Aiming at the problems of low accuracy and random noise in angular velocity signals of conventional Micro-Electro-Mechanicals gyroscopes, a signal processing method based on multivariate variational mode decomposition (MVMD) optimized by four-vector intelligent metaheuristic (FVIM) is proposed in this paper. Firstly, the collected angular velocity signals are processed by MVMD, and divided into two components of frequency and energy. Secondly, the FVIM is utilized to optimize the two key parameters of modal decomposition, namely the number of modes k and the penalty factor α, so as to obtain the optimal combination of parameters. Then, the signals are decomposed into multiple intrinsic mode function (IMF) components, and sample entropy (SPN) combined with cosine similarity is taken as the joint decision criterion to classify the IMF components into noise components, aliasing components and effective components. Furthermore, the wavelet transform (WT) algorithm is adopted to process the aliasing components, and the processed components are reconstructed with the screened effective components to obtain the denoised angular velocity signals. Finally, experimental verification is conducted by applying the proposed method to both simulation signals and actual measured signals. The results show that the root mean square error is reduced by 57.1% during simulation signal processing, and the random walk values of the measured Y-axis and Z-axis angular velocity signals are decreased by 76.9% and 56.2% respectively.