ZHANG Shuyang. Sovereign Cloud-Sovereign Artificial Intelligence Conundrum: Policy Actions to Achieve Prosperity and Security[J]. Technology of IoT&AI, 2026, 58(8): 1-7.
Sovereign cloud and sovereign artificial intelligence have emerged as critical policy junctures in countries' efforts to balance digital sovereignty, national security, and economic competition. Driven by data sovereignty policy demands, geopolitical competition, economic nationalism, and the rapid advancement of artificial intelligence, states increasingly prioritize the transparency, controllability, and resilience of cloud infrastructure. Sovereign clouds have consequently expanded from data localization to the full stack encompassing software, operations, and governance. Generative artificial intelligence's deep dependence on cloud computing has intensified demands for regulation and autonomous control of cloud platforms, bringing sovereign artificial intelligence to the forefront of digital strategy. Although sovereign cloud can strengthen data control and governance autonomy, it entails enormous construction costs, limited service capacity, potential suppression of innovation, and fragmentation of the technological ecosystem, making it difficult to strike a balance between security and development. Rather than building a closed sovereign cloud, the key to advancing sovereign artificial intelligence lies in independent mastery of models, data, and talent systems, not necessarily in reliance on dedicated cloud infrastructure. Even on mature commercial clouds, risk tiering, enhanced transparency, and robust mutual trust and collaborative governance mechanisms are likewise preconditions for ensuring the autonomous and controllable development of artificial intelligence. These mechanisms safeguard national security and digital sovereignty while sustaining technological innovation and economic prosperity.
CHENG Junfeng, BAI Baoming, DING Wenhui, ZHOU Xin, GAO Shuilian, ZHAN Yue, XU Zhiping, LIU Bing. Research on Network Architecture Design and Key Technologies for Air-Space-Ground-Sea Integrated Network Towards 6G[J]. Technology of IoT&AI, 2026, 58(8): 8-11.
The 6G will achieve a leapfrog development from traditional terrestrial cellular networks to ubiquitous coverage integrating air, space, ground, and sea. By integrating satellite networks, aerial networks, terrestrial networks, and maritime networks, the air-space-ground-sea integrated network builds a three-dimensional communication infrastructure featuring full coverage, continuous connectivity, and all-scenario services. This paper systematically analyzes the design principles of the network architecture for the air-space-ground-sea integrated network towards 6G, proposes a layered decoupled and intelligently coordinated 4-layer 4-domain network architecture, and deeply explores key technologies such as multi-dimensional heterogeneous network cooperative access and dynamic resource scheduling, air-space-ground-sea integrated networking and cross-domain routing mechanisms, distributed edge computing and intelligent control domain architecture. This study provides a reference for theoretical research and engineering practice of the air-space-ground-sea integrated network.
WANG Botao, LUO Jing, WANG Xinnian. Research on Large Language Model Inspired Multimodal Decision Framework for Internet of Things[J]. Technology of IoT&AI, 2026, 58(8): 12-16.
A lightweight multimodal decision-making framework MindIoT inspired by a large language model is proposed to address the problems of difficult unified modeling of logs, images, and sensor timing in Internet of Things (IoT) scenarios, as well as asymmetric decision costs. This framework draws on the ideas of Low Rank Adaptation (LoRA), Retrieval-Augmented Generation (RAG), and hierarchical prompting to construct a “multimodal encoding gated fusion knowledge enhanced reasoning Bayesian action selection closed-loop feedback” process, enhancing edge decision-making ability without directly deploying a complete Large Language Model (LLM). The experiment was conducted on three simulation scenarios: smart home, industrial fault warning, and smart transportation, with accuracies of 95.0%, 88.5%, and 92.0%, respectively. The ablation results showed that RAG and LoRA brought about gains of approximately 3.9 and 1.6 percentage points, respectively. The experimental results indicate that using the lightweight mechanism of LLM to enhance traditional multimodal models is a feasible path, and “calibration before decision-making” should be the basic principle of IoT intelligent decision-making systems.
