2026 Volume 58 Issue 9
Published: 25 September 2026
Published: 25 September 2026
-
Select all|
-
2026, 58(9): 1-4.This report reviews the overall operating status of U.S. tech enterprises in China, compiling core operational indicators such as corporate revenue, capital expenditure, R&D investment, and employment scale. It objectively analyzes the multiple core benefits that operating in China brings to the United States, including talent absorption, market revenue growth, and reverse technology spillovers, while also acknowledging the practical difficulties and industry controversies faced by U.S. enterprises in China, such as declining profitability and production capacity relocation. The United States has no need to push for a full withdrawal of U.S. enterprises from China. Instead, it should strictly uphold the boundaries of security regulation, refrain from suppressing the business activities of U.S. enterprises in China that serve the local Chinese market, and consolidate the competitive advantages of the U.S. tech industry through market-based cooperation.
-
2026, 58(9): 5-9.Aiming at the problems of strong concealment of Internet of Things abnormal traffic and insufficient expression of artificial features, an artificial intelligence-based feature extraction method is proposed. Through traffic preprocessing, sample construction, local feature extraction by Convolutional Neural Network (CNN), temporal modeling by Bidirectional Gated Recurrent Unit (BiGRU) and attention enhancement, highly discriminative representations of abnormal traffic are formed. The results show that the proposed method can improve the feature expression ability of abnormal traffic and provide support for subsequent detection.
-
2026, 58(9): 10-14.The power cloud platform carries business systems such as dispatch support, measurement collection, and marketing services. Information Technology (IT) equipment and power environment equipment have long relied on independent monitoring systems, and protocol heterogeneity and alarm link fragmentation make it difficult to detect cross domain related faults in a timely manner. Based on this, the equipment monitoring requirements were analyzed, and the structural deficiencies of traditional solutions in collection coverage, perception latency, and disposal linkage were pointed out. A cross domain correlation monitoring system was proposed using Internet of Things technology, covering three levels: multi-source parameter collection and edge aggregation, equipment status recognition and fault prediction, and alarm grading and operation linkage. Through engineering verification, the application of Internet of Things technology has increased the perception coverage rate from 59.3% to 99.1%, and the average fault detection time has been shortened from 32.4 min to 4.6 min.
-
2026, 58(9): 15-18.To address the issues of signal fluctuations and computational limitations in real-time electroencephalogram classification for motor imagery tasks at edge nodes, research was conducted on edge-side classification methods. Explained the processes of online preprocessing, time-frequency feature compression, and lightweight model construction, and introduced the confidence smoothing discrimination mechanism. Combined with the BCI Competition IV 2a and BCI Competition IV 2b datasets, the results show that the method has an accuracy of 83.7%, an F1 value of 0.835, an average latency of 52.6 ms, and a model size of 1.7 MB.
-
2026, 58(9): 19-23.The intelligent transformation from digital to networking in the manufacturing industry have long remained fragmented, with decision-making processes overly reliant on static data, leading to mutual blockages among data flow, business flow, and decision-making flow, resulting in a pronounced coordination gap. To address this issue, a closed-loop collaborative advancement mechanism anchored in real-time Internet of Things sensing is proposed. This mechanism incorporates three sub-mechanisms: perception data-driven operations, intelligent transformation coordination, and networked collaborative execution, along with a comprehensive implementation pathway from sensing deployment to security governance. Through this mechanism, real-time sensing data is transformed into cross-tier collaborative drivers, enabling the transition from static planning to dynamic self-adaptation, providing a systematic solution for perception-driven collaborative transformation in manufacturing.
-
2026, 58(9): 24-27.To improve the accuracy of test-force analysis and the stability of loading control in internal loading tests of coal mine hydraulic supports, an analysis method based on an adaptive Internet of Things architecture is proposed. A multi-source sensing system involving pressure, displacement, strain and temperature is established. Data synchronization, dynamic fusion and adaptive regulation are implemented at the edge computing node. The experimental results show that the proposed method can reduce loading error, maximum overshoot and pressure-holding drift, shorten response delay, and enhance abnormal-state identification capability, providing a technical reference for intelligent testing and internal loading test optimization of hydraulic supports.
