As another answer has stated, tides can broadly be predicted far ahead by harmonic analysis (which is similar to taking a fourier transform of the tidal signal, but only allowing frequencies that correspond to various astronomical periods of the sun, the moon, and interactions between the two).
There are some caveats to this.
As David Hammen has noted, is that water depths may change over time due to changes in sediment. In many of the tidal sites that are currently being developed the speed of the current means that the seabed is scoured rock, without sediment, but in some areas this change may nevertheless become significant, perhaps over a 25-year timescale.
Superimposed on this astronomical signal can be significant (up to approx. 0.5 m/s) short-term variations due to meteorological effects such as surface winds and storm surges. This limits the accuracy with which one can predict the speed of tidal flow more than a few days ahead, but does not give a rationale for the 25-year figure.
As noted by this paper and others, while the astronomical (rather than meteorological) component of water elevations resulting from tides can be fully described using a sufficient number of harmonic constituents, the same is not always true of flow velocities. This is because the shape of landforms and of the seabed can cause additional periodic changes in the velocity at a given location which do not correspond to astronomical frequencies. An example is the tendency of a "jet" to form downstream of a constriction at certain phases of the tide. Nevertheless, harmonic analysis can still usually provide a close approximation and, once again, this does not provide a rationale for the 25-year limit given.
I suspect that 25 years is simply a number plucked from the air to mean "a long time", perhaps informed (as noted in a comment) by that being the intended lifetime of the turbines.