YUAN Zhigang, YIN Hesong, ZHAO Qiang, YAN An, LI Qi. Control Technology of Construction Defects of PHC Pipe Piles for Photovoltaic Supports Based on Internet of Things Monitoring[J]. Technology of IoT&AI, 2026, 58(8): 17-22.
During the static pressure pile driving construction of Prestressed High-strength Concrete (PHC) pipe piles, defects such as pile body inclination, sudden changes in the bearing stratum, and pile top damage are difficult to identify in real time. In response to the consistency requirements of verticality for photovoltaic array pile foundations, a multi-source sensing monitoring system based on Internet of Things is constructed. Inclinometers, hydraulic pressure sensors, and vibrating wire strain gauges are arranged corresponding to defect types. Comprehensive defect index Dt and risk index Rt are calculated in real time by edge nodes. Graded early-warning thresholds are calibrated through Receiver Operating Characteristic (ROC) curves of test piles, driving closed-loop control of pressure stopping and deviation correction. Verified with 1 422 piles at a 200 MW photovoltaic power station in North China, the qualified rate of verticality has increased to 96.8%, and the rework rate has decreased to 0.9%, indicating that the proposed technology has good engineering applicability in scenarios with large stratum lateral variation.
DING Bojun. Research on Anomaly Condition Identification of Electromechanical Systems Under Industrial Internet of Things Framework[J]. Technology of IoT&AI, 2026, 58(8): 23-27.
To solve the problem of precise detection of multiple faults in industrial electromechanical systems and meet the needs of anomaly recognition under complex industrial conditions, an industrial Internet of Things sensing system is built to collect multi-source operational data of electromechanical equipment and construct a multidimensional state parameter representation model. Using attention weight optimization method to fuse multi-source data, combined with the introduction of residual connections and dilated convolutions in Temporal Convolutional Network (TCN) to extract abnormal features, and relying on Auto Encoder (AE) reconstruction error and dynamic threshold to complete anomaly detection. The experimental results show that the algorithm has an industrial field detection accuracy of 98.7%, a detection delay of 68 ms, and excellent performance in identifying four typical faults such as circuit overload and temperature anomalies. Its comprehensive performance is superior to traditional AE, Isolation Forest (IF), and other comparative algorithms. This algorithm can effectively adapt to industrial edge deployment scenarios and provide technical support for intelligent detection of faults in electromechanical systems.
DONG Xingqiao. Coordinated Response Technology for Mine Disaster Rescue Based on Multi-sensor Fusion[J]. Technology of IoT&AI, 2026, 58(8): 28-31.
This study employs multi-sensor collaboration, reliability-weighted fusion, joint positioning of personnel and equipment, and dynamic path planning methods to analyze the effectiveness of disaster identification and coordinated response. The results show that the method achieves a disaster identification accuracy of 94.6%, an average positioning error of 0.82 m, and a response success rate of 95.4%.
NIE Xiangqian, LI Qimeng, XIAO Fan. Evaluation Model of Power System Information Communication Digitization Capability Based on Data Fusion[J]. Technology of IoT&AI, 2026, 58(8): 32-35.
Research on a digital capability evaluation model based on the diversity of multi-source data in power system information communication. The article analyzes the demand for data fusion and constructs an evaluation method that includes indicator system, feature processing, weight allocation, and evidence fusion. The research results show that the integrity of the model data reaches 97.42%, the consistency score reaches 96.88%, and the accuracy of level recognition reaches 94.16%, which can support the quantitative judgment of the digital capability of power information communication.
HUANG Bozeng, CHEN Xianghui. Data Automatic Collection Technology for Smart Water Meter Calibration Based on Internet of Things[J]. Technology of IoT&AI, 2026, 58(8): 36-40.
Aiming at the problems of reading mismatch, manual recording lag, and insufficient traceability in the multi location calibration of smart water meters, an data automatic collection technology for smart water meter calibration based on Internet of Things is proposed, and synchronous collection, edge verification, and error archiving methods are introduced. The experimental results show that the data recovery rate of the proposed scheme reaches 99.3%, the abnormal pulse removal rate is 96.5%, the mismatch interception rate is 98.9%, and the duplicate recording rate is 0.3%, all of which are superior to the comparative scheme.