-
2026, 58(9): 28-31.Aiming at the strong concealment of grounding faults in distribution lines and the lagging response of centralized positioning, a grounding fault localization method under edge perception was studied. By analyzing the zero sequence current, zero sequence voltage, and transient energy on-site through edge nodes, a multi electric quantity joint section scoring and topology verification mechanism is constructed. Compared to traditional centralized methods, this method can shorten the data transmission and computation links, enhance the identification of high impedance, intermittent ground faults, and improve the stability of localization under complex disturbances.
-
2026, 58(9): 32-37.To address the issues of high continuous operation intensity, hidden fault evolution, and lagging traditional maintenance response in petrochemical electrical systems, this study studies the predictive maintenance mechanism driven by intelligent Internet of Things. Through multi-source status data perception, operational feature extraction, fault risk prediction, and closed-loop feedback on maintenance strategies, risk identification and graded handling of equipment such as power transformers and distribution equipment, motors, frequency converters, cables, and Uninterruptible Power Supply (UPS) systems are realized. Application results show that this mechanism can enhance fault detection in advance and reduce the risk of unplanned downtime.
-
2026, 58(9): 38-41.To address issues such as insufficient sensing coverage, severe data silos, and delayed fault response in the operation and maintenance of distribution transformer area, an intelligent operation and maintenance method based on Internet of Things technology is proposed. By deploying sensing terminals, it enables monitoring of various state parameters such as temperature, current, and partial discharge. A normalized processing mechanism for operational data is established to support the fusion analysis of multi-source heterogeneous data. Abnormal state evaluation metrics are defined to determine whether equipment deviates from normal operating conditions. Combined with a remote command issuance mechanism, a closed-loop operation and maintenance response chain is formed. Test results demonstrate that the proposed method significantly improves data collection coverage and recognition accuracy while reducing fault response time, showing broad application prospects.
-
2026, 58(9): 42-47.Aiming at the problems of hidden terminal behavior and complex risk link of the Internet of Things, a network full risk identification method based on terminal behavior analysis is proposed. By constructing a behavioral feature matrix, combined with autoencoder anomaly detection and graph correlation analysis, the unified identification of access, communication, service, propagation and control risks is realized, and the basis for risk early warning and hierarchical disposal is provided.
-
2026, 58(9): 48-51.To address the issues of small targets in transmission line inspection images, delayed positioning, and high review pressure, a drone Internet of Things inspection defect identification system was designed. Firstly, the end-edge-cloud collaborative architecture was analyzed. Secondly, the methods of image acquisition, improved detection by YOLOv8s, and alarm push were introduced. Finally, the system was tested and applied for verification. The test results show that all the indicators of this system are superior to those of ordinary drone inspections, and it improves the accuracy of identifying small defects and the efficiency of inspection and disposal.
-
2026, 58(9): 52-55.Aiming at the problems existing in many current defect detection optimization algorithms in practical industrial product defect detection, such as complex detection environment, limited test samples and long training period, a novel one-shot data comparison method is proposed. This method combines image binary logical operation with numerical operation to compute the grayscale image data of the sample and the object under test, realizing rapid comparison of difference data between the sample and the object. Considering the subtle errors inherent in image data, the algorithm is applied to defect detection on a public Printed Circuit Board (PCB) dataset, which verifies the effectiveness of the proposed method.
