This paper proposes a new sampled-data-based non-PDC controller for autonomous vehicle lane-keeping. The controller is developed within the Takagi–Sugeno (T–S) fuzzy system framework using a non-quadratic Lyapunov approach. It is designed for nonlinear vehicle dynamics and addresses external disturbances such as road curvature and wind forces, as well as actuator saturation due to steering angle limits. The H_∞ criterion is adopted in the controller design, while the robust stability requirements of the closed-loop system are expressed as an optimization problem with LMI constraints. The proposed method demonstrates robust performance under various disturbances and driving scenarios. Simulation results using inputs from the TuSimple lane dataset confirm its effectiveness in maintaining accurate lane-tracking. These findings highlight the method’s potential for practical autonomous driving applications that integrate vision-based perception with robust nonlinear control.
The growing popularity of Electric Vehicles (EVs) calls for sophisticated fault diagnosis systems to guarantee dependable and effective operation. An Artificial Intelligence (AI) -driven fault diagnosis framework for EV powertrains is presented in this research, with a focus on DC-source and battery-powered Switched Reluctance (SR) motor drives. To address fault diagnosis issues, a hybrid deep learning system combining a Quasi-Newton Neural Network and a Vortex-Transformer Dense Block (QNVorT-DenseNet) is proposed. For precise fault diagnosis, this model makes use of both local and global feature learning capabilities. For ensuring ideal model performance, hyperparameter tuning is also done using the Lion Fish Swarm Optimization (LFSO) technique. The proposed approach entails a methodical process that begins with collecting of fault data, preprocessing including type conversions and imputation of missing values and exploratory analysis using correlation heatmaps and distribution plots. The data representation is improved by integrating feature engineering model, which includes Min–Max scaling and new feature extraction. The QNVorT-DenseNet structure is then used for model training, and LFSO technique is used for hyperparameter optimization to guarantee the best learning results. The Python-based validation shows that the model has high diagnosis accuracy of 98.45
Lung cancer represents the leading cause of cancer-related mortality, largely due to its tendency to be diagnosed at advanced, metastatic stages. Early diagnosis enhances clinical outcomes in lung cancer, as it enables therapeutic intervention at stages when treatment efficacy is substantially greater. This paper presents a novel ResNeXt model with the novel Hiking Lyrebird Optimization (ResNeXt_HLyO) model for diagnosing lung cancer. HLyO is a hybrid algorithm integrating the Lyrebird Optimization Algorithm (LOA) and the Hiking Optimization Algorithm (HOA). This research involves the acquisition of CT images, which are subsequently processed via homomorphic filtering. Furthermore, the Pyramid Non-local UNet (PN-UNet) is employed for segmentation, and augmentation strategies are applied to improve the model’s ability to generalize and to mitigate overfitting. Moreover, shape-based features are extracted along with the Local Boundary Summation Pattern (LBSP), combined with entropy. Furthermore, lung cancer is identified using ResNeXt_HLyO, where HLyO represents a hybrid of the Lyrebird Optimization Algorithm (LOA) and the Hiking Optimization Algorithm (HOA). ResNeXt_HLyO achieved high values of True Positive Rate (TPR) at 92.767
The increasing sophistication of cyber threats aimed at Industrial Control Systems (ICS) necessitates stringent, real-time defense mechanisms. This research presents a hybrid architecture combining hardware-accelerated anomaly detection with lightweight cryptography to safeguard smart grid infrastructure. Using a comprehensive ICS dataset (covering DoS, injection, and scanning attacks), we deploy a quantized Random Forest model on the Google Coral Edge TPU. This edge-native approach achieves 99.85
Financial systems often exhibit chaotic behavior due to nonlinear interactions, long-memory effects, and sensitivity to uncertainties and external disturbances, which cannot be adequately captured by integer-order models. This paper proposes a unified robust control framework for stabilizing fractional-order (FO) chaotic financial systems subject to norm-bounded parametric uncertainties and external disturbances. Based on fractional Lyapunov stability theory, a nonlinear controller is designed to accommodate both 2-norm and infinity-norm uncertainty bounds, thereby ensuring robust stabilization of the closed-loop system and effective suppression of chaotic behavior, while providing a practical tool for mitigating instability in financial markets. Numerical simulations of the FO financial system validate the effectiveness of the proposed controller in achieving stabilization and maintaining robustness against perturbations.