ZHANG Zhen, WU Xiaoyu, LI Zhuo, LIU Jiayi. Design of Automatic Voltage Regulation System for Factory Power Supply Based on Internet of Things[J]. Technology of IoT&AI, 2026, 58(8): 41-45.
To improve the stability of the output voltage on the 10 kV side of the factory’s main transformer, a automatic voltage regulation system is designed using Internet of Things sensing, iFix centralized control platform, TAPCON 230 automatic voltage regulator, and on-load tap changer linkage method. The results showed that after being put into operation, the width of voltage fluctuations during the same period was reduced by 74.17% to 86.33%, and the average duration of fluctuations decreased from 176.9 s to 15.0 s.
PENG Weijia, LI Qingqing. Improved Wavelet-Adaptive VMD Joint Optimization Method for Biomedical Signal Denoising and Feature Extraction in IoMT[J]. Technology of IoT&AI, 2026, 58(8): 46-58.
Aiming at the prominent problems existing in biomedical signals under the distributed acquisition and transmission scenarios of the Internet of Medical Things (IoMT), including weak non-stationary and nonlinear characteristics, severe multi-source composite noise interference, the inherent difficulty of traditional algorithms in balancing noise suppression with feature fidelity, and excessive computational redundancy, this paper presents a joint optimization algorithm that integrates improved wavelet thresholding with adaptive Variational Mode Decomposition (VMD). Firstly, a composite interference model of physiological signals in the IoMT environment is constructed, and a nonlinear continuously adjustable wavelet threshold function is derived to achieve adaptive retention of key pathological features during noise suppression. Secondly, an adaptive VMD decomposition strategy driven by the dual criteria of Pearson correlation coefficient and information entropy is proposed, enabling automatic optimization of the mode number K and central frequencies, thereby addressing mode aliasing, baseline drift, and motion artifacts at the mechanism level. Finally, a multi-domain dynamic feature fusion mechanism based on mutual information contribution is established to enhance the representational capacity of pathological features and classification accuracy. Experimental results demonstrate that the proposed algorithm achieves an output Signal-to-Noise Ratio (SNR) of 28.64 dB after denoising, reduces mean square error, and maintains a waveform similarity of 0.986. The comprehensive accuracy of heartbeat classification reaches 98.72%, with a single-frame delay of only 9.4 ms. Under the stringent constraints of low bandwidth, low power consumption, and high interference in IoMT, the algorithm exhibits significant performance advantages. These findings provide theoretical foundations and algorithmic support for smart medical remote monitoring and intelligent pathological screening.
YIN Hesong, ZHAO Qiang, MA Yan, ZHU Bohan, LI Yinan. Construction Control Technology of Photovoltaic Pipe Pile Driving in Sandy Loam Based on High-Precision Positioning Technology[J]. Technology of IoT&AI, 2026, 58(8): 59-64.
To address the problems of large pile position deviation, missing resistance model, and low closed-loop control accuracy in the construction of photovoltaic pipe pile driving under sandy loam stratum conditions, a pile driving construction control technology based on high-precision positioning is proposed. Taking Real-Time Kinematic (RTK) and Inertial Measurement Unit (IMU) fusion calculation as the core, a high-precision solution model of pile position and attitude is constructed. A dynamic resistance modeling method jointly characterized by sandy loam particle size distribution and moisture content is introduced. The Proportion Integration Differentiation (PID) adaptive control strategy combined with deviation classification alarm is adopted to realize the full closed-loop precise control of the pile driving process. Experimental results show that after applying the proposed technology, the average plane deviation of pile position is reduced to 6.2 mm, the verticality error is controlled within 0.09%, and the response delay is less than 80 ms. All indicators are superior to conventional construction methods, showing good engineering applicability.
QIU Xingtong, LI Yudachuan. Research on User Viewing Behavior Feature Extraction Technology Based on Big Data Analysis[J]. Technology of IoT&AI, 2026, 58(8): 65-68.
