I am doing research on genetic algorithms for solving the bin packing problem. I can understand the process now, but since the final output is a set of solutions for one list of items, I cannot figure out why do we need a set of solutions for one list of items when one solution should be enough? What are the applications where a GA solution would be better than classical approaches?
It would be great if someone could refer me to any scholarly/non-scholarly links that explain some practical applications of bin packing using genetic algorithms. I have visited the link for list of GA applications on wikipedia but it is not specifically for bin packing.
Background
The classical version of bin packing is a well understood problem for which relatively large instances can be efficiently solved to optimality or near-optimality using methods such as integer programming with column generation.
However, these models may not be as effective for solving special cases of bin packing which exhibit complex constraints or objectives (e.g., bin packing with conflicts or profits, multidimensional bin packing, bin packing with fragile objects, bin packing with load balancing, etc.).
In your case
You don't need a set of solutions, it's just that the way the genetic algorithm (GA) is designed, you end up with a set of solutions (your current population) once you stop its execution. You just pick the best of those solutions.
One advantage of GAs over classical methods for bin packing would be its ability to efficiently solve problems with complex constraints. For instance, here is a paper which uses GAs to solve the three dimensional single container arbitrary sized rectangular prismatic bin packing optimization problem (isn't that a mouthful to say!). While classical methods are usually very efficient on convex problems such as traditional bin packing, once you add non-convex constraints they become much harder to solve. For such problems, other methods such as GAs (among others) tend to do very well.