Multi-Terminal Direct Current (MTDC) transmission systems, developed from DC transmission infrastructure, face dual challenges: abrupt inertia reduction and cross-regional frequency disturbance amplification. Moreover, the communication networks supporting distributed control are vulnerable to Denial of Service (DoS) attacks, potentially causing control failures. To address these issues, this study proposes a distributed collaborative control strategy that ensures both frequency stability and communication security under time-convergence constraints and DoS attack risks, thereby providing an efficient, attack-resistant power transmission and distribution solution for smart grids. Specifically, the study first designs a fully decentralized secondary controller with zero communication to achieve rational allocation of active power. Subsequently, the distributed cooperative control theory is introduced into the secondary frequency control of the transmission system, embedded within a multi-agent framework, to provide a fixed-time upper bound on the convergence time for frequency restoration. Then, under the risk of DoS attacks, attack detection and communication repair mechanisms are introduced, and resilient distributed secondary control is employed to ensure frequency restoration and rational power allocation. Experimental results are validated using a simulated MTDC transmission system. The study finds that the secondary frequency regulator can rapidly restore the subnetwork frequency to its rated value, with a convergence time of less than 0.5 s and minimal steady-state error. The secondary control strategy effectively achieves proportional allocation of active power, with a frequency restoration time increase of no more than 1 s under attacks and a response accuracy exceeding 90
Conventional grids frequently suffer from inadequate prominence, ineffective energy distribution and interruption vulnerability. These difficulties highlight the necessity of a shift towards smart grid based on Internet of Things (IoT). As a result, this study offers a Hybrid Renewable Energy Sources (HRES) based smart grid system that uses a novel shortest routing path algorithm and security approach. In order to effectively manage energy generation and storage, the smart grid system is developed by merging RESs, such as Photovoltaic (PV) systems, wind and batteries with an IoT web interface for monitoring and display. This study introduces a unique Recurrent Neural Network (RNN) with Glowworm Swarm Optimisation (GSO) algorithm for efficiently transporting data in the shortest path to the target. Furthermore, data security is ensured by using the Enhanced Elliptic Cryptographic Curve (EECC) technique, which authenticates and encrypts data exchange against unwanted access. For effective validation of this topology, the framework is validated through MATLAB Simulink and the experimental analysis. Moreover, the comparative analysis is made with the recently utilized topologies for proving the effectiveness of proposed smart grid system.
In this study, a fractional-order mathematical model is developed to describe the dynamic behavior of an armature-controlled DC motor by incorporating memory-dependent effects inherent in its electrical and mechanical subsystems. The classical integer-order motor equations are generalized using Caputo fractional derivatives, allowing a more flexible and realistic representation of armature current, angular velocity, and rotor position dynamics. The resulting fractional model reduces to the conventional DC motor model as a special case when the fractional orders approach unity, thereby ensuring consistency and physical interpretability. Numerical simulations are carried out using the explicit Toufik–Atangana (TA) numerical scheme, and the obtained results are rigorously validated through comparison with the benchmark Adams–Bashforth–Moulton (ABM) predictor-corrector method. Time-domain responses and sensitivity analyses are conducted to investigate the influence of fractional orders on the transient and steady-state characteristics of the motor, revealing that reduced fractional orders introduce non-local damping that effectively suppresses transient overshoots. Comprehensive computational profiling confirms the theoretical convergence of the TA scheme, while demonstrating exceptional efficiency-executing up to 75
In this paper, a distributed state estimation algorithm is proposed for systems under cyber-attacks in the presence of communication channel time delays. By considering non-identical, unknown, and time-varying delays in communication channels, a worst-case modeling approach is developed utilizing known upper bounds to guarantee system stability. To detect cyber-attacks, a detector is constructed to check the data and eliminate invalid data from the state estimation. All of the agents are locally equipped with a detector. The proposed discrete time estimator consists of 2 steps: In the first step, consensus is obtained on the information received from the neighbors with a delay, and in the second step, the measurement update is performed. By utilizing the concepts of augmented matrix, Kronecker multiplication, and stochastic stability, sufficient conditions to guarantee tracking and stability can be achieved. Additionally, the estimator’s convergence is proved in the networks with strongly connected graph and limited delays. Finally, numerical simulations demonstrate the performance and efficacy of the proposed estimator in the presence of cyber-attacks as well as time delays in the communication channels.