-
2026, 58(9): 56-60.Model Predictive Control (MPC) is widely used in mobile robot path tracking. However, its standard form suffers from limitations such as restricted prediction accuracy, inability to adaptively adjust weights, and fragility of the feasible region due to hard constraints. To address these issues, this paper proposes an improved strategy that incorporates feedforward curvature compensation, state-dependent dynamic weighting, and constraint relaxation. By introducing reference path curvature information to correct the prediction model, designing a continuously differentiable weight function to adapt to different operating conditions, and adopting slack variables with penalty terms to ensure the optimization always remains feasible. Theoretical analysis shows that the improved algorithm, while preserving the convex optimization structure, enlarges the attraction domain of the closed-loop system, reduces the upper bound of steady-state error, and exhibits better robustness against model mismatch and external disturbances.
-
2026, 58(9): 61-65.To address the issues of low segmentation accuracy and poor real-time performance caused by the highly variable hand poses and significant scale differences in complex backgrounds, we propose a multi-scale hand segmentation network based on a lightweight encoder and ShuffleNet With Coordinate Attention UNet (ShuffleCA-UNet). Firstly, we reconstruct the encoder using the channel segmentation and shuffling units from ShuffleNetV2, significantly reducing the model’s parameter count. Secondly, we embed the Coordinate Attention (CA) module into each basic unit and downsampling unit of the encoder to enhance spatial position features of the hand and suppress interference from complex backgrounds. Finally, we introduce an Atrous Spatial Pyramid Pooling (ASPP) module in the deep layers of the network to capture multi-scale contextual information, effectively addressing the scale-sensitive issues caused by changes in hand distance. The decoder employs bilinear interpolation upsampling combined with feature concatenation to effectively restore hand details. Experimental results on the public hand dataset EYTH (EgoYouTubeHands) show that this method achieves an F1-score of 85.54% and a Mean Intersection over Union (MIoU) of 87.45%, with only 1.409×107 model parameters. Compared to UNet, it improves by 3.65 and 3.35 percentage points, respectively, while reducing the number of parameters by 54.6%, striking a good balance between segmentation accuracy and inference speed.
-
2026, 58(9): 66-70.Defect identification for power equipment relies on large-scale annotations and closed-set assumptions, leading to overfitting, poor transferability, and failure to handle novel types of rare defects. A few-shot identification method is constructed for rare defects. By embedding phase-period awareness and thermal field smoothness priors into the feature network and integrating prototype metric meta-learning, class boundaries are delineated with extremely few samples. A physically constrained cross-modal generative network is designed to augment samples; Mahalanobis distance boundary detection is introduced to identify unknown defects, and prototype fusion enables incremental learning without retraining. Supported by only 5 samples per class for 7 types of rare defects, the proposed method achieves effective performance in terms of accuracy and the Area Under the Receiver Operating Characteristic Curve (AUROC), breaking the limitation of scarce samples and providing a feasible approach for rare defect diagnosis.
-
2026, 58(9): 71-74.Aiming at the problem that Large Language Model (LLM) deployment is constrained by computing power, memory, delay and energy consumption in the edge-end collaboration scenario, a joint optimization method of structured pruning, low-bit quantization and error compensation is proposed, and a dynamic scheduling architecture is constructed. Experiments show that the collaborative reasoning delay is reduced to 610 ms and the energy consumption is 11.4 J under 2 048 token, which takes into account the model compression and reasoning efficiency.
-
2026, 58(9): 75-80.Based on the PointNet++ framework, to address the issues of uneven density distribution and insufficient local topological information representation in indoor scene point cloud semantic segmentation, a method for indoor point cloud semantic segmentation that integrates density-aware collaborative optimization and topological enhancement is proposed. The density-aware collaborative module and topological feature branch are introduced to respectively deal with the problems of uneven data density and missing local structural information. Through the fusion of geometric-density-topological branch features, multi-dimensional information is achieved for collaborative complementarity. Experimental results show that the Overall Accuracy (OA) of the proposed method reaches 84.7%, the mean Accuracy (mAcc) reaches 68.1%, and the mean Intersection over Union (mIoU) reaches 57.4%, all of which are superior to the comparison methods.
