Existing　fault　diagnosis　of　converters　needs　to　train　a　large　number　of　fault　samples　to　realize　fault　identification　and　location,　but　fault　data　acquisition　is　difficult　and　the　training　process　is　complicated.　This　paper　presents　an　algorithm　combining　quadratic　B　spline　wavelet　analysis　based　on　Mallat　(fast　recursive　algorithm)　and　kernel　density　estimation　to　solve　above　problems.　Firstly,　the　output　voltage　of　the　converter　is　pre-processed　with　quadratic　B　spline　wavelet　analysis　based　on　Mallat　to　enhance　anti-noise　ability　and　reduce　data　dimension.　Then,　the　classifier　based　on　kernel　density　estimation　is　used　to　identify　and　locate　fault　samples.　The　proposed　method　can　identify　fault　samples　accurately　without　fault　samples　in　fault　identification　stage,　and　only　a　few　number　of　fault　samples　in　fault　location　stage　need　to　train　for　realizing　fault　location.　The　method　has　advantages　of　high　reliability,　simple　implementation　and　high　classification　performance.　Simulation　and　experiment　verify　feasibility　and　effectiveness　of　the　proposed　method.　©　2019,　Power　System　Technology　Press.　All　right　reserved.