89 results about "Pruning algorithm" patented technology

The invention relates to the technical field of model pruning algorithms, in particular to an unmanned aerial vehicle target detection model lightweight method based on a pruning algorithm, and the method comprises the steps: initializing a training model; sparse training; pruning the model; finely adjusting the model; according to the method, pruning optimization of parameters and structures is carried out on the STUAV-YOLO model, redundant and unimportant connections and weights in a neural network are removed, so that the operation efficiency of the model is improved, the model is more suitable for being deployed on unmanned aerial vehicle equipment with limited resources, and real-time and accurate high-altitude target detection tasks are achieved.

The invention discloses an operation and maintenance workflow cooperation system and method, and relates to the technical field of business process.The method comprises the steps that after a natural language operation and maintenance requirement is received, a subtask set containing task attributes is extracted through a semantic model built based on a pre-training operation and maintenance field language model; inputting the sub-tasks into a causal mining model, capturing an implicit dependency relationship between the tasks through an attention mechanism which takes task types and resource demands as weight regulation factors, and generating an operation and maintenance relationship graph which contains dependency confidence coefficients and dependency types and does not have cyclic conflicts; splitting the atlas into a task chain set and a free task point set by adopting a causal-oriented greedy pruning algorithm based on a directed edge association subtask maximum aggregation and task chain set scale minimization principle; task chains are distributed through a weighted matching algorithm in combination with the chain overlap ratio and the to-be-handled task amount of the intelligent agent, remaining free task points are distributed according to the balance principle, and accurate disassembly and efficient cooperation of operation and maintenance tasks are achieved.

The invention provides a system-level fault analyzing and tracing method and system based on multi-layer causal diagram extraction, and belongs to the technical field of fault diagnosis. Using a multi-level convolutional neural network to convert the time sequence monitoring data features into a feature matrix; by introducing a hierarchical adjacency pruning algorithm and an elastic network regularization constraint, sparse modeling of a multi-level causal matrix is realized, and a causal matrix graph, namely a prediction contribution matrix graph, is obtained; according to a proposed score quantization algorithm, direct propagation and indirect propagation effects are comprehensively considered, prediction information provided by each variable is quantified, a reason score is provided, and a fault reason variable is determined. According to the method, multi-dimensional feature information of the system-level fault can be compared, all useful information is fully utilized, the contribution degree of the system variable fault is accurately evaluated, and the high-level fault reason detection rate is obtained.

Disclosed in the present invention are a method for constructing a channel fingerprint twin, and a system. In the present invention, a coarse-grained channel fingerprint and a fine-grained channel fingerprint are respectively regarded as a physical object and a digital twin object, and image super-resolution technology is used to construct a relationship between the coarse-grained channel fingerprint and the fine-grained channel fingerprint. In the present invention, on the basis of variational inference and a re-parameterization theory, an evidence lower bound of a fine-grained channel fingerprint twin likelihood is derived to serve as an objective function, and the coarse-grained channel fingerprint is introduced as side information to design a conditional generative diffusion model for generating the fine-grained channel fingerprint, wherein the conditional generative diffusion model can be deployed in a core computing center of a channel fingerprint twin. In addition, in the present invention, a one-shot pruning algorithm and multi-objective knowledge distillation technology are further introduced to acquire a lightweight conditional generative diffusion model. The method for constructing a channel fingerprint twin provided in the present invention not only ensures the reconstruction accuracy, but also has relatively strong scalability and generalization capability in wireless communication scenarios with different fine-grained channel fingerprints.

The invention discloses a hybrid expert model structured pruning and acceleration method and system based on micro expert sorting, belongs to the technical field of large language models, and solves the problems that an existing MoE pruning method cannot consider the requirements of fine-grained pruning, reasoning acceleration and structural analysis generalization at the same time, coarse-grained expert-level pruning damages the performance of a model, and the reliability of the model is poor. And fine-grained compression lacks speed increase and lacks a unified micro-analysis method. The method comprises the following steps: splitting each expert network in a hybrid expert model into a plurality of micro experts, and modeling the plurality of micro experts, so that the micro experts in different expert networks have comparability; sorting all the micro experts according to the energy indexes of the micro experts by adopting a micro expert sorting algorithm; and processing the sorted micro-experts by adopting a pruning algorithm, selecting a core micro-expert for reservation, and directly deleting the rest of the micro-experts. The method is suitable for application scenes such as edge calculation and multi-task learning.

