Scholarlink

<p>Integration of an inertial navigation system (INS) with a global navigation satellite system (GNSS) is known as a significant strategy for reducing time-growing errors of position, velocity, and attitude information in moving objects over land, sea, and air. A deep-learning-aided filter is proposed to solve the issue of interrupted/blocked signals of GNSS while the limitations imposed by long-term errors of MEMS-grade inertial sensors lead to time-increasing positioning errors in the INS. Under concentrated noisy conditions and delayed measurements together with a low update rate of data epochs, a 12-layer long short-term memory (LSTM) together with a recurrent fuzzy wavelet (RFW) network is proposed for enhancement of the Kalman filter (KF). Therefore, successful extraction of long-term dependency of INS yields improved velocity and localization data during GNSS disruptions. Correct prediction of autonomous vehicle motion trajectory as target performance of the designed LSTM-RFW-KF method is assessed through real-world flying test data of an agricultural drone. Investigation of tests shows that the proposed deep learning algorithm can effectively capture the long-term chronological dependency of INS outputs. Appropriate accurate predictions in INS velocity and position data during GPS interruption are obtainable.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Increasing performance of INS/GNSS using LSTM-recurrent fuzzy wavelet kalman filter

  • 期刊论文

  • 发表日期：

    2025年6月30日

查看全文
  • Reza Safvat,
  • Jafar Keighobadi

摘要

Integration of an inertial navigation system (INS) with a global navigation satellite system (GNSS) is known as a significant strategy for reducing time-growing errors of position, velocity, and attitude information in moving objects over land, sea, and air. A deep-learning-aided filter is proposed to solve the issue of interrupted/blocked signals of GNSS while the limitations imposed by long-term errors of MEMS-grade inertial sensors lead to time-increasing positioning errors in the INS. Under concentrated noisy conditions and delayed measurements together with a low update rate of data epochs, a 12-layer long short-term memory (LSTM) together with a recurrent fuzzy wavelet (RFW) network is proposed for enhancement of the Kalman filter (KF). Therefore, successful extraction of long-term dependency of INS yields improved velocity and localization data during GNSS disruptions. Correct prediction of autonomous vehicle motion trajectory as target performance of the designed LSTM-RFW-KF method is assessed through real-world flying test data of an agricultural drone. Investigation of tests shows that the proposed deep learning algorithm can effectively capture the long-term chronological dependency of INS outputs. Appropriate accurate predictions in INS velocity and position data during GPS interruption are obtainable.