Knowledge Graph Question Answering (KGQA) has gained significant attention in recent years, but many existing methods struggle to effectively reason over temporal constraints. This limitation reduces accuracy in temporal contexts and creates challenges in maintaining consistency and detecting conflicts within temporal knowledge graphs. To address this, a temporal constraint-based multi-hop reasoning approach is proposed, consisting of four modules: temporal information parsing, fact retrieval, constraint extraction and subgraph temporal reasoning. By incorporating temporal constraints, the model identifies implicit time-related information, thereby improving answer accuracy. Unlike conventional methods that focus on entities, this approach applies the attention mechanism at the relationship level to enhance reasoning performance. Experiments conducted on two public datasets, CronQuestions and ComplexCronQuestions, demonstrate the method’s superiority over state-of-the-art models. On CronQuestions, the model improves Hits@1 by 4.8

Multivariate time series anomaly detection holds significant value in domains such as information security and industrial control systems. However, due to the sparsity and complexity of multidimensional data, existing methods struggle with potential data scarcity scenarios encountered in the real world. To address these challenges, this paper proposes Pretrained Large Model Enhanced Limited-Sample Multivariate Time Series Anomaly Detection (ELimSD), an anomaly detection method adaptable to diverse data coverage scenarios, by leveraging frozen pre-trained large models. First, to obtain anomaly-free time series data, we propose a large model-based data augmentation strategy for generating anomaly-free samples. This strategy mitigates data contamination caused by anomalies by learning patterns from normal time series. Second, Discrete Cosine Transform (DCT) and Kolmogorov-Arnold Networks (KAN) are employed to extract frequency-domain and time-domain features, respectively, from the augmented time series data, uncovering inherent contextual dependencies within the sequences. Finally, anomaly detection is performed based on the reconstruction error between the real data and the reconstructed data. Experiments are conducted on six real-world datasets spanning different domains. Under limited sample settings, ELimSD achieves the highest F1 scores across all six datasets compared to state-of-the-art methods. Furthermore, it demonstrates an average F1 score improvement of 4.4% over the best existing baseline methods on four benchmark datasets from the anomaly detection field.

The rapid growth of electroencephalogram (EEG)-based applications has led to an exponential rise in EEG data volume, particularly unlabeled data, which challenges supervised learning techniques like EEG classification. Although several unsupervised EEG clustering studies have emerged to provide solutions for EEG labeling, they focus on shallow models and overlook practical scenarios possessing partial prior information. To this end, we propose a semi-supervised deep clustering method for handling partially labeled EEG signals, termed semi-supervised deep EEG clustering with union constraint learning and contrastive learning (EEGcUCC). To the best of our knowledge, this work represents the first attempt to address semi-supervised EEG clustering with deep learning. In EEGcUCC, an unsupervised contrastive learning-based pre-training paradigm, contrastive pre-clustering, is developed for initializing cluster-friendly embeddings and end-to-end output of cluster assignments. To mitigate label scarcity-induced confirmation bias, we introduce a union pairwise constraint loss that integrates prior hard constraints with soft pseudo-constraints derived from high-confidence samples. Furthermore, the proposed union contrastive learning utilizes a semi-supervised contrastive loss to globally integrates prior samples, class prototypes, and pseudo-labeled samples, facilitating the learning of robust embeddings. Comprehensive experiments on 12 EEG datasets demonstrate EEGcUCC yields superior clustering results, compared to advanced unsupervised and semi-supervised clustering methods.

Temporal knowledge graphs (TKGs) with evolving event representations serve as critical infrastructure for AI applications like information retrieval and rumor detection. Current TKG completion methods face three key limitations: (1) Conversion of quadruple structures to triplets constrains temporal information representation, (2) Cross-space interactions among entities, relations, and timestamps are challenging, and (3) Neighborhood contexts are underutilized and disambiguation mechanisms for polysemic elements are absent. To address these challenges, we propose the Quadruple-based Neural Discourse-Aware (Q-NDA) network. Our framework employs space-specific encoders for entity, relation, and timestamp representations, coupled with an Information Discourse Module enabling bidirectional cross-space communication. The architecture incorporates contextual disambiguation and neighborhood attention mechanisms to resolve semantic ambiguity while aggregating structural patterns from adjacent quadruples. This constitutes the first systematic approach addressing completion in TKGs through integrated neighborhood exploitation. Evaluations on benchmark datasets demonstrate Q-NDA’s superior performance, particularly achieving state-of-the-art results on the geometrically complex GDELT dataset - yielding 19.3