The invention provides a user consumption behavior multi-dimensional portrait analysis method and system based on a neural network, and relates to the technical field of data analysis, and the method comprises the steps: obtaining user historical consumption behavior data, and constructing a basic feature vector; key time sequence features are determined through a sub-sequence dynamic pruning algorithm and entropy value weighted mapping; extracting sequence features by adopting a bidirectional long-short-term memory network; constructing a multi-task adversarial feature extraction network to obtain scene invariant features; performing feature fusion to obtain multi-dimensional combined features; constructing a feature index tree to calculate user similarity; and hierarchical clustering is carried out to obtain a consumption behavior portrait. According to the invention, high-precision user portraits are realized, and scene adaptability and calculation efficiency are improved.

The invention relates to the field of environment-friendly packaging boxes, and discloses a multi-modal data fusion environment-friendly packaging box intelligent design auxiliary system which comprises a packaging full-life-cycle heterogeneous atlas database module, a design constraint parameter analysis module, a topological variation index engine module, a compliance and multi-objective optimization module and a parameterization scheme generation module. According to the method, a time dimension is introduced through a topological variation index engine, instantaneous stress in a folding process is calculated in combination with a nonlinear viscoelastic model, physical evolution of a structure is simulated in a virtual design stage, and the fracture risk is predicted; meanwhile, multi-objective optimization is carried out by utilizing a Hash mask mechanism based on laws and regulations and a pruning algorithm, and a compliance design scheme containing production process parameters is output, so that the problems of lack of physical simulation and low compliance verification efficiency in environmental protection material design are solved, and the physical feasibility and the production yield of the design scheme are improved.

The invention discloses an alkaline electrolytic cell time-varying operation domain calculation method and system based on a dynamic model, and belongs to the field of power system operation regulation and control. The dynamic model-based alkaline electrolytic cell time-varying operation domain calculation method comprises the following steps: establishing an alkaline electrolytic cell operation domain model according to an electrochemical steady-state model, a thermal dynamic model and an oxygen-in-hydrogen dynamic model of an alkaline electrolytic cell; converting the operation domain model of the alkaline electrolytic cell into an optimization model; traversing and solving the optimization model under different initial temperatures and hydrogen concentrations in oxygen to obtain an operation domain; an electrochemical steady state model, a thermodynamic dynamic model and a hydrogen-in-oxygen dynamic model are introduced, so that a comprehensive time-varying operation domain model is established; therefore, the operation domain can reflect the time-varying characteristics of the electrolytic cell more accurately, a reliable boundary is provided for optimization and regulation of the system under different environmental conditions, and the production efficiency and the safety are improved; and a polyhedron pruning algorithm and a geometric constraint decoupling algorithm are adopted, so that the solving efficiency of the operation domain is remarkably improved.

The invention relates to the technical field of model lightweight, and particularly discloses a deep neural network model optimization method based on hierarchical reinforcement learning and multi-agent collaborative distillation, and the method comprises the steps: building a structured pruning searcher based on an ABC algorithm, constructing a pruning combination reduction strategy dynamic artificial bee colony pruning algorithm, and carrying out the optimization of a deep neural network model. Performing fitness evaluation to guide a search process, and outputting an optimal pruning network structure under resource constraint; establishing a staged distillation architecture and a multi-dimensional hierarchical loss function, and realizing smooth and progressive knowledge transmission between the teacher model and the assistant model; a fine-grained quantization scheme based on parameter classification is designed, differential bit widths are configured for weights, batch normalization parameters and activation output respectively, a quantization perception training loss function fusing a hardware delay look-up table model is constructed, hardware perception joint fine tuning of the network weights and quantization parameters is achieved, and the quantization precision of the network weights and the quantization parameters is improved. Therefore, the effect of remarkably improving the model compression efficiency on the premise of keeping the precision is achieved.

The invention provides a real-time incremental RAG method and system based on a dual-tense knowledge graph, and belongs to the technical field of artificial intelligence and knowledge graphs. The method comprises the following steps: through dual-tense decoupling modeling of event effective time and system input time, constructing a versioned knowledge graph and establishing a tense index; efficient retrieval is carried out by adopting a time sequence pruning algorithm based on indexes, and dynamic expansion of a graph pattern is realized through a declarative compiler; realizing abstract-free high-precision result sorting by utilizing a multi-modal concurrent retrieval and attention fusion mechanism; incremental updating of the knowledge graph is realized through streaming data processing and a distributed engine; and closed-loop optimization is formed according to user feedback. The system effectively solves the problems that a traditional RAG scheme is high in data updating delay, tense information is lost, retrieval efficiency is low and expansibility is poor, tense query accuracy, retrieval efficiency and system throughput are remarkably improved, and meanwhile operation and maintenance complexity is reduced.

