Wei Guo’s research while affiliated with Tianjin University and other places

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.

Stock trend prediction is a significant challenge due to the inherent uncertainty and complexity of stock market time series. In this study, we introduce an innovative dual-branch network model designed to effectively address this challenge. The first branch constructs recurrence plots (RPs) to capture the nonlinear relationships between time points from historical closing price sequences and computes the corresponding recurrence quantifification analysis measures. The second branch integrates transposed transformers to identify subtle interconnections within the multivariate time series derived from stocks. Features extracted from both branches are concatenated and fed into a fully connected layer for binary classification, determining whether the stock price will rise or fall the next day. Our experimental results based on historical data from seven randomly selected stocks demonstrate that our proposed dual-branch model achieves superior accuracy (ACC) and F1-score compared to traditional machine learning and deep learning approaches. These findings underscore the efficacy of combining RPs with deep learning models to enhance stock trend prediction, offering considerable potential for refining decision-making in financial markets and investment strategies.

Prolonged fatigue not only affects people’s learning and work efficiency, but also leads to a series of symptoms such as insomnia and forgetfulness. Timely and accurately identifying fatigue state is crucial in various industries. However, existing fatigue detection methods either rely on manually extracted features which is unable to fully utilize the deep level information of signals, or are complexly designed and hard to be implemented. In this paper, we propose spectral group-guided lightweight CNN (SGL-Net), which is an ultra-lightweight CNN model for fatigue detection. The design concept of SGL-Net is closely related to the mechanism of information processing in human brain. Firstly, the spectral space embedding decomposes the EEG signal into various rhythms, we also enrich the decomposition tree using wavelet convolution, where the complex rhythm information is decoupled. Secondly, we propose a novel spatial-temporal modality encoder, which captures the relationship among electrodes and evaluates the power spectrum of different rhythms. Finally, a kernel-restricted multi layer perceptron (MLP) is adopted for fatigue detection, ensuring the parameter sparsity simultaneously. We also design a well-suited hardware system for EEG acquisition on forehead. Experimental results have demonstrated the robustness, effectiveness, and practicability of the SGL-Net in real-world applications.

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.

Underwater glider (UG) plays an important role in ocean observation and exploration for a more efficient and deeper understanding of complex ocean environment. Timely identifying the motion states of UG is conducive for timely attitude adjustment and detection of potential anomalies, thereby improving the working reliability of UG. Combining limited penetrable visibility graph (LPVG) and graph convolutional networks (GCN) with self-attention mechanisms, we propose a novel method for motion states identification of UG, which is called as visibility graph and self-attention mechanism-based graph convolutional network (VGSA-GCN). Based on the actual sea trial data of UG, we chose the attitude angle signals of motion states related sensors collected by the control system of UG as the research object and constructed complex networks based on the LPVG method from pitch angle, roll angle, and heading angle data in diving and climbing states. Then, we build a self-attention mechanism-based GCN framework and classify the graphs under different motion states constructed by a complex network. Compared with support vector machines, convolutional neural network, and GCN without self-attention pooling layer, the proposed VGSA-GCN method can more accurately distinguish the diving and climbing states of UG. Subsequently, we analyze the variation of the transitivity coefficient corresponding to these two motion states. The results suggest that the coordination of the various sensors in the attitude adjustment unit during diving becomes closer and more efficient, which corresponds to the higher network measure of the diving state compared to the climbing state.

Characterizing the nonlinear dynamic behaviors of gas-liquid two-phase flow is a challenging but rewarding research topic. In this paper, a novel multi-layer visibility graph-based ordinal network (MVGON) is proposed for exploring the gas-liquid flow behaviors. In particular, the vertical upward gas-liquid flow experiments are first conducted in a 50 mm inner-diameter pipe and multivariate measurement data is acquired. Then, MVGON is inferred from experimental signals. MVGON not only can effectively fuse multivariate time series but also enables to further excavate concise key information based on inheriting system features extracted by visibility graph. For each derived projection network of MVGON, graph energy and average clustering coefficient are both calculated for quantitatively characterizing the topological structure of MVGON. In addition, permutation entropy and multivariate pseudo Wigner distribution-based time-frequency distribution are calculated for the three typical flow patterns to support our findings. The results indicate that our MVGON framework allows effectively characterizing the nonlinear dynamic behaviors during the evolution of different gas-liquid flow patterns. Meanwhile, it provides a novel approach for characterizing the complex system dynamics based on multivariate time series analysis.

In this paper, we design and fabricate dual-tunable waveguides in a two-dimensional periodic plate with threaded holes. Dual tunability is realized by using rods held with nuts as well as assembly prestress of the nuts. A straight waveguide, a bent waveguide, and a wave splitter are designed by changing the distribution of rods and nuts in different circuits. The experimental and numerical results show that the frequencies of guided waves can be tuned by the assembly prestress. By increasing the amount of prestress, the frequency range of the passing band can be shifted upward. Confinements, guiding, and splitting of Lamb waves are clearly observed in both experimental measurements and numerical simulations. This work is essential for the practical design of reconfigurable phononic devices.