[Objective]Driven by the “dual carbon” goals, air circuit breakers are increasingly applied in new energy power grids due to their environmental friendliness and reliability. As the energy storage capacity improves, the breaking capacity of air circuit breakers faces severe challenges, making it vital to improve their load-bearing capacity. Air circuit breakers generally adopt grid plates to generate near-electrode voltage drop for limiting breaking current and assisting arc extinction. Understanding the motion law of charged particles, knowing about the sheath formation process, and calculating near-electrode voltage drop serve as the basis for research on the breaking of air circuit breakers. Compared to the cathode sheath, the anode sheath features short formation time, high simulation difficulty, and insufficient relevant research. Building microscopic models for the solution can improve both air arc theory and enhance breaking performance from both microscopic and macroscopic perspectives. [Methods]This study conducted reasonable assumptions for the physical breaking process of air circuit breakers. By taking the tiny spatial region near the anode as the research object, a microscopic transient dynamic model characterizing the dynamic formation process was built. Meanwhile, the collision and non-collision terms in the model were decoupled. Subsequently, the decoupled equations were solved by adopting the particle-in-cell Monte Carlo collision (PIC-MCC) method to obtain the spatial distribution of the electric potential, electric field, and charged particle phase space during anode sheath formation in the air medium. The near-anode voltage drop at various stages was obtained by analyzing potential diagrams and particle phase space distribution. Furthermore, the role of electronegative oxygen ions in the formation of the anode sheath during arc extinction was discussed, thus elucidating the causes of each stage of anode sheath development. [Results] Simulation results indicate that the mobility difference between electrons and ions leads to space charge accumulation, alters the potential distribution between electrodes, and forms strong near-electrode electric fields. The strong electric fields restrain electron absorption by the anode and lead to the formation of the anode sheath. Additionally, the formation of the anode sheath transitions from the initial “negative anode sheath” state to a typical “positive anode sheath” state. The accumulation characteristics of space charge near the electrodes and mobility differences at various stages are decisive factors for the formation of the negative sheath, transition to the positive sheath, and anode voltage drop variation. Electronegative oxygen ions accelerate the formation and stage transition of the anode sheath. [Conclusions]This study provides not only a microscopic charge-based interpretation of the formation of near-electrode voltage drops, but also an important theoretical reference for optimizing the design of circuit breakers and improving their breaking reliability and performance in the context of the “dual carbon” goals.

[Objective]Converter stations serve as critical interfaces between direct current (DC) and alternating current (AC) transmission systems, and their stable operation is crucial for ensuring the overall safety and power supply reliability of the power grid. However, key equipment, such as converter valves, transformers, and smoothing reactors, is prone to failure due to material aging, mechanical wear, degraded electrical insulation performance, or local overheating during long-term high load operation. These failures not only lead to a decrease in equipment performance, but also may cause system-level faults, seriously affecting the safe and stable operation of the power system. Therefore, effective monitoring of the operation status of converter stations, accurate evaluation of equipment performance, and early prediction of faults have become important research topics in the current power system operation and maintenance field. In this context, this paper proposed a method for main feature mapping of full service data of converter stations based on a Bayesian network. [Methods]Firstly, rough set theory was used to preprocess the massive business data of the converter stations. A condition attribute set that had a significant impact on the operation status and performance evaluation of the equipment was identified by calculating the dependency and importance between attributes. This process effectively reduced the dimensionality of data and avoided the interference of redundant information on subsequent analysis. Secondly, the attribute set was further reduced based on the logical relationship between the condition attributes and the decision attributes to ensure that the retained attributes could accurately reflect the change characteristics of equipment status. On this basis, a Bayesian network model was constructed to extract the full business data features after attribute reduction, which utilized its uncertainty processing ability. The probability dependency relationship between features was constructed to provide model support for equipment status evaluation and fault prediction. Lastly, a self-organizing feature mapping neural network was introduced. By continuously adjusting the connection weights and learning factors through a neuron competition learning mechanism, the network can achieve adaptive clustering of input data features, forming a business main feature mapping network. This method can effectively preserve the main feature information of data while reducing dimensionality of high-dimensional data, improving the accuracy and efficiency of feature extraction. [Results]The results indicate that the proposed method can efficiently extract key features from complex business data of converter stations, and the missing and false mapping rates of the mapping results remain at low levels, proving the effectiveness and reliability of the method. [Conclusions]The proposed method can provide technical support for the status monitoring and fault warning of converter station equipment, and offer a new implementation approach for the intelligent operation and maintenance management of power systems. This method not only improves the intelligence level of operation and maintenance management of converter stations, but also provides a reference for data analysis and feature extraction of complex industrial systems.

