基于重力梯度辅助定位的概率神经网络改进方法  

Improved probabilistic neural network method based on gravity gradient aided location

作　　者：袁赣南[1] 张红伟[1] 袁克非 吴简彤[1] 

机构地区：[1]哈尔滨工程大学自动化学院,哈尔滨150001

出　　处：《中国惯性技术学报》2013年第3期369-374,共6页Journal of Chinese Inertial Technology

基　　金：国家自然科学基金项目(61001154)

摘　　要：为了提高潜器导航定位精度,针对等值线算法在惯导系统初始误差较大时易发散的问题,提出基于概率神经网络调优的等值线改进方法。首先,在搜索区域内,利用概率神经网络算法对惯导系统航迹进行调优,并经过卡尔曼滤波器与惯导系统航迹进行信息融合形成待匹配航迹;在此基础上利用实时等值线算法得到最佳匹配位置。分别在不同初始条件下进行仿真分析,得出概率神经网络算法在大的初始误差下不易发散但定位精度不高的结论,然后在潜器行驶6 h后,初始误差为5.438的条件下进行仿真验证,结果表明,改进方法定位精度均值优于0.537,从而证明改进方法是有效的,即使在大的初始误差下仍然能够达到较高的定位精度。The iterated closest contour point（ICCP） algorithm is liable to divergence if the Inertial Navigation System（INS） initial error is large. In order to solve this problem and improve the navigation location precision, an improved ICCP method based on probabilistic neural network（PNN） optimization was proposed. First, in the search area, a PNN algorithm was used for INS track optimization and an awaiting matching track was formed by INS information fusion through Kalman filter. Based on this, a real-time ICCP algorithm was used to obtain the best matching position. In variable initial error conditions, it is proved that the PNN algorithm does not arouse divergence easily when with large initial error but has no high location precision. Then the improved method is verified by simulation after vehicle has run 6 h and the initial matching error is 5.438？ , and the results show that the average positioning accuracy of the improved method is superior to 0.537？ , which proves that the improved method is effective, and a high location precision can still be achieved even when with large INS initial error.

关 键 词：辅助导航 重力梯度 概率神经网络算法 等值线算法 潜器 

分 类 号：U666.1[交通运输工程—船舶及航道工程]