Abstract: For forecasting the gyro drift tendency of a missile, a prediction model based on LSSVM was established taking the time series of gyro's drift as study object, and an online algorithm for nonlinear system based on hermite matrix inversion was proposed. The mathematical model of regression LSSVM, and online learning algorithm were introduced. Based on the characteristic that the reproducing kernel matrix is Hermite and positive definite, a new online learning algorithm was proposed by matrix block-inversion. The algorithm was applied to prediction research on real gyro drift data of certain missile. The experiment results show that the algorithm fully utilizes the historical training results, reduces storage space and calculate, has a fast operation speed and a high prediction precision.

Key words: nonlinear system, least squares support vector machine, adaptive identification, gyro drift prediction, Hermite matrix inversion