ABSTRACT Critical infrastructure systems such as the electric grid are increasingly cyber–physical; however, despite their inherent cyber–physical characteristics, the physical process system and communication/control network system are traditionally analysed in silos. As these systems become more cyber–physical, it is crucial that models and methods are available to assess the cyber–physical system (CPS) interdependencies, characteristics and event propagation for improved planning, operation and response. Thus, we propose an integrated structural and temporal CPS interdependency analysis framework that provides insight into the CPS function during normal operation as well as disturbances. This structural and temporal analysis of real‐time techniques (START) framework is uniquely designed for assessing CPSs by accounting for the challenges of analysing cyber and physical data streams together due to data availability, data type and time scale differences. By leveraging both structural (e.g., graph analysis) and temporal (e.g., data analytics) techniques, different CPS behaviours and configurations can be accounted for. This paper presents the START framework, a subset of techniques applied to exemplar transmission and microgrid systems and analysis of resulting structural and temporal CPS interdependency insights.
Unmanned aerial vehicles have been widely adopted owing to their compact size, high maneuverability and low cost. Consequently, research on unmanned aerial vehicles guidance, navigation and control systems has emerged as a prominent topic. Path planning constitutes a critical component of guidance, navigation and control systems. To overcome the limitations of existing unmanned aerial vehicles path planning algorithms, including restricted environmental perception, low planning efficiency and diminished learning efficacy-we propose a novel method that integrates the artificial potential field with deep reinforcement learning. We devise a reward strategy that incorporates artificial potential field and path fitting, and construct a hierarchical mechanism that fuses reinforcement learning with classical algorithms. This mechanism effectively optimises the search path, mitigates inefficient exploration and substantially accelerates the path planning process. Finally, experiments conducted in a simulation environment, along with comparative analyses against state-of-the-art techniques, demonstrate the superiority of the proposed method in terms of path planning efficiency.
ABSTRACT Rotor imbalance in wind turbines can lead to significant safety risks and economic losses, which makes accurate fault analysis essential for reliable operation. However, accurate and robust analysis under varying wind conditions remains challenging. To address these issues, a multistage rotor imbalance analysis framework is proposed, which includes fault detection, fault type classification, fault severity classification and blade localisation. First, a knowledge‐based expert system with a multinode decision structure is developed, where nacelle vibration is used for rotor status detection and output power is used for imbalance type classification. Second, a convolutional neural network with training interference that incorporates variable dropout and microbatch training strategies is introduced to improve noise tolerance. Moreover, an adaptive deep interpolation convolutional neural network (DI‐DCNN) is proposed for handling wind‐induced variations through a multisubnetwork weighting and fusion mechanism. Third, the harmonic components of the nacelle vibration are compensated for using information on the rotor azimuth to achieve blade localisation. Experimental results under various wind conditions demonstrate that the proposed method achieves high accuracy across all the tasks and maintains stable performance under varying turbulence intensities. Moreover, the framework does not require additional sensors and has good generalisation ability. The proposed method provides a unified and robust solution for rotor imbalance detection, classification, and localisation in wind turbines.
ABSTRACT This study examines the scientific literature on digital twin technology (DTT) in supply chain management (SCM), utilising bibliometric analysis to investigate the annual publication trend, keyword co‐occurrence, bibliographic coupling and co‐citation regarding foundational themes in SCM, as well as international collaborations. The Scopus dataset is analysed for the period from 2009 to 2024, encompassing a total of 1256 publications, which are analysed using VOSviewer software. This study reviews trends in the application of DTT within SCM, featuring tables, graphs and maps that illustrate key performance metrics for article production. The findings indicate a rising number of publications over the years, with the United States, United Kingdom, Germany and India being the most prolific nations in terms of publishing and collaboration leaders. The most influential keywords related to DTT in SCM include supply chains (SCs), additive manufacturing (AM), 3D printing, digital twin (DT) and sustainability. The authors identify four distinct thematic clusters: AM and SC innovation; shaping parts and service SCs; sustainability and global impact; and DT and Industry 4.0, which are relevant for current and future researchers in this area. The study is useful for researchers, practitioners and policymakers interested in comprehending, implementing and further studying DTT in various sectors of SCM.
