Aiming　at　the　problem　of　colored　noise　interference　generated　by　wide　area　measurement　signal　through　the　filter　in　process　of　low　frequency　oscillation　identification　in　power　system,　a　method　combined　with　covariance　matrix,　second　order　derivatives　method　and　total　least　squares-estimation　of　signal　parameters　via　rotational　invariance　techniques　(TLS-ESPRIT)　for　identification　of　low　frequency　oscillation　in　power　system　was　proposed.　Firstly,　signal　was　collected　through　the　band-pass　filter　to　remove　trend　and　high　frequency　noise.　Then,　the　covariance　matrix　of　sample　matrix　was　constructed　as　a　new　sample　to　eliminate　the　effect　of　colored　noise.　Finally,　TLS-ESPRIT　algorithm　was　used　to　identify　modal　parameters.　The　second　order　derivative　method　was　adopted　in　the　problem　of　determining　order,　making　order　adaptive,　without　artificially　designating　threshold.　The　simulation　results　show　that　compared　with　other　methods,　COV-TLS-ESPRIT　algorithm　has　the　advantages　of　anti-noise　performance　and　excellent　fitting　accuracy,　which　can　effectively　and　accurately　identify　the　dominant　modes　from　environmental　noise　with　strong　practicability.　This　algorithm　can　also　realize　on-line　identification　conveniently.　©　2017,　Editorial　Board　of　Journal　of　Huazhong　University　of　Science　and　Technology.　All　right　reserved.