The problem can be slightly simplified if you ignore the home/away  
requirement.  The second part of the season can be the same as the first  
part but with the teams switched.

This leaves you with 17 dates on each of which there must be 9 fixtures,  
with each team playing once on each date.

Firstly generate all of the required matches for the season (ignoring who  
is at home).  This is the set of all combinations of length 2 from a set  
of 18 elements.

This will give you 153 pairs of teams that you need to divide into 17  
separate pots ensuring that each team only appears once in each pot.

I'm not sure of the best way to do this, but I think the following  
approach should work reasonably well (although somebody else can probably  
suggest a more efficient way of doint it):

Firstly shuffle the list of 153 pairs (assuming you don't want the  
fixtures to be predictable).  Then take each pair in turn and allocate it  
to the first available pot (fixture date) in which neither team is already  
present.  I haven't tried it, but I think this should give you what you  
need.

Once you've done this you can just duplicate the fixtures for the second  
half of the season, switching the home and away teams for each match.

Dan.