In Roll-to-Roll (R2R) precision coating systems, the unwinding section is a critical component that determines the stability of the entire tension control system, which primarily comprises unwinding and traction units. To address the limitations of conventional tension control methods, such as insufficient control accuracy and poor disturbance rejection during the unwinding process, this study proposes a cascaded dual-loop control strategy optimized by an improved artificial lemming algorithm (IALA). The outer loop employs a super-twisting sliding mode controller (STSMC), whereas the inner loop utilizes a proportional-integral controller with a nonlinear gain (NLPI). First, a nonlinear coupled dynamic model of the unwinding and traction units is established based on their tension transmission characteristics and operational mechanisms. Considering the nonlinearity and strong coupling of the system, a cascaded dual-unit control structure is developed. The outer tension loop generates the reference angular velocity via the STSMC, and the inner velocity loop performs angular-velocity tracking through the NLPI controller and generates the motor torque command, thereby achieving coordinated dual-loop control. The closed-loop stability and tracking convergence of the proposed cascaded control system are analyzed using Lyapunov theory. Because the STSMC contains multiple interdependent parameters that are difficult to tune using traditional empirical methods, this study introduces the IALA for controller-parameter optimization. The simulation results demonstrate that the proposed IALA exhibits superior optimization accuracy and convergence performance. Compared with conventional PID and ADRC controllers, the proposed IALA+STSMC-NLPI strategy exhibits superior dynamic response, disturbance rejection capability, and robustness. Under variable operating conditions, for the unwinding unit, the ITAE and IMSE are reduced by approximately 15.7
This article proposes a prescribed-time fault-tolerant attitude control scheme for flexible spacecraft subject to angular velocity constraints and actuator saturation. The developed controller is inherently continuous and ensures unwinding-free performance. To this end, an attitude control model for flexible spacecraft is first established by incorporating a selection mechanism between the Modified Rodrigues Parameters (MRPs) and their shadow set. The control framework is then designed using a backstepping approach augmented with a smooth time-varying scaling function to enforce prescribed-time convergence. Additionally, a Chebyshev Neural Network (CNN) is employed to estimate and compensate for modal vibrations induced by flexible appendages, as well as unknown external disturbances. A key innovation of the proposed fault-tolerant control law is its capability to simultaneously address the unwinding issue, suppress modal vibrations, and achieve practical attitude tracking to a prescribed residual set within a prescribed time, even under actuator faults and system constraints, without prior knowledge of the inertia matrix or disturbance bounds. The stability and prescribed-time convergence of the closed-loop system are rigorously established using Lyapunov theory. Comparative simulation studies verify the effectiveness of the proposed method and demonstrate its performance advantages over existing approaches.
Modern distributed and decentralised energy storage systems have many critical difficulties, one of which is the efficient and safe charging of batteries. For Li-ion, NiMH, and NiCd batteries, Multi-Stage Constant Current Charging (MSCCC) Federed Learning Based on Transformers is what we provide. Time series modelling using an attention-based architecture, privacy guarantee preservation via federated learning, and multiscale feature extraction using discrete wavelet transforms make up our technique. The suggested method relies on the encoding component of transformers to represent the interdependencies among the many states of a battery, including its charge level, voltage, current, and internal resistance. With federated learning, federated training of models can be performed within the battery management systems while keeping the data confidential and ensuring scalability and privacy. DWT allows creating reliable features based on consideration of dynamics of electrochemistry processes and trends observed during charging. Then the model outputs are incorporated into the adaptive MSCCC algorithm for tuning the current profiles under conditions of temperature and voltage constraints. When compared to existing centralised and federated approaches, the suggested DWT-Transformer-FL method outperforms them all in terms of prediction accuracy, Having a R^2 score of 0.90 and an average RMSE of 0.063, as determined by thorough examination of many datasets. When compared to the conventional methods of charging, which include a multistage process or constant current and voltage, Compared with the centralized and federated learning baselines, the proposed DWT–Transformer–FL model achieved an average cross-chemistry root mean square error (RMSE) of 0.063 and a coefficient of determination of R^2=0.90 . Under the common synthetic controller protocol, the proposed framework achieved competitive charging efficiency, complete attainment of the target state of charge (SOC), zero voltage and thermal constraint violations, and real-time-capable supervisory execution.