[Objective]In the construction of power grid engineering, there is a large amount of on-site equipment with intricate layouts and interwoven arrangements, which makes it difficult to accurately capture and extract some subtle and concealed risk features. This affects the image-based risk recognition effect and poses potential threats to the safe and stable operation of power grid engineering. To enhance the capability of on-site risk recognition and ensure the safe and orderly progress of power grid engineering construction, relevant research was conducted. [Methods]This paper proposed an intelligent image semantic coordination recognition algorithm for on-site risk features of power grid engineering. Smooth filtering technology was used to preprocess the images collected at the power grid engineering sites, thereby effectively suppressing noise interference. Meanwhile, key structural and detailed information was preserved as much as possible, so as to provide high-quality inputs for subsequent feature extraction. On this basis, the preprocessed image was input into a convolutional neural network (CNN), and the synergistic effect of convolutional and pooling layers was utilized to automatically extract multi-level image features. Specifically, the convolutional layer was responsible for extracting local and high-level semantic features, while the pooling layer performs dimensionality reduction on features and enhances the discriminative representation capability, enabling deep representation of complex scene information. Furthermore, natural language processing was integrated to perform semantic mapping and textual representation of image features. Through semantic analysis, subtle and concealed on-site risk features were identified, which thereby compensated for the limitations of a single visual model in semantic understanding. Moreover, the identified risk features were subjected to similarity calculation with the specified requirements of power grid engineering, and the risk zones on the power grid sites were accurately identified based on the matching results. [Results]To verify the effectiveness of the proposed algorithm, an image dataset comprising 10 000 high-resolution images was constructed, covering various typical construction scenarios, including cable laying, transformer installation, support operation, and grounding construction, under diverse temporal and meteorological conditions. The experimental results show that the proposed method can effectively identify on-site risk features, and the positive evaluation probability of the recognition results is stably maintained at over 96％, which means that in the vast majority of cases, the method can accurately identify risks. In addition, the method achieves a low false positive rate while maintaining a high recall rate, exhibiting superior overall recognition performance, which can provide reliable data support for the safety management of power grid engineering sites. [Conclusions]The proposed method achieves accurate recognition of risk features in complex power grid engineering sites by integrating multi-source technologies, effectively improving the accuracy of risk recognition, which is of important theoretical significance and engineering application value for reducing potential safety hazards during construction.

[Objective]As cloud computing and big data technologies advance rapidly, load balancing and resource allocation in edge computing scenarios have become a research hotspot. In the distributed system of Internet of Things in electricity, how to efficiently handle massive user requests and data service and ensure stable and reliable operations of tasks while reducing electricity costs and improving system resource utilization rate is a key technical challenge that urgently needs to be solved at present. In response, this paper proposed a load balancing and resource allocation method based on a distributed online algorithm. It meets the demands of real-time and efficient service operations by optimizing resource scheduling strategies in edge computing systems. [Methods]A distributed optimization model with edge computing capabilities was designed. This model could dynamically adjust the allocation of computing resources and task scheduling strategies through real-time situation analysis and load prediction. To solve the core optimization problem of load balancing and resource allocation, the Lagrangian dual decomposition method was employed to reconstruct and derive the original optimization problem. A distributed online algorithm based on mini-batch learning (MBL) was introduced. By integrating multi-time-scale optimization strategies, the algorithm could dynamic adaptive adjustment of multiple types of variables such as the service deployment location, the number of activated central processing units (CPUs), and computing resource allocation. A series of simulation experiments tested and confirmed the algorithm′s operation performance across different load conditions. [Results] The simulation experiment results indicate the significant performance advantages of the distributed MBL algorithm. Compared to the classic round robin (RR) and shortest job first (SJF) algorithms, the distributed MBL algorithm is superior in all key evaluation indicators. In terms of electricity cost control, the distributed MBL algorithm effectively controls the rising trend of electricity costs through intelligent scheduling and optimal allocation of resources, and the operating state is more stable under high-load conditions. In terms of resource utilization rate, the utilization rates of CPU, memory, and storage resources by the distributed MBL algorithm exceed those of the compared algorithms, which shows higher operational stability and resource utilization efficiency. In terms of task discard rate, the distributed MBL algorithm, with its advanced intelligent queue management and load prediction mechanism, stabilizes the task discard rate around 0.2％, a value well below the industry standard threshold, which fully reflects its high reliability in high-load scenarios. [Conclusions]In conclusion, the distributed MBL algorithm shows significant performance superiority and broad engineering application prospects in load balancing and resource allocation in the edge computing architecture. The algorithm can effectively reduce power consumption, improve the resource utilization rate, and suppress the task discard rate, providing an efficient and reliable technical solution for resource scheduling in edge computing environments.

