AI 中文总结
针对多表面空间物体姿态估计的不适定问题，本文采用交互式多模型算法结合光曲线闪光约束，提升了收敛率并验证了混合步骤的必要性。
AI 中文摘要
光曲线反演可用于估计空间物体的轨道、姿态、光学特性及形状。由于光曲线是标量视星等的时间演化，再现给定光曲线的姿态不具有唯一性，因此该估计问题可能是不适定的。卡尔曼滤波中的初始估计会受到影响，且取决于物体特性、观测几何及待估计参数数量，不准确的初始估计可能导致滤波器发散。此前有研究利用光曲线的突变（称为闪光）来约束姿态估计范围，本文将基于闪光的姿态估计方法扩展至多表面物体，此类物体存在多个产生闪光的姿态，姿态估计无法唯一确定。为解决该问题，本文采用交互式多模型（IMM）算法，该算法在估计序列中运行多个并行滤波器并进行模型交互，每个滤波器假设闪光发生在对应表面，通过各滤波器的似然更新模式概率以确定假设的正确性；此外，IMM中的混合步骤通过转移概率矩阵实现滤波器间的交互，使算法能够适应时变闪光源。本文对地球同步轨道上的箱式卫星开展数值模拟，初始姿态误差最高达80度的蒙特卡洛试验表明，所提方法将收敛率从单表面滤波器的10%提升至73%，且混合步骤至关重要，若移除该步骤，收敛率会降至40%。
英文摘要
Light curve inversion enables the estimation of orbit, attitude, optical properties, and shape of space objects. Because a light curve is the temporal evolution of a scalar apparent magnitude, the estimation problem can be ill-posed owing to the non-uniqueness of the attitude that reproduces a given light curve. The initial estimate in Kalman filtering can therefore be sensitive, and, depending on the object properties, observation geometry, and number of estimated parameters, an inaccurate initial estimate may lead to divergence of the filter. A previous study uses a sudden change of light curves, called glint, to constrain the range of attitude estimate. The current paper extends the attitude estimation method using glint for multiple-surface objects. Such objects have multiple attitudes to yield glint, and the attitude estimate is not uniquely determined. To address this issue, this paper employs the interacting multiple model (IMM) algorithm that runs multiple parallel filters with model interaction in the estimation sequence. Each filter assumes that the glint occurs on the corresponding surface. The mode probability is updated by the likelihood of each filter, determining the correctness of the hypotheses. Furthermore, the mixing step in the IMM allows interaction among the filters through a transition probability matrix, enabling adaptation to the time-varying glint source. Numerical simulations are conducted for a box satellite in a geosynchronous orbit. Monte Carlo trials with initial attitude errors of up to 80~deg show that the proposed method improves the convergence rate from 10\% for a single surface filter to 73%, and the mixing step is shown to be essential, since the convergence rate drops to 40% when it is removed.
CommentsAccepted for publication in Advances in Space Research
