In order to solve the problem that the initial alignment of low-cost micro inertial measurement unit (MIMU) on unmanned ships under severe sea conditions, which is not only affected by complex disturbances but also the random noise and bias instability of MIMUs, the variational Bayesian cubature H_∞ filter algorithm (VBCH) was proposed. The simulation results show that while both the system noise and measurement noise are "time-varying" and "thick-tail", the VBCH algorithm is still robust. It is also shown that as MIMU's random noise and bias instability increasing, which makes the system noise and state equations gradually uncertain, the VBCH algorithm can maintain alignment accuracy while the robust adaptive algorithms based on Kalman filter algorithm gradually fail. Besides, for unmanned ships with low accuracy MIMU, the alignment accuracy of upward misalignment angle can reach 1.5°, and the alignment accuracy of the eastward and northward misalignment angles can reach 0.1°, which can meet the requirements for initial alignment of unmanned ships with large misalignment angles.