Interpolation isn't probably relevant in this case. Bestfit curve (or approximation) which Abigail mentions is.
This is why: we want averages (read: good, useful values) in arbitrary Xs, while interpolation produces a function which will exactly evaluate to those Ys from those Xs we used as data for interpolation. Nothing said about Xs between original datapoints, they can be anything. Think of a sine wave which perfectly interpolates points (0, 0), (pi, 0), ..., (n*pi, 0).
Approximation minimizes distance between the curve and each of the datapoints (the datacloud) to produce exactly what you want: a function to calculate useful value in any point of the X axis.
Give Math::Approx a try. Like this:
my %data = (
mins_from_midnight('07:01') => 82,
mins_from_midnight('23:48') => 188,
# ...
);
my $approx = Math::Approx->new(undef, 3, %data);
print $approx->approx(mins_from_midnight('15:40'));
Update: and yes, the day you took a specific datapoint does not matter. Anyway, you want average and that means all the days are the same for your task. That's ok.