[Objective]During the live-line maintenance of low-voltage distribution networks, issues such as uneven load distribution, voltage fluctuations, and increased network losses often occur. Traditional load transfer methods carry risks of overlooking power outage losses and equipment damage losses, resulting in deficiencies such as low power grid efficiency and affected power supply quality. [Methods]This study proposed a load transfer method for faulty equipment during the live-line maintenance of low-voltage distribution networks. The risk losses of distribution network maintenance and faulty equipment in the distribution network were calculated. An outage scheduling optimization model was established with the aim of minimizing risk losses. Meanwhile, a load transfer optimization model was established for the distribution network with the objectives of minimizing network losses, switch operation costs, and electricity sales losses. Constraints were imposed on the node voltage. The bacterial foraging optimization (BFO) algorithm was employed to solve these two models. This algorithm simulated the foraging behavior of bacteria to explore the optimal load transfer plan in the search space. A plan that can both minimize risk losses and achieve minimum network losses was sought in the search space. [Results] Experimental results indicate that during the live-line maintenance of low-voltage distribution networks, the proposed method can effectively evaluate equipment and power grid fault losses. It can control the network losses of the distribution network below 40 MW·h, and keep the load transfer stably at a relatively low level below 1.0 MW. [Conclusions]The proposed load transfer method for faulty equipment can effectively handle problems such as uneven load distribution, voltage fluctuations, and increased network losses during the live-line maintenance of low-voltage distribution networks. By minimizing risk losses and network losses, the method achieves better power grid operation effects, improving power supply quality and power grid efficiency. This study establishes an outage scheduling optimization model by taking into account the risk losses of outage and faulty equipment and a load transfer optimization model by integrating multiple objectives, fully considering various factors in power grid operation. This method provides a more scientific and effective load transfer strategy for the live-line maintenance of low-voltage distribution networks, contributing to enhancing the stability, reliability, and economy of power grid operation and being of great significance for ensuring power supply and improving power supply quality.

[Objective]With the continuous expansion of power engineering construction scale and rapid technological iteration, massive multi-source heterogeneous data have been generated. Such data are characterized by diversity, incompleteness, inaccuracy, and inconsistency. As a result, traditional cost prediction models lack sufficient ability to mine data correlation, and generally suffer from low prediction accuracy and poor stability, which restrict the scientificity of engineering investment decision-making. Aiming at the complex application scenarios of multi-source heterogeneous power data, this paper proposed a targeted fusion prediction algorithm to improve the accuracy of cost prediction and enhance the noise robustness of the model, so as to provide quantitative support for power project cost management and control. [Methods]This paper proposed a heterogeneous power project cost data fusion and intelligent evaluation algorithm based on denoising autoencoder network. The denoising autoencoder network was adopted to fuse data of technical and economic factors and complete feature extraction of the two types of influencing factors. Meanwhile, a self-attention network was constructed to mine the internal correlation among data of technical, economic, and environmental factors, so as to realize the optimized prediction of power project cost. [Results] The proposed algorithm exhibits excellent comprehensive performance. In terms of prediction accuracy, the accuracy of the proposed algorithm reaches 87.70％ in substation project scenarios and 88.90％ in line project scenarios, which is significantly improved compared with the comparison algorithms. In terms of error control, the mean absolute error and root mean square error of the proposed algorithm are greatly reduced relative to the comparison algorithms. In terms of goodness of fit, the coefficient of determination of the proposed algorithm is 0.94 and 0.95 for substation project and line project scenarios, respectively. [Conclusions]The heterogeneous power project cost data fusion and intelligent evaluation algorithm based on denoising autoencoder network can effectively solve various complex problems caused by power project data. The algorithm realizes feature purification through denoising autoencoder network and completes multi-dimensional data correlation modeling with the help of self-attention network, showing superior performance in cost prediction of both substation project and line project. With excellent overall comprehensive performance, the proposed algorithm can effectively improve prediction accuracy, provide high-precision quantitative basis for power project investment decision-making, and possess outstanding technical superiority and project applicability.

[Objective]Accurate identification and monitoring of power consumption behavior are crucial for the scientific formulation of power system dispatching plans. Currently, algorithms for power consumption behavior identification are mainly divided into two categories:machine learning and deep learning algorithms. The classification precision of machine learning algorithms still has room for improvement, and the generalization ability of the models is relatively insufficient. Although deep learning algorithms have unique advantages, they do not fully exploit the temporal features of data and are prone to overfitting, which thereby reduces classification precision. [Methods]To address the problems of insufficient mining of temporal features and the need to improve classification precision in power consumption behavior classification algorithms, a power consumption behavior classification model based on an improved double-layer bidirectional long short-term memory network and attention mechanism (IDL-BiLSTM-Atten) was proposed. The core implementation process of the model was as follows:first, the fused cumulative sum (CUSUM) algorithm was adopted to accurately identify the switching events of electrical equipment; second, steady-state features, transient-state features, and non-electrical parameters were integrated to construct a comprehensive feature set of electrical equipment behavior. To overcome the performance bottleneck of traditional neural networks due to the layer-by-layer serial connection architecture, an improved double-layer bidirectional long short-term memory (IDL-BiLSTM) network was built by introducing a data cross-layer transmission strategy, which allowed data to skip some or all intermediate layers and be directly transmitted to deep networks for processing, thus effectively enhancing the stability and classification precision of the model. Meanwhile, to strengthen the driving effect of key information on output results, the attention mechanism was embedded into the double-layer bidirectional long short-term memory model, further improving the computational efficiency and classification precision of the model in complex task scenarios. [Results]A total of 1 500 sets of training samples were generated using typical electrical equipment and used as input data to train the IDL-BiLSTM-Atten model. The results show that adding transient-state features and non-electrical parameters as input data is conducive to improving the classification precision of the model. Compared with traditional bidirectional long short-term memory (BiLSTM) and long short-term memory (LSTM) models, the IDL-BiLSTM-Atten model has superior performance in convergence speed and classification precision. In practical application cases, the IDL-BiLSTM-Atten model can achieve accurate classification of power consumption behavior, which verifies its application reliability in practical scenarios. [Conclusions]Compared with traditional models, the proposed model shows improvements mainly in two aspects:first, expanding the dimension of input features by incorporating transient-state features and non-electrical parameters into the input data of the model, which effectively improves the classification precision of the model; second, optimizing the model structure design by improving the BiLSTM model and embedding the attention mechanism, which enables the model to have better performance in both convergence speed and classification precision.

