An　adaptive　genetic　algorithm　with　neural　network　function　is　presented　and　applied　in　the　optimal　allocation　of　power　system　filters　to　suppress　the　harmonic　pollution　of　power　grid.　A　simplified　model　of　filter　is　established　according　to　the　actual　situation.　With　the　minimum　filter　investment　as　the　objective　function　and　the　node　harmonic　voltage　index　as　the　restriction,　the　improved　genetic　algorithm　is　applied　to　optimize　the　location　and　quantity　of　filter　installation　and　the　parameters　of　filter　installed.　The　calculation　and　simulation　of　an　IEEE　case　prove　the　improved　algorithm　and　model　are　correct　and　feasible.