Cross-subject motor imagery (MI) Electroencephalogram (EEG) decoding remains challenging due to significant domain shifts caused by inter-subject variability. Despite the success of Euclidean domain alignment techniques via deep learning, they fail to capture the potential hierarchical structures inherent in EEG signals. To address this challenge, we propose MI-DGHCL, a domain generalization model for MI-EEG signals that leverages hyperbolic contrastive learning. MI-DGHCL introduces a Mamba-based feature extractor, incorporating slice feature embedding and slice-aware scanning modules that simultaneously capture both global information and local contextual features of EEG signals. Furthermore, it enforces both feature consistency and semantic consistency. Feature consistency is attained by aligning domains through the minimization of feature covariance in Euclidean space. Subsequently, hierarchical representations are derived via hyperbolic embeddings, and supervised contrastive learning pushes intra-class samples across subjects close in hyperbolic space, thereby ensuring semantic consistency. Comprehensive experiments on 3 MI-EEG datasets demonstrate MI-DGHCL yields superior results, compared to advanced methods.

There are many multi-objective optimization problems in dynamic environments (DMOPs), characterized by conflicting objectives and changing objective functions over time. Additionally, the dynamic nature of DMOPs may lead to continuous changes in the pareto front. However, existing methods experience significant issues such as severe loss of diversity and slow convergence, which make it challenging to track the dynamic pareto front both accurately and efficiently. To tackle these issues, a knee point guided prediction model is proposed in this article, oriented to population composition structure, which has three original components: (1) Based on the movement trend of previous knee points, the knee point generation strategy combines neighborhood search and step size exploration to identify them in response to environmental changes; (2) Depended on knee point classification, historical non-dominated solutions are reused to cultivate high-quality individuals in new environments, thereby expediting population convergence; (3) Diversity individuals are generated through uniform interpolation between predicted knee points, which increases the distribution of the population. These three strategies are integrated to establish a comprehensive prediction model to direct the generation of initial populations in changing environments, enhancing both the diversity and convergence of population. The effectiveness analysis and performance comparisons with some state-of-the-art algorithms demonstrate that the proposed algorithm exhibits significant advantages in enhancing solution quality. Furthermore, experimental results based on real-world applications validate the practical significance of this study.

The dynamic nonlinear characteristics and scarcity of anomalous data in real-world scenarios pose dual challenges of error propagation and pattern confusion for anomaly detection. Moreover, the disruption caused by continuous missing values to the inherent structure of the data further significantly degrades the effectiveness of traditional detection methods. The majority of current solutions employ the conventional two-stage procedure, where error propagation during the imputation process hinders the effective capture of complex interactions between missing and anomalous patterns. Therefore, we first construct the gradient-penalized Wasserstein generative adversarial imputation network (GW-GAIN) to address continuous missing challenges. This network adopts the gradient penalty mechanism and Wasserstein distance strategy to mitigate pattern collapse while maintaining the consistency between generated data and original data features, thereby reconstructing missing data with optimal diversity and fidelity. Subsequently, leveraging the constructed imputation network, we propose an end-to-end unsupervised anomaly detection method for continuous missing values by integrating continuous imputation with anomaly detection (CIAD). This method unifies data imputation and anomaly detection into a cohesive optimization framework, enabling the automatic learning of pseudo-anomalous samples during training without external data to mitigate imputation bias. Utilizing nine publicly available datasets and five commonly applied metrics, we evaluated anomaly detection performance across various missing threshold and missing window conditions. Comparison results with eleven state-of-the-art methods demonstrate that the proposed method outperforms all baselines across multi-scenario anomaly detection tasks within each dataset. In particular, the Recall and AUPR metrics improved by 8.038% and 7.873% compared to the suboptimal model. Furthermore, the ablation analysis performed on typical datasets with low, medium, and high anomaly proportions reveals that all components enhance the overall performance to varying degrees. At the same time, the weight of each contribution differs across scenarios with distinct anomaly proportions. The experimental outcomes demonstrate that the proposed method delivers satisfactory performance for time series anomaly detection in scenarios with continuous missing values, thus providing an effective solution for unsupervised anomaly detection in datasets featuring continuous missing values.

