Industrial wireless sensor networks (IWSNs) have to continuously monitor process and transmit signals from the industrial environment with strict energy limitations. The impact of energy depletion–induced signal interruptions is significant on the reliability of tasks and the real-time industrial operation. To overcome this challenge, a signal-aware reliant task scheduling scheme (RTSS) for rechargeable IWSNs with energy harvesting mechanisms is proposed in this paper. The RTSS is based on a research gap–driven principle: For every researched limitation, a corresponding innovation is developed: a real-time signal–energy coupling model that continuously monitors signal traffic density and node energy levels during periods of active signal processing; a knowledge-learning mechanism that dynamically classifies nodes as available, idle, or low-energy, allowing for proactive task allocation decisions without reliance on static energy assumptions; a dynamic task reallocation strategy that seamlessly transfers tasks from energy-degraded nodes to available alternate processing nodes before signal interruptions; and an embedded energy harvesting module that recharges low-energy nodes during idle processing periods to ensure long-term network operability. The RTSS continuously updates its knowledge base over a scheduling interval, thereby enhancing the accuracy of the scheduling over time and continuity of its signal. The evaluation through simulation with a real-world computer numerical control (CNC) machine dataset of 200 tasks and 60 sensor nodes verifies the superiority of the proposed RTSS over the state-of-the-art RTSS baselines: joint power function–network-in-box (JPF-NIB), low-complexity channel resource allocation algorithm (LCCRAA), and enhanced energy optimization model (EEOM), in all critical performance metrics. In particular, the RTSS yields improvements of 10.48%, 10.87%, and 10.38% in the signal awareness task scheduling rate, successful signal delivery ratio, and energy harvesting rate, respectively, and reduces the average energy consumption by 8.63% and signal interruptions by 13.66% compared with the best performing baseline. The findings demonstrate the adaptability, scalability, and reliability of the RTSS for handling energy constraints in industrial applications with the requirement of uninterrupted signal processing and task execution.
Edge computing provides an effective solution for delay-sensitive services in the industrial Internet of Things (IIoT). However, the dependencies between tasks in the IIoT present challenges for task offloading and resource allocation in edge networks. This paper addresses the problem of computation offloading for dependent tasks in IIoT environments. We propose a dependent task model and jointly optimize subtask offloading decisions, transmit power, and local computation frequency for each terminal device, considering inter-regional mobile edge computing (MEC) collaboration. The objective is to minimize the weighted sum of task execution delay and energy consumption for terminal devices. Due to the combinatorial characteristic of offloading decisions among all tasks and their coupling with resource allocation, the proposed optimization problem is formulated as a mixed-integer nonlinear programming problem. To solve this problem, we introduce a single-offload candidate-space reduction rule under fixed transmit power and local CPU frequency. Based on this rule, a low-complexity Gibbs sampling algorithm is proposed to update the offloading decisions. Subsequently, for a given offloading decision, a deep deterministic policy gradient (DDPG)–based joint optimization algorithm is introduced to obtain the optimal transmit power and local CPU computation frequency. Simulation results indicate that the proposed algorithm achieves better performance than the considered benchmark schemes under the evaluated small- to moderate-scale IIoT scenarios.
The rapid proliferation of internet of things (IoT) devices has dramatically expanded the cyberattack surface, enabling sophisticated botnets to compromise large numbers of interconnected endpoints for coordinated malicious campaigns. Traditional signature-based defences are increasingly inadequate against these evolving threats, yet critical gaps remain in understanding how feature selection jointly optimises detection accuracy and computational efficiency across diverse botnet families under real-world deployment constraints. This paper presents a systematic evaluation of random forest (RF) and decision tree (DT) classifiers for multi-class IoT botnet detection using the N-BaIoT dataset—a large-scale benchmark comprising approximately 7.06 million network traffic instances from nine commercial IoT devices, spanning 115 features and 11 traffic classes, including Mirai and Bashlite/Gafgyt attack variants. An RF-driven feature selection framework based on Gini impurity scoring was applied to reduce the feature space from 115 to the 35 most discriminative attributes, enabling rigorous preselection and post-selection performance comparison. RF achieved 97.8% classification accuracy, and DT achieved 97.03%, with near-perfect precision, recall and F1 scores across Mirai attack categories. Critically, feature reduction decreased computational overhead by approximately 2 min and 46 s while yielding a marginal accuracy gain, confirming that lightweight models can satisfy the real-time constraints of resource-limited IoT environments without sacrificing detection reliability. These results establish a scalable, interpretable ML framework for proactive IoT botnet defence, with direct implications for the design of efficient intrusion detection systems in large-scale, heterogeneous IoT deployments.
