Aiming　at　the　noise　interference　and　incorrect　order　in　the　low-frequency　oscillation　identification　of　wide-area　measurement　system,　a　method　based　on　the　improved　wavelet　threshold　de-noising　and　RCRSV-MP(Relative　Change　Rate　Singular　Value　ordered　Matrix　Pencil)　algorithm　is　proposed　to　identify　the　low-frequency　oscillation　modes　of　power　system.　Based　on　the　wavelet　de-noising,　the　improved　threshold　varies　with　the　increase　of　decomposition　levels　to　effectively　suppress　the　noise　of　low-frequency　oscillation　signals.　The　de-noised　signals　are　then　identified　by　the　RCRSV-MP　algorithm　to　obtain　the　parameters　of　each　oscillation　mode.　The　order　set　by　the　RCRSV　is　adaptive　and　no　manual　threshold　is　needed.　Research　results　of　simulation　example,　test　system　and　actual　grid　case　show　that,　compared　to　other　methods,　the　proposed　method　has　better　anti-noise　performance,　higher　fitting　accuracy　and　stronger　practicability　in　the　on-line　identification.　©　2017,　Electric　Power　Automation　Equipment　Press.　All　right　reserved.