-
2026, 58(9): 81-84.Combining Non-Orthogonal Multiple Access (NOMA) with cooperative relaying can enhance wireless network performance. To address the shortcomings of Orthogonal Multiple Access (OMA) technology, a scheme is introduced where the destination node directly decodes the remote users and combines the relay signal for selection. The closed-form expression of the user outage probability is derived. Simulation results show that, compared to OMA, the outage probability of the remote user in NOMA is significantly lower; increasing the power allocation coefficient decreases the outage probability of the near user but increases that of the far user; the larger the channel fading parameter, the lower the probability of interruption.
-
2026, 58(9): 85-89.To address the challenges of heterogeneous multi-source data, cloud response latency, and high false positive rates at individual edge nodes in industrial Internet of Things anomaly detection, we developed a collaborative edge-cloud perception method. This method employs lightweight feature extraction, operating condition-aware attention fusion, structured pruning—INT8 quantization—knowledge distillation compression, and collaborative inference across edge nodes to achieve joint discrimination of multi-source anomaly features. Experimental results show that the comprehensive optimization reduces model inference latency to 39 ms, with the edge-cloud collaborative perception achieving an accuracy rate of 94.3% and an F1 score of 93.7%, while the false positive rate is reduced to 3.6%. This method balances recognition accuracy with edge real-time performance, thereby enhancing the stability of anomaly detection in industrial Internet of Things under complex operating conditions.
-
2026, 58(9): 90-94.To address issues such as multiple types of intelligent equipment in universities, scattered operational data, delayed abnormality detection, and insufficient operation and maintenance scheduling efficiency, a Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) based intelligent monitoring and operation and maintenance optimization method is proposed. Through multi-source data collection, convolutional feature extraction, and time-series dependency modeling, it achieves equipment status recognition, fault warning, cause localization, and optimization of operation and maintenance tasks. Experimental results show that this method improves anomaly identification accuracy and operational response efficiency.
-
2026, 58(9): 95-98.Aiming at the problems that Internet of Things terminals are resource-constrained and vulnerable to attacks, a lightweight artificial intelligence-based method for abnormal behavior recognition of terminals is proposed. Firstly, a multi-dimensional behavior profile is constructed based on terminal traffic and system call features. Secondly, the neural network model with pruning and quantization is deployed on the gateway to realize real-time detection of malicious traffic. Finally, the Isolation Forest algorithm is adopted to identify unknown anomalies. Experimental results show that the proposed method achieves excellent accuracy, with Central Processing Unit (CPU) usage stabilized at 25%～30% and false positive rate lower than 3.5%. It balances edge detection efficiency and energy consumption, providing a new approach for Internet of Things security protection.
-
2026, 58(9): 99-102.Aiming at the problems of insufficient utilization of multi-source vital sign data and low efficiency of manual analysis in forensic pathological cause-of-death determination, an artificial intelligence-assisted cause-of-death determination method based on Internet of Things monitoring data is proposed. A multi-sensor fusion acquisition architecture is constructed to complete time-series data segmentation and event alignment preprocessing. Kernel Principal Component Analysis (KPCA)and graph neural networks are adopted to extract time-frequency features and cross-modal correlation features of vital signs.A deep classification network combining few-shot augmentation and class-weighted loss is designed, and attention visualization is introduced to achieve interpretable output of the determination process. Experimental results show that the proposed method achieves a F1-score of 0.86 in multi-class cause-of-death determination tasks, outperforming the baseline methods, and can provide technical support for assisting forensic pathological practice.
-
2026, 58(9): 103-106.In Wi-Fi-based action recognition scenarios, Wi-Fi signal transmission is susceptible to environmental and individual variations, while direct data transmission may lead to user privacy leakage. To address this issue, this paper proposes a novel personalized Federated Learning (FL) scheme enhanced by knowledge distillation. The scheme applies knowledge distillation on both the client and server sides to train personalized models and extract more generalizable public knowledge. Experimental results demonstrate that the proposed scheme achieves significant performance advantages and effectively mitigates the degradation of model recognition accuracy caused by data heterogeneity.
