The invention discloses a fast starting method for an MEMS gyroscope based on composite learning, which is used for solving a technical problem that the existing MEMS gyroscope modal control method ispoor in practicability. The technical scheme is that firstly a composite adaptation law of a fuzzy logic weight is designed according to a fuzzy prediction error and a tracking error, and the weightcoefficient of the fuzzy logic is modified to achieve effective dynamic estimation for the unknown dynamics; and meanwhile, a sliding mode controller is designed according to a sliding mode hypersurface and a double-exponential approach law so as to achieve feedforward compensation for the unknown dynamics and enable a vibration error of a detection mass block to be quickly converged. According tothe invention, the prediction error and the tracking error are considered, the composite learning updating law of the fuzzy logic weight is designed, the weight coefficient of the fuzzy logic is modified, and effective dynamic estimation for the unknown dynamics is achieved. The sliding mode controlled designed according to the sliding mode hypersurface and the double-exponential approach law isintroduced, so that the vibration error of the detection mass block is enabled to be quickly converged, thus a requirement of fast start of the gyroscope is met, and the practicability is good.