I would think of the next two approaches:

1. Binary search. For example, you could use a binary tree, or a B-tree, or anything similar, as the data structure. What it should do, if it doesn't find a key in the tree, is return the record for the next smaller value. Make sure the dates are easily sortable, for example by using the ISO-8601 combined date/time format: "20031012T132000". This setup allows for variable gaps between measurement times. I'm not sure what Perl modules are available for this means... Keywords to search for include tree, trie, heap.
2. Since your times are always exactly 30 minutes apart, just convert it to epoch times, subtract some starting value, divide by 30*60 and use the resulting value as an integer array index, in a normal array.