Aiming at the problem of dispersed multi-terminal view log, inconsistent behavior label, and user’s preference recognition precision, this paper researches on the feature extraction of user’s viewing behavior. In this paper, the source and characteristic of viewing behavior data are described, and the methods of data collection, behavior order construction, label standardization, preference weight calculation and image fusion are presented. The test results show that the recognition accuracy is 91.6%, and the image stability is 89.7%. The technique is able to support detailed analysis of the viewing behavior.
LIN Wanfang, CHEN Fengzhen. Path Planning for Welding Robot Based on Improved ACO-GA Algorithm[J]. Technology of IoT&AI, 2026, 58(8): 69-74.
With the expansion of the application scope of welding robots, their path planning problem has received attention. To improve work efficiency and reduce time consumption, Ant Colony Optimization (ACO) algorithm is adopted to solve the shortest distance problem, and Genetic Algorithm (GA) is introduced to solve the shortest time planning problem of fixed-point paths. Firstly, ACO is combined with parameter optimization and random perturbation factors to improve the global search performance of the algorithm. Then, a method combining GA and B-spline curve interpolation is designed to achieve segmented path planning in the shortest time path planning. The simulation results show that the ACO path planning strategy can improve the iterative calculation efficiency and path planning performance; The optimized welding robot has less fluctuation in joint angle and angular velocity. Compared with the traditional particle swarm algorithm, the improved GA method takes 96.30 s, saves 17.61 s compared to the particle swarm algorithm, and saves 10.91 s compared to the A* algorithm. The simulation results verify the effectiveness of this path planning algorithm for welding robots.
WANG Pengyuan, Qin Kaimin, ZHUO Sheng. Digital Decision-Making Method of Power Generation Enterprises Based on Multi-agent Collaboration[J]. Technology of IoT&AI, 2026, 58(8): 75-79.
The operation of power generation enterprises involves unit scheduling, equipment maintenance and production optimization. Traditional empirical decision-making is unable to meet the requirements of rapid response and multi-data source analysis. A decision support system based on multi-agent collaboration is constructed, and a hierarchical structure is adopted to collect, clean and integrate data from Supervisory Control And Data Acquisition (SCADA) systems, Distributed Control System (DCS), Enterprise Resource Planning (ERP) systems and equipment monitoring data, generating a unified data structure. The intelligent agents for scheduling, operation, analysis and optimization process the task data respectively, and calculate the score of candidate solutions within a unified framework to generate the comprehensive decision result. The results show that the agent collaboration method outperforms the traditional method in terms of decision response time, scheduling efficiency, operation accuracy and operation cost control, providing decision support for power generation enterprises.
QIU Shu. Research on linkage Control Technology of Distributed Emergency Risk Warning Terminals[J]. Technology of IoT&AI, 2026, 58(8): 80-83.
In response to the high false alarm rate of traditional point source threshold warning in toxic gas leakage scenarios in chemical industrial parks and the difficulty of actively linking downstream terminals based on wind field evolution, a spatiotemporal correlation model for warning terminals is constructed, which integrates wind direction correction factors and time delay correlation. A linkage criterion based on local risk gradient and a neighborhood consistency collaborative control protocol are proposed to achieve adaptive election and distributed driving of terminal groups. The experimental results show that the proposed method outperforms the comparative scheme in all indicators, significantly improving the robustness and timeliness of emergency control.
YAO Xiangyu. Deep Learning Image Processing Method for Internet of Things Cameras Based on Multi-scale Feature Fusion[J]. Technology of IoT&AI, 2026, 58(8): 84-88.
To address the problems of detail loss and noise interference in Internet of Things camera images under complex scenes, a Transformer-based image processing method based on multi-scale feature fusion is proposed. By constructing a feature pyramid structure and introducing an attention mechanism, dynamic cross-scale feature fusion is achieved. Experimental results show that the proposed method outperforms traditional methods in both Peak Signal to Noise Ratio (PSNR) and Structural Similarity (SSIM) metrics, effectively improving image detail restoration capability and robustness in complex scenes.