The invention relates to the technical field of purchase demand prediction, and discloses a part purchase demand prediction method and system based on machine learning, and the method comprises the steps: obtaining modular product basic data, and generating a BOM graph structure; calculating prior probability distribution through a Bayesian inference algorithm; executing an adaptive probability pruning algorithm to solve the problem of combinatorial explosion; quantizing uncertainty by using a Bayesian deep learning network; calculating a differentiated safety inventory coefficient based on the value-at-risk model; executing an importance sampling algorithm to carry out Monte Carlo simulation on high-risk low-frequency configuration, and verifying a demand coverage rate in an extreme scene; an incremental learning mechanism is utilized to update probability distribution and a pruning threshold according to the new order data, and a self-adaptive optimization demand prediction result is output; according to the method, the inventory cost is remarkably reduced, the stockout risk is reduced, and the balance between the calculation efficiency and the prediction accuracy is realized.

The invention discloses a cross-border logistics service quotation system and a real-time dynamic quotation adjustment method thereof, and the method comprises the steps: dynamically obtaining multi-source heterogeneous data, and constructing a door-to-door service chain graph through a knowledge graph technology; business rules are compiled into a calculation unit capable of being subjected to hot updating through a dynamic rule engine, and automatic compliance auditing is carried out; generating a candidate logistics path combination based on a service chain graph and the dynamic rule, and filtering by using a path pruning algorithm; and carrying out parallel real-time calculation on the effective path by utilizing a stream-oriented calculation framework to obtain a plurality of quotation schemes, and carrying out multi-dimensional visual output on the quotation schemes. Through technology fusion, the beneficial effects of remarkably improving the quotation precision, greatly increasing the calculation efficiency, enhancing the system flexibility, reducing the user decision cost and the like are realized.

A method may comprise: converting one or more functions or function call instructions of a first neural network (NN) model into one or more graph modules; analyzing a relationship between one or more inputs and one or more outputs of the one or more graph modules; generating a second NN model including the one or more graph modules as one or more nodes of a directed acyclic graph (DAG) by coupling the one or more inputs and outputs of the graph modules based on the relationship; adding one or more markers corresponding to a weight parameter of one or more layers of the second NN model; and updating the one or more markers according to a pruning algorithm that removes at least a portion of the weight parameter.

The application provides a multiple layout splitting method, system, medium and program product. First, an initial splitting coloring operation is performed based on a saturation priority strategy to generate an initial layout splitting scheme. Then, the global exploration ability of a tabu search algorithm is utilized, and a preset target function is combined to perform multiple rounds of iteration update operations on an initial conflict node set to generate a multiple layout splitting optimization scheme. For ultimate conflict nodes that cannot be completely eliminated by the tabu search algorithm, a backtracking pruning algorithm is started to maximize the attempt to eliminate all ultimate conflicts to generate a final multiple layout splitting target scheme. The application combines the initial optimization ability of the saturation priority strategy, the global exploration ability of the tabu search algorithm, and the quality guarantee of the backtracking pruning algorithm to realize accurate splitting of multiple layouts, guarantee splitting quality and lithography yield, improve splitting efficiency, reduce computational complexity, and adapt to the needs of multiple layout splitting.

The present disclosure discloses a redundant backup deployment method for microservice instances in edge computing environments. Redundant backup deployment of the microservice instances and the selection of a primary instance are acquired using Transformer-based deep reinforcement learning (T-DRL). For a service failing to satisfy the service level agreement (SLA), the actual service reliability is improved by downgrading the switching priority of the microservice instances; for a service satisfying the SLA, the deployment cost of the microservice instances is reduced using a pruning algorithm. The present disclosure effectively reduces the deployment cost of microservice instances and minimizes the number of active edge nodes while ensuring the SLA, thereby effectively reducing the resource consumption in edge computing environments.

