ABSTRACT. The Dynamic Communication Network (DCN) describes the interactions over time among various communication nodes, and it is widely used in Big-data applications as a data source. As the number of communication nodes increases and temporal slots accumulate, each node interacts in with only a few nodes in a given temporal slot, the DCN can be represented by an High-Dimensional Sparse (HDS) tensor. In order to extract rich behavioral patterns from an HDS tensor in DCN, this paper proposes an Adaptive Temporal-dependent Tensor low-rank representation (ATT) model. It adopts a three-fold approach: a) designing a temporal-dependent method to reconstruct temporal feature matrix, thereby precisely represent the data by capturing the temporal patterns; b) achieving hyper-parameters adaptation of the model via the Differential Evolutionary Algorithms (DEA) to avoid tedious hyper-parameters tuning; c) employing nonnegative learning schemes for the model parameters to effectively handle an the nonnegativity inherent in HDS data. The experimental results on four real-world DCNs demonstrate that the proposed ATT model significantly outperforms several state-of-the-art models in both prediction errors and convergence rounds.

ABSTRACT. Deep neural networks typically impose significant computational loads and memory consumption. Moreover, the large parameters pose constraints on deploying the model on edge devices such as embedded systems. Tensor decomposition offers a clear advantage in compressing large-scale weight tensors. Nevertheless, direct utilization of low-rank decomposition typically leads to significant accuracy loss. This paper proposes a model compression method that integrates Variational Bayesian Matrix Factorization (VBMF) with orthogonal regularization. Initially, the model undergoes over-parameterization and training, with orthogonal regularization applied to enhance its likelihood of achieving the accuracy of the original model. Secondly, VBMF is employed to estimate the rank of the weight tensor at each layer. Our framework is sufficiently general to apply to other convolutional neural networks and easily adaptable to incorporate other tensor decomposition methods. Experimental results show that for both high and low compression ratios, our compression model exhibits advanced performance.

ABSTRACT. Data characterized by high dimensionality and sparsity are commonly used to describe real-world node interactions. Low-rank representation (LR) can map high-dimensional and sparse (HDS) data to low-dimensional feature spaces and infer node interactions via modeling data latent associations. Unfortunately, existing optimization algorithms for LR models are computationally inefficient and slowly convergent on large-scale datasets. To address this issue, this paper proposes an Accelerated Asynchronous Parallel Stochastic Gradient Descent A2PSGD Algorithm for High-Dimensional and Sparse Data Low-rank Representation with three fold-ideas: a) establishing a lock-free scheduler to simultaneously respond to scheduling requests from multiple threads; b) introducing a greedy algorithm-based load balancing strategy for balancing the computational load among threads; c) incorporating Nesterov's accelerated gradient into the learning scheme to accelerate model convergence. Empirical studies show that A2PSGD outperforms existing optimization algorithms for HDS data LR in both accuracy and training time.

ABSTRACT. Today, numerous web services with similar functionalities are available on the Internet. Users often evaluate the Quality of Service (QoS) to choose the best option among them. Predicting the QoS values of these web services is a significant challenge in the field of web services. A Canonical Polyadic (CP)-based tensor network model has proven to be efficient for predicting dynamic QoS data. However, current CP-based tensor network models do not consider the correlation of users and services in the low-dimensional latent feature space, thereby limiting model’s prediction capability. To tackle this issue, this paper proposes an Extended Canonical polyadic-based Tensor Network (ECTN) model. It models the correlation of users and services via building a relation dimension between user feature and service feature in low-dimensional space, and then designs an extended CP decomposition structure to improve prediction accuracy. Experiments are conducted on two public dynamic QoS data, and the results show that compared with state-of-the-art QoS prediction models, the ECTN obtains higher prediction accuracy.

ABSTRACT. A second-order-based latent factor (SLF) analysis model demonstrates superior performance in graph representation learning, particularly for high-dimensional and incomplete (HDI) interaction data, by incorporating the curvature information of the loss landscape. However, its objective function is commonly bi-linear and non-convex, causing the SLF model to suffer from a low convergence rate. To address this issue, this paper proposes a PID controller-incorporated SLF (PSLF) model, leveraging two key strategies: a) refining learning error estimation by incorporating the PID controller principles, and b) acquiring second-order information insights through Hessian-vector products. Experimental results on multiple industry HDI datasets indicate that the proposed PSLF model outperforms four state-of-the-art latent factor models based on advanced optimizers regarding convergence rates and generalization performance.

ABSTRACT. Cryptocurrency transaction networks (CTNs) are frequently encountered in real applications. Generally, it is impossible to observe the full interactions among all nodes at each time slot due to practical limitations, so there are numerous missing links in CTNs. A Non-negative Latent Factorization of Tensors (NLFT)-based link prediction model has proven to be efficient in predicting the missing links of a temporal network. However, existing NLFT models fail to model the evolution relationship of a temporal network, which restricts model prediction capability for temporal links. To address this issue, this paper proposes an Autoregressive-incorporated Non-negative Latent-Factorization of Tensors (ArNLFT) model. Its core idea is to adopt an autoregressive model to represent the evolution relationship of temporal network, thereby building an autoregressive-incorporated objective function. Moreover, a non-negative parameters learning scheme based on single latent factor-dependent, non-negative and multiplicative update rule is designed to ensure the nonnegativity of the proposed model. Eventually, the effectiveness of the ArNLFT model is verified by performing temporal links prediction task on two real CNTs, and the results show that the ArNLFT achieves significant accuracy gain compared to its peers.

