The invention discloses a DBN (deep belief network)-GA neural network-based fault detection method of a high-voltage circuit breaker. The method specifically includes the following process: using current data, which are obtained by monitoring of an online monitoring system, as input variables; then using a deep belief network-based deep learning algorithm to construct a fault type prediction model, determining restricted Boltzmann machine models, marking the same as RBM, extracting partial current data samples to send the same to the constructed models, and carrying out training; carrying outtraining learning on the entire deep belief network model after training on the restricted Boltzmann machines; and finally, inputting all the data into the trained fault type prediction model, and processing the input coil tripping/closing current data by the fault type prediction model to complete fault detection of the high-voltage breaker. According to the method disclosed by the invention, a fault type of the circuit breaker can be judged more accurately and efficiently while compensating for the deficiencies of artificial neural network detection is realized, and then efficient inspectioncan be performed.