To reduce random drift of the fiber optic gyro

a model was set up using time series analysis method

aimed at weak non-stationary characteristics of random drift. A global optimum algorithm for selecting model order was put forward. It changed two-dimensional search program into one-dimensional one. Consistent estimation of model order was achieved. An improved parameter estimation algorithm was put forward. It transformed the non- linear parameter estimation process into a linear one. Both measured sequence and reverse one were used to estimate parameters

so data information was fully utilized and the accuracy of parameter estimation was improved. The goal of parameter estimation is to get the minimum sum of forward and backward filtering error squares that was found in（ p + 1） -dimensional space. The trend terms

periodic terms

autocorrelation and partial autocorrelation of measured drift were analyzed. The statistical results of non-stationary models established from three ways verified the validity of above algorithms. And the results show that the way to identify drift model directly works better than the way to identify random model and non-stationary drift model respectively commonly used in engineering.