一种新的基于LSTM的航位推算算法  

A Novel Dead Reckoning Algorithm Based on LSTM

作　　者：涂杰 黎杰 刘凯 TU Jie;LI Jie;LIU Kai(No.92246 Troops of PLA,Shanghai 200000;Naval Aviation University,Yantai 264001)

机构地区：[1]中国人民解放军92246部队,上海200000 [2]海军航空大学,烟台264001

出　　处：《舰船电子工程》2023年第9期111-115,共5页Ship Electronic Engineering

摘　　要：针对传统航位推算(DR)算法中构建的运动模型简化程度较高,对运动信息的预测精度有限,制约DR算法性能的问题,提出一种新的基于长短时记忆网络(Long Short Term Memory Network,LSTM)的DR算法,即LSTM-DR算法,该算法根据仿真实体的序列运动信息,以及仿真实体运动信息的预测误差,利用深度神经网络的表征能力,实现对仿真实体运动规律的拟合,从而提升仿真实体运动模型的预测精度。实验结果表明,所提出的LSTM-DR算法,使运动模型训练后能获得良好的预测精度,从而在保持仿真实体运动的连续性和平滑性的同时,大幅降低对通信资源的消耗。A new DR algorithm based on long short term memory(LSTM)Network,namely LSTM-DR algorithm,is proposed to solve the problem that the motion model constructed in the traditional DR algorithm is highly simplified and the prediction accuracy of motion information is limited,which restricts the performance of DR algorithm.According to the sequence motion information of the simulation entity and the prediction error of the simulation entity motion information,the algorithm uses the characterization ability of the deep neural network to achieve the fitting of the motion law of the simulation entity,so as to improve the prediction accuracy of the simulation entity motion model.The experimental results show that the proposed LSTM-DR algorithm can achieve good prediction accuracy after the motion model is trained,thus greatly reducing the consumption of communication resources while maintaining the continuity and smoothness of the motion of the simulation entity.

关 键 词：LSTM DR 交互仿真 仿真实体 循环神经网络 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]