ABSTRACT External connectivity for smart grid involving internet, data equipment, relays and breakers is essential to provide reliable and secure power supply. However, their interconnectivity also makes the grid susceptible to external threats with potential to damage equipment and cause power supply disruptions and safety hazards like false data injection attacks (FDIAs). In this paper, intrusion detection for FDIAs is proposed using three approaches. (1) A stack‐based model containing an expansion decision tree and neural network, (2) recurrent neural metworks (RNNs) and (3) convolutional neural networks (CNNs). The experiment evaluation is performed using a publicly available dataset of Mississippi State University and Oak Ridge Nation Laboratory. On the provided dataset, the top performance is achieved on image‐based classification using ResNet‐18 with accuracy, precision, recall and F1 score of 97%, 97%, 95% and 96%, respectively. The DNN‐GRU framework achieved accuracy, precision, recall and F1 score of 88%, 78%, 72% and 75%, respectively. Similarly, a version of the stack model of expansion decision tree and neural network combination achieved accuracy, precision, recall and F1 score of 95%, 95%, 95% and 95%, respectively. Each of these proposed methods has different preprocessing steps with different results. ResNet 18 has outperformed the hybrid model and recurrent neural network in precision, recall, F1 score and accuracy, which results in correct predictions, better identifying true positives (recall), avoiding false positives (precision) and achieving a robust balance between them (F1 score).
ABSTRACT Cyber‐Physical Systems (CPS) are central to modern smart infrastructures, enabling intelligent processes that handle large volumes of data while ensuring security, safety, reliability, and resilience. The integration of Artificial Intelligence (AI), particularly deep learning, with CPS and Internet of Things (IoT) technologies is driving advancements in Industry 4.0, smart grids, and intelligent transportation systems. AI enhances cybersecurity by enabling anomaly detection, continuous monitoring, network scanning, log analysis, and classification of data into legitimate and malicious categories, allowing real‐time threat identification and proactive mitigation. At the same time, the adoption of CPS and Industrial Internet of Things (IIoT) introduces significant cybersecurity challenges, as security incidents can directly affect humans and critical assets. This convergence of AI and cybersecurity provides a dual advantage: improving operational efficiency while reinforcing defences against evolving cyber‐physical threats. By enabling real‐time threat detection, resilient response, and robust protection of both infrastructure and AI algorithms, CPS can maintain safety, reliability, and performance under complex cyber and physical stressors. This special issue highlights research at the intersection of AI, cybersecurity, and CPS/IIoT, focusing on innovations that enhance resilience, protect critical systems, and support the sustainable growth of smart industrial and urban ecosystems.
ABSTRACT To address the detection issue of false data injection attacks (FDIAs) in networked microgrids (NMGs) with renewable energy integration and secondary frequency regulation, this paper proposes a novel detection scheme based on delay unknown input observers (DUIOs). Firstly, a comprehensive dynamic NMG model is developed under the cyber‐physical integration framework, incorporating renewable energy systems, hybrid energy storage systems and dynamic frequency control for multi‐microgrid interactions. This model establishes a foundation for analysing multi‐target FDIAs across different channels and control areas. Secondly, an innovative DUIO is designed that incorporates renewable energy uncertainties and load prediction errors into the observer design, achieving accurate system state estimation under unknown disturbances through the introduction of an observer delay parameter. Finally, a multi‐target detection framework based on residual analysis is proposed, enabling precise attack localisation across different channels and regions through multiple parallel observers and a moving accumulation residual evaluation mechanism. Simulations conducted on a three‐area NMG validate the effectiveness of the proposed scheme under various attack scenarios, including single‐area multi‐target attacks, multi‐area multi‐target attacks and coordinated attacks, while maintaining robustness against renewable energy fluctuations.
ABSTRACT Accurately screening the characteristic factors that influence short‐term power load forecasting is an effective means to improve prediction accuracy. Non‐critical features in multidimensional datasets can make it difficult for the prediction model to distinguish electrical loads, thereby reducing model accuracy. To tackle this challenge, a novel Non‐Intrusive Load Monitoring (NILM) framework for appliance recognition is proposed to overcome the problem of distinguishable electrical load features, which combines a Variational Mode Decomposition (VMD) module, a convolutional Neural Network (CNN), and a Bidirectional Long Short‐Term Memory (BiLSTM) network. First, the original electrical power data is first decomposed by the VMD module, which excellently achieves noise reduction and stationary processing of non‐stationary load data, effectively separating valid load features from interference components and laying a high‐quality data foundation for subsequent feature extraction and forecasting. Second, CNN is adopted to extract the local spatial features of the decomposed load data for accurate appliance recognition and classification, which excels in automatically mining hidden spatial features of electrical loads and greatly improves the distinguishability of load features among different electrical equipment. Meanwhile, BiLSTM is used to capture the bidirectional temporal dependencies in the load data for short‐term power load forecasting, which surpasses traditional unidirectional time‐series models in mining long‐term and bidirectional temporal correlation of load data and makes the forecasting results more consistent with the actual operation law of electrical loads. To further exploit the model's potential, an optimised strategy based on the sparrow search algorithm (SSA) is developed to optimise the key parameters of the CNN‐BiLSTM model, which optimises the model's parameter configuration efficiently and adaptively and avoids the accuracy loss caused by manual parameter adjustment. The results show that the proposed method effectively improves the accuracy of short‐term power load forecasting.