Stress significantly impacts human well-being and is closely linked to sleep disturbances. This study proposes a system for detecting stress levels using physiological signals collected during sleep via IoT-enabled sensors. Key parameters-including heart rate, respiration, skin temperature, oxygen saturation, and heart rate variability-are analyzed to detect stress-induced variations. A seven-level stress classification model was developed using six machine learning algorithms: support vector machine, decision tree, na & iuml;ve Bayes, k-nearest neighbors, random forest, and logistic regression. After preprocessing, these models were evaluated using accuracy, precision, recall, and F1-score. The results demonstrate the potential of sleep-based physiological monitoring for early stress detection, offering a noninvasive and continuous assessment method. Future work includes extended validation on larger and more diverse datasets to improve model generalizability.
Passive localization is an important technology within Internet of Things (IoT) applications that allows for tracking and monitoring of people without needing active transmitters. This feature proves useful in privacy-conscious areas like security and healthcare, where active tracking or wearable devices are not practicable. Traditional radio tomographic imaging (RTI) methods suffer limitations due to vulnerability to environmental noise and rough spatial resolution. To overcome these issues, we suggest a new framework that combines a wireless mesh of 14 ESP32-transceiver nodes, multipath-sensitive RTI modeling, and a deep convolutional neural network (CNN) for accurate localization. The system records bidirectional received signal strength indicator (RSSI) measures to produce RTIs, which are then optimized by CNN to provide submeter accuracy. Experimental tests illustrate a precision of 92.81%, outperforming traditional RTI methods and other machine learning baselines. The suggested solution successfully alleviates line-of-sight (LOS) and nonline-of-sight (NLOS) problems, ensuring scalability and cost-effectiveness for real-world IoT applications.
Sensor placement is important for wireless sensor networks (WSNs) to obtain high coverage and energy efficiency. This paper presents a hybrid sensor deployment approach, in which circle packing theory is combined with an energy-efficient clustering algorithm (LEACH-C) for enhanced sensor coverage and energy efficiency. The proposed scheme is based on leveraging the geometric compactness of hexagonal spatial arrangements to decrease redundancy and communication overhead rather than conventional region-based or metaheuristic-based approaches. By applying LEACH-C clustering, the network lifetime is improved via load balancing between nodes. Based on extended statistical and simulation-based data, this approach shows better performance than classical placement methods in coverage efficiency, network longevity, and scalability. This method allows connecting geometric deployment models and practical applications of WSN while providing a scalable foundation for smart cities, environmental monitoring, and industrial automation applications.
Smart healthcare edge networks should be able to serve two purposes at once: to train federated machine learning models across a range of devices without violating patient privacy and to schedule other activities with latency constraints, like real-time patient events. Such methods as FFL-ANN attempt this by using fixed fuzzy rules, which do not work in the situation where the conditions of the network change in an unforeseen manner. In this paper, the framework FSI-ANN is introduced to combine particle swarm optimization to quality-aware model aggregation with ant colony optimization to adaptive real-time task scheduling and ANN-based predictions into a single framework. We experimented with FSI-ANN on 200 edge devices. It achieved 0.825 precision compared with 0.82 with FedAvg and 0.80 with FFL-ANN and reduced inference latency by 18%, 0.37-0.45 s. Throughput was maintained at 33 tasks/sec as compared with 27 of FedAvg. At burst load, the miss rate of the critical deadline was decreased by 90.2 percent and the energy consumed was decreased by 14.8% per round. The results suggest that adaptive learning using swarm is superior to the fixed rule-based approaches and simple averaging in the distribution of resources at the sustainable healthcare advantage.
