125 results about "Complete graph" patented technology

The invention relates to the technical field of three-dimensional modeling in engineering surveying, in particular to a point cloud building component modeling method and system based on feature extraction. Three-dimensional point cloud data of building components are obtained, a multi-scale local neighborhood calculation covariance matrix is constructed with each point as the center, eigenvalues are decomposed, an edge probability graph is generated, corresponding component categories are semantically output, and corresponding key feature point sets are screened; taking the feature point cloud as a control point, constructing a three-level B-spline surface model, and adjusting the corresponding spatial distribution density; a weighted complete graph is formed through a key feature point set, a non-planar area is identified and processed through a Kurtowski theorem, a QEM algorithm is applied to carry out lightweight processing on a three-dimensional model, screening and optimization are carried out through automatically extracting key feature points of building components, and a concise and accurate lightweight model is constructed based on feature point topological optimization and a graph theory algorithm. The automation degree and efficiency of modeling are improved, and the contradiction between model lightweight and precision is effectively solved.

The invention relates to the fields of industrial design, constructional engineering, energy and the like, discloses a topological graph structure information processing system based on image recognition, and overcomes the problems of insufficient topological relation restoration, low recognition accuracy, lack of optimization and correction mechanisms, difficulty in editing operation, poor data reusability and the like in the prior art. An efficient, accurate and easy-to-edit solution is provided, powerful support is provided for digital processing, informatization management and data accumulation of drawings, and the system comprises an image recognition result acquisition module, a preliminary topological information analysis module, a line segment detection and duplicate removal module, a graph structure information construction module, an element fitting module and a bending fitting module. The point merging module is used for merging two identified non-coaxial points into one point; and the final graph structure generation module is used for integrating the connection relation between the points formed by the element fitting module and the bending fitting module into a complete graph structure.

The invention provides a medical image information alignment method and device based on graph network guidance and a medium, and belongs to the technical field of medical image processing. Converting the single-mode feature graph into a complete graph, and utilizing information transmission and aggregation characteristics of a graph neural network; updating edge weights and removing low-quality edges and nodes to obtain a plurality of sub-graphs containing important objects; constructing a cross-modal diagram; a cross attention mechanism is configured based on the patch level, patch level information matching is achieved, and differentiation between different medical images is found; defining an internal learning kernel function and an external learning kernel function; and calculating the distance between the two sample pairs in the high-dimensional space, and carrying out clustering analysis by using the distance. According to the method, attention of the model to important objects and a multi-mode matching mode are more in line with a human thinking mode, and meanwhile, a cross attention mechanism is introduced to achieve alignment of finer-grained information.

The invention belongs to the related technical field of path planning, and discloses a line patrol planning method and system considering route information age, and the method comprises the steps: judging whether a route network graph G has an Euler loop or not, finding out all nodes with odd degrees in the graph G if the Euler loop does not exist, and constructing a complete graph Gf; solving a minimum weight perfect matching problem for the graph Gf to obtain a set OM of optimally matched edges, taking the graph G as an initialized expansion graph G ', traversing each edge (v, u) of the set OM, searching a shortest path Rs from a node v to a node u in the graph G, adding each edge in the path Rs into the current graph G', and obtaining a final graph G 'after traversing is completed; and searching an Euler loop as a line patrol path R from the graph G '. According to the method, the Euler loop is constructed as the line patrol path, so that the average age information of the to-be-patrolled network diagram can be optimized to ensure that each part in the line is patrolled in time, and the patrol reliability is ensured.

The invention relates to the technical field of artificial influence weather, and discloses an artificial influence weather knowledge graph construction method and system based on a large model. The method comprises the steps that multi-source meteorological data are collected and preprocessed to generate a standardized data set; extracting a meteorological feature set by using the pre-trained large model; detecting a data dynamic change frequency, and starting a dynamic strategy to switch and mark an abnormal data segment when the data dynamic change frequency exceeds a threshold value; in combination with a similarity matching result of the historical knowledge graph and the current features, decomposing the feature set of the non-abnormal data segments to generate knowledge feature subsets; correcting the knowledge feature subset based on the coupling relationship between the meteorological kinetic parameters and the environmental parameters; and reversely backtracking the key influence factors along the weather influence propagation map, and constructing the knowledge map. According to the method, multi-source data are integrated through standardized processing, abnormal data are dynamically identified, historical knowledge and physical mechanism are combined to correct features, key factors are finally backtracked to complete graph construction, and the method is suitable for knowledge integration and analysis in the field of artificial influence weather.