Quantum cryptography with the combination of random number generation and quantum key distribution has recently emerged as a key technology for ensuring secured communication in next-generation IoT sensor networks. This technology proposes advanced encryption and decryption standards when compared to classical methods. However, the existing IoT networks require adaptive mechanisms to defend against photon-based quantum attacks and entropy operation within sensor networks. The proposed quantum cryptographic security system is designed for building an attack-resilient sensor network with distributed sensors. Here, each sensor incorporates quantum random number generators and performs key exchange between sensors using photonic units to maintain a secure communication. The proposed work presents key research gaps by detecting and mitigating photon-number-splitting (PNS) attacks, random number generator (RNG) compromise, and fake-state attacks, which are not adequately handled by the classical methods. A hierarchical clustering framework support secure and energy-aware routing through a multi-parameter neighbour selection process that evaluates entropy health, node link trust, and delay. This research work proposes a set of adaptive algorithms to generate secure re-keying, quantum entropy validation with decoy-state attack verification. Simulation results show higher entropy-health preservation, lower overhead, and attack detection and mitigation when compared to the traditional methods.
Short-term electric load forecasting is critical for grid dispatch, yet existing Transformer-based models suffer from the quadratic complexity of self-attention and over-reliance on time-domain features. This paper proposes TF-iSTARformer, which innovatively replaces the distributed multi-head attention with a centralized aggregate-redistribute module that compresses all variate information into a global core representation, achieving linear complexity and robustness against abnormal channels. Concurrently, a time-frequency dual-domain fusion method is designed, where a Temporal Convolutional Network extracts transient dynamics and a Frequency Enhanced Channel Attention Module based on Discrete Cosine Transform mines periodic patterns without Gibbs artifacts, distinguishing our approach from previous methods that rely solely on time-domain modeling. Experiments on the Tetouan dataset demonstrate that TF-iSTARformer consistently outperforms iTransformer and other benchmarks across 24 h, 48 h, and 72 h forecasting horizons. For the Boussafou substation, TF-iSTARformer achieves the lowest MSE, MAE, and MAPE of 0.2074, 0.2526, and 1.7921 at the 24 h horizon, outperforming iTransformer by 6.3
The need for battery charging systems that are compact, highly efficient and intelligent management has become crucial due to the quick acceleration of the deployment of Electric Vehicles (EVs). In order to improve power quality, guarantee galvanic isolation, and provide accurate charge control for multi-battery configurations, this study proposes an architecture for an EV battery charging system. The system uses a Bridgeless Power Factor Correction (PFC) High-Resonant PWM converter, which removes rectifier diode losses and incorporates resonant switching for low-loss operation, to optimize input-side efficiency and guarantee unity power factor. This stage greatly lowers harmonic distortion while converting the AC input into a regulated DC link.The inverter provides high-frequency AC by converting the rectified DC, ensuring higher power density and effective galvanic isolation. The isolation transformer, a crucial part guarantees user and system safety while permitting a compact magnetic design because of the high working frequency. The parallel synchronous rectifier on the secondary side transforms high-frequency AC back into a regulated DC that is used to charge batteries. Multiple EV batteries charged simultaneously with its parallel arrangement, which improves temperature distribution and minimizes ripple.A Proportional Integral (PI) control approach is for dynamically regulating the parallel synchronous rectifier. The evaluation of this presented study is done via MATLAB, and the comparative analysis outcomes prove the importance of proposed model in terms of efficiency (96