Dynamic distributed production scheduling poses significant challenges to modern manufacturing systems due to environmental complexity and the need to respond to real-time disruptions. This paper studies the dynamic distributed hybrid flow shop problem (DDHFSP) with simultaneous dynamic job arrivals and machine breakdowns, aiming to minimize makespan. To address this problem, a learnable iterated greedy algorithm (LIG) is proposed within the classical IG framework. LIG is designed to learn to learn how to adaptively configure its operational components by extracting knowledge from search trajectories, thereby effectively handling dynamic scheduling environments. Based on the IG structure, four rescheduling strategies for dynamic events, four destruction-reconstruction strategies, and four local search strategies are designed. The algorithm employs a long short-term memory (LSTM) network to extract temporally dependent production features, and adopts proximal policy optimization (PPO) to build an agent that adaptively selects the most promising combination of strategies according to the current state for execution within the IG framework. Experimental results demonstrate that LIG outperforms several state-of-the-art metaheuristic and reinforcement learning-based methods across various problem scales and dynamic scenarios, exhibiting faster convergence, better solution quality, higher stability, and stronger generalization capability. Ultimately, the proposed learnable mechanism establishes a new paradigm for tackling dynamic disturbances in DDHFSP.

Building a question answering system with professional knowledge understanding has become a research hot-spot. In highly specialized fields such as medicine, the question answering task faces challenges such as complex terminology and strong knowledge dependence, forcing most methods to rely on additional domain-specific pre-training (e.g., BioBERT) over large-scale domain corpora to improve model performance. However, such methods are costly in computing resources and time. To this end, this paper proposes a cross-source alignment and dynamic knowledge injection framework CADI for medical question answering, aimed at enhancing model performance while significantly reducing the reliance on computationally exorbitant domain-specific pre-training. The CADI framework integrates two key mechanisms to achieve effective injection and fusion of multi-source medical knowledge: (1) A Cross-source Semantic Alignment (CSA) mechanism with Siamese structure aligns medical knowledge graph embedding and language model representation, which significantly reduces semantic deviation between multi-source embeddings. (2) A Dynamic Attention Knowledge Injection Network (DAKI-Net) is designed, which can adaptively adjust the fusion weight of structured medical knowledge graph embedding and definition embedding according to the context. Experimental studies on two authoritative medical question answering datasets, COVID-QA and PubMedQA, show that the proposed CADI framework has significant advantages in accuracy, flexibility and transferability of introducing medical knowledge. Notably, in certain medical QA tasks, CADI achieves performance comparable to, or even surpassing, that of domain-specific models when built upon general-purpose language models. This provides an effective path for constructing subsequent low-resource medical question answering and significantly reducing reliance on domain-specific pre-training.

Multivariate time series in modern industrial systems exhibit complex spatiotemporal coupling, high-frequency noise, and dynamic operational uncertainties, while existing models often ignore the instantaneous effects in fault propagation, making it difficult to extract true physical mechanisms to cope with rapidly propagating sudden anomalies. This paper proposes an industrial fault root cause analysis framework based on mixed causal discovery and dual residual disentanglement (MDRCA). First, a dynamic weight generation module integrating contrastive learning is developed, utilizing a dual-head decomposition architecture to explicitly extract nonlinear lagged causal dependencies, effectively addressing uncertainties and filtering out collinearity noise. Second, the DirectLiNGAM algorithm is introduced into the residual space after removing historical trends to capture instantaneous interactions, thereby constructing mixed causal relationships of both lagged and instantaneous effects. Finally, a dual residual disentanglement algorithm is designed for the inference stage, which sequentially separates historical lagged biases and instantaneous network interferences across temporal and spatial dimensions, reconstructing structured residuals to achieve the quantification and ranking of anomaly root causes. Comprehensive experiments demonstrate that the framework not only possesses strong causal topology reconstruction capabilities on synthetic datasets and industrial benchmarks but also exhibits high stability and practical value in its fault root cause tracing performance when applied to real on-orbit satellites. Furthermore, robustness evaluations confirm that the framework possesses strong anti-interference capabilities when facing uncertainties, and the low-order polynomial complexity of its online inference can satisfy the real-time response requirements of physical systems to sudden faults.

