Pitch　system　fault　prediction　and　improvement　of　prediction　accuracy　are　key　technologies　for　wind　power　development,　which　ensure　safe　operation　of　the　grid　and　effectively　reduces　operation　and　maintenance　costs.　The　Supervisory　control　and　data　acquisition　(SCADA)　system　data　is　analyzed　and　processed　to　extract　the　associated　parameters,　i.e.　output　power,　wind　speed,　pitch　angle,　and　rotor　speed.　A　Back　Propagation　(BP)　neural　network　is　used　to　train　the　system,　taking　into　account　the　volatility　and　uncertainty　of　wind　turbine　parameters,　and　a　regression　prediction　model　with　a　support　vector　regression　(SVR)　algorithm　is　also　used　for　training.　A　pitch　failure　prediction　model　is　established　to　predict　the　operation　of　the　pitch　system,　which　is　used　to　develop　a　reasonable　operation　and　maintenance　plan.　Through　the　system　simulation,　the　prediction　model　performance　index,　error-index,　and　output　data　graphics　are　compared　and　analyzed.