ABSTRACT. This paper presents a novel approach to electric load forecasting by integrating multifactor prediction methods with time series analysis. The proposed CNN-LSTM model leverages historical electric load data for time series forecasting while incorporating various external factors to improve prediction accuracy. We construct a comprehensive feature pool, combining original load data with empirical variables such as temperature, seasonal trends, and holidays. The CNN component extracts salient features from the data, while the LSTM component captures temporal dependencies. Our experimental results, based on data from 26 transformers over 41 months, demonstrate that the CNN-LSTM model outperforms traditional methods, achieving higher accuracy in electric load forecasting.

ABSTRACT. Accurately identifying key quality features can significantly streamline the manufacturing process by reducing the number of controlled variables and enhancing product quality prediction. To maximize prediction accuracy while minimizing the number of features, this study proposes a Machine-learning and Discrete Multi-verse-optimizer-based Hybrid method called MDMH. MDMH combines machine learning and the discrete Multi-verse optimizer, utilizing hierarchical clustering for effective feature grouping and employing a wrapper technique to find the optimal solution. After comparing its results with those of the exact solver and other intelligent optimization methods, the proposed method demonstrates superiority in terms of accuracy.

ABSTRACT. Human resource allocation in parallel assembly processes (PAPs) is an important optimization problem. It can be regarded as a resource-constrained project scheduling problem with specific constraints. In this paper, a mixed-integer programming (MIP) model is established for it to minimize the maximum completion time (i.e., makespan) by comprehensively analyzing the characteristics of a PAP, including temporal constraints: worker periodic unavailability and the tail time of operations; and spatial constraints: workstation and human resource constraints. To address this problem, a heuristic rule for prioritizing idle workers that considers the precedence relationships of operations is designed as a baseline method, and an iterative greedy algorithm (IGA) is further proposed to provide high-performance solutions. These methods aim to shorten the cycle and costs of a PAP. The effectiveness of the proposed algorithms is validated through extensive experiments.

ABSTRACT. The flexible job shop scheduling problem is a typical NP-hard problem, which has strong guiding significance for production scheduling in enterprises. In this paper, we propose an improved genetic algorithm based on knowledge driven and variable neighborhood search to solve the problem. On the one hand, the proposed algorithm repeatedly extracts the knowledge, which contains the information of the current best machine selection solution, from an updated elite solution set and uses it to guide the mutation of machine selection. On the other hand, a threshold on the number of iterations is set and a variable neighborhood search is employed to explore better solutions for the elite class once the threshold is reached. The experimental results on benchmark test instances indicate the superiority of our proposed algorithm over those reported in recent literature.

ABSTRACT. To achieve high diagnostic efficiency and precision, this paper proposes a novel fault diagnosis method that integrates multi-domain feature extraction and BPNN. First, the characteristic curves of photovoltaic (PV) array current-voltage (I-V ) and power-voltage (P-V ) are analyzed to identify key feature parameters such as voltage, current and power. Traditional fault detection methods often rely on unimodal information in either the time domain (Dt) or frequency domain (Df ), leading to insufficiently comprehensive characteristics. Therefore, this study employs multi-domain feature extraction across the time domain (Dt), frequency domain (Df ) and time-frequency domain (Dtf ) to construct a multi-domain feature set. Variational Mode Decomposition (VMD) is used to avoid over-decomposition during feature extraction in the Dtf . Second, a fault diagnosis method is developed by optimizing the neural network weights and thresholds using the Circle Adaptive-Grey Wolf Optimizer (CA-GWO) algorithm. Experimental results demonstrate that the proposed multi-domain BPNN-based approach effectively and accurately detects faults such as short-circuit (Fsc), open-circuit (Foc), abnormal aging (Faa) and local shading (Fls) in PV arrays. Furthermore, extensive testing shows that the proposed method significantly improves the accuracy and speed of fault diagnosis compared to classical BPNN methods.

ABSTRACT. With the improvement of people's living standards, the annual global production of waste continues to rise, but the traditional household waste classification methods are burdened with a heavy task due to the wide variety of waste types and the difficulty of identifying them. Deep learning based waste image classification methods can accurately classify the waste images. In this paper, Visual Geometry Group (VGG16) is used to handle multi-category household waste classification. In order to further improve its classification accuracy, an attention mechanism is incorporated into it, and thus a VGG16 deep learning model with attention mechanism (VGG16-AM) is proposed. Experimental results show that the classification accuracy of our proposed model on the waste dataset is significantly improved to 93\% compared to other deep learning algorithms.

ABSTRACT. Most 3D detection methods relying solely on LiDAR are inevitably affected by the sparsity of point clouds. Given that color images can provide abundant additional details to enrich LiDAR data, many studies have explored the fusion of LiDAR points and color images to improve 3D object detection accuracy. However, these approaches have not fully addressed the issue of LiDAR point sparsity, and different data representations between images and point clouds pose challenges for effective fusion, often leading to suboptimal performance. This work proposes a novel 3D object detection framework, which integrates two key components: pseudo point generation and object detection by fusing pseudo and LiDAR points. For the first component, we employ a twin surface estimation method to realize depth completion, which applies a fuzzy model based on the binary ambiguity hypothesis, resulting in the estimation (foreground and background surfaces) of twin surfaces. By fusing these twin surfaces with appropriate weighting, this work generates a dense depth map to produce pseudo-points enriching a sparse LiDAR point cloud. For the second component, this work introduces a fusion strategy called 3D grid-wise focused fusion, which effectively combines the pseudo-point cloud and the original LiDAR point cloud for object detection. We evaluate our approach with a benchmark KITTI dataset, which outperforms traditional 3D object detection methods relying solely on LiDAR data.