Electroencephalography (EEG) is extensively employed for supervised analysis across various scenarios, including disease diagnosis, neurocognition, and brain-computer interfaces. It has heightened the demand for effective EEG labeling. However, the manual labeling process is labor-intensive and time-consuming. Furthermore, there are few studies that attempt to decode information from EEG in an unsupervised manner using deep learning. In this paper, we present a novel end-to-end EEG clustering approach via a state space model (SSM), termed EEGcMamba. EEGcMamba introduces a universal backbone, BrainMamba, for EEG feature learning, and incorporates both a weighted instance-level contrast head and a dual-branch cluster-level contrast head for contrastive learning. Specifically, BrainMamba utilizes a slice-aware scanning mechanism to input segmented EEG slices through multiple sequences, capturing fine-grained contextual connections between slices. To mitigate risks of pushing similar EEG samples apart further in the embedding space, we introduce weight terms from the data space during calculating the instance-level contrastive losses. Furthermore, in the cluster-level contrast head, the assignment-discrimination branch accounts for the clustering distribution consistency, while the semantic-aware branch employs pseudo-labeling semantics to establish group-instance discrimination. Extensive experiments on 11 EEG benchmark datasets demonstrate the superiority of EEGcMamba over existing advanced methods. The code will be available at Github.

Competitive many-task optimization (CMaTO) is a special many-task optimization paradigm whose purpose is to find the best optimal solution for all tasks. However, the existing CMaTO algorithms perform poorly in the design of knowledge transfer from auxiliary tasks to the main task, resulting in a prolonged period of stagnant optimal fitness for the main task. To address these shortcomings, a competitive many-task optimization algorithm is proposed, based on reinforcement learning and meta-knowledge transfer, leveraging differential evolution as a foundational evolutionary strategy. This algorithm employs a reinforcement learning algorithm to select auxiliary tasks that can accelerate the convergence of the optimal value or jump out of the stagnation state according to the evolutionary state. Meanwhile, a stagnation detection operator is designed to switch the main task when the optimal value stagnation threshold upper limitation is reached. Furthermore, the meta-knowledge migration algorithm is embedded to judge the evolutionary state of the population based on the distance between the optimal solution and the centroid of the population. The migration radius is adaptively adjusted, and the knowledge is utilized to facilitate the evolution of high-quality solutions for the source task, which can assist the population in accelerating convergence or escaping a local optimum. To evaluate the performance of the proposed algorithm, three CMaTO benchmark test suites and a real-world Unmanned Aerial Vehicle (UAV) task allocation problem are chosen to compare it with other state-of-the-art strategies. The results show that the proposed algorithm achieved better performance.

As the production environment becomes increasingly complex, the integration of soft computing techniques becomes essential for addressing resource-constrained scheduling problems. This paper delves into a worker constrained hybrid flow shop scheduling problem (WHFSP) that integrates worker resources at each processing stage. A mixed-integer linear programming (MILP) model is constructed which enables the use of mathematical solvers to obtain optimal solutions for small-scale instances. Additionally, a novel soft computing-based scheduling framework, namely a proximal policy optimization-based hyper-heuristic algorithm (PPO-HH), is proposed. It automatically selects the most suitable low-level heuristic strategies based on the current state and historical data, facilitating efficient exploration and exploitation of the complex decision space. Several low-level heuristics including perturbative and local search operators are developed to explore the solution space. Subsequently, a high-level control strategy based on proximal policy optimization is proposed. A solution quality evaluation function and a reward mechanism based on problem characteristics are formulated. This mechanism provides feedback to PPO-HH based on the degree of alignment between the actions taken by the agent and the objectives, gradually optimizing the selection of low-level heuristic strategies. Eventually, it generates a probability distribution for each low-level heuristic in the given environment. Comprehensive numerical experiments are conducted to evaluate the performance of both the MILP model and the components of the PPO-HH algorithm. The comparison results show that PPO-HH is effective and efficient for solving the WHFSP.