-
2026, 58(9): 107-112.To address bottlenecks such as severe occlusion of unordered workpieces, difficulty in positioning highly reflective surfaces, and low fault tolerance of rigid grippers, an integrated closed-loop grasping system of “perception - decision - execution” is designed. A hierarchical cross-modal network combining “3D global coarse positioning + 2D local fine positioning” is utilized to overcome occlusion and reflection, and the spatial coordinate system is unified through hand-eye calibration. An EtherCAT-based compliant gripper inspired by the Fin-Ray effect is developed to achieve millisecond-level hardware synchronous triggering and physical fault tolerance. A motion planning algorithm fusing Signed Distance Field (SDF) collision cost and grasping quality is proposed. Experimental results show that the proposed system achieves a target recognition rate of 98.7%, a grasping success rate of 96.5%, and a cycle time reduced to 3.1 s, balancing high compliance and high real-time performance.
-
2026, 58(9): 113-117.To enhance the automation and intelligence of badminton posture recognition, this study employed a deep learning framework, utilized OpenCV for video acquisition and preprocessing, and leveraged convolutional neural networks as the core technology to develop a posture estimation module. By extracting key skeletal point coordinates and temporal features, the system employs classifiers to identify and categorize typical movements such as high clears and smashes. The model was tested on a Graphics Processing Unit (GPU) server using the PyTorch framework, and a visual interface was developed. Results demonstrate that the system achieves high accuracy in recognizing specialized badminton movements.
-
2026, 58(9): 118-121.To solve the problem of independent perception models and fragmented information across different stages of the device lifecycle, which leads to decreased perception accuracy, a 4-layer state perception path based on Artificial Intelligence (AI) decision-making technology was constructed. Baseline modeling, degradation trend prediction, and failure mode discrimination were implemented in stages, and model parameters were continuously inherited through cross stage data transmission. Validation was conducted on a compressor unit in a petrochemical enterprise. The baseline modeling coverage of the proposed path scheme increased from 81.3% to 96.7%, the average absolute error of degradation prediction decreased from 0.183 to 0.094, the warning advance increased from 18.4 h to 43.7 h, the accuracy of failure mode discrimination increased from 79.6% to 93.8%, and the misjudgment rate during the aging stage decreased from 16.3% to 5.9%. Cross stage baseline inheritance and degradation trajectory reuse can effectively improve the device’s full lifecycle state perception performance.
-
2026, 58(9): 122-125.To address the imaging misalignment issues caused by optical axis eccentricity, focal plane drift, and stage positioning errors in wafer infrared defect detection, a stage reference acquisition, deviation modeling, closed-loop iterative compensation, and edge residual weighting correction method are constructed. The experimental results show that this method can reduce positioning errors, grayscale fluctuations, and missed detection rates, and improve the stability of weak contrast and edge defect recognition.
-
2026, 58(9): 126-130.To address the issue of time delay jitter caused by uneven allocation of heterogeneous network resources in the regulation process of virtual power plants, this study proposes a Heterogeneous Graph Attention-based Deep Reinforcement Learning (HGAT-DRL) strategy. By establishing a mapping model between power regulation services and communication service quality, it quantifies the physical characteristics of 5G, carrier, and optical fiber resources to achieve deep alignment between communication resources and power fluctuation rhythms. Experimental results show that the end-to-end average delay is reduced to 14.5 ms, the regulation command response success rate reaches 99.1%, and the maximum frequency deviation is 0.08 Hz. The proposed solution provides an effective pathway to enhance the real-time control accuracy of virtual power plants and offers theoretical support for optimizing the communication architecture of new power systems.