ABSTRACT. In response to the increasing complexity of inventory management in large-scale warehouses, we present a drone-assisted quick response (QR) code recognition technology for warehouse management. It combines panoramic stitching and image preprocessing for QR code recognition. The experiments are conducted and performed in two phases. First, we compare three basic QR code recognition methods based on Pyzbar, OpenCV, and dynamsoft barcode reader (DBR), with experimental results indicating that DBR performs the best. Second, building on the DBR method, we compare the recognition effects of different image preprocessing techniques, namely the method proposed in this paper, geometric correction, sharpening (SH) and adaptive histogram equalization (AHE), and the median filter algorithm. The results show that the proposed method excels in average processing time per batch (τ ), stitching success rate (η), and the number of recognized QR codes (ζ). The average processing time per batch is 3.13 seconds, the stitching success rate (η) is 63.43%, and totally 746 QR codes are recognized. It approaches the performance of geometric correction methods with its advantage in processing speed. It provides significant practical value and innovative solutions for enhancing efficiency in modern warehouse management.

ABSTRACT. In recent years, Dynamic Vision Sensors (DVS) have garnered significant attention within the research community, primarily due to their rapid response times, adaptive perceptual capabilities, and notably low power consumption. However, the inherent frameless nature of DVS poses a challenge, rendering traditional camera-based rain removal methodologies inapplicable. This paper introduces an innovative event density-based clustering approach tailored for rain removal in DVS environments. The proposed method entails segmenting the neighborhood surrounding DVS output events into multiple regions, followed by the computation of event density within each region. These computed densities serve as pivotal features for clustering purposes. Notably, the class exhibiting the minimal variance and mean of event density is delineated as the rain class. This classification strategy enables the method to adeptly accommodate diverse hardware configurations and scene variations, thereby facilitating efficient rain information removal. To further enhance the methodology, a novel Skeleton Extraction Density Segmentation (SEDS) technique is proposed to calculate event density. The efficacy of the approach is substantiated through applying the Gaussian Mixture Model (GMM) clustering method, which is systematically compared against the K-means clustering method. This comparative analysis provides a comprehensive quantitative evaluation of clustering performance across an array of images.

ABSTRACT. Human-machine collaborative assembly is commonly used in industrial assembly process, while the existing assembly guidance methods have fixed processes, complex guidance procedures and low guidance accuracy. Based on this, in order to improve the success rate and quality of assembly, a digital twin-driven human-machine collaborative assembly guidance method system is proposed, and an inverse guidance mechanism for the assembly process is established, which realizes the on-line accurate identification of the parts to be assembled through the improved lightweight YOLO model, and generates the geometric reference features of the parts for the assembly through the OpenCV. The effectiveness of the proposed method is verified by taking the assembly process of a bipolar expansion helical cylindrical gear reducer as an example, and the advancement of the proposed method is verified by comparing it with the algorithms of the same type. The experimental results show that the proposed assembly guidance method is easy to implement, low cost and high accuracy, which is of great significance for improving the accuracy and flexibility of the human-machine collaborative assembly process.

ABSTRACT. Point Cloud Completion aims to reconstruct the complete 3D shapes based on partial 3D point clouds. Existing methods typically require either complete point clouds or multiple partial views of the same object. However, complete 3D point clouds or diverse partial point clouds for large-scale aerospace components are extremely challenging to acquire in practical industrial scenarios due to their complex geometries and substantial sizes. In this paper, we present a novel self-supervised framework with feature augmentation called FASSC for the aerospace field. Unlike traditional methods, FASSC only operates on a single incomplete point cloud per object. Specifically, this method divides the incomplete point cloud into different patches as input and predicts the missing patches by learning prior information from various parts of objects. By incorporating the Region-Aware Chamfer Distance and Normal Consistency Constraint, FASSC eliminates the need for complete point clouds. We also propose an Attention Feature Augmentation Module (AFA) that establishes spatial relationships among points by introducing self-attention mechanism to reveal the detailed geometry of 3D shapes. Experimental results demonstrate that our approach exhibits notable enhancements on multiple datasets.

ABSTRACT. Image-based cyberbullying introduces added complexity compared to text-based forms, presenting new challenges that traditional methods struggle to address. Owing to the rapid advancement of deep learning, computer vision technology has become a powerful tool to address this issue. In recent years, Convolutional Neural Networks (CNNs) have gained prominence for image processing and recognition because of their excellent feature extraction abilities. Initially, this paper conduct a comprehensive review of the literature in the fields of computer vision and cyberbullying detection. This paper provides an extensive comparison of the performance among different CNN architectures, including VGG, ResNet, and WRNs, on a dataset of cyberbullying images. It shows and discusses that the VGG architecture outperforms other architectures in terms of solution quality and computational efficiency.

ABSTRACT. This work studies an emergency routing scheduling problem for the forest fire patrol. It aims to plan the optimal routes for patrolling the high-risk places with priority levels. The objective of the problem is to minimize the total travel time of all the patrol vehicles, so that the forest fire can be discovered at early time if it happens. To study the problem, an integer programming (IP) model is formulated for it. Then a fast tabu search based algorithm is proposed to solve the problem. Numerical experiments are conducted and computational results are compared with CPLEX. It is shown that the formulated model is effective and the proposed algorithm is efficient to obtain near-optimal solutions, which can provide useful information for emergence management for the decision makers.