Consumer Internet of Things (CIoT) is an emerging technology that took the traditional consumer electronics to next level with smarter devices and higher connectivity. However, the rapid boom of this technology has captured a great deal of hacker’s attention in the past decade. Due to resource restricted nature of CIoT devices and limited computational abilities, security professionals are unable to strengthen the security and privacy measures. Since the existing security methods suffers from data imbalance and overfitting issues, this paper proposes a distributed and dynamic weighted boosting method named Federated-Boosting for the accurate and timely detection of cyber-attacks in CIoT. Specifically, a weighting strategy is designed to overcome the data imbalance issue by dynamically adjusting the weights of weaker class samples. Additionally, a regularized loss function is designed that helps in controlling the overfitting of the model and provide enhance performance and generalization. Finally, a time-based and performance-based dynamic aggregation scheme is designed using distributed training model that keeps the privacy of the CIoT devices intact to the local server. Extensive experiments on two real-world datasets validates the superior performance of Federated-Boosting model against state-of-the-art detection models.

Optimizing the milling parameters for complex surface can improve the machining quality. However, existing methods are oriented toward a simple machining path and single machining process. In this paper, a milling parameter optimization method for complex surface is proposed by combining the data-driven models and physical constraints. Initially, the chatter indicator is derived from the tool vibration using Variational Mode Decomposition. Subsequently, multiple few-shot prediction models are developed based on real machining data using deep neural networks. A mathematical optimization model is then constructed by combining multiple prediction models and physical constraints, which aims to minimize the machining failure rate and maximize the material removal rate under multiple constraints. The spindle speed, feed speed, cutting depth, and path spacing are the optimization parameters. Finally, a Hypervolume-based Multi-Objective Optimization (HMOO) algorithm is proposed to solve the optimization model. The solution set produced by HMOO exhibits superior convergence and diversity compared to the S-Metric Selection Based Evolutionary Multi-Objective Algorithm (SMS-MOEA). In experiments, a five-axis machine tool is employed to mill turbine blades with complex surface. Experimental results demonstrate that the proposed prediction models achieve higher accuracy than widely used regression algorithms. Integrating the high-precision prediction models with HMOO significantly enhances blade machining quality while guaranteeing reliable machining efficiency, resulting in a 5.25% reduction in the machining failure rate.

Tool wear prediction is crucial for maintaining machining quality and improving production efficiency. Adapting to variable cutting conditions in a dynamic manufacturing environment remains a significant research topic. Existing methods for tool wear prediction under variable cutting conditions require retraining with labeled samples collected from new cutting conditions, necessitating trial cutting experiments to collect labeled samples and the storage of a separate model for each new cutting condition. This results in high manual labeling costs and substantial storage requirements, and prevents models trained on known conditions from being directly or efficiently deployed to new cutting conditions. To overcome these limitations, we propose a novel setting for tool wear prediction under variable cutting conditions. In this setting, the prediction model is not exposed to any data from the new cutting conditions during training and must be directly deployed to these new cutting conditions after training. To address this challenge, we propose an online, real-time quantitative tool wear prediction method under variable cutting conditions. The proposed method employs a domain generalization (DG) algorithm that incorporates classwise gradient consistency (CGC) and domain-invariant feature-level sample mixing to train the prediction model. This strategy enables the prediction model trained on known cutting conditions to be directly deployed to new, unknown cutting conditions without retraining, thus allowing online quantitative tool wear prediction under new cutting conditions. Experimental results indicate that the variable cutting conditions can degrade the performance of traditional data-driven methods and models. The proposed method reduces the mean absolute error (MAE) by at least 33.02% compared to traditional data-driven methods and by at least 10.03% compared to the state-of-the-art DG algorithms.

Mobile edge computing (MEC) significantly boosts the computing power and reduces the energy consumption of Internet of Things (IoT) devices, serving as a valuable complement to cloud computing. The application of unmanned aerial vehicle (UAV) for MEC systems can effectively alleviate the issue of insufficient or damaged communication facilities in remote areas, further expanding the scope of MEC applications. In this article, we present a system model for UAV-assisted wireless-powered MEC systems in a dynamic environment with the objective of maximizing the computation rate of user devices. Due to the complexity of the optimization objective in dynamic environments, we propose a swarm intelligence-based optimization framework with a mechanism for responding to environmental changes, which is intended to enhance population diversity in both static and dynamic environments with the aim of overcoming premature convergence. We integrate particle swarm optimization and harmony search into the proposed framework, naming them DOPSO and DOHS, respectively. Simulation results for two offloading modes in UAV-assisted MEC systems indicate that the proposed framework significantly outperforms other dynamic optimization algorithms.

