A Hybrid Long Short-Term Memory and Kalman Filter Model for MIMU Inertial Data Fusion
IEEE Sensors Journal · 01.05.2026 · 26 (9) · 13666-13676
DOI: 10.1109/jsen.2026.3677375 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Sensors Journal, erschienen am 01.05.2026. Verfasst von Jixiang Lu und 4 weiteren Autor:innen. Die Arbeit nennt 40 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Natural Science Foundation of Shaanxi Province.
Abstract (Originalsprache)
Microinertial measurement unit (MIMU) is playing an increasing role in multiple domains, but performance limitations constrain widespread deployment in high-end applications. In this study, we present an innovative data fusion strategy for MIMU array that enhances the performance of micro inertial sensors. The introduced hybrid model combines the unique strengths of long short-term memory (LSTM) and Kalman filter (KF). LSTM analyzes long-term dependencies in MIMU output data, while KF can generate accurate short-term state estimations. The LSTM generated prediction serves as new observation, which is then ingeniously integrated with KF to obtain the high-quality state estimation. Furthermore, leveraging the nonlinear modeling capability of neural network, the hybrid model enhances the bias accuracy of MIMU array under unknown disturbances. The experimental results demonstrate that LSTM-KF model is remarkably superior to conventional KF and LSTM. Under static conditions, root-mean-square error (RMSE) evaluation results across three axes are $0.0013~^{\circ }$ /s, $0.0015~^{\circ }$ /s, and $0.0012~^{\circ }$ /s, respectively, representing a maximum improvement of 7.0 times compared with KF. The efficacy of LSTM-KF model is fully verified via dynamic tests. In the field test, it also effectively mitigates uncharacterized disturbances stemming from some factors in actual environment. The improvement achieved with LSTM-KF is better than KF and Sage-Husa KF. This method provides a low-cost solution for high-precision inertial measurement.
Förderung
- National Natural Science Foundation of China
- Natural Science Foundation of Shaanxi Province
Zitieren
APA
Lu, J., Xue, L., Cai, G., Yang, B., & Wang, X. (2026). A Hybrid Long Short-Term Memory and Kalman Filter Model for MIMU Inertial Data Fusion. IEEE Sensors Journal, 26(9), 13666-13676. https://doi.org/10.1109/jsen.2026.3677375
Vancouver
Lu J, Xue L, Cai G, Yang B, Wang X. A Hybrid Long Short-Term Memory and Kalman Filter Model for MIMU Inertial Data Fusion. IEEE Sensors Journal. 2026;26(9):13666-13676. doi:10.1109/jsen.2026.3677375
BibTeX
@article{Lu2026,
author = {Lu, Jixiang and Xue, Liang and Cai, Guangbin and Yang, Bo and Wang, Xinguo},
title = {A Hybrid Long Short-Term Memory and Kalman Filter Model for MIMU Inertial Data Fusion},
journal = {IEEE Sensors Journal},
year = {2026},
volume = {26},
number = {9},
pages = {13666--13676},
doi = {10.1109/jsen.2026.3677375}
}
EUVIMED