An on-line forecasting method based on the wavelet-suboptimal multiple fading kalman filtering algorithm is proposed in order to advance the forecasting precision of the gyro′s drift coefficients for its data mutation and strong trend according to the characteristics of the gyro′s drift coefficients non-stationary time series. The experimental results indicate that the gyro′s drift can be precisely forecasted, which can effectively improve the problem of inaccurate results brought by the data mutation and the strong trend.