Temporal Knowledge Graphs (TKGs) extend traditional knowledge graphs by incorporating temporal information, enabling reasoning over time-dependent facts. However, many real-world knowledge graphs are incomplete, requiring effective reasoning methods to infer missing information and improve their overall quality. Multi-hop reasoning is a promising method for this task, but existing methods often fail to fully utilize temporal information and relationship modeling, resulting in less accurate inference. Additionally, many approaches lack interpretability, making it difficult to explicitly trace the reasoning pathways. To address these challenges, this paper proposes reinforcement learning-based multi-hop path reasoning for TKGs (RLPR), a novel model designed to enhance inference accuracy and interpretability by integrating temporal and relational information. RLPR introduces a timestamp decomposition strategy to better capture temporal dependencies while reducing model complexity. Unlike conventional entity-focused attention mechanisms, RLPR applies attention at the relationship level, allowing for more precise modeling of relational interactions. Furthermore, RLPR employs a reinforcement learning-based strategy network that explicitly constructs multi-hop reasoning pathways, improving both the interpretability and adaptability of the inference process. Compared to SOAT methods, the RLPR model achieved the best results on open-source datasets. For the link prediction results, the metrics improved by 15.07 %, 13.56 %, 4.26 % on the ICEWS14 datasets. When the path length setting for the RLPR model during reasoning is set to 2, the RLPR model improved various metrics on the ICEWS14-2 datasets by 13.04 %, 60.61 %, 38.67 %, and 27.08 %.

With the rapid development of industrial globalization and diversification, the application of distributed and heterogeneous systems in production scheduling has become increasingly widespread. This paper studies a distributed heterogeneous flexible job shop scheduling problem under nonidentical time-of-use electricity tariffs (DHFJSP-NTOU) that uniquely integrates three underexplored dimensions: heterogeneous factories with timedependent processing capabilities, geographically varying time-of-use electricity tariffs, and dual optimization of production efficiency and energy sustainability. A mixed-integer linear programming (MILP) model for DHFJSP-NTOU is established. To solve the DHFJSP-NTOU, a graph-reinforced multi-objective optimization algorithm (GRMO) is developed, which features three innovations: a hybrid initialization strategy balancing greedy heuristics and solution diversity, a graph neural network (GNN) framework dynamically encoding operational interdependencies and factory-specific constraints, and a reinforcement learning-driven adaptive operator selection mechanism using proximal policy optimization (PPO) for intelligent search guidance. Finally, comprehensive experiments are carried out to assess the performance of both the MILP model and the components of the GRMO. The experimental outcomes indicate that the GRMO outperforms several of the most recent high-performance methods in solving the DHFJSP-NTOU problem. The structural analysis further validates that the GNN-based feature extraction enhances search efficiency compared to conventional methods. These innovations provide a new paradigm for addressing the challenges of sustainable scheduling in distributed manufacturing systems with heterogeneous resources.

With the speedy progression and adoption of IoT devices in modern self-driving vehicles (SDVs), autonomous driving industry is gradually reforming its capabilities to provide better transportation services. However, this domain faces enormous security and privacy challenges and thus has become an attractive target for attackers due to its rapid growth and market worth. Furthermore, the rapid transformation in technological tools in transport industry and speedy evolution of cyber-attacks paved the way for designing efficient IDSs. Motivated by these challenges, we put forward a new secure and efficient IDS approach for the security of SDVs. The propose approach utilizes an emerging strategy to mitigate security vulnerabilities and cyber attacks detection using zero trust (ZT) model. Through this work, we put forward a context-aware zero trust security framework for IoT-based SDVs. The proposed framework utilizes a context-aware design to evaluate the trustworthiness of the devices using multi-source trust and reputation model. Then, to make the framework more effective and reliable, we introduce crawler system into the context of the IoT-devices in SDVs to make the system unbiased. Additionally, an observer module is developed that employs state-of-the-art machine learning algorithm to detect malicious actions. Empirical results on two standard benchmark datasets (i.e., Car Hacking and ToN_IoT) validate the practicality and robustness of propose framework in real-world transport systems with enhanced security and trust management against evolving cyber-threats. Detection results demonstrate that the proposed framework secured the best performance by achieving 99.43% and 99.52% accuracy for Car Hacking and ToN_IoT, respectively. The findings of this study will help the security professionals and researchers to comprehend the importance of ZT architecture in developing effective and robust security solutions for modern IoT-based SDVs.