ABSTRACT. Generalized Voronoi Diagrams (GVD) are extensively utilized in path planning, particularly in algorithms based on RRT*, where they guide the sampling process and thereby enhance the search efficiency of RRT*. Traditional methods for constructing GVD, however, often demand substantial computational resources, especially in complex and large-scale environments. In this article, we propose an efficient GVD generation algorithm leveraging Convolutional Neural Networks (CNN). By harnessing the learning capabilities of CNN, the model can swiftly predict the corresponding GVD probability map from a input image containing obstacle information. The training dataset is derived from traditional GVD computation methods, with GVD pruning applied to complex environmental scenarios. The training outcomes indicate that for simple environments, the CNN-based GVD generation aligns closely with the results obtained via traditional methods. In complex environments, the model is also capable of accurately predicting smooth and efficient GVD, consistent with manually pruned GVD results. Simulation results demonstrate that the model’s predictions can be seamlessly integrated into the RRT* algorithm, providing effective sampling heuristics. This integration leads to a significant improvement in GVD computation speed compared to traditional methods and exhibits robust generalization capabilities in complex environments.

ABSTRACT. With the increasing prevalence of automated guided vehicles (AGVs) in industry, the AGV dispatching has emerged as a research hotspot. Recently, matrix manufacturing systems (MMS) have been favored for their suitability in mass-customized production. However, fluctuating and diverse transport demands in MMS pose challenges for effective AGV scheduling. This paper investigates a multi-AGV dispatching problem (AGVDP) in MMS, involving time-constrained pick-up and delivery tasks. A mixed-integer linear programming model is developed to minimize total transportation costs, including travel distance, early arrival time, and utilized AGVs. To achieve this, an effective adaptive large neighborhood search algorithm (ALNS) is proposed. In ALNS, a scope-based nearest neighbor heuristic is introduced for initial construction. Diverse problem-specific operators are designed to generate large neighborhoods. To enhance ALNS's robustness to instance variations, an adaptive simulated annealing acceptance criterion is proposed. Computational experiments using a well-established benchmark validate the superiority of the proposed ALNS.

ABSTRACT. In this study, we address a new dynamic scheduling-location problem (DSLP), which is that incorporates movable machines . The objective is to designed to optimize job scheduling and machine placement dynamically. For this strongly NP-hard problem, we formulate the problem as a mixed-integer linear programming (MILP) model. To further improve its computational efficiency and tighten the lower bounds for faster convergence to optimal solutions, we propose several valid inequalities based on the characteristics of the DSLP. Numerical experimental results for a case and randomly generated instances demonstrate significant improvements in system efficiency by allowing machine movement. The findings highlight the potential of extending traditional location scheduling frameworks to include dynamic elements, offering new insights and practical implications for complex industrial applications.

ABSTRACT. In semiconductor manufacturing, a cluster tool consists of loadlocks, several processing chambers and a robot for transporting wafers. During wafer fabrication, a chamber needs cleaning for removing residual gases and solid particles to reduce risk of contamination wafer surface. Chamber cleaning makes the scheduling problem of cluster tools more complicated, which leads to deadlock. The method based on virtual wafer can avoid such problem, but this method also reduces chamber utilization. Aiming to improve the throughput of cluster tools with chamber cleaning, this paper modifies robot sequences and adjust the processing time of virtual wafer. Then, based on the time characteristics of the system, a linear programming model is established to achieve the optimal solution of scheduling. Finally, illustrative examples are given to verify the feasibility and effectiveness of the proposed method. Compared with the existing virtual wafer method, this method can effectively reduce the system cycle time and improve the production efficiency of dual-arm cluster tools with chamber cleaning.

ABSTRACT. Robotic cluster tools have been applied for wafer fabrication. To improve the productivity of such a tool, one method involves increasing the number of spaces within a process module and the fingers on a robot arm, leading to multi-finger-robotic cluster tools with multi-space process modules. This study addresses the scheduling issue of four-finger-robotic cluster tools with four-space process modules, a commonly seen configuration in modern semiconductor fabs. Residency time constraints make the scheduling issue of such a tool more challenging. This study conducts an analysis of the steady-state operations of such a tool under two presented robot task sequences. Then, based on the temporal properties of the tool, two linear programs are developed to find feasible schedules to minimize the system cycle time. Experiments test the performance of a tool under different robot task sequences by considering different parameters.

ABSTRACT. As the underlying network of the industrial control system (ICS), the fieldbus network can prevent attacks from the network. However, attackers can bypass physical defenses and physically connect intrusion devices to the fieldbus network to carry out various attacks. Many existing methods focus on detecting active intrusion devices by extracting their signal characteristics, but they struggle to detect inactive intrusion devices that are performing eavesdropping attacks without sending signals. This paper proposes a reflection-based channel-state group fingerprint to detect inactive intrusion devices. We theoretically analyze the reflection signals generated by the access of the intrusion device and observe that these reflection signals impact the signals of benign devices. Based on this, we extract the signal from a benign device before the intrusion and utilize it as a channel-state fingerprint. We detect inactive intrusion devices by analyzing the channel-state differences before and after intrusion. Additionally, we combine the channel-state fingerprints of multiple groups of devices to improve detection performance. The experimental results show that our method outperforms 99% in all detection metrics when detecting inactive intrusion devices.