ABSTRACT Cyber‐physical microgrids are vulnerable to stealthy cybersecurity threats that disguise their actions through the exploitation of system knowledge. Such actions can severely impacts microgrids deployed in defense bases, slowing the response time of military forces during national emergencies. Several machine‐learning algorithms have been proposed to detect intrusions in the grid networks; however, these traditional machine‐learning algorithms lack data privacy and are subject to several adversarial machine‐learning threats. This paper proposes a novel federated machine learning (FML)‐based three‐model framework to detect and identify stealthy data‐integrity attacks while ensuring data privacy in microgrid networks. The proposed architecture uses a variational mode decomposition technique to extract derived features from incoming measurement and control datasets. The extraction of these derived features allows FML models to learn minute variations in data patterns that allow them to perform significantly better than the models trained with generic datasets consisting of raw features. Our experimental results show the efficient performance of the proposed methodology against different types of data integrity attacks while considering primary and secondary controllers in microgrids. Further, the applied FML‐integrated random forest ensemble algorithm outperforms the existing generic FML algorithms during noisy and noise‐free datasets with prediction latencies of only 91–134 µs per sample within the 0.1 s sampling interval and requires communication bandwidth of around ∼8.25 KB/s at the control center and ∼2.7 KB/s per edge client for communication.
ABSTRACT Human activity recognition (HAR) with the help of wearable sensors has become a major research focus because of its broad application areas, such as healthcare monitoring, smart homes and human computer interaction. Yet, it is not easy to recognise activities accurately by using multivariate sensor data because the sensors can produce noisy signals, there can be redundant features and complex temporal dependencies make the task difficult. In our paper, we suggest a deep learning method that combines sensor‐to‐image conversion, feature‐level fusion, dimensionality reduction and multi‐scale classification to solve the above issues. Firstly, raw multivariate sensor signals are transformed into structured image representations with the use of spectrogram‐based encoding, thus enabling convolutional neural networks to grasp spatial patterns in temporal data quite well. Two deep architectures that complement each other, namely Inception and Xception, are used to obtain significant features from the generated images. Next, as a way of feature‐level fusion, the feature vectors extracted from the two networks are joined to harness the complementary information contained in both networks. After that, principal component analysis (PCA) is used to get a small reduced fused feature (RFF) representation in order to minimise feature redundancy and computational complexity. This reduced feature space is later processed through a common multi‐scale convolutional front‐end with kernel sizes of 3, 5 and 7, and then CNN, LSTM and RNN classifiers are used to represent spatial temporal activity patterns. As shown by the tests on the WISDM, UCI‐HAR and PAMAP2 datasets, the proposed method can achieve excellent results with accuracies of 98.88%, 98.71% and 98.71%, respectively.
ABSTRACT Short‐term hydropower scheduling is inherently affected by uncertainties in both inflow and electricity demand, which challenge the reliability of management strategies. Deterministic approaches struggle to maintain feasible and economically efficient schedules, especially when a rare or extreme scenario occurs. In this paper, an integrated framework combining probabilistic forecasting with safe reinforcement learning is proposed to enhance the short‐term hydropower scheduling strategy. LightGBM and temporal fusion transformer are combined into a multimodel forecasting layer to generate calibrated probabilistic predictions. The forecasts are transformed into risk‐aware constraints for a deep reinforcement learning agent to optimise reservoir operations. Experiments under three different scenarios are conducted to demonstrate the effectiveness of the proposed framework. Probabilistic forecasting provides well‐calibrated uncertainty bounds that adapt to both stationary and highly volatile conditions. The proposed approach can achieve higher cumulative rewards, maintain operational feasibility under compound disturbances and exhibit strong adaptability to nonstationary and biased forecasting regimes.