[Objective]With the large-scale integration of new energy into power grids, power system structures are becoming increasingly complex, and dynamic changes in operating states are intensifying. As an important technical means to ensure safe and stable operation of power grids, graphical dispatching operation simulation and rehearsal impose higher requirements on the real-time performance and accuracy of dynamic simulation. However, traditional central processing unit (CPU) serial simulation is limited by hardware parallel computing capability, making it difficult to meet the real-time requirements of large-scale power grid graphical simulation. This problem is particularly prominent in electromagnetic transient simulation of complex power grids, where the contradiction between high computational load and low simulation efficiency seriously restricts the timeliness and reliability of dispatching decisions. Therefore, it is necessary to construct a power system dynamic simulation scheme that considers both high accuracy and high efficiency. [Methods]To meet the requirements of graphical dispatching operation simulation and rehearsal, this paper proposed a power system dynamic simulation scheme integrating graphics processing unit(GPU) acceleration. First, the shifted frequency analysis (SFA) method was used to construct a dynamic phasor model of the power grid. An analytic signal was generated through the Hilbert transform and subjected to frequency shifting, mapping the original power signal to the zero-frequency domain, thus reducing computational complexity while retaining key characteristics of the transient process. Second, the delay-based linear multi-step composite (LB-LMC) method was used to decouple simulation tasks at the component level and network level. At the component level, parallel computing kernels were designed according to the single instruction multiple threads(SIMT) mode, in which a single thread was responsible for computing one component instance of the same type to ensure thread homogeneity. At the network level, a GPU high-performance linear algebra library was called to solve algebraic equation systems. Finally, a two-layer parallel architecture of “data parallelism+task parallelism” was constructed, and task scheduling and thread safety control were implemented using the OpenCL framework. Thread contention was resolved through a graph-based association matrix reduction method. [Results] The IEEE 14-bus system provided by DIgSILENT was used as the test system. A disturbance was introduced by disconnecting Line_0001_0005 at t=0.1 s for simulation verification. The results show that, compared with the electromagnetic transient simulation results of DIgSILENT PowerFactory, the proposed method has small relative errors in key parameters such as voltage, current, and generator rotor speed. In terms of simulation efficiency, in the IEEE 14-bus system scenarios, the total time consumption of the proposed method is 4.9 s, and the real-time factor is 0.245. The efficiency is improved by approximately 90％ compared with traditional CPU serial simulation and by approximately 67.8％ compared with basic GPU parallel simulation. As the network scale expands, the efficiency advantage of the proposed method is further enhanced, and ultra-real-time simulation capability is maintained in large-scale power grid scenarios. [Conclusions]The GPU-accelerated dynamic simulation method proposed in this paper effectively balances simulation accuracy and computational efficiency by combining the SFA dynamic phasor model with the LB-LMC decoupling method, and can meet the ultra-real-time simulation requirements of graphical dispatching operation simulation and rehearsal for large-scale power grids. The two-layer parallel architecture of “data parallelism+task parallelism” constructed in this paper breaks through the computing power limitations of traditional CPU serial simulation and effectively solves thread contention in parallel computing through graph-based thread safety design. This method can provide high-precision dynamic data support for dispatching operation visualization and intelligent deduction, and has important engineering application value for improving power grid dispatching decision-making and ensuring safe and stable operation of power grids with a high proportion of new energy sources.

[Objective]Nuclear power generation has brought tremendous convenience to human production and life. However, the rapid development of the nuclear power industry has also posed challenges to the safe storage of spent fuel. Thus, it is urgent to improve the neutron absorption performance of spent fuel storage racks in nuclear power plants. Boron (B) has a large neutron capture cross-section, which makes it a preferred component for neutron absorption materials of spent fuel. Iron-based amorphous alloys can accommodate more than 20% (atomic fraction) of B atoms and possess both excellent comprehensive properties and an advantage of low costs, which endows high-B iron-based amorphous alloy coating materials with broad application prospects. Meanwhile, gadolinium (Gd) is also an excellent neutron-absorbing element that can synergistically enhance the neutron absorption effect in B-containing materials. Therefore, high-content B element and an appropriate amount of Gd element are incorporated in the composition design. [Methods]A series of (Fe_0.72B_0.24Nb_0.04)_100-xGd_x(x= 2.5, 3.0, 3.5, 4.0, and 4.5) amorphous alloys were prepared via copper mold casting by adjusting the composition with the addition of Gd element with the atomic fraction ranging from 2.5% to 4.5%. The effects of Gd content on the amorphous forming ability, microstructure, and mechanical properties of Fe-B-Nb-based amorphous alloys were investigated. The amorphous forming ability of the amorphous alloys was characterized by X-ray diffractometer (XRD) and scanning electron microscope (SEM). The compressive properties were tested using a mechanical property testing machine. [Results]When the atomic fraction of Gd is 2.5% and 4.5%, the critical diameters of the amorphous alloys reach 2.5 mm; when the atomic fraction of Gd is 3.0%, 3.5%, and 4.0%, the critical diameters of the amorphous alloys reach 3 mm. In addition, when the atomic fraction of Gd is 3.5%, the amorphous alloy exhibits the optimal amorphous forming ability, with the full width at half maximum (FWHM) of the XRD peak reaching 0.425 4°, the corresponding supercooled liquid region width being 88 K, and the reduced glass transition temperature being 0.582. The fracture strength is 2 197 MPa. [Conclusions]With the increase of Gd atomic fraction from 2.5% to 4.5%, the amorphous forming ability of the alloys first increases and then decreases. This indicates that the addition of an appropriate amount of Gd element can effectively improve the amorphous forming ability of the alloys. The above-mentioned amorphous alloys exhibit the brittle characteristics, and their fracture strengths decrease with the increase of Gd atomic fraction. The novel amorphous alloys prepared in this paper have excellent mechanical properties and good glass forming ability, which can provide a new alternative for the development of neutron absorption coating materials used in spent fuel storage racks.

