There　are　redundancy　and　poor　separation　ability　in　the　power　quality　disturbance　feature　sets,　which　leads　to　the　low　classification　accuracy　of　the　power　quality　disturbance　signals.　Aiming　at　this　problem,　a　feature　selection　for　the　power　quality　disturbance　signals　is　proposed.　First,　the　Hilbert-Huang　transform　is　used　to　extract　the　frequency　domain　features,　and　the　set　of　all　features　of　power　quality　disturbance　signals　is　constructed.　Then,　the　rules　of　the　feature　subset　selection　are　constructed　based　on　the　indexes　of　the　intersection　degree,　the　redundancy　degree　and　the　separation　degree,　and　the　selected　feature　subsets　are　obtained　by　the　improved　cuckoo　search　method.　After　that,　the　cost　factor　is　defined　based　on　the　subset　dimension　and　the　classification　accuracy　of　the　feature　subset,　so　as　to　evaluate　the　performance　of　the　feature　subset　in　different　dimensions,　and　then　the　feature　subset　with　the　lowest　cost　factor　is　selected　as　the　optimal　feature　subset.　Finally,　the　optimal　feature　subset　is　used　to　train　the　classification　model　and　classify　the　power　quality　disturbance　signals.　The　simulation　results　show　that　the　proposed　feature　selection　has　better　performance　in　obtaining　a　subset　of　features　with　smaller　dimensions　and　in　classifying　the　power　quality　disturbance　signals.　©　2023　Power　System　Technology　Press.　All　rights　reserved.