QIAN Kunhong, CUI Xianliang, SHANG Penghui, XU Jiachen. Research on Collaborative Development Technology of Artificial Intelligence Technology Driven by Multi-modal Perception[J]. Technology of IoT&AI, 2026, 58(8): 89-93.
To enhance the recognition stability and collaborative processing capability of multi-source perception data in intelligent Internet of Things scenarios, a multi-source heterogeneous collection, cross modal deep semantic alignment, dynamic weight fusion, and edge cloud collaborative reasoning method are constructed. The experimental results show that the accuracy of the dynamic fusion model reaches 95.3%, with an F1 score of 94.5%. The false positive rate and false negative rate are reduced to 2.6% and 3.1%, respectively, which is better than the static fusion model and has good robustness and engineering adaptability.
ZHAO Yujia, XIONG Yanning. Application of Artificial Intelligence Technology in Computer Network Fault Detection[J]. Technology of IoT&AI, 2026, 58(8): 94-97.
To improve the efficiency of computer network fault detection, this study employs multi-source data fusion and intelligent recognition methods to analyze the application of artificial intelligence technologies in network fault detection. Focusing on traffic, logs, link status, and device operational data, the paper establishes mechanisms for identifying anomalous traffic, classifying fault types, and determining detection results, and validates these mechanisms using a continuous 30 d operational dataset. The results demonstrate that this method can reduce fault discovery time, improve fault identification accuracy, and provide support for optimizing network operations and maintenance responses.
QIN Kaimin, ZHUO Sheng, JING Hua. Research on Interface Adaptation of Power Generation Enterprises Based on Machine Learning[J]. Technology of IoT&AI, 2026, 58(8): 98-101.
For the multi-system interface connection scenario of regional power generation enterprises, aiming at issues such as field expression differences, semantic inconsistency, and low configuration efficiency, a field semantic matching model based on machine learning was constructed, and an interface adaptation technical path was designed. Firstly, starting from the characteristics of interface data, a unified feature representation method was established; secondly, the vector space semantic determination mechanism was introduced, combined with mapping rules to generate strategies to complete the interface configuration process; finally, a feedback data iteration mechanism was constructed to enable the model to continuously adjust the determination boundaries in the actual operating environment. The application results show that the proposed method has obvious advantages in matching accuracy and processing efficiency, and can adapt to complex interface environments.
CHEN Zhen. Water Resources and Hydropower Construction Progress and Quality Monitoring Technology Based on Deep Learning[J]. Technology of IoT&AI, 2026, 58(8): 102-105.
To improve the timeliness and accuracy of construction progress control and quality defect identification in water resources and hydropower projects, a collaborative monitoring technology integrating video, unmanned aerial vehicle imagery, sensor data and construction logs is developed using deep learning and multi-source data fusion methods. Validation is carried out based on 2 221 groups of progress images and 1 269 defect samples, covering 6 types of construction processes and 5 types of defects. The results show that the F1-score of the proposed technology for progress recognition reaches 91.5%, and the mAP@0.5 for quality defect detection reaches 90.5%, which can support refined management on the engineering site.
ZHOU Ke, CAO Wenqi. Research on Wine Quality Prediction Based on GA-XGBoost[J]. Technology of IoT&AI, 2026, 58(8): 106-110.
Aiming at the problems of strong subjectivity and low efficiency in manual evaluation of wine quality, as well as insufficient prediction accuracy and generalization ability of existing machine learning models, a wine quality prediction model based on GA-XGBoost is proposed. The Genetic Algorithm (GA) is adopted to optimize the optimal parameter combination of XGBoost. Multi-model comparison experiments show that the proposed model achieves comprehensive performance improvement on the test set, with accuracy, precision, recall and F1-score reaching 0.961, 0.958, 0.964 and 0.959 respectively, which is significantly superior to Support Vector Machine (SVM), random forest and XGBoost models. It realizes objective, fast and high-precision evaluation of wine quality, and provides reliable technical support for brewing process optimization and data-driven quality control in the wine industry.