[Objective]Welded joints of aluminum/titanium dissimilar metals can combine the advantages of aluminum alloys and titanium alloys, having low density, high specific strength, corrosion resistance, and other characteristics, and exhibiting excellent electrical conductivity and heat transfer capabilities. Thus, they have been widely used in aerospace, automotive, electronics, ships, and other fields. The technical difficulties in the manufacture of aluminum/titanium dissimilar material components lie in the significant differences in physical and chemical properties and metallurgical incompatibility between aluminum alloys and titanium alloys. Direct welding will produce a large number of brittle and hard intermetallic compounds such as TiAl₃, TiAl₂, and TiAl in the weld, resulting in welding hot cracking during the solidification process of the molten pool. To improve the performance of the welded joints of aluminum/titanium dissimilar metals, a pure nickel transition layer was adopted to the weld, which regulated the metallurgical reaction in the weld and reduced the welding heat input on the low-melting-point base material side. [Methods]The base materials were 4 mm thick 5052 aluminum alloy and Ti₆Al₄V titanium alloy. A 0.1 mm thick pure nickel strip, as a transition layer, was applied to the welding end surface of the aluminum alloy, and Ti₆Al₄V welding wire with a diameter of 1.2 mm was used as the filling material. The cold metal transfer (CMT) welding method was adopted. After multiple rounds of parameter optimization, the optimal welding parameters were determined to be the wire feed rate of 6 m/min and the welding speed of 15 cm/min. Comparative tests were carried out with and without a pure nickel transition layer, and the microstructure and mechanical properties of the joints were analyzed after welding. [Results]The results show that complete and effective joints cannot be obtained after welding without the use of the transition layer, and there are welding defects such as non-fusion and non-penetration in the weld. Dendrites formed from the high-melting-point base material extend into the low-melting-point base material in the solidification process, which results in the formation of penetrating cracks. Complete and good welded joints are obtained after welding of aluminum/titanium dissimilar metals with the use of the pure nickel transition layer. No welding defects are found in macroscopic testing and microscopic analysis, and the transition layer is completely melted during the welding process. After welding, the transition layer forms an element mixing zone of about 300 μm in situ. By using X-ray diffraction (XRD), scanning electron microscopy (SEM), transmission electron microscopy (TEM), and other modern analysis and detection methods, it is determined that a ternary reaction occurs in the region, producing AlNi₂Ti, NiTi₂, Al₃Ni₂, and Al₃Ni as the main phase products. The joints are all broken in the weld zone during the tensile process, with an average tensile strength of 62 MPa. [Conclusions]Aluminum/titanium dissimilar metals were welded using the CMT welding technology. The welding end surface of the aluminum alloy base material was coated with a 0.1 mm thick pure nickel strip as a transition layer. No welding defects exist in the welded joint. The transition layer completely melts during the welding process and participates in the metallurgical reaction in the weld, forming an approximately 300 μm thick element mixing zone in situ, where the major metallurgical reactions occur and the primary phases are formed. The addition of the transition layer can effectively reduce the heat input on the low-melting-point base material side, and avoid the generation of a large number of Ti-Al brittle intermetallic compounds in the weld, which can result in welding defects. Compared with the welded joint without the addition of the pure nickel transition layer, the use of the transition layer improves the fluidity of the molten pool, and achieves welding penetration on the back of the welded plate.

