An Approach Based on BiLSTM-Transformer for Marine Inertial Navigation

In the Global Navigation Satellite System (GNSS) denial environment, for the problem of fast dispersion of the positioning error of pure inertial navigation system (INS), this paper proposes a hybrid deep learning model based on bidirectional long and short term memory network (BiLSTM) and Transformer. The model establishes the mapping relationship between IMU timing data and location information through a multi-level feature fusion architecture.First, a causal convolutional layer is used to extract local dynamics features. Second, BiLSTM is utilized to capture the bidirectional long-term dependency of sensor data. Finally, the cross-channel feature interaction and dynamic weight allocation are realized by the multi-head self-attention mechanism, which improves the accuracy of the model.In order to verify the effectiveness of this method, this study conducted sea trials and collected data under different navigation conditions. The experimental results show that the proposed scheme effectively reduces the error accumulation of the inertial navigation system compared with the traditional method in pure inertial autonomous positioning, which proves the reliability and universality of the method, and has the value of popularization and engineering application.