Chuang Shi

In global navigation satellite system-denied underground environments, developing high-precision positioning technologies faces dual measurement challenges: magnetic aliasing effects and structural repetitiveness. Although magnetic fingerprinting provides an effective alternative, its measurement reliability is often compromised by sensor noise and environmental magnetic interference. This paper proposes multi-scale dilated dynamic time warping (MS-DDTW)-Net, an artificial intelligence-enhanced framework designed for robust two-stage positioning. The first stage employs a MS-DDTW algorithm to capture hierarchical temporal patterns, effectively mitigating positioning deviations caused by magnetometer drift. The second stage utilizes a Siamese network based on a CNN-GRU architecture for deep semantic verification, mapping measurement sequences into a discriminative metric space to resolve spatial ambiguity and suppress erroneous matching. Experimental results in real-world underground facilities demonstrate that the proposed method achieves a mean absolute error (MAE) of 3.472 m and a root mean square error of 5.771 m, outperforming traditional dynamic time warping and k-nearest neighbors methods by 44.74% and 65.39% in terms of MAE, respectively. This hybrid framework significantly enhances measurement reliability and positioning robustness, offering a novel technical solution for localization in complex underground environments.

Indoor geomagnetic positioning utilizes the pervasive magnetic field as a natural signal source for navigation. However, the nonuniqueness of magnetic signatures and environmental interference introduces significant measurement ambiguity, hindering high-precision instrumentation in complex environments. Furthermore, the lack of quantitative metrics for assessing the “localizability” of a magnetic map limits the reliability of existing positioning systems. To address these measurement challenges, this article proposes a quality-driven positioning framework that integrates a robust signal matching strategy with a novel fingerprint quality evaluation system. First, a dual-index evaluation system comprising a discrimination index (DI) and a variation index (VI) is established to quantitatively characterize the spatial distinguishability and temporal consistency of the magnetic field data. This system demonstrates a strong coefficient of determination (R ${}^{2}~=0.91$ ) with the mean overall positioning error (MOPE), providing a theoretical basis for predicting measurement uncertainty. Second, to mitigate signal mismatches in repetitive environments, we introduce a modulus-weighted constrained derivative dynamic time warping (CDDTW) method. This method incorporates physical constraints of the magnetic modulus into the matching process to enhance coarse-level signal retrieval. Subsequently, a Multiscale BiGRU-Attention Siamese Network is employed as a fine-grained metric learning model to extract robust features from noisy sensor data. Experimental results indicate that the proposed framework achieves a localization accuracy of better than 3 m. Compared to traditional baseline methods, the proposed approach improves accuracy by over 35% and significantly reduces large error outliers, validating its effectiveness in enhancing the robustness and reliability of geomagnetic measurement systems.

The geocenter motion (GCM), associated with the degree‐1 component of surface mass redistribution in the Earth's fluid envelope, is difficult to observe with sufficient precision. Estimating GCM through the degree‐1 deformation approach assumes that seasonal Global Navigation Satellite System (GNSS) variations are primarily induced by surface mass loading. However, this is not the case for GNSS displacements due to the presence of prominent non‐loading errors. For a refined estimation of GCM, we modeled and mitigated three types of non‐loading errors, including bedrock thermoelastic deformation, GNSS draconitic errors (DRE), and background noises, in GNSS displacements derived from the International GNSS Service third reprocessing. Results demonstrate that thermoelastic deformation is a significant contributor to annual variations in the Z component with an amplitude of approximately 1.8 mm. Prominent non‐seasonal scatters in the X and Y components are also significantly reduced by removing DRE and filtering out background noises. Besides, an abnormal fluctuation in the X component over the period 2012–2014 has also been mitigated. Overall, by accounting for non‐loading errors, the GNSS‐derived GCM becomes more consistent with independent GCM estimates from the geophysical loading model, the method combined Gravity Recovery and Climate Experiment and Ocean Bottom Pressure data, and Satellite Laser Ranging. Taking the geophysical loading model as an example, the percentages of GNSS‐derived GCM variances that can be explained are remarkably improved from (35%, 60%, and 48%) to (75%, 68%, and 73%) in the X, Y, and Z components, respectively. Accurate modeling of non‐loading errors can provide a perspective for obtaining refined geocenter estimates solely relying on GNSS displacements.

The ionosphere poses a significant source of error in satellite‐based navigation systems for aviation and radio communication applications. Accurate estimation of the total electron content (TEC) can effectively mitigate the impact of such errors. However, constrained by observational techniques, the acquisition of global ionospheric TEC in practical applications relies heavily on high‐precision forecasting products. In this study, we construct a global ionospheric forecasting model based on the global ionospheric TEC products disseminated by the International GNSS Service (IGS) and deep learning. We incorporate an attention module to extract global spatiotemporal features from historical ionospheric data and employ these features to predict the TEC values over the next 24 hr. Additionally, we select a long short‐term memory (LSTM) model and a ConvLSTM model as baseline models for comparison, conducting experiments under varying solar activity conditions. The experimental results demonstrate that RA‐ConvLSTM model outperforms the other two models in quantifying the performance of the models. During high solar activity years, the bias and Root Mean Square Error (RMSE) of RA‐ConvLSTM model are −0.0298 TECU and 3.8980 TECU, respectively, while during low solar activity years, these values are 0.0905 TECU and 1.5059 TECU, marking a notable improvement over the comparative models. Furthermore, by contrasting the precision of the three forecasting models during geomagnetic storms, the RA‐ConvLSTM model exhibits the least fluctuations in accuracy, indicative of a higher degree of stability in its forecasting outcomes.