Grid problem ?

If this fluctuation is due to poor measurement technique, it is pretty bad. The method suggested by Mike is only workable if there is an average of a number of readings. Traditionally really accurate measurements of low frequencies have been by phase locked frequency multiplication, the loop filter providing a suitable level of averaging. I designed such a device many years ago, and it was fairly widely used  for film and television discharge lighting where exact frequency control is required to prevent flickering. Multiplying to 500 kHz allowed readings to 0.001 Hz at 10 readings a second. Of course the input waveform needed to be filtered to reduce the noise bandwidth somewhere near the expected frequency range, but that is fairly obvious. Loop bandwidth was about 1Hz, so it followed "wobbles" quite well. I expect that this reading is not instrumented well, probably the period is just measured with a processor timer with little filtering etc. A similar technique to that above must be used by the synchronous wind generators to control the blade pitch, perhaps assisted by the reactive power phase from the alternator. I bet that gave someone a serious headache to design, as the error is probably chaotic with wind speed! I would doubt that 25% wind is properly stable, there are just so many time delays and the whole system would have far too many similar but not identical responses that normal control analysis would be impossible.

If this fluctuation is due to poor measurement technique, it is pretty bad. The method suggested by Mike is only workable if there is an average of a number of readings. Traditionally really accurate measurements of low frequencies have been by phase locked frequency multiplication, the loop filter providing a suitable level of averaging. I designed such a device many years ago, and it was fairly widely used  for film and television discharge lighting where exact frequency control is required to prevent flickering. Multiplying to 500 kHz allowed readings to 0.001 Hz at 10 readings a second. Of course the input waveform needed to be filtered to reduce the noise bandwidth somewhere near the expected frequency range, but that is fairly obvious. Loop bandwidth was about 1Hz, so it followed "wobbles" quite well. I expect that this reading is not instrumented well, probably the period is just measured with a processor timer with little filtering etc. A similar technique to that above must be used by the synchronous wind generators to control the blade pitch, perhaps assisted by the reactive power phase from the alternator. I bet that gave someone a serious headache to design, as the error is probably chaotic with wind speed! I would doubt that 25% wind is properly stable, there are just so many time delays and the whole system would have far too many similar but not identical responses that normal control analysis would be impossible.