ABSTRACT. The advent of big data and cloud computing has ushered in a new era of complexity in cybersecurity. Analysis of network security situations has emerged as a pivotal aspect of safeguarding networks. However, with the escalating volume of alarms and the enhancement of attack techniques, the task of network security analysis has grown increasingly daunting. The power system faces numerous challenges in defending its information security, including analyzing vast amounts of data, mining alarm correlations, and detecting high-risk alarms. In this paper, we propose a long-periodicity detection method for alarms over a long time span, aiming to reduce the number of false alarms. Experimental results on real datasets have demonstrated that we can effectively detect long-period alarms over a long time span. This can enhance the understanding of the overall network security situation and help staff pinpoint high-risk alarms with real threats.

ABSTRACT. As the scale of power systems continues to expand, the number of security alerts generated by intrusion detection systems (IDS) has surged, posing a significant challenge for security operations personnel in identifying genuine network intrusions among vast alerts. Existing methods typically rely on machine learning to classify alerts, but these models often lack interpretability. To address this issue, we propose URL-SemCom, a novel framework that employs a language model to understand the semantic information within the URLs in alerts. We expanded the vocabulary of the language model with commonly used URL vocabulary and designed a tailored enhancement task. We proposed a cost-sensitive strategy to address the poor performance caused by the imbalance of positive and negative samples in the real-world power system data during the model training process. Finally, we used an Adaboosting ensemble classifier to improve the model's accuracy in classifying high-dimensional vectors. Comprehensive experiments demonstrate that our method significantly enhances the effectiveness of alert identification, providing a robust tool for improving cybersecurity measures in power systems.

ABSTRACT. To improve the energy efficiency of the dryer section in the paper making process, it is essential and significant to predict the key parameters of the dryer section accurately. Compared with conventional mechanism-based prediction models, data-driven methods effectively establish the parameter prediction models. This work selects the feature variables from the production data, and then adopts four kinds of data-driven prediction models for comparative analysis. All the prediction models are performed on two testing cases with high and low correlation degrees respectively. The testing results provide useful experience and insights for the operators and managers in the pulp and paper industries.

ABSTRACT. Security constrained unit commitment (SCUC) problem with uncertainty has become an important research task with the increasing scale of renewable energy access in the power system. Multi-stage robust optimization method has been proved to be effective for solving the UC problem with uncertainty, but it has to deal with the difficulty of excessive computation. The complexity of the UC problem is also increasing due to the rapidly expanding overall size of the power system. A common method to improve the efficiency of solving UC problems is to eliminate redundant security constraints. However, this method is mainly applied to the deterministic UC problems at present. In this paper, a solution framework based on a multi-stage robust optimization method with eliminating redundant security constraints is established. Numerical testing is performed for IEEE 24-bus system and 118-bus system. Most of the security constraints are identified as redundant constraints in different scenarios. The computational efficiency is obviously improved after redundant constraints are eliminated.

ABSTRACT. With the increasing penetration of renewable energy sources (RESs) such as wind, photovoltaic, and hydrogen energy, microgrids face significant challenges due to the inherent intermittency and uncertainty of the output of RESs. The core difficulty in microgrid scheduling problems is how to effectively mitigate the uncertainty with operation constraints guaranteed. To tackle this issue, this paper proposes a multi-stage stochastic scheduling model based on the Scenario-Tree method. Specifically, a basic operational model of the microgrid is first established, which includes multiple stages and multiple types of energy storage devices. Secondly, a multi-stage stochastic optimization model based on the scenario tree approach is further developed, employing a real-time rolling economic dispatch strategy. Numerical tests conducted on the proposed model validate its feasibility and economic efficiency in dealing with uncertainties.

ABSTRACT. Petri-net-based information flow analysis is an effective approach to check non-interference and analyze information security. However, Petri nets generally utilize the interleaving semantics, and easily suffer from the state space explosion problem. In this paper, we utilize unfolding techniques in checking non-interference. Specifically, we introduce some theorems to depict the structural behaviors of non-interference in the unfolding, and propose the finite interference-complete prefix (FICP) for checking non-interference. The theorems and FICP model benefit the checking of non-interference, which can alleviate the state space explosion problem.

ABSTRACT. Skeleton-based action recognition is a hot topic, with a focus on developing specialized graph convolutional networks (GCNs) to model 3D skeleton data. In order to facilitate the capture of spatial and temporal dependencies in graph convolutional networks for extracting discriminative features, we propose auxiliary tasks in this paper, including a Spatial-Temporal Masked Reconstruction (STMR) mechanism and a Skeleton Prototype Contrastive (SPC) learning strategy. Firstly, Spatial-Temporal Masked Reconstruction mechanism utilizes the spatial structure and temporal continuity of skeleton joints to reconstruct masked joints, allowing GCNs to capture more valuable patterns and semantics for action recognition. Then, Skeleton Prototype Contrastive learning strategy (SPC) mines the characteristic skeleton features of each action and learn discriminative skeleton representations. Extensive experiments on NTU RGB+D and NW-UCLA datasets demonstrate that the proposed method achieves significant action recognition performance.

ABSTRACT. Continual learning is an emerging field of artificial intelligence (AI) that focuses on equipping models with the ability to adjust to new tasks while retaining previously acquired knowledge. This capability is critical for the development of versatile AI systems. It enables AI to handle dynamic real-world data effectively. Traditional machine learning models often struggle to meet efficiency and resource demands when dealing with changing datasets. Thus continual learning becomes a promising alternative. In this paper, we introduce Knowledge Distillation and Combined Loss Enhanced Continual Learning Network (KDCL), which aims to mitigate catastrophic forgetting and balance the stability and plasticity of continual learning. KDCL combines knowledge distillation and combined loss functions to improve learning efficiency. Through experiments on the CIFAR-100 dataset, KDCL significantly improves the average accuracy compared to existing models, highlighting its capability to retain past knowledge and effectively integrate new information.

