The string did not change. The temperature did, and open-circuit voltage moves with it.
Panel specifications are printed at Standard Test Conditions, which include a cell temperature of 25 degrees C. Open-circuit voltage, Voc, goes up as the cell gets colder. Every crystalline module datasheet publishes a temperature coefficient for Voc, and it is a negative number in percent per degree C, typically somewhere around minus 0.3 percent per degree C depending on the cell technology.
Negative coefficient with a falling temperature means rising voltage. Cold, clear, sunny mornings are the worst case, because irradiance comes up while the panels are still cold.
Why it shows up as an intermittent fault
A string sized comfortably against the 25 degree figure can sit under the controller’s maximum input voltage all summer and then overshoot it for twenty minutes on a January morning. The controller protects itself, throws an overvoltage fault, and by the time you walk outside to look, the panels have warmed and everything reads normal. This is why it looks like a mystery fault rather than a sizing error.
The calculation to run
You need three numbers from your panel datasheet: Voc at STC, the temperature coefficient of Voc, and the number of panels in series. Then you need the coldest cell temperature you expect, which is roughly your record low ambient temperature, because a panel at dawn is not warmer than the air around it.
Work out the temperature difference below 25 degrees C, multiply by the coefficient to get the percentage increase, apply that to Voc, then multiply by the number of panels in series. Compare the result against the controller’s maximum PV input voltage, not its operating range.
If the number lands close to the ceiling, shorten the string by one panel. Headroom here is cheap and an overvoltage event is not.
Two mistakes worth avoiding
The first is using average winter lows instead of record lows. The event you are protecting against is the coldest morning in the equipment’s lifetime, not a typical one.
The second is checking against Vmp, the voltage at maximum power, because it is the figure you see the array producing day to day. Vmp is not the number that damages a controller. Voc at open circuit, before the controller starts drawing, is the number that matters, and it is always higher.
Status: this describes the calculation method using coefficients that every crystalline module datasheet publishes. The exact coefficient and Voc for your panels must come from your own panel datasheet, because they vary by model. No single manufacturer figure is cited here.