ABSTRACT Time‐Sensitive Networking (TSN) is a foundational technology for deterministic communication in industrial automation, automotive and aerospace applications. However, its complexity increases significantly when handling heterogeneous traffic with multiple shaping mechanisms and when networks span multiple administrative domains. This paper presents a targeted state‐of‐the‐art review of analysis and optimisation methods for such multishaper and multidomain TSN systems. We selected representative works through keyword searches on scholarly databases and expert‐driven citation analysis, focusing on papers that directly contribute analysis or optimisation methods for these configurations. This review examines four multishaper combinations, that is, time‐aware shaper (TAS) with credit‐based shaper (CBS), asynchronous traffic shaper (ATS), cyclic queueing and forwarding (CQF) and frame pre‐emption (FP), and four multidomain scenarios, that is, TSN‐to‐TSN federation, TSN‐to‐Deterministic Networking (DetNet) integration, TSN‐to‐5G and TSN‐to‐Wi‐Fi convergence. We synthesise 44 representative works in two comparative tables that categorise each study by analysis method, optimisation approach, configuration scope and validation scale. Based on the review and experimental evidence, we summarise 14 open challenge categories across multishaper and multidomain configurations, including the scalability of formal analysis methods, the impact of hardware constraints and the need for unified runtime configuration. Case studies further show that formal analyses become impractical inside large optimisation loops, that joint multishaper optimisation is necessary to avoid infeasible configurations and that large‐scale multidomain orchestration remains computationally expensive. We outline future research directions to address these gaps for next‐generation industrial systems.
ABSTRACT Remote environmental monitoring covers large and heterogeneous regions, yet large target areas lack communication stations, creating a persistent backhaul constraint. To provide connectivity, monitoring‐oriented cyber–physical systems adopt low earth orbit satellite links. However, a LEO satellite offers only short, intermittent contact windows, and both window availability and in‐window link quality vary with satellite pass geometry and weather driven disturbances. In this paper, a multimodal fusion model, Residual Multi‐Modal Temporal Fusion Network (RMM‐TFNet) that integrates numerical meteorological variables, historical communication statistics, and sky images is proposed to predict access window availability and signal quality. The architecture employs two modality specific encoders and combines their representations through two complementary fusion modules. A series of controlled data transmission experiments were conducted, yielding a corpus of communication link measurements in practical situations. The proposed model is found to achieve higher predictive accuracy than competitive deep learning baselines by evaluating on these measurements.
Cyber‐physical systems have been proposed for Industry 5.0 and Society 5.0. One such CPS is the Social Co‐OS (cyber–human social co‐operating system). Social Co‐OS is a co‐operating system between cyber and human societies that views the social system as a dynamic circular structure composed of three layers: individual behaviour, interindividual interaction and institutional formation. Within this framework, the cyber system supports collective decision‐making and individual cooperative behaviour across these layers. The objective of this study is to define a novel application architecture based on the Social Co‐OS concept, design a user interface flow and implement it in practice. Specifically, we develop a social impact evaluator, a pluralistic policy simulator and a consensus‐building facilitator, which constitute the deliberative and political loop of Social Co‐OS. Additionally, we implement a personality estimator and a behaviour change promoter, which constitute the operational and administrative loop, along with a common mediator that serves as the cyber–human interface. Through these implementations, we demonstrate that Social Co‐OS applications can effectively support human social systems and offer practical utility for policy co‐making and cooperation as evidenced by examples grounded in real‐world challenges. In the future, we aim to promote the wider adoption of digital democracy and cooperative platforms.
Microgrids rely on communication networks for reliable operation, which makes them inherently vulnerable to cyberattacks. Such attacks can destabilise system dynamics and drive states away from their nominal operating trajectories. Although several physics-informed and machine learning-based strategies have been developed to counter these threats, the rapidly evolving cyber landscape enables adversaries to bypass static defences or rules-based mitigation approaches. This paper proposes a dynamic, online-trained and fully decentralised reinforcement learning (RL)-based cyberattack response framework to protect microgrids from evolving cyberattacks. The proposed framework deploys multiple deep Q-networks (DQNs), each associated with a distributed energy resource (DER), to enable localised and adaptive attack mitigation. In this framework, each DQN processes local voltage and frequency measurements-combined with intrusion detection system (IDS) alerts-as observations and rewards to guide decision-making. Extensive simulation studies demonstrate the robustness of the proposed framework under diverse attack scenarios and varying IDS-induced detection delays. Comparative analysis highlights its superiority over existing static or preexisting rules-based mitigation approaches. Finally, we present an analysis that shows the framework's scalability to real-life microgrids with more interacting agents.
