Abstract:  State estimation is an important part of the control part of the dynamic positioning system. In order to keep the ship in the target position, the accurate ship status information is needed. In view of the strong nonlinear problems such as extended Kalman filter and unscented Kalman filter, which are filtered by Gauss approximation and cannot adapt to ship motion, an improved particle filter algorithm based on Bayesian estimation is adopted. In order to reduce the degeneracy and impoverishment of standard particle filtering, MCMC(Markov Chain Monte Carlo) is introduced to construct Markov chains to produce samples from target distribution, and to reduce the correlation between particles. The simulation results show that the improved particle filter can separate the low-frequency motion information from the measurement information containing high-frequency motion information and noise. The filtering accuracy is higher and the stability is better.

Key words: MCMC, particle filter, dynamic positioning, state estimation

摘要： 状态估计是动力定位系统控制部分重要的一环，要让船舶保持在目标位置就需要获得准确的船舶状态信息。针对扩展卡尔曼滤波和无迹卡尔曼滤波等滤波依赖高斯逼近而无法适应船舶运动这类强非线性问题，选用基于贝叶斯估计的改进粒子滤波算法。针对标准粒子滤波的退化和贫化问题，引入马尔科夫链蒙特卡罗算法，构造马尔可夫链产生来自目标分布的样本，降低粒子间的关联性。仿真结果表明，改进的粒子滤波能够从包含高频运动信息和噪声的测量信息中分离船舶低频运动信息，滤波精度较高，稳定性较好。

关键词: MCMC, 粒子滤波, 动力定位, 状态估计