LIU Yiwen, GAO Yajun, WANG Tao, PING Zengliang. Research on Application of Intelligent Warehousing AGV Scheduling Strategy Under Task Priority Constraints[J]. Technology of IoT&AI, 2026, 58(8): 111-114.
Taking the scheduling of Automated Guided Vehicle (AGV) in intelligent warehousing under task priority constraints as the research object, a task priority quantification, AGV state perception, scheduling decision-making, and path conflict handling model is constructed to address issues such as multi task parallelism, vehicle state differences, and path conflicts. The model is validated through simulation. The simulation results show that this strategy can shorten the average completion time to 14.8 min, the waiting time for high priority tasks to 4.9 min, the on-time completion rate for high priority tasks to 93.6%, the AGV utilization rate to 79.2%, and the number of path conflicts to 39, effectively improving the efficiency and operational stability of intelligent warehouse scheduling.
YIN Lichen. Design of Local Area Network Risk Monitoring System Based on Edge Computing Architecture[J]. Technology of IoT&AI, 2026, 58(8): 115-118.
To address the issues of hidden risks within local area networks and delays in centralized monitoring and response, this study designs a network risk monitoring system based on an edge computing architecture. It describes the system’s layered deployment approach, multi-source data integration methods, lightweight risk assessment mechanism, and alarm coordination process. Validation conducted in a simulated enterprise local area networks environment showed that the proposed method achieved an F1 score of 0.923, with the false positive rate reduced to 3.9%. At an input rate of 1 000 records per second, the average total latency was 31.4 ms, and Central Processing Unit (CPU) utilization was 15.6%, demonstrating strong near-real-time monitoring and edge deployment capabilities.
LI Hong. Intelligent Integration of Multi-channel Resource in Computer Networks Based on Multi-task Autoencoder[J]. Technology of IoT&AI, 2026, 58(8): 119-122.
To address the challenges of redundant information verification and low integration efficiency in directional target allocation for multi-channel resource processing in computer networks, this study proposes an intelligent integration method based on multi-task autoencoders. Under multi-channel task-driven scenarios, a channel resource integration task-specific encoder is constructed as the foundational resource processing framework. The Fixed Channel Allocation (FCA)+Dynamic Channel Allocation (DCA) synchronous allocation mechanism is designed to enhance information verification efficiency, enabling initial integration and achieving final intelligent integration through adaptive scheduling. Experimental results demonstrate that this method achieves throughput exceeding 1.2 Mb/s during multi-channel resource allocation, with significantly improved overall efficiency and superior performance.
SHENG Danning. Application and System Optimization of 5G in Intelligent Transportation Internet of Things—Case Study of Tianjin Transportation[J]. Technology of IoT&AI, 2026, 58(8): 123-127.
With the advancement of new infrastructure and the modernization of urban traffic governance, the traditional 4G-based transportation Internet of Things faces problems such as data silos, single application models, insufficient capabilities in dynamic road network regulation and complex weather adaptation. It can hardly meet the requirements of large-scale terminal concurrency and millisecond-level vehicle-road interaction, leaving intelligent transport long confined to shallow monitoring applications. Benefiting from its advantages of low latency, massive connectivity and edge computing, 5G enables the construction of a terminal-edge-cloud collaborative intelligent transportation Internet of Things system. Taking Tianjin’s transportation system as a research sample, this paper analyzes the local road network layout, hub traffic flow and coastal-specific meteorological characteristics from the perspectives of data fusion, road network scheduling and digital twin applications. It identifies the shortcomings in the current construction of 5G-enabled intelligent transportation Internet of Things, and proposes system optimization strategies suitable for coastal hub cities, providing a theoretical reference for the digital upgrading of regional transportation.
ZHUANG Zhiliang. Temporal Correlation Fault Location Method for Concurrent Alarms in Program Broadcasting Links[J]. Technology of IoT&AI, 2026, 58(8): 128-132.