The invention provides a logistics pickup and delivery path optimization method and system, and the method comprises the steps: constructing a complete graph according to the obtained customer order data and an arc set between each pair of customer nodes, modeling a vehicle path problem VRPSPDT corresponding to the complete graph into a multi-target optimization problem, decomposing the multi-target optimization problem into a plurality of sub-problems, and carrying out the modeling of the sub-problems, and the plurality of sub-problems are iteratively solved for many times until a logistics goods taking and delivering path corresponding to the customer order data is obtained. In the nth iteration, operation of an evolutionary operator is carried out on current iteration information, a filial generation solution is obtained, the evolutionary operator comprises sequential crossover and random variation, the filial generation solution is input into a pre-trained deep reinforcement learning DRL model, the filial generation solution is optimized based on the DRL model, an improved solution is obtained, and the current population is updated based on the improved solution, and updating the current reference point according to the sub-target value of the filial generation solution. The efficiency, the accuracy and the reliability of determining the logistics goods taking and delivering path are improved.

The invention discloses a milling path planning method and system for stress gradient steepest descent, and the method comprises the steps: carrying out the discretization of a residual stress field based on a lightweight bearing member, and mapping the discretized residual stress field to a milling model, so as to form a high-dimensional matrix; calculating a gradient vector and stress streamline distribution of the residual stress field, and according to the gradient vector and the stress streamline distribution, based on a minimum potential energy principle and a stress steepest descent strategy, performing milling path planning degradation to obtain an optimized baseline; and according to the high-dimensional matrix and the optimized baseline, milling path planning is converted into graph complete traversal with the graph theory as guidance, a steepest descent scheme of strain energy and rigidity is optimized, and an optimal milling path scheme is obtained. The method serves for precision machine manufacturing process design, and the problems of array geometric feature milling path planning and machining deformation control of current lightweight force bearing components are solved; and the method is particularly suitable for light-weight bearing components with array geometric characteristics and high dimensional precision.

The invention discloses a decentralized trust relationship collaborative prediction method and a related device, and the method comprises the steps: starting a block chain system, publishing a public safety parameter and an initialization model on a block chain, and loading a consensus algorithm at a consensus node; constructing a local complete graph by using local data, pre-training the local graph to obtain a user feature extraction and relationship prediction model, and obtaining features of user nodes; establishing a public graph based on the public knowledge, sending the public graph to a block chain consensus node, and performing consensus confirmation on the public graph; and outputting a final trust relationship prediction model. The invention discloses a decentralized trust relationship collaborative prediction method based on federated learning and a graph neural network, and aims to solve the limitation of a traditional trust evaluation method in a multi-source distributed scene and improve the accuracy of trust relationship prediction.

The current application is directed to optimal, locally optimal, and / or near-optimal path determination that defers edge-weight computations, avoids redundant edge-weight computations, and that does not require initial complete graph construction. Many real-world problems are not efficiently solved by current graph-traversal-determination methods because of the overheads and computational complexities involved in edge-weight computations. Moreover, it may be difficult or impossible to construct complete graphs to which current graph-traversal-determination methods can be applied. The currently disclosed optimal, locally optimal, and / or near-optimal path-determination methods and systems construct incomplete graphs during optimal, locally optimal, and / or near-optimal path determination, deferring node generation and edge-weight computations until necessary to expand incomplete graphs during the search for optimal, locally optimal, and / or near-optimal graph-traversal paths.

The invention discloses a multi-unmanned aerial vehicle cooperative regional patrol path planning method based on hierarchical optimization, and the method comprises the steps: firstly, abstracting a task space containing a base station and a plurality of target points into a weighted undirected complete graph, and building a mathematical model with a target of minimizing the total completion time of a system and the standard deviation of task time; then solving is carried out through a hierarchical strategy; firstly, k-means clustering is adopted to carry out initial task allocation; solving a traveling salesman problem for each task cluster by using a simulated annealing algorithm, generating an initial path, and performing local optimization by using a 2-OPT algorithm; and finally, carrying out global collaborative optimization by adopting adaptive large-scale neighborhood search, dynamically adjusting task allocation and paths through a cross-unmanned aerial vehicle damage and repair operator, and carrying out iterative optimization according to a composite acceptance criterion and an adaptive mechanism to generate a near-optimal scheme. According to the method, through tight coupling of task allocation and path planning, the defect that a traditional method is prone to local optimization is effectively overcome, and joint optimization of completion time and load balancing is achieved.

