So over the last few weeks, I have noticed some strange graphs from the monitoring system.
If you remember the architecture with a small microcontroller (PIC10F220) feeding a pulse train to the main controller, you can see what is happening with monitoring the panel. As the sun sets the panel produces less and less voltage, until it reaches about 5v. At this time, the regulator for the ADC chip (a 78L33) drops out of regulation and the '220 stops running. When the controller stops running, the pulse train stops and the interrupts that the main controller uses to time the pulses no longer happen. This means the values are no longer updated and the reading flatlines at whatever the last value was.
To fix this, I have decided that when I read a set of values out of the box buffer, I will reset them to 0. Under daytime circumstances, the incoming pulse train will overwrite them in very short order (just a few milliseconds) and there will be no noticable difference in the readings. However during the night, the lack of pulses will force the reading to zero, and clamp it at that level.until the sun rises and the panel starts to generate power.