Shupe effect will affect the gyro bias in variable temperature environment, and establishing a linear model to make compensation is a commonly auxiliary method in engineering. In this paper, firstly, the correlation between gyro output drift and temperature was analyzed, and then temperature, temperature gradient and high-order terms and cross terms were used as the independent variable set. For the increase of the number of independent variables in the compensation model, the range of the output drift after the compensation of the fiber gyro has the lowest value, it is proposed to determine the number of independent variables based on the optimal subset regression. The multivariate polynomial regression model was established to approximate the relationship between zero-bias output and temperature and its correlation. The experimental and simulation results show that when the ambient temperature changes from -40℃ to 60℃, the model reduces the final drift range by 71.05% and the zero offset output by 94% after 100s moving average processing. It effectively reduces the influence of temperature on the gyro bias, and has the advantages of occupying less resources and presenting good real-time performance.