ABSTRACT. Graph data are well suited to represent complex interactions and dependencies owing to its inherent relational structure. Graph neural networks (GNNs) can effectively explore the topological structure and node feature information of graph data to perform perceptual inference tasks. In GNNs, the state of each node is iteratively updated by aggregating information from its neighbors. Note that some noise may be introduced due to the information aggregation of irrelevant neighbor nodes, leading to poor performance with higher computational overhead. This paper proposes a dynamic graph generation method based on neighborhood constraints, which learns the degrees of the nodes and imposes the neighborhood constraints to limit the neighborhood space of the nodes, resulting in a sparse graph with dynamic neighborhoods. Based on the sparse graph, the dynamic graph neural network selectively aggregates the information of the most important neighbors to improve the performance of the model. Furthermore, to ensure neighborhood constraints satisfaction during the learning process, a loss function is designed to penalize the nodes whose degrees exceed a specific threshold. The evaluation results demonstrate that our proposed method achieves a higher inference accuracy with the same or higher sparsity compared to state-of-the-art methods.

ABSTRACT. Ultra-wideband (UWB) has been widely studied due to its broad prospects in indoor positioning. However, due to the complexity of indoor environments, UWB signals are inevitably obstructed by objects such as walls, metal doors, and wooden boards during transmission. This causes the UWB signal to introduce non-line-of-sight (NLOS) resulting in degradation of positioning accuracy. The recognition of NLOS is of great significance for improving the positioning accuracy of UWB. Therefore, this paper proposes a simple method to identify NLOS caused by different obstacles blocking UWB signals. Firstly, the channel impulse response (CIR) of UWB signals is extracted and analyzed. Further, the CIR data are transformed into a two-dimensional feature image using a method of Markov transition field (MTF), which is then recognized by a convolutional neural network (CNN) model. Finally, experiments are conducted to verify the reliability of the proposed method in dynamic and static environments. The experiments results show that the recognition accuracy of the MTF-CNN model proposed in this paper is 98.67% under static experiment, which is 13.18% higher than the original CNN algorithm and 92.5% under dynamic experiment.

ABSTRACT. This study investigates the application of the DconnNet model for the segmentation of CT images of visceral organs. Experimental results show that DconnNet significantly outperforms U-Net across various metrics, highlighting its superior capability in accurately delineating complex structures within medical CT images. The DconnNet model improves feature representation by effectively disentangling directional subspaces from the shared latent space, leading to more anatomically consistent segmentation. This method performs exceptionally in maintaining spatial coherence and robustness amidst noise and artifacts, which makes it highly suitable for clinical applications. Despite its higher computational demands, DconnNet’s performance highlights its potential as a powerful tool for medical image segmentation. Further research can focus on reducing computational load, extending its application to other medical imaging modalities, and incorporating advanced learning techniques to further improve its generalizability and efficiency.

ABSTRACT. Accurately predicting the concentration of PM2.5 in the air is of great significance for air quality control and pollution prevention. Due to the high complexity of the air environment and the time series nature of the data, this work proposes combining a Butterworth filter with an attention-based bidirectional long short-term memory network to predict the concentration of PM2.5 in the air. The model employs a Butterworth filter to smooth the sequence, reducing high-frequency noise interference, and uses an attention-based bidirectional long short-term memory network to extract effective information from complex long-term dependencies. Experimental results show that this method outperforms other similar methods.

ABSTRACT. Liquid metal-based pressure sensor is a kind of flexible tactile sensor that senses external pressure stimuli based on piezoresistive effects by encapsulating gallium-based liquid metal in flexible microchannels. Considering many advantages such as high flexibility and good stability, liquid metal pressure sensors are of interest in a variety of fields including health monitoring, medicine, and soft robotics. Due to the low Young modulus of liquid metal, both applied pressure and bending generate nonnegligible strain to change their morphology, which triggers the drift of resistance signals in pressure perception. Thus, this paper proposes a crease-based sensor, which has the characteristic to suppress bending interference. Based on the inspiration of using creases for stress absorption, the groove structure is designed and integrated between microchannels, forming an array of equally spaced grooves. The sensor is made by casting silica gel in two 3D-printed molds that can form the designed structure, injecting and encapsulating liquid metal connecting to electrodes. The cross-sectional area changes of microchannels under different loading conditions are simulated by the Finite element method. Then the resistances of liquid metal can be calculated with the cross-sectional area changes using the law of resistance, revealing the relationships between the detected resistances and applied pressures. The sensing performance of the sensor was tested under various loading conditions, including pure-bending, folding, and compressed on curved surfaces. Simulations and electrical tests prove that the structure can suppress the signal draft of the sensor placed on curved surfaces. The sensor is made of cheap materials and can be directly placed on curved surfaces to detect pressure, avoiding the influence of bending, which shows great application potential in flexible gripper perception, wheeled robot force control, and other aspects.