The invention discloses a subway map construction and assembly method based on dynamic configuration, and relates to the technical field of computer software. Comprising the following steps: configuring a metro map, and converting the metro map into a main process priority queue and a sub-process priority queue through a metro map construction method; assembling data into a main process priority queue and a sub-process priority queue according to a job historical record through a variant sliding window algorithm and a data assembling method; and pruning the main process priority queue and the sub-process priority queue through a dynamic jump type post pruning algorithm, and converting into a metro map of the target operation process. According to the method, metro map construction and assembly methods are packaged, metro map configuration is opened for users to use, the problem that a large number of codes need to be written for each operation process in diversified operation scenes is solved, the method supports dynamic adaptation to display requirements of different operation processes, meanwhile, the processing speed and the memory utilization efficiency are improved, and the user experience is improved. And high flexibility of configuration driving is realized.

The application discloses an operation and maintenance workflow cooperation system and method, relates to the technical field of business processes, and comprises the following steps: after receiving a natural language operation and maintenance requirement, a semantic model based on a pre-trained operation and maintenance field language model is used to extract a subtask set containing task attributes; the subtasks are input into a causal mining model, the hidden dependency relationship between the tasks is captured through an attention mechanism taking the task type and resource requirement as a weight adjustment factor, and an operation and maintenance relationship graph containing dependency confidence, dependency type and no circular conflict is generated; a causal orientation greedy pruning algorithm is used to split the graph into a task chain set and a set of free task points based on the principles of maximum aggregation of the subtasks associated with the directed edges and minimization of the size of the task chain set; the task chains are distributed by a weighted matching algorithm combined with the chain coincidence degree and the to-do task amount of the intelligent agent, the remaining free task points are distributed according to the balance principle, and accurate disassembly and efficient cooperation of the operation and maintenance tasks are realized.

The application relates to a model pruning method of an interpretable CNN classification model, belongs to the field of image compression, and solves the problems of high operation complexity, large time and memory consumption and difficulty in deployment on terminal equipment of an existing deep CNN model, and solves the problem of lack of interpretability of an existing model pruning algorithm. The method comprises the following steps: inputting a training picture into a neural network model to be pruned, and extracting a feature map matrix of each convolution layer; upsampling the feature map matrix to the size of the input picture, and then performing a normalization operation to construct a saliency map; multiplying the saliency map and the input picture element by element to construct a weighted input picture; subtracting the input picture from the weighted picture element by element to construct an attention region occlusion map; inputting the attention occlusion map into the model to be pruned, observing the change of model accuracy as an importance score of the channel, and pruning the channel to obtain a pruned lightweight model. The application realizes high pruning rate of the model and improves the interpretability of the pruning process.

The application provides a mechanical equipment residual service life prediction method and device, and relates to the technical field of equipment management.The application adopts an adaptive pruning algorithm for light processing, and continuously optimizes model performance through error calculation and model evaluation, so that the prediction speed and prediction accuracy are improved; the pruning process automatically prunes redundant elements through an adaptive structured pruning strategy, so that unnecessary operations and searches are avoided; and when different neural network layers are used, different pruning rates are used for automatic structure pruning, so that an optimal light network model is finally obtained; the residual service life prediction accuracy is significantly improved, and less storage space is occupied, so that the application can be conveniently deployed on platforms such as small embedded systems, and timely maintenance and maintenance of maintenance personnel are reminded, so that the application is suitable for preventive maintenance and management of various mechanical equipment, and has good practicability and popularization value.

The invention provides a target detection method and device based on automatic driving, a new energy automobile and electronic equipment, and the method comprises the steps: obtaining a road image; preprocessing the road image to obtain a target detection data set; training the target detection data set according to a pre-constructed first target detection model to obtain output features; optimizing the first target detection model according to a model pruning algorithm to obtain a second target detection model; obtaining distillation loss according to the first target detection model and the second target detection model; and performing knowledge distillation on the output characteristics according to the distillation loss to obtain a detection result. According to the invention, the method can improve the detection capability of a detection model for a target, improves the detection precision and efficiency, reduces the redundancy degree of the model, reduces the calculation amount, and reduces the consumption of calculation power.

