With　the　growing　penetration　of　renewable　power　generation,　the　problem　of　power　quality　disturbances　is　becoming　increasingly　serious.　The　correct　classification　of　power　quality　disturbances　is　a　prerequisite　and　basis　for　managing　and　controlling　power　quality　problems.　Random　forest　is　a　classification　method　with　the　same　weights　of　each　classifier,　which　may　cause　low　accuracy　in　classification.　Therefore,　this　paper　proposes　a　method　to　change　the　weight　according　to　the　accuracy　and　similarity　of　sub-decision　trees,　which　can　solve　the　overfitting　problem.　Firstly,　the　signals　are　preprocessed　by　Fast　Independent　Component　Analysis　(FastICA).　Furthermore,　the　windowed　short-time　Fourier　transform　(STFT)　is　used　to　extract　the　features　from　the　denoised　signals.　Moreover,　four　STFT　features　and　other　four　statistical　features　are　classified　by　weighting　random　forest.　The　weights　of　sub-decisions　tree　are　changed　by　the　classification　accuracy　and　similarity　of　decision　trees.　Finally,　numerical　and　simulated　experiments　have　been　carried　out　based　on　preprocessed　disturbance　signals　from　the　renewable　energy　systems.　The　results　illustrate　the　accuracy　and　effectiveness　of　the　proposed　method.　©　2022　IEEE.