As　the　main　protection　and　control　equipment　of　the　power　system,　the　high-voltage　circuit　breaker　are　required　to　be　disconnected　instantaneously　within　a　few　milliseconds.　Once　it　fails,　it　will　seriously　threaten　the　safety　of　the　power　grid.　In　this　paper,　a　new　high-voltage　circuit　breaker　fault　diagnosis　algorithm　based　on　denoising-stacked　autoencoder　is　proposed.　Firstly,　the　acceleration　sensor　is　used　to　collect　the　vibration　signal　of　the　high　voltage　circuit　breaker.　The　high　voltage　circuit　breaker　fault　signal　data　are　collected　during　equipment　failure　in　the　laboratory　simulation　experiment　and　site　field　operation.　This　non-stationary　random　vibration　signal　is　then　denoised　and　processed　using　the　Hilbert-Huang　transform.　Since　the　on-site　vibration　signal　is　derived　from　data　from　different　voltage　levels　and　equipment　manufacturers,　it　is　necessary　to　clean　the　data　firstly.　Finally,　the　denoising-stacked　autoencoder　is　used　to　perform　automatic　feature　extraction　and　pattern　recognition　classification　on　the　preprocessed　data.　Automatic　feature　extraction　reduces　the　dependence　of　traditional　artificial　feature　engineering　on　expert　knowledge　as　much　as　possible,　and　makes　full　use　of　fault　features,　thus　improving　the　accuracy　of　diagnosis　and　the　generalization　ability　of　the　model.　©　2019　IEEE.