基于COMGRU的AUV航路轨迹预测方法  

An AUV route trajectory prediction method based on COMGRU

作　　者：徐鹏[1] 徐东[2] 李腾涛 赵宏瑞 赵佳媛 XU Peng;XU Dong;LI Tengtao;ZHAO Hongrui;ZHAO Jiayuan(China Shipbuilding Industry Group Co.,Ltd.,Lianyungang 222006,China;College of Computer Science and Technology,Harbin Engineering University,Harbin 150001,China)

机构地区：[1]中国船舶重工集团公司,江苏连云港222006 [2]哈尔滨工程大学计算机科学与技术学院,黑龙江哈尔滨150001

出　　处：《哈尔滨工程大学学报》2024年第7期1384-1390,共7页Journal of Harbin Engineering University

基　　金：装备预研重点实验室基金项目(6142215200309).

摘　　要：针对采用神经网络预测自主水下机器人航迹存在滞后性的问题,本文提出一种基于信息压缩的改进门控循环神经网络,用于水下自主机器人航路多步轨迹预测。该算法将水下自主机器人航行轨迹附近的障碍物位置信息、海流信息以及时空轨迹信息共同构成的地理位置信息进行数据压缩处理,作为本文预测网络的输入,以提高网络训练效率。实验验证该算法减少了水下自主机器人航迹多步预测的滞后性且具有较高的准确率。To address the issue of hysteresis in predicting the trajectory of autonomous underwater vehicles(AUVs)using neural networks,this paper proposes an improved gated recurrent unit(GRU)network based on information compression for multi-step voyage path prediction.The algorithm compresses geographical location information,which includes obstacle location information,sea current information,and space-time trajectory information near the AUV′s voyage path.This compressed information is used as the input for the prediction network,enhancing the network′s training efficiency.Experiments confirm that the algorithm effectively reduces hysteresis and achieves high accuracy in multi-step AUV trajectory prediction.

关 键 词：水下自主机器人 航迹预测 门控循环神经网络 数据压缩 时空轨迹 多步预测 滞后性 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]