The invention discloses a high-voltage circuit breaker fault diagnosis method based on an unsupervised learning model. The high-voltage circuit breaker fault diagnosis method comprises the following steps: obtaining a coil current waveform in a switching-off and switching-on process of the high-voltage circuit breaker, and obtaining the state parameter of the high-voltage circuit breaker according to the waveform; sampling a vibration signal of the high-voltage circuit breaker, and selecting the time node t of the high-voltage circuit breaker; classifying and numbering the faults of high-voltage circuit breaker as the output of a fault model identification system, and taking coil current, the vibration signal and the time node as the input of the fault model identification system; taking a denoising self-decoding algorithm as a typical unsupervised learning model, training the model, adopting a SVM (Support Vector Machine) structure, , and obtaining a regression expression of a circuit breaker fault through a loss function obtained by the denoising self-decoding algorithm; and according to the regression expression of the circuit breaker fault, obtaining a corresponding relationship between coil current data and a fault type when a fault happens, and judging the fault type through fault data to be determined. The training optimization process of the high-voltage circuit breaker fault diagnosis method can avoid a local optimal solution.