基于COMGRU的AUV航路轨迹预测方法
An AUV route trajectory prediction method based on COMGRU
徐鹏 1徐东 2李腾涛 2赵宏瑞 2赵佳媛2
作者信息
- 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
- 折叠
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.
Abstract
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Key words
autonomous underwater vehicle(AUV)/trajectory prediction/gated recurrent unit(GRU)network/data compression/space-time trajectory/multi-step prediction/hysteresis引用本文复制引用
基金项目
装备预研重点实验室基金项目(6142215200309)
出版年
2024
国家科技期刊平台
NSTL
万方数据