The rapid development of Internet of Things (IoT) devices contributes to specific difficulties in ensuring the privacy and security of interconnected systems. Considering that cyberattacks have become more prevalent, there is a need for an efficient and scalable Intrusion Detection System (IDS) based on deep learning (DL) techniques for IoT, which can be turned into a highly complex system. Even though numerous studies lack implementation and structural details for combining IDS with Zero Trust (ZT). This paper introduces a Trust-Aware Zero Trust Intrusion Detection System using Multimodal Transformer Architecture (TAZTID-MMTA). The main purposes of the proposed TAZTID-MMTA framework for IoT environments are to integrate behavioral and network data for robust and adaptive threat detection. The features are normalized using z-score standardization and transformed into temporal sequences via sliding window segmentation. Subsequently, a dual-branch Transformer architecture is introduced, where behavioral and network modalities are encoded independently and fused using bidirectional cross-attention. Furthermore, a comprehensive trust computation mechanism integrates behavioral trust, historical trust, prediction confidence, and model-derived trust into a unified trust score. The experimental results of the TAZTID-MMTA technique are evaluated using the benchmark ToN-IoT dataset. The simulation outcomes demonstrated the superiority of the TAZTID-MMTA technique under different measures, with an Accuracy of 99.26
In this paper, we introduce a classification system for freezing of gait (FOG) detection of Parkinson’s patients using a single accelerometer and machine learning. The contributions of this work are that, first, the O’Day dataset of Parkinson’s patients with different genders, ages, and disease durations is applied to our machine learning framework, where FOG is detected using three-axis accelerometer data. Second, to develop a low-cost, low-power wearable sensor system for FOG detection, we then implement and conduct experiments where signals are collected from a participant who simulates walking with FOG and follows the O’Day recommendation. The low-energy-consumed IEEE 802.15.4 ZigBee module with a three-axis accelerometer is implemented, and the proposed machine learning framework from the first part is tested. Third, test protocols and evaluation cases are also proposed, where how the different datasets and the training and testing data sizes affect the classification accuracy can be investigated. The results indicate that using the O’Day dataset with only one sensor attached to the chest, we can achieve 99.4
Motion interpretation is a key part of advanced video analysis. Traditional models face problems such as network redundancy and weak modeling of spatio-temporal semantic interaction, making it difficult to achieve refined deep motion understanding. Therefore, a spatio-temporal motion understanding network based on the multi-layer perceptron is proposed. It achieves temporal fusion by extracting limb dynamic features and encoding global spatial features through a two-stream architecture. For insufficient single-modal information, a multi-modal spatio-temporal collaborative fusion algorithm is further proposed. The algorithm combines skeleton and visual images, uses dynamic target clipping to extract visual semantics, and achieves cross-modal feature alignment and deep coupling through dynamic modal fusion modules and multi-scale spatio-temporal reasoning mechanisms. Experiments showed that the proposed perceptron model had an accuracy rate of up to 89.6
Workplace safety is a major criterion in different organizations which need better monitoring systems for preventing accidents and provide conformity with safety regulations. Existing models for safety monitoring is generally based on manual analysis and conventional Machine Learning (ML) models. This work presents a workplace safety monitoring based Human Activity Recognition (HAR) model using Capsule Network (CN) with Optimized Vision Transformer (CN-OViT) for spatial and temporal feature extraction. The hyper-parameters of the ViT are optimized by the Enhanced Arctic Puffin Optimizer (E-APO). The proposed model improves feature extraction and spatial–temporal representation learning. The CN model captures hierarchical spatial relation effectively and the OViT model captures temporal features and HAR process. The approach is evaluated on Safe and Unsafe Behaviour dataset and provides better performance. The experimentation shows the robustness of the suggested model in attaining high accuracy and precision of 99.8
Accurate quantification of hydration and fluid-induced tissue variations is required for clinical assessment and performance monitoring. This work presents a machine-learning-assisted microwave resonator biosensor for quantitative non-invasive hydration assessment. Operating in the 2.2–2.8 GHz band, the proposed two-port metamaterial resonator detects water-dependent dielectric variations through the transmission coefficient ( S_21 ). The star-octagon fractal geometry incorporates optimized split, inter-ring, and coupling gaps, which provide strong electromagnetic confinement and maintain a stable resonance near 2.42 GHz. Both simulation and phantom measurements show a consistent resonance downshift as hydration increases. A corresponding reduction in the quality factor is observed because of increased dielectric loss. The sensor achieves a sensitivity of 5.7 MHz per 1 R^2 = 0.90 ) and MAE = 0.05 . Furthermore, SHAP analysis identifies the resonant frequency and quality factor as the dominant predictors of hydration variation.
