The attached dataset provides the GB grid frequency every second for two hours (02.00-04.00) on 1 September 2015 (at some location that I do not remember).  I used it to test some linear stochastic process models for the frequency as a function of time.  On the 2-hour timescale the trace fit reasonably well to an Ornstein-Uhlenbeck (OU) process with a decay time of around 30 minutes.  On a shorter timescale, in particular the first 3 minutes 20 seconds, the trace looked smooth so I fitted it to a first-order filtered OU process, which gave me time constants of 11 minutes and 1.9 seconds.  This was important evidence that a useful model for power imbalance as a function of time is a filtered white noise.  Furthermore, the power spectrum of the 2-hour trace showed a regime with slope around -4 for periods shorter than 25 seconds, consistent with being a first-order filtered OU process, until the Nyquist cutoff at 2 seconds.

The data were publicly available from National Grid (http://www2.nationalgrid.com/Enhanced-Frequency-Response.aspx) at the time.  I downloaded them.  The link is no longer active, so I have made the data available here.  I made no changes.

The data are in an excel file with three columns: date (DD/MM/YEAR), time (HH:MM:SS), frequency (Hz).  A graph is also provided.