Weeds are a significant challenge to crop quality and quantity and therefore there is a need to adopt effective weed control and management systems. Nowadays, object detection has found extensive applications in the agricultural field such as the detection of weeds through deep learning, machine learning, image processing and IoT. The idea in this paper is to present the proposal of an autonomous rover that can identify and classify weeds in real time using the YOLO object detection method. The dataset that will be utilised in the current research is a collection of 5997 images of weed instances, allowing even more accurate detection and classification of weeds. We also combined the Convolutional Block Attention Module (CBAM) with YOLO to enable the model to pay attention to the useful spatial and channel-wise features, as an evaluation of the performance of various YOLO models is based on inference time and weed detection accuracy. Based on the experiment, YOLOv8 and its variant YOLOv8-X demonstrated the best mean average precision (mAP) of 93.6% with that inference times of 3.4 and 2.2 ms per image, respectively. YOLOv9-E (an extension of YOLOv9) using CBAM, on the other hand, had better mAP of 99.5% with inference times of 10.6 and 2.5 ms, respectively. These findings indicate that YOLOv8 and YOLOv9 hold a good prospective of automatic field-level weed detection and emphasise the significance of high-quality datasets, efficient model architectures and attention mechanisms to the efficient and correct autonomous weed management.
This paper proposes a new framework for the analysis of cyber-physical system security against denial-of-service (DoS) attacks using generalised stochastic Petri nets. Although cyber-physical systems, through increased integration of computational and physical processes, offer great advantages, they are subject to cyber threats that can disrupt their critical operations. Among them, DoS attacks, which overload communication channels and prohibit the exchange of data between system components, are a major concern. Traditional methods of security assessment are inadequate given the unique complexities of cyber-physical system architectures. This research presents a generalised stochastic Petri net-based model able to capture the dynamics of a cyber-physical system under attack scenarios for the comprehensive analysis of system vulnerabilities and defencive mechanisms. The model incorporates immediate and timed transitions, thus mapping both continuous operations of the cyber-physical system and the discrete-event nature of cyber threats. Simulation experiments validate the effectiveness of the model in demonstrating how DoS attacks can degrade system performance. The results reflect the need for improved methodologies for security testing in order to enhance the resilience of cyber-physical systems, particularly in safety-critical applications.
Behaviours and activities are natural concepts (found, e.g., in UML and SysML) for model-driven design of cyber-physical systems (CPS). These concepts are formalised in the activity framework , a model-based framework incorporating a model of activities with determinate timing and behaviour, and a strong mathematical foundation based on max-plus algebra that allows efficient timing analysis and optimisation. It provides a layered view of the system's actions and events, activities built from them, and sequences of activities that capture the overall behaviour of the system. Implementations of supervisory control for CPS to govern the system behaviour are often made by hand. Preserving the specified behaviour and the model-predicted timing in an implementation is challenging, due to the need to simultaneously handle action timing, synchronisation, concurrency, pipelining and plant feedback . We introduce an execution architecture and engine to automatically synthesise an implementation of a supervisory controller directly from a model specification. The execution engine is guaranteed to execute a specification in a time- and behaviour-preserving fashion, even in the presence of action timing variations and including event feedback in a physical execution. We prove that the architecture and engine preserve the specified ordering of actions and events of the model as well as the timing thereof, up to a known bound. We validate our approach on a prototype production system.
The mining sector increasingly adopts digital tools to improve operational efficiency, safety, and data-driven decision-making. One of the key challenges remains the reliable acquisition of high-resolution, geo-referenced spatial information to support core activities such as extraction planning and on-site monitoring. This work presents an integrated system architecture that combines UAV-based sensing, LiDAR terrain modeling, and deep learning-based object detection to generate spatially accurate information for open-pit mining environments. The proposed pipeline includes geo-referencing, 3D reconstruction, and object localization, enabling structured spatial outputs to be integrated into an industrial digital twin platform. Unlike traditional static surveying methods, the system offers higher coverage and automation potential, with modular components suitable for deployment in real-world industrial contexts. While the current implementation operates in post-flight batch mode, it lays the foundation for real-time extensions. The system contributes to the development of AI-enhanced remote sensing in mining by demonstrating a scalable and field-validated geospatial data workflow that supports situational awareness and infrastructure safety.
The security of industrial control systems (ICSs) is crucial due to their integral role in critical national infrastructure. This study tackles the escalating challenges posed by sophisticated cyberattacks, especially those that are unknown and evade existing detection mechanisms. Despite extensive research, there is a notable gap in systematically comparing supervised and unsupervised learning models for anomaly detection, leading to inconsistent evaluations of their effectiveness. To bridge this gap, we developed a comprehensive anomaly detection framework to systematically evaluate these models, focusing on their capability to detect unknown attacks. Utilising operational data from the Secure Water Treatment (SWaT) testbed, we assessed six unsupervised and five supervised learning methods. Our findings reveal significant performance disparities: supervised models excel in precision but have higher undetected rates, whereas unsupervised models achieve superior recall at the expense of increased false alarm rates. This study provides critical insights into the strengths and limitations of both approaches, guiding the development of more robust ICS security solutions.
