Instead of forcing meta-data (i.e. pseudo-indices) into the fieldnames, why are you not simply using a non-scalar structure? Then this task could be achieved using some very basic indexing.
"Is it possible to retrieve the struct fields based on a part of their name and to insert them into a new struct?"
Obtain the FIELDNAMES, identify the parts of the text you want, use something like STRUCT2CELL or a loop to select the fields that you want, then convert them back to structure. Certainly possible if rather fiddly... but better data design (e.g. non-scalar structure with simple fieldnames) would make this task much easier:
S(1).Hs = 1;
S(1).Ts = 11;
S(2).Hs = 2;
S(2).Ts = 22;
S(3).Hs = 3;
S(3).Ts = 33;
T = S([1,3]) % indexing
You will have much simpler, more efficient code when you do not force meta-data into fieldnames.
