where it could actually read cell temperature instead of ambient air.
Nobody bills for that. It cost us maybe $80 in parts and half a day of labour to do properly instead of slapping it together.
And it changed how Nasir talked about the whole project. He sent photos of the inside of that enclosure to two other clinic administrators in the region. One of them called me in March.
To be fair, a tidy enclosure doesn't fix a badly sized bank, and I'd take correct engineering with messy wiring over the reverse every time. But when the engineering is roughly equal — and between competent brands it usually is — the presentation is what the client uses to decide whether they trust you. The $80 difference per project showed up in feedback long before it showed up in anything I could measure on a meter.
Our post-install survey scores moved up noticeably after we formalized the enclosure standard. I want to say it was in the low twenties percentage-wise, but I'd have to pull the actual numbers. The direction isn't in doubt.
The email that arrived in December
Three weeks after commissioning, Nasir wrote to ask how to store LiFePO4 batteries in winter, because the backup 48v 20ah lifepo4 unit sat in an unheated outbuilding and he was worried the cold would ruin it.
Fair question, and the answer is more forgiving than most people expect:
- Charge level: around 50%. For a 48V pack that's roughly 53V resting, or about 3.25–3.3V per cell. Full charge combined with heat is what ages a pack. Cold alone is not the villain here.
- Temperature: 0°C to 25°C if you can. Cold storage is fine. The damage comes from charging below 0°C, not from being cold. Any decent BMS will block charge current below freezing, and that's exactly the behaviour you want.
- Disconnect it. Open the DC breaker. A BMS with Bluetooth or a display draws a small parasitic load, and over four months of −15°C nights that adds up more than you'd think.
- Top up every three to six months. Check voltage, bring it back toward 50% if it has drifted. Don't leave it at 20% for an entire winter.
- Before spring use: let the pack come up above freezing, then charge slowly the first time.
That's the standard guidance for the LiFePO4 packs we work with — check your own BMS cutoffs, though, because they do vary. There are budget units on the market whose low-temperature charge protection doesn't actually function. Test it once, deliberately, before you rely on it for a winter.
The 20Ah unit was fine, for what it's worth. It sat at about 55% SOC in a −10°C shed all winter and came back in April with cell balance inside 15mV.
What I'd do differently
Looking back, I'd have sent the string diagram before I sent the quote. The wiring confusion cost two days I had budgeted for something else, and it was entirely avoidable — not a technical problem, a communication one. I said 50A. He heard 50A. Both of us were right and the job still went wrong.
I'd also have written 50% SOC into the shipping documents instead of assuming the warehouse would handle it sensibly. They did handle it sensibly. It still meant four weeks at low charge, and I got lucky.
And I'd have built the enclosure standard two years earlier, because it turns out clients judge you on the thing you weren't optimizing for.
Nineteen days from that Thursday call to a commissioned system. The dead generator got picked up on day twenty. Snow came four days later.
Good outcome. Not one I'm eager to repeat on that timeline — but the checklist exists now, which means the next one leans a little less on luck.
