The invention discloses a sub-health prediction method for an urban mass transit vehicle door system. The method comprises steps that motor acquisition data of the urban mass transit vehicle door system is pre-processed; segmental characteristic extraction of the pre-processed motor acquisition data is carried out; data standardization and discreteness processing on extracted segmental characteristics are carried out; a sub-health state model for the urban mass transit vehicle door system is acquired; according to the sub-health state model, online prediction is carried out, the online clustering result is compared with the sub-health state model established in an offline mode, and a health state of the present urban mass transit vehicle door system is acquired through prediction. The method is advantaged in that before the door system is in a fault, the sub-health degeneration state can be acquired, a modeling method is simple, multiple of known sub-health states can be predicted, during online application, unknown sub-health states can be discovered through a clustering algorithm, and prediction of the sub-health states of the urban mass transit vehicle door system can be realized.