This invention relates to a user interface layout generation method based on a latent diffusion model, belonging to the field of image generation. This invention extracts components and constraints from interface requirement description text using the BERT model, utilizes a graph convolutional network module to predict complete constraint relationships and borders to generate a complete graph, and generates the user interface layout based on the latent diffusion model. On one hand, introducing a latent diffusion model into user interface generation can meticulously capture the complex relationships and details between UI elements, thereby generating a more accurate user interface that conforms to interaction specifications. On the other hand, introducing a graph convolutional network into user interface generation can effectively predict complete constraint relationships between components, effectively capturing and modeling the complex dependencies and interactions between user interface elements; through graph structures, multi-dimensional constraints such as the position, alignment, and function of elements are applied, generating a more coordinated layout that conforms to design specifications; improving the aesthetics and usability of the generated interface, and enhancing the user experience.

The invention relates to a financial data anomaly detection method based on multi-channel graph contrast learning, and belongs to the field of financial data anomaly detection. According to the method, a financial information relation graph with abnormity is input, and abnormal information and behaviors of each user in the graph are evaluated to detect abnormal users. Firstly, the multi-view-angle judgment module adopts a view-level comparison method based on a complete graph, so that information in an attribute network can be comprehensively displayed; then, the graph reconstruction module fully integrates attributes and topological information; finally, wide experiments are performed on six common graph data sets. The method also adopts an AUC index to measure the anomaly detection performance. According to the method, the comprehensiveness of data reconstruction can be effectively considered, the unreliability of neighborhood information is reduced, the complexity is not increased, and compared with other methods, the method has better performance on financial data anomaly detection tasks and has certain practical value.

The invention discloses a graph state segmentation and distribution method and system based on a quantum communication network, comprehensively considers finiteness of node quantum resources and quality of distributed graph states, provides a feasible solution for effectively distributing any graph state in a large-scale quantum communication network, and provides a graph state segmentation scheme based on segmentation-integration. According to the method, a large-scale graph state is divided into a plurality of sub-graph states to be prepared and distributed respectively, graph state related operation is executed on part of nodes, so that the nodes are restored to the required complete graph state, quantum resources required by a single node can be effectively reduced when the large-scale graph state is generated, the smoothness of a whole quantum communication network is kept, and the generation efficiency of the large-scale graph state is improved. Congestion caused by a certain node is avoided; meanwhile, a reliable transmission strategy and conflict-free multi-destination routing are adopted, so that the success rate of graph state distribution is improved.

The invention discloses a redundant space robot inverse kinematics solving method based on a graph neural network, and relates to the technical field of robot motion control and artificial intelligence crossing. A distance geometric graph is constructed to represent the configuration of the mechanical arm, an inverse kinematics problem is converted into a completion problem of a partial graph, joint configuration is sampled, and data pairs of a complete graph and the partial graph are stored as a data set; training based on an isotropic graph neural network and a conditional variation auto-encoder, and training a model based on a loss function of an evidence lower bound; and in the reasoning process, constructing a partial graph, inputting the partial graph into the trained prior network, outputting hidden variables meeting Gaussian mixture distribution, extracting sampling points, and generating a reconstructed complete graph to obtain joint angle information. The mechanical arm structure is represented through the distance geometric diagram, the inverse kinematics problem is converted into the complementation problem of partial diagrams, probability distribution of a solution space is learned through a conditional variation auto-encoder frame, multi-solution generation of different mechanical arm configurations is supported, and the solving precision and efficiency are high.