-
2026, 58(9): 131-134.To enhance the security and deployability of information encryption programming, the Advanced Encryption Standard (AES) block computation, parameter whitelist, interface encapsulation, and key identification management methods were employed to analyze database fields, interface communication, and Internet of Things terminal encryption applications. The results indicate that the Galois/Counter Mode (GCM) is suitable for interface authentication transmission and device security reporting, the Counter Mode (CTR) is appropriate for streaming data caching and high-throughput caching encryption, and the Cipher Block Chaining (CBC) mode is effective for legacy systems and large file compatibility encryption.
-
2026, 58(9): 135-139.The security level of services carried by China’s Optical Transport Network (OTN) is increasingly improving, which puts higher demands on the quantum computing resistance of key distribution. However, traditional encryption systems based on computational complexity have security risks under the threat of quantum computing. To address this issue, Quantum Key Distribution (QKD) technology is introduced into OTN, and a fusion architecture based on shared fiber transmission of quantum and classical channels is constructed. The security enhancement is achieved by combining wavelength division isolation and dynamic management of key pools. The research results indicate that compared to classical encryption and Post-Quantum Cryptography (PQC) schemes, QKD technology can effectively improve the key security, key update frequency, and quantum computing resistance of OTN, providing strong support for OTN security enhancement.
-
2026, 58(9): 140-145.In response to the problem of insufficient bandwidth between narrowband transmission networks with bandwidth less than 2 Mb/s and high-definition video transmission requirements of 720p and above in emergency communication scenarios, this study aims to explore methods for achieving stable and continuous transmission of high-definition emergency videos under existing bandwidth resources, in order to enhance emergency communication command and dispatch capabilities. In terms of methodology, specific solutions have been researched and developed from three aspects: encoding simplification, transmission protocols, and network regulation. In terms of encoding, based on H.265/High Efficiency Video Coding (HEVC), the encoding efficiency is improved by simplifying the intra prediction mode, controlling the bitrate, and implementing simple transformations. In terms of transmission protocols, design reliable transmission protocols based on User Datagram Protocol (UDP) and Real time Transport Protocol (RTP), and implement selective retransmission and packet loss coverage mechanisms. At the same time, the control layer constructs a network adaptive mode to monitor real-time parameters such as bandwidth, packet loss rate, and latency, and dynamically adjust the bitrate and frame rate. The research has achieved stable transmission of 720p, 30 f/s high-definition videos in a 1 Mb/s narrowband transmission network environment, with end-to-end latency controlled within 200 ms and packet loss recovery rate exceeding 85%. The verification through the real-case emergency communication drill for the Yellow River ice flood in the Inner Mongolia Autonomous Region shows that, compared with the traditional Transmission Control Protocol (TCP)/H.264 scheme, the subjective video quality is significantly improved, the video stuttering rate is reduced by more than 60%, and the smooth playback duration is increased from less than 30% to over 95%.The proposed integrated encoding transmission control technology scheme can effectively break through the bottleneck of high-definition video transmission in narrowband environments, significantly improve the reliability of video command in emergency communication scenarios, and has important engineering application value for improving emergency communication support capabilities.
-
2026, 58(9): 146-149.To address the bottlenecks of poor real-time performance, data redundancy, and identification difficulties in bridge structural health monitoring, a three-layer collaborative system of end edge cloud is constructed. The perception layer achieves high-precision synchronous sampling; the edge layer identifies anomalies and compresses data in real time through wavelet packet energy entropy; and the cloud layer completes multi-source fusion and damage identification. Field measurements show that the response time of the proposed system is reduced to within 3.5 s, the power consumption is significantly lower than that of traditional wireless schemes, and the accuracy is close to the wired benchmark, providing an efficient and reliable solution for intelligent operation and maintenance of bridges.
