For the finiteness and nonstationarity of the gyro testing data, The Autoregressive ( AR) model and Hidden Markov Model (HMM) are employed to forecast the gyro drift. Firstly, AR model is used to derive the AR coefficients of the gyro drift data as eigen element, for its sensitivity to the change of condition; and then the Hidden Markov Model with mixture-of-Gaussians output is trained by the preceding results; finally, the gyro drift tendency is predicted by the modified weighed prediction method, which resolves the problem of gyro drift prediction under small sample condition. The influence of different rank on weighed model and number of states on HMM are also investigated in simulation, which also demonstrated the validity of the proposed method.%针对陀螺仪实验数据的有限性和非平稳性,提出了基于自回归(AR)模型和隐马尔科夫模型(HMM)的陀螺漂移预测方法.首先利用AR模型参数能够敏感状态变化规律的特性,提取陀螺漂移数据的自回归系数作为特征量；然后对具有混合高斯输出的HMM进行训练；最后对陀螺仪的状态进行加权预测,改进了趋势预测的方法,解决了陀螺漂移在小样本数据条件下的预测问题.实验分析了加权模型阶数和HMM状态数对陀螺漂移预测结果的影响,并验证了预测方法的有效性.
展开▼
