基于改进隐马尔可夫模型的水下目标搜索及投放策略研究  

Underwater target search and placement strategies based on an improved hidden Markov model

作　　者：钱龙霞 李汉霖 王红瑞[2] 洪梅[3] 韩佳 QIAN LongXia;LI HanLin;WANG HongRui;HONG Mei;HAN Jia(School of Science,Nanjing University of Posts and Telecommunications,Nanjing 210023,China;College of Water Sciences,Beijing Normal University,Beijing 100875,China;College of Meteorology and Oceanography,National University of Defense Technology,Changsha 410073,China;Some Unit,Beijing 100081,China)

机构地区：[1]南京邮电大学理学院,南京210023 [2]北京师范大学水科学研究院,北京100875 [3]国防科技大学气象海洋学院,长沙410073 [4]某部队,北京100081

出　　处：《中国科学:技术科学》2025年第3期520-539,共20页Scientia Sinica(Technologica)

基　　金：国家自然科学基金(批准号:52279005);国家重点研发计划(编号:2018YFC0407900);教育部人文社会科学研究规划基金(编号:23YJAZH111)资助项目。

摘　　要：动态海洋环境条件下,基于隐马尔可夫模型(HMM)的水下目标搜索技术仍然存在搜索效率低、计算耗时高和无法高效投放搜索设备等问题.在深入研究确定性寻路算法优势基础上,本文提出HMM-Dijkstra水下目标搜索模型,并设计了两种搜索投放算法.主要建模步骤为:首先确定投放个数,对水下目标的隐蔽性评估数据做逆序排序并依据分位数选择栅格点作为投放位置;其次,基于K近邻(KNN)原理对投放位置上的搜索者进行分区;最后,对代价函数进行调整,提出改进Dijkstra算法,该算法可以自动确定终点并不断迭代至结束,输出一组搜索路径.三种地图规格条件下的模拟实验均表明,本文算法在绝大多数情况下比现有算法的搜索性能提升10%~50%,并且各情况下均节省耗时1~2个数量级.基于水下目标隐蔽性评估数据和声场数据的仿真实验表明,投放算法提高了搜索性能的稳定性.最后进行敏感性分析,研究各种场景下的投放数量参考值.对于10×10栅格地图:(1)如果隐蔽性评估数据可靠,水下目标不动(或规律移动)时,投放数量建议值为3(或4);(2)如果隐蔽性评估数据不可靠,则相应投放数量建议值变为3和5(或4).对于20×20栅格地图:若为情况(1),相应投放数量建议值为5和2(或4);若为情况(2),相应投放数量建议值为6和5(或1).Under dynamic marine environmental conditions,underwater target search algorithms based on hidden Markov models(HMMs)still have problems such as low search efficiency,long computation time,and inability to efficiently deploy search equipment.This paper presents the HMM-Dijkstra underwater target search model and designs two search and placement algorithms based on in-depth research on the advantages of deterministic routing algorithms.The main modelling steps are as follows:First,the number of targets to be placed is determined,the concealment evaluation data of the underwater targets are sorted in reverse order,and grid points are selected as placement locations based on quantiles;Second,based on the K-nearest neighbour(KNN)principle,the searchers at the placement locations are partitioned;Finally,the cost function is adjusted,and an improved Dijkstra algorithm is proposed,which can automatically determine the endpoint and continuously iterate to the end,outputting a set of search paths.Simulation experiments under three different maps show that the proposed algorithm improves the search performance by 10%to 50%compared with existing algorithms in the vast majority of cases and reduces the time by 1–2 orders of magnitude in each case.Simulation experiments based on underwater target concealment evaluation data and sound field data show that the placement algorithm improves the stability of the search performance.Finally,a sensitivity analysis is conducted to study the reference values for the number of placements in various scenarios.For a 10×10 grid map,(1)if the concealment assessment data are reliable and the underwater target is stationary(or moves in a regular way),the recommended value for the number of placements is 3(or 4);(2)if the concealment assessment data are unreliable,the recommended values for the corresponding number of placements is 3 and 5(or 4).For the 20×20 grid map,if case(1)above holds,the recommended values for the corresponding placement quantity are 5 and 2(or 4);in case(2),the reco

关 键 词：隐马尔可夫模型 水下动态目标搜索 路径规划 栅格地图 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]