ABSTRACT. Multi-robot patrolling systems have been studied for guaranteeing maritime safety under severe weather conditions. Their patrolling path planning problem can be established as a large-scale multi-objective optimization problem (LSMOP) due to conflicting objectives and huge numbers of decision variables. Currently, evolutionary algorithms based on divide-and-conquer strategies and dimensionality reduction are frequently used methods to solve such problems. However, the ways that they seek to simplify the problems may ignore the characteristics of decision variables and relevance among them. In this work, we propose a novel mutation operator based on a non-dominated ranking and TextRank-inspired ranking, which forms a dual-level ranking mutation, to enhance the optimization performance of our previous immune algorithm. The new improved algorithm can guide population to mutate towards a more favorable direction by means of the non-dominated ranking and investigate correlations among individuals and genetic positions inspired by TextRank. Comparisons to the state of the art on both of the concerned problem and benchmarks illustrate the desired performance of our proposed method.

ABSTRACT. In the production process of the process industry, precise adjustment of working conditions presents a challenge due to the complexity of processes and unknown disturbances. Central control operators need to adjust setpoints based on deviations in process parameters and monitor target values to maintain system stability. However, many operating procedures excessively rely on human experience, increasing the uncertainty of the production process. In addition, the expert knowledge is not fully embedded in accumulated operations, limiting its potential in decision support. Therefore, data-driven modeling of production processes is essential for developing industrial expert systems to realize intelligent manufacturing. This work proposes a work condition prediction framework based on an Operation Mode Library (OML) to realize Working Condition Prediction (WCP), called for OML-WCP short. Taking the cement rotary kiln adjustment process as an example, a stable OML is constructed using Gaussian mixture clustering technology. Experimental results with real-life operation data of a cement plant reveal that the prediction accuracy of OML-WCP outperforms the existing methods. Moreover, the continuous accumulation of operating mode libraries can improve prediction accuracy in practical applications.

ABSTRACT. Efficient human-robot collaboration necessitates bidirectional perception between humans and robots. For robots, understanding the operator′ s behavior is crucial for enhancing safety and work efficiency in human-robot collaboration. Human limb movements encompass diverse behavioral information, and accurate human motion prediction has attracted significant attention. Existing human motion prediction methods can extract some behavioral features, but they fall short of fully capturing the dynamic and complex interactions across different time points and body joints in human motion sequences. In this study, a Spatio-Temporal Transformer Network model (STTFN) is proposed to automatically learn the spatio-temporal dependency relationships in human motion sequence data for prediction. The spatio-temporal knowledge embedding block employs an attention mechanism and a graph attention network to extract spatio-temporal behavioral features from raw data. An encoderdecoder network based on Transformer and Long Short-Term Memory (LSTM) is constructed for further analysis to acquire motion prediction data. This study conducts an experiment utilizing a human-robot collaborative assembly dataset to validate the effectiveness of the proposed model. The results demonstrate that our model outperforms classical models, thereby advancing the field of human motion prediction.

ABSTRACT. Bus scheduling is vital to ensue the service quality and save operational cost. current studies typically consider the problem as an optimization problem and solve it using exact or heuristic approaches. In this paper, we consider the problem as a sequence decision problem, and propose an Evolutionary Neural Network based Approach (ENNA) for it. A neural network is used to make a decision at each departure time in the bus timetable,to select a bus to depart at the time. By this means, a complete bus scheduling scheme is generated by a series of decisions of the neural network. The structure and weights of the neural network are optimized by a genetic algorithm. To meet the real-world operational needs of bus duty types and improve bus utilization, we design a bus duty type transition rule that dynamically decides the duty type of buses. Experimental results demonstrate that the proposed ENNA outperforms manual scheduling schemes, and can generate bus scheduling schemes with fewer buses.

ABSTRACT. In the electricity-carbon blockchain trading system, with the rapid expansion of blockchain technology applications, intrusion events targeting blockchain nodes are becoming increasingly frequent, and security issues are becoming more prominent. The security situation of blockchain systems is increasingly complex and severe. To meet the requirements for real-time and accurate anomaly detection in the electricity-carbon blockchain trading network, this paper proposes an active-passive-combined anomaly detection method. This method comprises two parts: passive anomaly detection and active anomaly detection. Passive anomaly detection collects and analyzes blockchain nodes’ network traffic and triggers active anomaly detection when passive anomaly detection identifies an anomaly. In contrast, active anomaly detection collects and evaluates log data from suspicious nodes. This framework can process and analyze large-scale network alerts and host logs under limited bandwidth. Experimental results on real blockchain nodes demonstrate the effectiveness of the proposed method.

ABSTRACT. With the rising labor costs, traditional manual disassembly methods can no longer meet the demand. The collaborative work of multi-skilled robots in disassembly lines can fully utilize their multi-skilled characteristics and high operational efficiency. This paper investigates the line balancing problem of multi-skilled robot disassembly lines, where the hybrid disassembly lines consist of linear and U-shaped structures. A mathematical model aiming to maximize profit is established. This paper proposes an improved fireworks algorithm to address the aforementioned problem. The validity of the model is verified using CPLEX. The superiority of the algorithm in solving this issue is validated through comparisons with peer swarm intelligence algorithms.

ABSTRACT. Virtual coupling presents a promising solution for enhancing urban rail transit efficiency through collaborative operations facilitated by information communication rather than physical connections. However, effectively controlling virtual coupling trains presents substantial challenges, particularly in managing implicit and nonlinear safety constraints. This paper proposes a novel adaptive hierarchical control approach to address these challenges, modeling the control problem as a Constrained Markov Decision Process (CMDP). The proposed approach integrates an upper controller utilizing safe reinforcement learning (RL) and a lower controller employing Proportional Integral Derivative (PID). Safe RL incorporates an unrolling safety layer (USL) into the RL algorithm, effectively addressing safety constraints inherent in the system. Additionally, dynamic adjustment of PID parameters based on the state of the virtual coupling trains enables precise control of train operations. This hierarchical architecture capitalizes on the adaptability of safe RL while leveraging the stability of PID, facilitating dynamic strategy adjustments in response to environmental changes to ensure virtual coupling stability and reliability. Simulation experiments validate the efficacy of the proposed approach, demonstrating significant advantages in controlling virtual coupling trains and thereby enhancing operational efficiency and safety.

