基于智能多模型粒子滤波的运动物体状态估计研究

项目名称： 基于智能多模型粒子滤波的运动物体状态估计研究

项目编号： No.60801056

项目类型： 青年科学基金项目

立项/批准年度： 2009

项目学科： 金属学与金属工艺

项目作者： 杨宁

作者单位： 上海电力学院

项目金额： 20万元

中文摘要： 运动物体状态估计是涉及传感器、运动模型、测量模型、误差和算法等多方面内容的一个研究方向。合理的利用运动模型、分析噪声和采用适当的算法对于状态估计精度有很重要的影响。本项目主要研究如何通过粒子滤波、交互式多模型算法和智能优化算法提高运动物体状态估计的精度。 采用tabu recorder和某些约束规律来确定每个解算循环中的初始路径，增强全局搜索，采用角度区分的方法确定领域，增强局部搜索，改善禁忌算法搜索性能。采用遗传算法和禁忌算法在测量数据处理之前进行初始粒子优化，改善初始粒子质量，用较少的粒子获得较好滤波效果，并在车辆导航中的应用中得到了较好的验证； 针对模型噪声的不确定，采用交互式多模型算法构建多个不同过程噪声值的车辆定位模型进行参数辨识，并应用于全球定位系统和航位推算GPS/DR组合定位系统中，结果显示系统能较好的实现参数识别，并保证了较好的定位精度。

中文关键词： 运动物体; 状态估计; 粒子滤波; 交互式多模型; 智能优化算法

英文摘要： The state estimation of moving object is an important research orientation which refers to sensors, motion model, measurement model, error and algorithms. The right models, appropriate noise analysis and right algorithms would influence the estimation precision. In this research, particle filter, interacting multiple model and intelligent optimization algorithms are used to improve the state estimation precision of moving object. An improved angle-based crossover Tabu search is proposed. The tabu recorder and some constrains are used to confirm the initial routes during every resolution cycle, which improve the global search. The angle-based method is used to confirm the neighborhood which improves the local search. Genetic algorithm and Tabu search are used in particles optimization. A set of particles is chose from a lot of initial sample particles, and kept the similarity randomicity, means and covariance as the initial sample particles. In a land vehicle navigation test, the result shows the improved algorithm can get better results with fewer particles. Because some of parameters, such as process noise can't be confirmed rightly and only can be approximate estimated, the process noise is replaced by three parameters and IMM (Interacting Multiple Model) is used to resolve this problem. The method is applied in a GPS/DR land navigation system. The results prove the method has good performances.

英文关键词： Moving objet; state estimation; particle filter; interacting multiple model; intelligent optimization algorithm