your "new ids should be as similar as possible to the old ones" is poorly defined and -sorry - your code is far too undocumented to be read or published (especially for such a delicate use case)

I'll give you a concept instead of code:

For what I understand you need 5 steps

1. split /_/
2. split /(\d+)/
3. count nondigit substrings in a hash %strings
4. shorten keys of this hash "as similar as possible"
5. recompose

The magic is of course in step 4 (and you have to care about about different lengths of numbers in step 2)

for instance you could start to investigate the first n chars of each \w+ string in a loop.

As long as a grep(/$shortend\w*/) keys %strings counts more than 1 match you have to try again with a larger $n.

you can extend this method to  grep(/$pre\w*$post/) ...

To further shorten the recomposed string you might consider using CamelCase instead of _ as delimiter (i.e. uppercase first character)

To be clear, all of this doesn't guaranty len(strings) <= $length, so you might be forced to skip readability for some keys in an extra step of shortening.