-
2026, 58(9): 150-155.Aiming at the problems in the teaching process of university computer laboratories, such as the difficulty in timely sensing students’ in-class behaviors, the dependence of classroom status statistics on manual inspection, and insufficient granularity of teaching feedback data, a student behavior analysis and management system based on multimodal large models is designed. The system adopts a hierarchical architecture of “client collection, server analysis, and teacher-side display”. The student clients regularly collect data including screen captures, active windows, process names, and keyboard-mouse activity levels. After compression, desensitization, and encrypted transmission, the server uses multimodal prompts and invokes large models to complete behavioral semantic recognition. The system classifies student behaviors into five categories: learning-related, auxiliary learning, non-learning, idle, and abnormal. Furthermore, evaluation indicators are constructed, including learning concentration, task relevance, idle duration, and abnormal behavior frequency. Experimental results show that the proposed method achieves an overall recognition accuracy of 91.2% on the programming experiment course test set, which is 22.9 percentage points higher than traditional rule-matching methods. It exhibits significant recognition advantages especially in boundary scenarios such as auxiliary learning and abnormal behaviors, and can more accurately understand the semantic correlation between screen content and classroom tasks. It can provide data support for experimental teaching supervision, student behavior profiling, and teaching process improvement.
-
2026, 58(9): 156-159.
-
2026, 58(9): 160-163.To improve the reliability of multi-service communication in underground coal mines, this study takes the industrial ring network of a coal mine as an example, and adopts methods including dual-core redundancy, underground closed-loop networking, explosion-proof and battery-backed power supply, and Quality of Service (QoS) isolation to investigate the bearer of fully-mechanized mining remote control and video backhaul. The results show that the single-link switchover recovery time of the ring network is 186 ms, the packet loss rate of mixed services is 0.05%, and self-healing monitoring and hierarchical scheduling are realized.
-
2026, 58(9): 164-169.In the practice of security governance for cloud-based Internet of Things platforms, the requirements of cybersecurity level protection generally encounter problems such as unclear object definition, scattered control bearing, and lack of systematic implementation paths during specific implementation processes. By focusing on key aspects such as the access layer, operating environment, data processing process, and interface exposure, a technical construction path for level protection based on the platform architecture is formed. The implementation effect is verified and analyzed through indicators such as the degree of security control coverage, changes in security operation monitoring capabilities, and level protection assessment results. The experimental results show that this path can effectively enhance the overall coverage level of security control, improve the ability to detect and handle abnormalities during platform operation, and demonstrate good compliance at the assessment level.
-
2026, 58(9): 170-174.A distribution operation and maintenance optimization scheme based on the Internet of Things architecture is proposed to address the problems of difficult fault location, low efficiency of manual inspection, and state misjudgment caused by wind power output fluctuations in 66 kV wind power collection lines. We designed a cloud edge end Internet of Things data collection architecture for three types of operation and maintenance problems, constructed a dynamic evaluation model for line status, developed a fault location and diagnosis method based on multi node collaborative measurement, and established a cloud based historical data-driven degradation trend warning mechanism. The practical application of the proposed system in a wind farm shows that it has good engineering application value.
-
2026, 58(9): 175-178.In response to the problems such as poor real-time performance, data silos, and delayed decision-making in the traditional manual inspection of process industrial parks, a visual inspection management platform based on digital twin technology is designed. The platform comprehensively utilizes Internet of Things, big data, Building Information Modeling (BIM), and 3D virtual reality technologies to construct a digital twin of the park’s physical entities and virtual space, achieving real-time mapping of the physical entities and virtual space. It realizes core functions such as online monitoring of equipment status, intelligent inspection along free and fixed routes, association of equipment ledgers, real-time tracking of automatic guided transport vehicles, automatic generation of inspection reports, and historical data traceability. Project practice has shown that the proposed solution significantly improves inspection efficiency and management refinement level, enhances fault prediction capabilities, optimizes inspection processes, ensures production safety, and provides a reusable technical path for the digital transformation of process industrial parks.
