@Roylang I have an approach which may yield the unit at the other end of the formation, for example my data says that 91319 was at one end of a composition with 91155 at the other on Sunday 10/12/2017 - they worked train 1 that day. This data source doesn't have the middle unit numbers. Is that any help?
I've checked how far it goes back and the earliest I could find was the first 2011 record with 91339 being accompanied by 91089 on train 16 on Thursday 07/07/2011. This only works for 'leading' units, so won't help with the 92XXX vehicles listed (or 91XXX if they were running as middle cabs).
I've done the others for 22/11/2017:
91325 was with 91029
91285 was with 91153
91147 was with 91183
91145 and 91095 were together on that date.
On 20/06/2017 91197 was with 91053
On 13/12/2016 91153 was with 91149
On 1/12/2016 91239 was with 91323
No data returned for 91311 on 5/8/2016 I'm afraid - not sure why that is - may be it was running as a middle cab?
On 10/09/2015 91305 was with 91229
On 10/07/2015 91007 was with 91061
On 19/06/2014 91061 was with 91207
On 5/06/2014 91029 was with 91235
On 3/10/2013 91087 was with 91137
On 5/09/2013 91343 was with 91127
On 18/07/2013 91117 was with 91291
On 14/07/2013 91113 was with 91217
On 11/10/2012 91099 was with 91227
On 13/09/2012 91019 was with 91347
On 23/06/2012 91215 was with 91029
On 11/05/2012 91013 was with 91031
On 16/03/2012 91287 was with 91221
I think that's all the ones I can help with...