[Objective]In the machining of complex surfaces with five-axis CNC machine tools, frequent variations in tool orientation induce nonlinear motion of the tool tip, making the control and compensation of contour errors extremely challenging. This is particularly evident when machining trajectories with large curvature, where frequent tool-orientation adjustments reduce the tracking accuracy of rotary axes, thus significantly enhancing the disturbance effect of orientation contour errors on position contour errors. [Methods]To address this issue, this paper proposed a cross-coupled control (CCC) structure based on online contour error estimation. The five-axis CNC machine tool was divided into an orientation layer and a position layer, corresponding to the rotary axes and translational axes, respectively. A periodic update method for the tool tip position was designed. While tool-axis direction error compensation was performed in the orientation layer, the tool tip position offset caused by orientation variations was fed back to the position layer in real time. On this basis, a cross-coupled controller was designed and variable cross-coupled gains were introduced. This dual-layer CCC structure made the five-axis system equivalent to a three-degree-of-freedom (3-DOF) position control system subjected to bounded disturbances. [Results] To verify the effectiveness of the proposed method, two sets of simulations were conducted using a three-dimensional helical trajectory with constant curvature and a petal-shaped curve trajectory with significant curvature variation. The simulation results show that, compared with other methods, the proposed dual-layer CCC contour error control method reduces the tool tip position contour error by approximately 14.11％ and 17.51％, respectively. Meanwhile, the tool-axis orientation contour error remains basically stable and maintains high accuracy, which is consistent with the conclusion derived from the tool tip position update method. [Conclusions]The proposed dual-layer CCC method effectively addresses the online control and compensation of contour errors in five-axis CNC machine tools. While maintaining the accuracy of tool-axis orientation contour errors, it significantly reduces tool tip position contour errors and improves trajectory tracking accuracy, fully verifying the effectiveness of this control structure in online contour error control of five-axis CNC machine tools.

[Objective]The operational adaptability of modern aeroengines should be urgently improved to adapt to supersonic cruising and extreme flight conditions. Conventional rotational speed regulation methods face challenges such as aerodynamic losses and efficiency degradation. Variable geometry turbines (VGTs) can significantly optimize flow field distribution and enhance the engine′s operational adaptability and fuel efficiency by dynamically adjusting the guide vane throat area (such as by changing the stagger angle or thickness). This paper aims to systematically summarize current research on configuration design, tip flow leakage suppression techniques for adjustable vanes, motion accuracy, and reliability of guide vane adjustment mechanisms, elucidate simulation analysis methods for adjustment mechanisms under the effect of multi-factor coupling, and provide theoretical support for developing lightweight, high-reliability mechanisms. [Methods]This study compared the topological structures and kinematic characteristics of mainstream guide vane adjustment mechanisms, summarizing the advantages and disadvantages of two structural types. Meanwhile, the study presented structures and methods for reducing guide vane tip clearances and reviewed the effectiveness of different geometric configurations of adjustable guide vanes in suppressing tip clearance flow leakage. It introduced modeling approaches involving error models and rigid-flexible coupling dynamic models and analyzed the effects of dimensional tolerances, kinematic pair clearances, aerodynamic loads, and thermal deformation on adjustment accuracy. [Results]After summarizing various mechanism configurations and research methods, the corresponding research content and key technologies were provided. This reveals that the primary current research directions for simulation analysis of guide vane adjustment mechanisms include mechanism configuration design, tip flow leakage suppression, building of accurate multi-factor coupling models, and multi-dimensional reliability analysis (such as motion accuracy reliability, seizure prevention reliability, and structural strength reliability) based on simulation results. [Conclusions]Current research on VGT guide vane adjustment mechanisms focuses on guide vane adjustment schemes, transmission mechanisms, leakage control, and accuracy and reliability analysis, effectively optimizing their aerodynamic performance, motion accuracy, and reliability. Future research should focus on the following aspects to promote the further development of this technology:developing novel topological structures to enhance mechanism stability and adjustment ranges, conducting in-depth studies on the adjustment accuracy and aerodynamic performance of variable-thickness guide vane configurations in actual operating conditions, building more rational simplified multiphysics coupling models that simultaneously incorporate key practical factors such as dimensional deviations, clearances, thermal deformation, and aerodynamic loads, and improving the reliability assessment system by integrating methods like fault tree analysis (FTA) and optimizing failure-prone components. By carrying out multidisciplinary collaborative innovation, the development of guide vane adjustment mechanisms toward light weight and high accuracy can be promoted to provide critical technical support for performance breakthroughs in advanced aeroengines.