The invention provides a CTSP solving method and system based on a convolution enhanced agency attention mechanism, and relates to the technical field of data processing, and the method comprises the steps: obtaining city data and salesperson data; based on the city data and the salesman data, constructing a complete graph used for describing CTSP problem instances; based on the complete graph, constructing a constrained Markov decision process model, and modeling a path generation process as a step-by-step decision mechanism; a neural strategy network based on convolution enhancement and an agency attention mechanism is constructed, and the neural strategy network is used for solving a constrained Markov decision process model; training the neural strategy network based on the convolution enhancement and agency attention mechanism by using a reinforcement learning algorithm; generating a preliminary optimal path scheme through the trained neural strategy network based on convolution enhancement and an agency attention mechanism; and inputting the optimal path scheme into a traditional optimizer for fine optimization to obtain a final optimal path scheme.

The invention discloses a big language model Cypher generation method and system based on graph pattern alignment enhancement, and the method comprises the steps: constructing a data set which comprises a plurality of types of training data; supervising and finely tuning the large language model; sampling the supervised and fine-tuned large language model to obtain a sampling data set; training the large language model after supervision and fine tuning; obtaining a question and a graph database mode input by a user; related sub-graph modes are selected from the complete graph modes according to user questions and are verified; the large language model generates a corresponding Cypher query using a selected mode and a given question. According to the big language model Cypher generation method and system based on graph pattern alignment enhancement, after the big language model is enabled to better adapt to a Cypher generation task through supervision and fine tuning of various data, the model is trained by model sampling after supervision and fine tuning, the capability of the big language model to generate the Cypher is enhanced, and the big language model and the graph pattern are effectively aligned.

The present invention provides an entity data association method, system, device and medium based on a knowledge graph, and relates to the field of computer application technology. The entity data association method includes: obtaining data to be entered into the graph, the data to be entered into the graph includes multiple initial fields; obtaining the target field that needs to be fuzzy matched in the multiple initial fields; creating an ES index containing the target field according to the data to be entered into the graph, and importing the data to be entered into the ES index after initialization; traversing each data in the data to be entered into the graph, fuzzily matching the target field in each data with the ES index to obtain fuzzy associated data; generating side information with an associated relationship according to the fuzzy associated data, and storing the fuzzy associated data and side information in a graph database. The present invention can identify and associate the fuzzy relationship between entity data, and establish a complete graph data association relationship.

The present invention discloses a path planning method based on the large-scale sparse Chinese Postman Problem, including: obtaining the positions of the endpoints and intersection points of the roads in the path to be optimized, and establishing an undirected road network graph with the road endpoints and intersection points as nodes; screening out the set V of odd-degree nodes in the undirected road network graph odd , constructing a complete graph formed by the shortest paths between all nodes in V odd ; clustering the odd-degree nodes in V odd to obtain multiple sub-node groups; merging the sub-node groups so that the coverage rate of the adjacent node sets of each merged sub-node group itself is above the adjacent node set coverage rate threshold; performing an even-degree processing on the merged sub-node groups; solving the shortest matching edge set within each even-degree processed sub-node group, and adding the corresponding edges in the shortest matching edge set to the undirected road network graph to form an Euler graph with all node degrees being even; determining the path starting point, and obtaining an Euler circuit starting from this starting point in the Euler graph as the optimized path.

The invention discloses a semantic space topology driven multistage SLAM closed-loop detection method and device, a medium and a product, and relates to the field of computer vision. The method comprises the following steps: preprocessing image data corresponding to a key frame by using a region growing algorithm and a Laplacian operator; constructing a semantic feature vector based on a preprocessing result, calculating the similarity and a similarity threshold value of the current key frame and the non-connected key frame, and screening a first-level closed-loop candidate key frame; screening second-level closed-loop candidate key frames from the first-level closed-loop candidate key frames by utilizing a continuity condition; constructing a BFS tree of an undirected complete graph of the current key frame by using instance abstract space information, calculating the similarity with the BFS tree of the undirected complete graph of the secondary closed-loop candidate key frame, and taking the candidate frame with the highest similarity as a closed-loop key frame; and realizing closed-loop detection according to the relationship between the closed-loop key frame similarity and a threshold value. The problem that closed-loop detection cannot be accurately realized due to the influence of illumination and view angle change in a scene is solved.

