From: testa@ncat.edu   
      
   Greetings all.   
      
   I'm working with a TSP-like scheduling problem, and am interested in using   
   FL to improve the solutions generated by my algorithm.  The algorithm I'm   
   using is a GA, with a path representation for the solution (e.g., the   
   solution looks like "depot-A-B-D-C-depot").  Any pointers to relevant   
   literature would be greatly appreciated.   
      
   The algorithm generates a number of candidate solutions (doesn't matter   
   how), which are then evaluated in a two-step process.  The first step   
   calculates the cost of the solution, and the second step determines whether   
   the solution violates any constraints.  Penalties are assessed via penalty   
   method.  I imagine this is a fairly common method, which most folks here   
   have seen.   
      
   The main bottlenecks in the algorithm seem to be the cost and constraint   
   satisfaction sections of the code.  The entire algorithm, of course,   
   performs much better if more candidate solutions are evaluated.  I was   
   thinking that if I could somehow approximate (1) the cost of a candidate   
   solution, and (2) the likelihood of it violating a constraint, that would   
   allow the algorithm to examine more candidate solutions, thus increasing the   
   probability of a better solution being found.   
      
   The algorithm typically is run for a given amount of CPU time, say, 60   
   seconds, which further emphasizes the need for an efficient way to calculate   
   costs and constraint satisfaction.   
      
   I've done several web searches for solutions to TSP-like problems that   
   involve fuzzy logic, but don't see many things that would apply directly.   
   If anyone knows of a paper that may relate to this, I would greatly   
   appreciate a pointer.   
      
   Thanks very much,   
      
   Len   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)