In　order　to　deal　with　the　problem　of　slow　search　speed　and　premature　convergence,　a　flexible　particle　swarm　optimization　algorithm　is　proposed.　Simulations　have　been　done　to　illustrate　that　this　algorithm　can　not　only　significantly　speed　up　the　convergence,　but　also　effectively　solve　the　premature　convergence　problem.　Furthermore,　the　algorithm　is　applied　to　neural　network＇s　training　in　the　agent　model　in　comparison　shopping　and　the　simulation　experiment　not　only　shows　that　compared　with　related　algorithms,　the　hybrid　algorithm　which　is　based　on　the　flexible　particle　swarm　optimization　and　BP　algorithm　can　quickly　converge　to　a　reasonably　good　solution,　but　also　makes　the　agent　model　in　comparison　shopping　more　effectively.　©　2007　IEEE.