This work investigates the trade-offs among energy efficiency, privacy, and adaptability in wireless sensor networks (WSNs), bridging the gap between legacy clustering protocols such as LEACH and the requirements of modern IoT deployments. While LEACH offers simplicity, its static, probabilistic approach is limited in dynamic and privacy-sensitive environments. To address these challenges, we propose DQN-FL-DP-LEACH, a coupled framework that integrates decentralized deep Q-networks (DQNs) for energy-aware clustering, federated learning (FL) for decentralized policy refinement, and an adaptive differential privacy (DP) mechanism for privacy-protected aggregation under a trusted base station. Extensive simulations show that DQN-FL-DP-LEACH increases network lifetime (last node death) by 1.7%-28.8% and first node death by up to 13.6% compared to LEACH across a wide range of scenarios. Under heterogeneous energy conditions, the protocol reduces performance degradation by 58.6%, experiencing only a 6.3% reduction in network lifetime versus LEACH ' s 15.2% decline. The framework achieves up to 9.4% higher packet delivery ratio and maintains fairness, with Gini coefficients below 0.1 for more than 90% of rounds. The privacy results are reported as an adaptive round-wise privacy-control schedule together with final cumulative accounting bounds, avoiding the ambiguity between instantaneous privacy control and cumulative privacy loss. These findings demonstrate that DQN-FL-DP-LEACH combines privacy-aware aggregation, adaptive clustering, and improved communication reliability in diverse WSN conditions. The results highlight that context-aware machine learning protocols with explicit privacy-control mechanisms can provide measurable advantages for future IoT networks while still exposing practical limitations related to update compression assumptions and local training overhead.
With the advent of billions of end devices connected to the edge of the internet, a large amount of valuable edge data is generated in production and life. When raw data is trained centrally, the risk of privacy leakage exists. Federated learning protects local data privacy by uniting multiple computing nodes for efficient machine learning without sharing data. However, federated learning faces the problem of data sparsity in practical applications, which makes global models difficult to train. In addition, explosive data growth reduces the availability of training models and tends to bring network transmission pressure. For solving these issues, this paper proposed a privacy-preserving strategy based on joint learning in edge computing scenarios. First, the concept of parallel over-parameterization is utilized to dynamically sparsely train joint learning models to reduce the resources occupied by communication and computation. Second, the privacy preservation of model parameters is investigated by adding perturbations to model parameters to ensure the security of model uploading to edge servers. Eventually, this paper conducted comparison experiments with HFL, DP-FedAvg, and SAFL. The experimental results demonstrate that the method proposed in this paper has certain advantages in privacy protection.
Wireless Sensor Networks (WSNs) are composed of energy-constrained sensor nodes that often operate in dynamic and hard-to-access environments, where battery replacement is impractical and node mobility can significantly degrade network stability. Existing clustering-based solutions frequently address either energy efficiency or mobility but rarely both in a unified and adaptive manner. Motivated by the need to simultaneously reduce energy consumption and ensure stable Cluster Head (CH) selection in mobile WSNs, this paper proposes an intelligent Fuzzy logic-based clustering framework that integrates residual energy, sleeping time, and mobility factor into the CH election process. Sleeping time is dynamically adjusted to reflect energy-saving behavior, whereas mobility awareness prevents unstable nodes from being selected as CHs. MATLAB-based simulation results demonstrate that the proposed approach significantly reduces energy consumption and extends network lifetime, achieving delayed node death and improved network stability compared to existing methods. This integrated design enables a unified and lightweight fuzzy framework suitable for resource-constrained mobile WSNs.
Intrusion detection system (IDS) is designed to analyze and monitor the network traffic to identify unauthorized access or attacks in an Internet of Things (IoT). IDS assists in protecting IoT devices and networks by recognizing malicious activities and preventing potential breaches. However, IDS struggled with accurately identifying cyber threats in IoT environments due to the presence of redundant features that lead to overfitting and reduced generalization. Feature selection is essential to enhance model robustness, remove misleading information, and ensure that only the most appropriate features contribute to the detection process that enhances overall performance and security. This research proposes a feature selection-based Di-Strategy Black-Winged Kite Algorithm (DS-BWKA) to detect the intrusion system in an IoT environment. In BWKA, tent mapping and dynamic cosine learning factor (DCLF) are incorporated as a di-strategy to initialize population and enhance global search ability for feature selection that enhances model performance. The proposed DS-BWKA-LSTM framework effectively addressed the class imbalance by selecting highly discriminative features and learning temporal patterns that enhanced the detection of minority attack classes. Long short-term memory (LSTM) is employed to identify IDS and capture temporal dependencies in network traffic, allowing for better detection of sequential attack patterns. The proposed DS-BWKA achieves a high accuracy of 99.56%, 99.67%, and 99.94% for CSE-CIC-IDS2018, ToN-IoT, and IoT23 datasets compared with existing methods like Long short-term memory autoencoders (LSTM-AE) and secured automatic two-level intrusion detection system (SATIDS), respectively.