After a device-level failure in the program broadcasting chain, downstream nodes often trigger consecutive alerts in a short time, making manual troubleshooting prone to interference from secondary alerts. The Temporal Correlation-based Fault Localization (TCFL) method is proposed, which converts multidimensional monitoring metrics into node anomaly scores. Through Granger causality tests, a directed correlation graph is formed, and root causes are traced via reverse random walks. Simulation replay results for single root causes show that TCFL achieves an overall Accuracy@1 (Acc@1) of 91.3%±2.1%, with an Mean Localization Latency (MLL) of 38.2 s, providing a foundation for bypass pilot validation.
ZHANG Mingzhi. Internet of Things Network Intrusion Detection Technology Based on Deep Learning[J]. Technology of IoT&AI, 2026, 58(8): 133-136.
Aiming at the problems of heterogeneous Internet of Things terminals, complex traffic characteristics, and covert intrusion behavior, a deep learning based Internet of Things network intrusion detection technology is proposed. By collecting terminal traffic, protocol fields, gateway logs, and access behavior data, a security feature matrix is constructed, and a Convolutional Neural Network - Long Short-Term Memory (CNN-LSTM) model is used to extract local traffic features and temporal correlation features, achieving recognition of attacks such as scanning, blasting, denial of service, and abnormal access. The experimental results show that the proposed technology can improve detection accuracy, reduce false positive and false negative rates, and enhance the security protection capability of the Internet of Things network.
LI Qingrong, YU Peiqi, WANG Ge, LIU Pengfei, WEI Jinmin, HUANG Hefu. Health Assessment of AGV Drive System Based on Ordered Clustering of Vibration Signals[J]. Technology of IoT&AI, 2026, 58(8): 137-143.
The health assessment of the drive system in Automated Guided Vehicles (AGV) is a core aspect of implementing health management for AGV drive systems. This study evaluates the health status of the drive system by analyzing the vibration signals generated during AGV operation. Firstly, the Local Binary Pattern (LBP) operator is employed to characterize the local features of the vibration signals, and combined with the Symmetrized Dot Pattern (SDP) transformation, the SDP image shape features of the vibration signals are ultimately extracted as key representations. Secondly, an ordered clustering algorithm based on the Genetic Algorithm (GA) is utilized to analyze the vibration signal features of the AGV, and the health level of the AGV is determined based on the clustering results. Finally, according to the health level clustering data, corresponding health labels are assigned to the feature vectors; the vibration features and their health label datasets are used to train a Support Vector Machine (SVM), and the optimized SVM model is employed as a tool for evaluating the health level of the drive system. The results show that the proposed method can effectively distinguish the differences among various vibration signals, thereby improving the accuracy of health assessment for AGV drive systems.
ZHANG Jipeng, LI Bizheng, SU Jinzhi, WANG Kai. Research on Adaptive PID Servo Control System for Permanent Magnet Motors in Aviation Turntables[J]. Technology of IoT&AI, 2026, 58(8): 144-148.
To address the issue of low-speed tracking in aviation turntable permanent magnet motors affected by inertia variation and friction disturbances, an adaptive Proportion Integration Differentiation (PID) combined with load disturbance compensation control method is proposed. This approach updates parameters through error zone partitioning and employs feedforward compensation to suppress disturbance oscillations. Experimental results demonstrate that the proposed method outperforms comparative approaches in all metrics, effectively mitigating tracking performance degradation caused by inertia variation.
XIANG Zhi, CUI Chuanxu. Research on Collaborative Operation Technology of Intelligent Comprehensive Mining Face Equipment in Coal Mines[J]. Technology of IoT&AI, 2026, 58(8): 149-152.
To enhance the continuous operation level of multiple equipment in intelligent comprehensive mining faces in coal mines, a method combining process analysis, mechanical system design, position detection, and motion coordination control was employed to study the collaborative operation path of comprehensive roadheader equipment. The results show that the collaborative system, constructed based on operational process characteristics and equipment matching constraints, can achieve orderly arrangement of key mechanisms, stable position detection of equipment, and multi-equipment coordinated control, effectively ensuring operational continuity and running coordination.
