Abstract:Aiming at the drift problem of laser interferometer during long time measurement, a drift error prediction method based on Bayesian dynamic model is proposed. By collecting the drift error sequence of the non-measurement stage, the data characteristics are analyzed and a Bayesian dynamic model is established. Then, the initial information required for recursion is obtained by using the non-information prior distribution method, and the state parameters of the model are trained and estimated by the Bayesian recursive algorithm. Finally, the drift error prediction effect is verified by experiments based on the Michelson laser interferometer made in the laboratory. The results show that the mean square error of the compensated residual drift error is reduced by 67.85% and 99.08%, respectively, compared with the usual least squares fitting method and neural network modeling method. According to the prediction results of repeated experiments, the mean square error of residual drift error is reduced by more than 82% compared with that before compensation. The validity and robustness of the proposed drift error prediction method based on Bayesian dynamic model are verified.