[Objective]Against the background of the deep integration of 5G, the Internet of Things, and the industrial Internet, network attack methods have become increasingly complex and covert. Traditional intrusion detection methods are affected by non-stationary distributions caused by feature collinearity, making it difficult to achieve active defense with high coverage and high stability. To improve the detection capability for active intrusion behaviors in complex network environments, this paper proposed an efficient active intrusion detection method for computer networks to enhance the comprehensiveness, stability, and real-time performance of intrusion detection and provide technical support for the deep integration of artificial intelligence and network security. [Methods]This paper constructed a three-level progressive active intrusion detection framework. The isolation forest algorithm was used to perform outlier cleaning and feature normalization on historical intrusion data packets. By constructing multiple isolation trees and calculating anomaly scores based on sample path lengths, noisy samples were identified and removed, thus constructing a high-quality training dataset. A damped incremental statistical algorithm was introduced to aggregate the temporal features of intrusion behaviors through a dynamic weighting mechanism based on time decay, alleviating the non-stationary distribution problem of feature streams and enhancing the model′s adaptability to dynamic changes in the network environment. A semi-supervised anomaly detection model, GANomaly, based on a generative adversarial network was constructed. A generator with an attention mechanism, a convolutional decoder, and a discriminator were integrated for adversarial training, and joint optimization was performed by combining feature matching loss, contextual loss, and encoder loss. Meanwhile, an anomaly score function and an adaptive threshold mechanism were designed to realize real-time identification and precise alerting of active intrusion behaviors. [Results] Experiments were carried out based on public datasets and a simulation environment, and the proposed method was evaluated from three aspects:intrusion pattern discrimination capability, traffic loss perception capability, and detection coverage. The results show that the proposed method is significantly lower than comparison methods in terms of the Davies-Bouldin index, indicating that it has better discrimination capability for different types of intrusion behaviors. In the traffic loss detection experiment, the correlation coefficient between the predicted curve and the actual curve of the proposed method reaches 0.94, outperforming the comparison methods. In terms of detection coverage, the coverage indicator C value of the proposed method remains stable at 1, enabling comprehensive coverage of various active intrusion behaviors and demonstrating good detection stability and robustness. [Conclusions]The proposed method improves the quality of training data through the isolation forest algorithm, enhances the model′s adaptability to dynamic feature changes by using damped incremental statistics, and realizes semi-supervised high-precision anomaly detection with the GANomaly model, effectively alleviating the non-stationary distribution and sample imbalance problems caused by feature collinearity. Experimental results show that the proposed method outperforms comparison methods in terms of detection comprehensiveness, stability, and real-time performance. It has good engineering application value and provides a new technical path for building efficient and reliable active network intrusion detection systems. Moreover, it offers a reference for the development of intelligent security defense systems toward adaptivity and high coverage.

[Objective]With the continuous implementation and advancement of China′s medical insurance payment reform, the mortality of patients with low-risk diseases has become a core quantitative indicator for medical quality and safety assessments of tertiary public hospitals, as well as a critical difficulty in refined operation and management of healthcare institutions. Traditional screening of high-risk fatal cases based on manual experience is inefficient and subjective, failing to realize early risk warning for low-risk cases. This study constructed a Super Learner ensemble model to improve the prediction accuracy and interpretability of discharge outcomes for patients with low-risk diseases, so as to provide a novel approach for strengthening mortality management of low-risk diseases in hospitals, and facilitate optimization of individualized clinical diagnosis and treatment, rational allocation of medical resources, and reduction of unnecessary medical resource waste. [Methods]Medical record data of diagnosis related groups (DRG)-categorized low-risk diseases in a tertiary hospital in 2025 were enrolled. According to the catalog of 115 low-risk diseases, 37 973 valid samples were collected, including 33 829 cured cases, 4 103 non-cured cases, and 41 fatal cases. Raw medical record data were preprocessed with categorical variable encoding and continuous variable standardization. The Pearson correlation coefficient method was used to remove redundant features and retain key variables with strong correlations. Six machine learning algorithms were used to construct ensemble model learners, and the Super Learner ensemble model was developed via 5-fold cross-validation and hyperparameter optimization. Model performance was evaluated using area under curve (AUC), accuracy, sensitivity, specificity, and precision, combined with receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). The Shapley additive explanations (SHAP) method was adopted for interpretability analysis. [Results]The Super Learner ensemble model achieves the optimal performance on the validation set, with an AUC of 0.991 3, accuracy of 0.989 6, sensitivity of 0.791 9, specificity of 0.986 2, and precision of 0.978 3, outperforming all individual models in comprehensive performance. Visualization analysis demonstrates its excellent discriminative ability, predictive calibration, and clinical net benefit. SHAP analysis reveals that among the data on the medical record front page, whether the patient is hospitalized for the first time, whether surgery is scheduled, and whether a readmission plan within 31 days after discharge exists are the three most important factors influencing mortality in patients with low-risk diseases, which are consistent with clinical patterns. [Conclusions]The proposed model has high prediction accuracy and favorable robustness. The used method and feature set are versatile and referable, which can provide replicable evidence for early clinical intervention and medical quality control, and have promotion and application value in advancing the development of intelligent and precise diagnosis and treatment management for low-risk diseases.

