基于贝叶斯动态模型的激光干涉仪漂移误差预测

Drift Error Prediction of Laser Interferometer Based on Bayesian Dynamic Model

针对长时间测量时激光干涉仪产生的漂移问题,提出了一种基于贝叶斯动态模型的漂移误差预测方法。通过采集非测量阶段的漂移误差序列,分析了数据特征并据此建立了贝叶斯动态模型。然后,利用无信息先验分布法获取了递推所需的初始信息,运用贝叶斯递推算法训练并估计了模型状态参数。最后,基于实验室自制的迈克尔逊激光干涉仪对漂移误差预测效果进行了实验验证。结果表明:与常用的最小二乘拟合法和神经网络建模法相比,基于该方法预测补偿后的残余漂移误差的均方差分别减小了67.85%和99.08%;由重复实验的预测效果可知,残余漂移误差的均方差较补偿前都减小了82%以上。验证了提出的基于贝叶斯动态模型的漂移误差预测方法的有效性和稳健性。

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 has been 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.