Advances in digital sensors, digital data storage, and communications have resulted in systems being capable of accumulating large collections of data. In light of dealing with the challenges that large volumes of data present, this work proposes solutions to inference and filtering problems within the Bayesian framework. Two novel sequential Markov chain Monte Carlo (SMCMC) frameworks are proposed for nonlinear and non-Gaussian state space models, able to deal with large volumes of data (or observations). These are SMCMC frameworks relying on two key ideas: (1) a divide-and-conquer type approach computing local filtering distributions, each using a subset of the data, and (2) subsampling the large data and utilizing a smaller subset for filtering and inference. Simulation results highlight the large computational savings that can reach 90% by the proposed algorithms when compared with a state-of-the-art SMCMC approach.
Exploiting privacy-aware user task offloading in a multi-UAV-assisted edge computing system offers a new approach to reduce and balance energy consumption and latency. However, the complexity and variability of operating scenarios can make privacy-aware user task offloading strategies challenging. This paper examines a system with multiple ground users, UAV servers equipped with computational resources, and a cloud server. Specifically, to implement the optimization algorithm, we first model various behaviors and overheads during the interaction between UAVs and users, including UAV flight, energy consumption, delay, and privacy, and combine these four models to formulate the optimization objective as a minimization problem. Subsequently, a Markov decision process is established for this problem. The UAV flight trajectory, system resource allocation scheme, and user task offloading strategy are jointly optimized using the deep deterministic policy gradient algorithm to solve this minimization problem and determine the optimal task offloading strategy. Finally, simulation experiments demonstrate the convergence performance of the proposed algorithm, verifying its effectiveness in reducing energy consumption and delay while enhancing user privacy protection across different scenarios.
The sensor network consists of several low-powered electronic sensing devices that are strategically deployed in a predefined area to gather important information. The sensors have low computing capacity, constrained battery energy, and limited storage. Therefore, the sensed data should be communicated to the sink efficiently for further processing. In this work, a clustering-based multihop data transmission method is proposed using fuzzy logic. The proposed method, termed fuzzy-based cluster head (CH) selection for unequal clustering (FSUC), utilizes four sensor factors: residual energy, distance to the sink node, node density, and the average distance of neighboring nodes to select optimal CHs. The FSUC selects CHs locally through a distributed competition process, enhancing the scalability of the sensor network. The comparative analysis of FSUC is performed with six existing models that involve LEACH, EEUC, FL-SEP, MOFCA, FMCR-CT, and FLPSOC in three different scenarios. The performance evaluation metrics are the number of alive nodes, total energy of alive nodes, and energy consumption in each round. The performance is also evaluated for the metrics occurrence of first node dead, half node dead, and last node dead. The energy consumption prediction across all compared models is conducted using exponential weighted mean (EWM), autoregressive integrated moving average (ARIMA), and linear regression (LR). Furthermore, statistical validation through t -test is employed to examine the reliability of the proposed model. The computational complexity of the model is also analyzed along with the complexity–performance curve to assess its practical feasibility.
6G is transformational and brings revolutionary innovations in virtually all sectors, such as healthcare, autonomous systems, and manufacturing, through ultralow latency, massive connection, and far superior computing capabilities than any previous generation. It would, however, introduce several other challenges that even technology would not avoid, such as privacy, security, and network management. This paper proposes integrating blockchain technology into 6G wireless architecture to tackle the emerging threats above. We study how blockchain′s decentralized, immutable, and transparent features can strengthen access control and user privacy, as well as middleware efficiency in complex 6G environments. The study recognizes critical vulnerabilities coupled with 6G network slicing and virtualization and proposes the necessary solutions based on blockchain technology toward diminishing the problem. A case study showing how blockchain-enabled slices can integrate seams into the 6G core network completes the study. Ultimately, we highlight the future research paths of this convergence among blockchain and next-generation systems in wireless communications.
