Multivariate time series (MTS) anomaly detection is of great importance in both condition monitoring and malfunction identification within multi-sensor systems. Current MTS anomaly detection approaches are typically based on reconstruction, prediction or association discrepancy learning algorithms. These methods detect anomalies by learning hidden representations of entire sequences, modeling dependencies at a single time-step level, or calculating an association-based metric inherently distinguishable between regular and deviant points. However, most existing methods typically fail to leverage all three types of models simultaneously to enhance overall performance, as well as often disregard the correlations between different sensors. To address the issues above, this paper proposes a novel deep learning-based unsupervised MTS anomaly detection algorithm called Association Discrepancy Dual-decoder Transformer (AD2T). AD2T employs a dual-decoder architecture to accommodate reconstruction, prediction, and association discrepancy learning tasks, thereby effectively utilizing information across these tasks to better characterize MTS data. We further develop a minmax training strategy to jointly optimize all the aforementioned tasks. Additionally, we propose a compound embedding module based on dilated causal convolution to simultaneously capture correlations in both temporal and sensor dimensions. Extensive empirical studies on five multi-sensor system datasets from the aerospace, server, and water treatment domains have demonstrated the superiority of our method, achieving an average improvement of 1.96% in F1-score compared to state-of-the-art (SOTA) methods.

As a new type of unmanned autonomous marine observation platform, underwater gliders (UGs) offer advantages such as low energy consumption and long operational ranges. However, during the gliding process, the complex marine environment often leads to abrupt changes in short-term control parameters, complicating the data and rendering them more challenging to predict. This typically poses difficulties in adjusting control parameters based on human experience, thereby significantly reducing UG control efficiency. To address this issue, this paper proposes a novel method termed DFFormer, aimed at enhancing the predictive accuracy of the rudder angles during UG motion. The proposed method integrates discrete wavelet transform (DWT) for rudder angle signal decomposition and employs a fast Fourier transform-based attention mechanism (FFT-Attention) to effectively capture and analyze its frequency- and time-domain characteristics. Notably, the method leverages a Transformer architecture to process the decomposed signals through multiple parallel pathways, substantially improving the capability to forecast the complex and variable control parameters of UGs. The effectiveness and practicality of the proposed method are demonstrated through actual sea trials. The experimental results indicate that the proposed method surpasses traditional approaches in terms of accuracy and computational efficiency, exhibiting superior performance in predicting UG control parameters and, to some extent, enhancing their heading-keeping ability.

Mesoscale eddies have attracted increased attention due to their central role in ocean energy and mass transport. The observations of their three-dimensional structure will facilitate the understanding of nonlinear eddy dynamics. In this paper, we propose a novel framework, the mesoscale eddy characterization from ordinal modalities recurrence networks method (MeC-OMRN), that utilizes a Petrel-II underwater glider for in situ observations and vertical structure characterization of a moving mesoscale eddy in the northern South China Sea. First, higher resolution continuous observation profile data collected throughout the traversal by the underwater glider are acquired and preprocessed. Subsequently, we analyze and compute these nonlinear data. To further amplify the hidden structural features of the mesoscale eddy, we construct ordinal modalities sequences rich in spatiotemporal characteristics based on the measured vertical density of the mesoscale eddy. Based on this, we employ ordinal modalities recurrence plots (OMRPs) to depict the vertical structure inside and outside the eddy, revealing significant differences in the OMRPs and the unevenness of density stratification within the eddy. To validate our intriguing findings from the perspective of complex network theory, we build the multivariate weighted ordinal modalities recurrence networks, through which network measures exhibit a more random distribution of vertical density stratification within the eddy, possibly due to more intense vertical convection and oscillations within the eddy's seawater micelles. These framework and intriguing findings are anticipated to be applied to more data-driven in situ observation tasks of oceanic phenomena.