[Objective]In order to investigate the wind load distribution characteristics on large-span cylindrical coal shed surfaces under moving downbursts, providing scientific basis and design reference for engineering practices, this study was carried out. [Methods]With a large-span cylindrical coal shed structure at a certain thermal power plant as the engineering background, based on the impinging jet model, the analysis focused on the characteristics of horizontal wind speed distribution in the transient wind field of a moving downburst. The study examined how downburst movement speed, wind direction angle, and the distance between the jet center and the center of the large-span cylindrical coal shed structure affected the wind load distribution on the structure′s surface. Based on the analysis results, conducted wind pressure zoning on the cylindrical coal shed surface and provided engineering design recommendations against downbursts. [Results]The moving downburst wind field tilts downward in the movement direction due to the jet center′s movement. It exhibits extreme wind speeds at radial distances r=Djet (D_jet signifies the jet port diameter) and near-ground height z=0.02D_jet. There is a significant difference in wind speeds between the leading and trailing edges of the downburst, and the wind field exhibits dynamic characteristics. When the jet center coincides with the coal shed structure center, the structure surface primarily experiences wind pressure, with the wind pressure coefficient reaching its maximum value of 1. When the jet center moves away from the coal shed structure center, the structure surface is influenced by airflow vortices, generating strong wind suction. As the distance between the two centers increases, the wind pressure coefficient on the structure′s surface decreases. Specifically, when the distance between the jet center and the center of the coal shed structure is 1D_jet, the difference between the positive and negative wind pressure coefficients on the structure′s surface reaches its maximum at 1.5. As the downburst moves faster, the airflow separation at the edge of the windward side of the coal shed structure intensifies, leading to a gradual increase in the wind pressure coefficient. Meanwhile, the degree of airflow separation on the leeward side decreases, resulting in a gradual reduction in the wind pressure coefficient. The difference in wind pressure coefficients at the same location on the structure′s surface reaches a maximum of 0.5. Additionally, the wind pressure distribution on the coal shed structure surface is highly sensitive to the downburst wind direction angle. As the wind direction angle increases from 0° to 90°, the influence range of the airflow vortices on the structure surface expands, and the wind pressure coefficient in the vortex center area decreases. The movement speed of the downburst significantly affects the wind pressure on areas A₁, B₁, and C₁ on the structure surface, the difference in wind pressure coefficients among these areas can reach up to 0.73. As the wind direction angle changes from 0° to 45°, the wind pressure coefficients in all areas except A₂ show minor variations. At a wind direction angle of 90°, the wind pressure coefficients in all areas change, with the largest difference in A₂, reaching 0.54. [Conclusions]The coal shed structure experiences higher wind pressures primarily at the edge regions, with the maximum wind pressure coefficients in zones B₂ and C₂ both being 0.98, and zone A₂ having a maximum wind pressure coefficient value of 0.95. These areas are the key sections for the design of such structures against downbursts, which require special attention and targeted reinforcement measures in engineering practice.

[Objective]This study aims to investigate the flexural behavior of carbon fiber-reinforced polymer (CFRP)-confined concrete-filled steel tubular (CFST) members, reveal their intrinsic mechanical mechanisms, and solve the related problems such as the theoretical calculation of the flexural capacity of circular CFRP-CFST flexural members. [Methods]Pure bending tests were conducted on 16 circular CFRP-CFST flexural specimens. Based on the experimental results, a comparative analysis method was adopted to explore the influence of the number of longitudinal CFRP layers on the neutral layer migration law under different steel tube diameters and the same circumferential CFRP. Additionally, the integral method was used to derive the theoretical calculation formula for CFRP-CFST flexural members. [Results] During the neutral layer migration process, longitudinal CFRP begins to fracture when the load is approximately 0.7P_u (P_u is the ultimate load). When the tensile strain at the mid-span of CFRP reaches about 8 000με, the longitudinal CFRP completely fractures and withdraws from work. When the tensile strain at the mid-span of the steel tube reaches about 9 000με, the steel tube enters the yield stage. Both the final migration position and the migration magnitude of the neutral layer decrease with the increase of the number of longitudinal CFRP layers, while the flexural capacity of the members is improved. When the load is less than 0.7P_u(i.e., before the longitudinal CFRP begins to fracture), the strain development of the steel tube and CFRP is basically consistent, which indicates that the two can work well collaboratively in the longitudinal direction. In the analysis of the neutral layer migration-bending moment relationship, with the increase in the number of longitudinal CFRP layers, the neutral layer migration curve shifts downward as a whole and can be divided into three stages: primary linear migration stage, horizontal stage, and secondary linear migration stage. The neutral layer migration rate shows a variation characteristic of “first accelerating, then slowing down, and then accelerating again”. When the mid-span bending moment is about 0-0.3M_u (M_u is the measured ultimate bending moment), the neutral layer migration rate is fast, the curve slope is large, and the steel tube, concrete, and CFRP are all in the elastic working stage, which is the first stage. When the mid-span bending moment is between 0.6M_u-0.7M_u, the neutral layer migration rate is slow, the curve is approximately horizontal, and the longitudinal CFRP begins to fracture. When the mid-span bending moment is greater than 0.7M_u, the longitudinal CFRP is completely fractured, and the steel tube in the tensile zone yields and enters the plastic strengthening stage. At this time, the tensile stress of the section is less than the compressive stress, and the stress state is unbalanced. In addition, the neutral layer migration is aggravated, and the curve slope increases, showing a secondary upward characteristic. In the analysis of the neutral layer migration-curvature relationship, with the increase in the number of longitudinal CFRP layers, the neutral layer migration curve shifts downward as a whole, and the migration rate shows a characteristic of “first fast and then slow”. The neutral layer migration-curvature curve can be divided into a linear elastic segment and a plastic segment, with a rate turning point appearing around 0.3M_u-0.7M_u. [Conclusions]Based on the neutral layer migration law and combined with the basic principle of the mechanical deformation process of flexural members, the calculation formula for the flexural capacity of circular CFRP-CFST flexural members is derived in this paper, and the calculation results are in good agreement with the experimental results. The revelation of the neutral layer migration law and the proposal of the corresponding calculation method further enrich the theoretical system of CFST structures and provide a theoretical basis for relevant engineering practices.