The invention discloses a vehicle obstacle detection method, device and equipment, a medium and a program product. The method comprises the steps of executing time synchronization operation and space calibration operation of a camera and a radar through a hardware-level timestamp synchronization module and a dynamic calibration compensation algorithm, and obtaining visual data and radar data after execution is completed; through a backbone network of the obstacle detection model, feature extraction is carried out on the visual data and the radar data by utilizing mixing precision quantification and combining an attention module, and a feature extraction result is determined; through a neck network of an obstacle detection model, feature fusion is carried out by combining environmental sensor data on the basis of the feature extraction result, and a feature fusion result is determined; and performing obstacle detection on the basis of the feature fusion result through the head network pruned by the obstacle detection model through a dynamic pruning algorithm, and determining an obstacle detection result. The defects of multi-sensor space-time alignment, extreme weather robustness and edge computing resource limitation in the prior art are overcome.

This invention discloses a method for solving power system time-series optimization problems, comprising: constructing a dynamic mathematical model of the power system time-series problem and determining the objective function and constraints; generating an initial solution based on the dynamic mathematical model, performing pruning preprocessing on the initial solution using a fast pruning algorithm to select high-quality initial feasible solutions; generating neighborhood solutions based on the high-quality initial feasible solutions using a simulated annealing algorithm, determining whether to accept the neighborhood solutions according to the Metropolis criterion, and marking the completion of one iteration after the determination; performing pruning feedback and search direction optimization after each iteration until a preset iteration convergence condition is reached, and outputting the optimal solution for power system time-series optimization. This invention can effectively improve the efficiency, accuracy, and global optimality of solving power system time-series optimization problems, and is suitable for the complex requirements of high-voltage and high-efficiency power systems.

The application discloses a kind of offshore wind farm group's decentralized coordination control method and device, its method includes: the tail flow wind turbine power and thrust load balance optimization and control model of considering tail flow influence of single wind turbine group are constructed;Original tail flow directed graph is constructed, and original tail flow directed graph is decomposed into completely uncoupled sparse sub-tail flow directed graph using graph weight pruning algorithm and graph depth-first search algorithm, sparse wind farm group-field-machine multi-layer decentralized control system is constructed, and decentralized wind farm power and thrust load balance optimization model is constructed, and the optimal value of power and thrust load balance control parameter is solved.The method and device of the application establish power and thrust balance optimization model, realize a kind of communication burden low, less and scalable wind turbine group decentralized coordination control method and device.

The application discloses a kind of wisdom legal inquiry method based on multi-wheel pruning Skyline algorithm, comprising: local equipment of legal department obtains local legal case according to the query command issued by central server;The local legal case obtained is uploaded to the central server in encrypted form, the central server carries out integration analysis, and the legal case with higher comprehensive document similarity is sent to local equipment.The scheduling strategy module of the central server obtains the physical node of local legal department or the corresponding legal department priority information according to the attributes such as name, gender, native place, address, case name, case-cracking event, capture event and report event in legal case.The application can make legal information retrieval personnel free from reading a large number of cases, save a lot of time and manpower and material resources, especially provides technical guarantee for finding similar cases, case-cracking clues and inducing crime trend.

The invention belongs to the technical field of communication, and discloses a pruning and fine tuning troposphere waveguide prediction method and system based on multi-granularity evaluation, and the method comprises the steps: calculating edge loss and task loss; fusing edge loss and task loss to construct a channel importance evaluation system, and performing quantitative evaluation and dynamic sorting on the contribution degree of the atmospheric waveguide prediction channel; on the basis of the sorting of atmospheric waveguide prediction channels, redundant channels are gradually eliminated by adopting an iterative pruning algorithm until a preset pruning rate is reached; and calculating the prediction confidence of the atmospheric waveguide prediction model before and after pruning, positioning a high-sensitivity atmospheric waveguide sample of which the prediction result is remarkably reduced, establishing a dynamic sample weighting mechanism, and performing fine adjustment compensation on the residual channel weight by using error back propagation. According to the method, a multi-granularity importance evaluation cutting mechanism and a prediction information guiding method are adopted, lightweight compression and fine adjustment are performed on the model, the effectiveness of the prediction model is improved, and accurate prediction and interaction of the non-uniform atmospheric waveguide are realized.