SUN Hainan. Design and Application of Real-Time Low-NOx Stable Combustion Optimization System for Boilers Based on Edge Computing[J]. Technology of IoT&AI, 2026, 58(8): 153-157.
A low nitrogen stable combustion system based on edge computing is designed to solve the problems of boiler low nitrogen operation delay, air distribution delay and stable combustion fluctuation. The field perception, edge computing and implementation architecture are described, and the combustion data preprocessing, nitrogen oxide (NOx) prediction, low nitrogen stable combustion decision-making and closed-loop control methods are introduced. The system application test is carried out against the background of a 240 t/h pulverized coal boiler. The test results show that after the system is applied, the NOx mass concentration decreases from 61.8 mg/m3 to 44.7 mg/m3, the carbon monoxide (CO) peak is 82 μL/L, the flame detection ratio is not less than 0.91, and the response time is 286 ms, providing technical support for the optimization of low nitrogen safety near the boiler.
LI Linping, MENG Xiaohui, HE Yanlong. Application of Digital Twin Technology in Fault Data Generation of Electromechanical System[J]. Technology of IoT&AI, 2026, 58(8): 158-161.
To address issues such as high costs of obtaining fault samples for electromechanical systems and insufficient coverage of operating conditions, a digital twin-based fault data generation method is proposed. Verification is conducted using cases of bearing wear, rotor imbalance, gear cracks, and winding degradation. The results show that the Root Mean Square (RMS) of bearing wear vibration increases from 0.68g to 1.57g, with a comprehensive sample quality score of no less than 0.86 and diagnostic accuracy consistently reaching 91.5% or higher.
LIU Wei, ZHANG Junhua. Data-Driven Intrusion Detection and Anomalous Behavior Early Warning Technology for University Networks[J]. Technology of IoT&AI, 2026, 58(8): 162-165.
To address heterogeneous terminals, long-tail services, and covert attacks in campus networks, we propose a data-driven intrusion detection and early warning framework. It employs a self-attention and Bi-LSTM cascade for fine-grained bidirectional flow semantic extraction, and incorporates a dynamic temporal graph to capture long-period threats like slow scanning and lateral movement, with an adaptive threshold adjusted to traffic tides. Real-network tests show over 23.2% higher detection rate than conventional machine learning methods, achieving 94.6% in complex scenarios and reducing false positives to below 1.8%, substantially outperforming mainstream graph neural network and boosting active network defense.
ZHAO Shuzhen, DU Xueying. Design and Implementation of Automatic Scoring System for Subjective Questions Integrating SIF and Doc2vec[J]. Technology of IoT&AI, 2026, 58(8): 166-171.
Aiming at the problems of low efficiency and strong subjectivity in manual scoring of subjective questions in online education, a text vectorization method SIF-D2V is proposed, which integrates Smooth Inverse Frequency (SIF) and Document to Vector (Doc2vec). Based on this method, an automatic scoring system for subjective questions is designed and implemented. SIF-D2V improves the semantic representation quality of text by integrating local keyword information and global topic semantics. The subjective question automatic scoring system is based on the Browser/Server (B/S) architecture and has functions such as teacher end exam management and automatic scoring, as well as student end online exams. The experimental results show that the SIF-D2V method has better average absolute error, root mean square error, and accuracy than single SIF or Doc2vec methods. The results verified the effectiveness of the method and provided reference for the research and application of automatic scoring of subjective questions.
PAN Dinglin, TAO Yilin. Design and Implementation of Intelligent Copywriting System for Corporate Brand Communication[J]. Technology of IoT&AI, 2026, 58(8): 172-176.
To improve the efficiency and consistency of corporate brand communication copy generation, an intelligent copy generation system is designed by integrating brand knowledge base, Transformer-based Bidirectional Encoder Representations from Transformer (BERT) vector indexing, Retrieval-Augmented Generation (RAG), and rule verification. Tests show that compared with general large language models, the proposed system increases the brand consistency pass rate by 26.4 percentage points and reduces the manual revision rate by 28.8 percentage points; compared with RAG, its comprehensive quality score is improved by 0.33 points.
