基于时变 TVAR模型和CKF 滤波的助推器落点预测

Impact Point Prediction with Combination of Time-varying AR Model and Cubature Kalman Filter

摘要:

为解决助推器难以精确回收的问题,提出了一种容积卡尔曼滤波(CKF)和时变自回归(TVAR)模型融合的助推器落点预测方法。针对外弹道观测数据的非平稳时序特点,利用TVAR模型对其建模,预测助推器脱落时和助推器落地之间一段时间的未来测量值,以离散化质点弹道模型作为状态方程,将未来测量值作为CKF滤波弹道位置估计的测量值。为普适非平稳序列,考虑时变TVAR对非平稳时间序列的时变参数和模型阶数的确定。该方法是预测助推器落点滤波外推法的一种新实践。实验数据结果表明,TVAR预测助推器落点与TVAR-CKF融合预测的助推器落点相比,融合后预测的结果与实际测量的助推器落点的偏差更小,可为实际应用提供参考。

Abstract:

To recover the booster accurately, when the naval gun was going to train the shooting toward the sea or land, wepropose an impact point prediction algorithm, which is a combination of time-varying AR model(TVAR) and cubature kal?man filter(CKF). Due to the non-stationary time series characteristics of observed data, TVAR was used to model the pro?jectile’s outer ballistic trajectory. TVAR model can forecast the future measurement values a period of time between thebooster falling off and the booster landing. The future measurement values were the input measurements of the discrete bal?listic particle trajectory model. the cubature kalman filter(CKF) would estimate the ballistic position, so a predict problemconverted into a filter estimation problem. For the study of the universal non-stationary series model, the time-varying pa?rameters and the order of the model were considered. The prediction method is a new practice in the field of the filtering ex?trapolation method. The experimental results of the deviation between the impact point and the actual impact point showedthat the method can be a practical application as a reference.