Many　regional　power　plants　do　not　possess　the　ability　to　compete　on　provincial　electricity　market　under　present　conditions,　which　are　normally　dispatched　by　regional　power　utility.　The　mathematical　model　of　daily　power　purchase　is　set　up　for　regional　power　utility,　which　takes　the　minimal　total　cost　as　the　objective　function　and　considers　constraints　of　power　balance,　supply　and　demand　balance,　generation　ability　and　transmission　ability,　as　well　as　the　influence　of　network　loss　allocation.　An　improved　chaos-genetic　algorithm　with　faster　evolution　speed　and　better　precision　is　brought　forward　for　it.　The　model　and　algorithm　are　applied　to　the　power　purchase　of　a　region.　Results　show　that,　compared　with　other　methods,　the　cost　with　consideration　of　loss　allocation　is　reduced　by　about　1.9%.