The invention discloses an improved YOLO11-based method for detecting breakage of a hair braid of an oil pumping unit in a low-light scene, and the method comprises the following steps: collecting an image data set of the hair braid on the oil pumping unit, and carrying out the preprocessing of the image data set, thereby obtaining a preprocessed image data set of the hair braid of the oil pumping unit; based on a YOLO11 model, a double-trunk module is adopted, and a CBAM attention mechanism and a loss function are fused to improve the model; carrying out training on the improved YOLO11 model; trimming the trained YOLO11 model by using an LAMP amplitude pruning algorithm, taking the trimmed YOLO11 model as a detection model after verification, and deploying the trimmed YOLO11 model on site; acquiring a camera real-time video stream of the hair braid of the oil pumping unit in operation, and detecting the camera real-time video stream in real time by adopting the detection model; real-time judgment of breakage of hair braid of oil pumping unit If so, a fault message is uploaded, and an alarm is given at a webpage end. According to the detection method, the accuracy of the breakage detection of the hair braid of the oil pumping unit under the low-light condition can be remarkably improved.

The invention belongs to the technical field of PCB circuit design, and discloses a netlist partitioning method based on neighborhood bias edge pruning, which comprises the following steps: constructing a PCB circuit netlist diagram data set for netlist partitioning based on a PCB circuit development board; constructing a netlist partition framework by using a graph neural network and an edge pruning algorithm based on netlist graph node neighborhood relation bias; and performing netlist partitioning based on the trained netlist partitioning framework and the PCB circuit netlist diagram data set. The edge pruning algorithm provided by the invention senses the nodes with high bias fields in the netlist graph data and performs edge pruning on the nodes, so that the influence of neighborhood bias on node classification is reduced, the node classification accuracy is remarkably improved, and high-confidence division of the PCB circuit netlist is realized. According to the method, a netlist division problem is abstracted into a node classification problem, training is carried out on a netlist graph data set, the problem of parameter sensitivity existing in a traditional division method is solved, and generalization of PCB netlist division is improved.

The invention provides a novel wafer layout and cutting calculation method, relates to the technical field of semiconductor chip manufacturing, and aims to optimize wafer layout by distinguishing a Shot interval and a Die interval and combining a graph periodic layout tree structure, remarkably improve the number of chips (DPW), effectively improve the wafer utilization rate, break through the limitation of a traditional method and improve the wafer cutting efficiency. Independent spacing constraints are input in Sp1, Sp2 constructs a periodic tree structure, optimal layout is searched in Sp3 and Sp4 through iteration and traversal, the DPW is improved from 624 to 637, the amplification is about 2.1%, benefited from innovative application of the self-adaptive step length and pruning algorithm, the marginal space is accurately captured, waste is eliminated through secondary fine tuning, geometric transformation of a non-rectangular Die shape is supported, and the method is suitable for large-scale popularization and application. The wafer area potential is further released, the breakthrough spacing optimization capability provides a higher-efficiency solution for semiconductor manufacturing, and the method has remarkable creative significance.

The invention discloses an urban power distribution network data driving distribution robust optimization scheduling method and device considering uncertainty. The method comprises the following steps: according to a traditional fuzzy set B1 based on a Wasserstein distance, constructing a fuzzy set B2 based on a Copula function and considering uncertainty correlation; according to the fuzzy set B2, constructing a micro-grid dispatching model of the urban power distribution network considering the uncertain relevance; deducing a worst case conversion method in the fuzzy set B2 through a duality theory, McCormick relaxation and conditional value-at-risk approximation of the micro-grid scheduling model, and obtaining a micro-grid linear scheduling model; trimming the collected original data set by adopting a sample trimming algorithm to obtain a trimmed data set; and based on the trimming data set, a solver is adopted to solve the micro-grid linear scheduling model, and an optimal scheduling strategy of the micro-grid is obtained.

This invention provides a continuous subgraph matching method for social networks based on dynamic pruning, belonging to the field of social network behavior detection. The invention proposes a dynamic pruning algorithm that establishes local equivalence theory by calculating the backward neighbors of query points and candidate points, defines subtree-level equivalence criteria in dynamic graph environments, and is applicable to highly isomorphic behavior patterns such as zombie account groups and terminal nodes of propagation chains. The dynamic pruning algorithm skips isomorphic subtree structures to achieve the pruning effect. A query graph decomposition and sorting mechanism is proposed to remove the binding constraint of the matching order on the update edges and candidate sets, actively identifying and prioritizing highly redundant candidate set vertices, further reducing the search space. Ultimately, while ensuring result completeness, it achieves a 16.9-fold performance improvement over RapidFlow and a 100% completion rate for complex queries, providing sub-second decision support for abnormal behavior detection in social networks and promoting the application of dynamic graph analysis in resource-constrained environments such as edge computing.