ABSTRACT. Train Relative Distance Braking Mode (RDBM) has great potential to increase line capacity by enabling shorter following distances based on train-to-train communication information. However, it is difficult for traditional train control methods to ensure avoidance of potential collision risks in RDBM while dealing with the nonlinearity and uncertainty of train dynamics. In this paper, a novel train safety control method combining Deep Deterministic Policy Gradient (DDPG) and Control Barrier Function(CBF) is proposed. Firstly, the safety constraint set for adjacent trains, considering their braking capabilities, is analyzed, and the uncertainty of train dynamics is modeled using a Gaussian process. Secondly, DDPG is used to explore the optimal control strategy under the action of uncertain train dynamics, and CBF provides compensation control input to ensure safety during the learning process. Then, a DDPG-CBF algorithm is presented to achieve efficient safety controller design and implementation. Finally, sufficient numerical simulations demonstrate that, compared with DDPG algorithm alone, the DDPG-CBF algorithm can not only ensure the safe and stable operation of trains, but also learn train control strategies expeditiously and accurately under different operating environments, with an average efficiency increase of 38.94%.

ABSTRACT. Focused on the challenge of high-density tracking control for urban rail transit trains, a cooperative control method for trains based on Distributed Model Predictive Control (DMPC) is studied. Firstly, the force of the train is analyzed and the state space model of train operation is established. Upon this foundational framework, the control objectives of the train control system and corresponding control strategies are proposed. Secondly, in pursuit of ensuring the safety of train operations, the DMPC is used to design a cooperative train consistency control algorithm that considers the dynamic parameters such as the position, speed, and acceleration of the front train. The stability of the train control system is analyzed and verified by the Lyapunov stability theorem. Finally, combined with the west section of Beijing Metro Line 11, the proposed train cooperative control method is simulated and verified. The experimental results show that the proposed train cooperative control method can ensure the safe operation of trains, and make the state of trains consistent and have good synchronization. Compared with the follow control method, the cooperative control method can reduce the train spacing and speed difference, which can increase by 85% and 76% respectively.

ABSTRACT. Chinese Train Control System level 3(CTCS-3) train control onboard equipment plays a crucial role in ensuring train safety and improving operational efficiency. And the on-board interface equipment enables the interaction between the on-board Automatic Train Protection (ATP), ground equipment, drivers, and trains. While its fault accounts for a relatively high proportion of faults, therefore, this paper proposed a fault diagnosis method for on-board interface equipment based on temporal knowledge graph completion. Firstly, this method introduces temporal series to integrate travel logs and fault statistics, which extract fault phenomena and perform entity alignment, construct the temporal knowledge graph. Secondly, on this basis, we constructed a fault diagnosis network based on knowledge graph completion，which incorporates the Temporal-Translating Embedding (T-TransE) vectorization algorithm, Bidirectional Long Short-Term Memory(Bi-LSTM), and Self-Attention (SA) mechanism for temporal feature extraction. Finally, the T-TransE vectorization model was pretrained using the on-board interface equipment fault data from a railroad station in recent years for selecting the most effective temporal integration method .In order to validate the superiority of the proposed fault diagnosis method and the effectiveness of the data combination method, the diagnostic network without data combination and temporal relationship introduction, and other common fault diagnostic networks, were tested using the on-board fault data. Compared with the follow fault diagnosis methods, the model proposed in this paper achieved the highest fault diagnostic accuracy of 96.8%.

ABSTRACT. Command and control (C2) is now widely applied in various fields. Whether in military, emergency response, traffic management, business operations, or healthcare, C2 systems play a crucial role. By integrating information, real-time monitoring, and efficient decision-making, C2 has significantly enhanced operational efficiency and response capabilities across these domains. The applications of C2 theory have not only enhanced overall coordination and rationality in resource allocation but also provided strong support for crisis management in various scenarios, ensuring the smooth completion of tasks and security. The OODA (Observe, Orient, Decide, Act) loop is a classic model in C2 field. Building upon the OODA loop and incorporating fractal theory, this paper proposes a nested OODA model and establishes an evaluation metrics for the nested OODA loop.

ABSTRACT. Situational awareness (SA) is a critical research topic in various fields, including military command and control, disaster response, industrial process control, and infrastructure monitoring. In military command and control, SA is particularly significant, representing the ‘Observe’ and ‘Orient’ phases of the OODA model. Commanders must acquire and comprehend information about the adversarial environment to make quick and accurate decisions and execute appropriate actions. Effective SA leads to an information advantage, which subsequently translates into decision-making and operational advantages. Real combat operations often assess overall command effectiveness without focusing on the impact of SA on command performance. Therefore, this paper discusses the importance of SA from the perspective of the OODA model in combat. Utilizing combat units and sensor data recorded in wargame simulations, we propose a method for analyzing SA effectiveness across three dimensions: overall situation, SA accuracy, and SA quality. We examine the impact of SA effectiveness on the simulation outcomes for commanders on both sides. By employing the wargame, this study highlights the significance of enhancing commanders’ SA capabilities in dynamic adversarial combat environments.