Based on Automatic Identification System (AIS) data in maritime settings, this paper explores the limitations of traditional Recurrent Neural Networks in extracting features from complex vessel trajectory sequences. We propose a fusion model, namely CNN-RGRU-Attention, for vessel trajectory prediction. The model integrates Convolutional Neural Network (CNN), Attention Mechanism, and Gated Recurrent Unit (GRU). The effectiveness of the model is validated using actual AIS data, demonstrating significant improvements in metrics such as Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE). The CNN-RGRU-Attention model provides crucial theoretical support for enhancing the safety management of maritime traffic services.

The magnetic tomography method (MTM) is a non-contact external inspection method for detecting metal magnetic memory signals. It has great potential for application in long-distance oil pipeline and subsea pipeline inspection. However, the spatial distribution characteristics and propagation laws of magnetic signals are not yet clear, which makes the MTM passive detection. In this study, a three-dimensional mathematical model of the magnetic field distribution of the stress concentration zone outside the pipe was established based on the boundary conditions. For the two cases in which the stress concentration zone was located at the top and bottom of the inner wall of the pipe, the model was solved by finite element analysis. The variation law of the magnetic signal outside the pipe was analyzed, and experiments were designed to verify the model. The results show that the shape of the magnetic memory signal remained unchanged after passing through the pipe wall. As the magnetic permeability of the pipe medium is much larger than that of air, the magnetic memory signal is significantly attenuated after penetrating the pipe wall. As the detection height increases, the magnetic induction outside the tube decays exponentially. The results also prove that the magnetic tomography method can detect the stress concentration zone at any position of the pipeline, and the detection accuracy is higher when it is located at the top of the pipeline.

Magnetic tomographic is a new and nondestructive testing technology based on the metal magnetic memory method, which can perform noncontact testing outside a pipeline. To identify the type of defect from the characteristics of the magnetic field signal detected by this method, we performed basic research on the propagation mechanism of a magnetic memory signal outside the pipe. In this paper, a magnetic field model of the stress concentration zone was first discussed based on the magnetic dipole theory. A 3-D formulation of the spatial propagation of the magnetic memory signals detected by the magnetic tomographic method was built, based on the Biot–Savart law and the magnetic vector potential. The reliability of the model was verified by comparing the field testing signal with the model-calculated values. The model can well describe the propagation law of the magnetic memory signal for the stress concentration zone generated on the inner wall of a pipeline.

Buried pipelines require reliable non-destructive inspection, whereas weak magnetic signals measured at the ground surface are strongly affected by lift-off values, material interfaces, and environmental noise. This study investigates the lift-off-dependent statistical dependence of magnetic field profiles generated by a controlled equivalent magnetic dipole source in an X80 steel pipeline. A current-carrying coil was used as a reproducible source surrogate, and axial magnetic-field profiles were measured at lift-off values of 0–150 mm using fluxgate magnetometers. Pearson correlation coefficient (PCC), wavelet squared coherence, mutual information (MI), and empirical attenuation modeling were combined to characterize the relationships between signals measured at different spatial layers. Three-dimensional finite element simulations were further performed to provide a field-distribution reference. Adjacent air-layer signals showed substantial similarity, with reported global PCC values of 0.61–0.83 and peak wavelet squared coherence values of 0.99–1.00 over the investigated conditions. Wall–air signal pairs exhibited lower linear correlation but retained measurable statistical dependence. Their estimated MI exceeded the value predicted by a bivariate Gaussian dependence model with the same PCC, indicating that the signal pairs cannot be fully characterized by a simple Gaussian-linear model. This observation may reflect a combination of interface-related field redistribution, non-Gaussian spatial waveform structure, preprocessing effects, and finite sample estimation characteristics; it should not be interpreted as direct proof of a nonlinear magnetostatic transfer law. An empirical exponential model provided a descriptive fit to the observed PCC variation over the investigated lift-off range. The results establish a controlled source baseline for understanding multi-layer signal similarity and lift-off effects in non-contact magnetic pipeline inspection. The observed inter-layer dependence may support future development of correlation-weighted signal enhancement and combined amplitude-correlation localization methods, which require validation using stress-induced magnetic memory sources and field measurements.