With the rapid development of artificial intelligence (AI) and the broad deployment of the Internet of Things (IoT), the smart grid as a novel grid has become an essential power management system. Forecasting a household's electricity consumption is crucial for optimizing the home energy management system. By collecting data and analyzing household electricity usage patterns, improper behaviors can be diagnosed more effectively. However, household electricity forecasting confronts many challenges due to the unpredictability of individual household consumption behavior. In this paper, we propose an intelligent electricity consumption prediction model based on CNN and a BiLSTM-Attention mixture model is used to forecast the short- term household load for a single family. First, we use the CNN network to extract the feature of the time series data of household load consumption. Then, the BiLSTM-Attention is used to predict the power consumption for each room and the total household based on the high dimensional feature data, which connects the time series data context and relationship. Next, we trained the BiLSTM-Attention network on the personal household electricity consumption dataset by the University of California. Finally, we compared the predicted value with the actual value and the prediction model with the traditional algorithm regarding MAE, RMSE, and accuracy performance. Experimental results demonstrate the efficacy of our proposed BiLSTM-Attention model in intelligent home electricity consumption prediction.
This paper proposes a novel hierarchical (HIE) sensor placement framework designed specifically for the structural health monitoring (SHM) of truss structures. The method strategically integrates zone-based modal sensitivity analysis, global optimization using D-optimality, and a modal assurance criterion (MAC)-based revival mechanism to overcome limitations of conventional sensor placement techniques—including spatial clustering, computational inefficiency, and disregard for initially excluded yet informative nodes. This four-stage methodology ensures spatial balance, enhanced modal observability, and robustness. A numerical study on a 19-node Howe truss shows that the proposed approach yields sensor configurations equal to or superior to those obtained by existing algorithms such as Fisher information matrix (FIM), effective independence (EI), and QR decomposition. The results highlight the practicality and computational efficiency of the proposed HIE scheme for sensor deployment in large-scale modular structures.
In 2021, Khan et al. suggested a scheme based on smart grid networks. They deployed the random oracle model to justify the security of their scheme formally. They also verified the security of the scheme using the AVISPA software tool. We studied this scheme and found some security issues. The scheme suffers from a confidentiality breach attack. In Khan et al.’s scheme, an adversary can track both the user and the server, and it does not provide user and server anonymity. An adversary can also impersonate a server. To overcome all these security issues, we propose a protocol for smart grid networks. In our proposed scheme, we maintain all the qualities of Khan et al.’s scheme and try to remove all its weaknesses.
Wireless sensor communication is a crucial technology nowadays; wireless sensor technology is not used only in terrestrial but also in underwater. Coverage and energy consumption are crucial challenges in underwater wireless sensor networks (UWSNs) due to the complexities of the aquatic environment and network connection costs. This study presents a hybrid fruit fly–bat optimization algorithm (FOA-BOA) for sensor node deployment in UWSNs. The technique combines both fixed and mobile sensor nodes to improve network connectivity and coverage rate. A novel multi-objective fitness function has been used for the optimization of coverage rate, connection cost, and network lifetime. It is used to optimize sensor node placement through modified bat and fruit fly optimization algorithms. Probability detection and a hybrid FOA-BOA-based multipoint search determine the optimal node deployment configuration. The hybrid FOA-BOA achieves an impressive coverage rate of 96.9%, far surpassing the whale optimization search (WOS) (74.4%) and moth flame optimization (MFO) (78.8%). Simulation results conducted in MATLAB reveal that the proposed hybrid algorithm significantly enhances network coverage, reduces energy consumption, and extends the network lifetime in comparison to two existing algorithms.
The drive train of the wind turbines needs to adapt to the rotational speed changes caused by wind speed fluctuations. The gearbox needs to maintain stable operation under nonsmooth working conditions, resulting in complex sensor signals and difficult extraction of fault characteristics. This paper explores the fault diagnosis method of rotating machinery based on acoustic emission records. Firstly, in order to reduce the dimensionality of the records while enhancing the component state information, the processing methods of framing, windowing, and periodic feature extraction are adopted to generate feature images. Secondly, the ResNet50 model is adopted for deep feature extraction. Meanwhile, an adversarial network is constructed to break the distributional difference between the source and the target domain. Further, the margin disparity algorithm is adopted between the classification and domain discrimination models to constrain the marginal distribution of the classifier and reduce the feature distribution differences between the source domain and the target domain. Experimental research and results show that this method has achieved relatively excellent recognition rates of component status across low-speed operating conditions. It provides new clues for the fault diagnosis of wind power transmission chains.