Disclosed in the present invention are a dynamic maximal clique enumeration device and method based on an FPGA with an HBM, the method including: the HBM stores a dynamic edge flow, a complete graph adjacency matrix, and candidate cliques; a matrix computing unit updates the complete graph adjacency matrix based on the dynamic edge flow, transmits the updated complete graph adjacency matrix to the HBM for storage, and determines header nodes, of which the corresponding candidate clique needs to be updated; a sequence computing unit constructs, according to the updated complete graph adjacency matrix and each header node to be updated, the sorted data set for reconstructing candidate cliques by data block sequencing; and an update computing unit executes, in parallel, an update task of the candidate clique corresponding to each header node to be updated based on the sorted data set, transmits the updated candidate cliques to the HBM for storage, and transmits the updated candidate cliques to the PC host to extract maximal cliques by means of a filtering operation. The present invention supports the computation of pipelined-type incremental maximal clique, thereby improving the overall computing efficiency of the task.

The invention relates to the field of quantum machine learning and high-energy physics, in particular to a particle flow classification method based on a quantum complete graph self-attention network, and the method comprises the steps: obtaining particle features in a data set, screening out four representative particles with the maximum transverse momentum from each particle flow to replace the characteristics of the whole particle flow so as to form a complete graph; performing self-attention coefficient calculation and weighted summation on particle features in the complete graph by using a QGAT model constructed by a self-attention mechanism to complete updating of the particle features; and the updated quantum state of the particle features is subjected to dimension reduction representation through the quantum convolutional network and then is used as the input of a quantum classifier for result prediction, so that the particle classification precision and the model expression ability are greatly improved.

Disclosed in the present invention are a dynamic maximal clique enumeration device and method based on an FPGA with an HBM, the method including: the HBM stores a dynamic edge flow, a complete graph adjacency matrix, and candidate cliques; a matrix computing unit updates the complete graph adjacency matrix based on the dynamic edge flow, transmits the updated complete graph adjacency matrix to the HBM for storage, and determines header nodes, of which the corresponding candidate clique needs to be updated; a sequence computing unit constructs, according to the updated complete graph adjacency matrix and each header node to be updated, the sorted data set for reconstructing candidate cliques by data block sequencing; and an update computing unit executes, in parallel, an update task of the candidate clique corresponding to each header node to be updated based on the sorted data set, transmits the updated candidate cliques to the HBM for storage, and transmits the updated candidate cliques to the PC host to extract maximal cliques by means of a filtering operation. The present invention supports the computation of pipelined-type incremental maximal clique, thereby improving the overall computing efficiency of the task.

The embodiment of the invention provides a wireless charging method for uncertain sensor node positions, and is applied to the technical field of wireless rechargeable sensor networks. The method comprises the following steps: actively detecting sensor node and obstacle distribution based on a transmitting coil array of the mobile charging equipment; a magnetic wave beam vector is formed by independently adjusting the voltage of each coil, and the accurate position of a node and the contour of an obstacle are determined by combining impedance measurement and wave beam intersection positioning; discretizing the region, constructing a vertex visible graph and generating a collision-free path complete graph; and calling a TSP solver to generate an initial charging path by taking minimization of movement consumption as a target, inputting the initial charging path to the trained reinforcement learning model, dynamically optimizing a track and a charging plan through real-time interaction, outputting an optimal path, and issuing the optimal path to the charging equipment. In this way, the problem of selecting a proper charging point in a complex scene with obstacles and uncertain node positions can be solved, accurate positioning of the nodes is achieved, and the charging efficiency is remarkably improved.

The application discloses an airport crowd route guidance scheme optimization method and device and a storage medium, and the method comprises the following steps: acquiring layout information of an airport, and abstracting an airport scene into a complete graph according to the layout information; screening edges in the complete graph according to a preset screening condition, so as to delete edges that do not meet the condition; abstracting an airport crowd route guidance scheme optimization problem into a combination optimization problem according to the screened complete graph; and using a discrete differential evolution algorithm to optimize the combination optimization problem, so as to obtain a final optimization result. The application performs mathematical modeling on an airport departure layer scene, and automatically generates and optimizes an airport crowd route guidance scheme by using a discrete differential evolution algorithm, so that a high-quality solution can be obtained, and the guidance of the crowd route in the airport can be effectively assisted. The application can be widely applied to the fields of intelligent transportation and evolutionary calculation.

The present invention provides a distributed graph convolutional neural network method for feature segmentation. First, the input features are divided into multiple parts and distributed to different computing nodes. The complete graph structure and a small part of node features are loaded on each computing node, thereby reducing memory requirements; then, the GCN slicing model on each computing node performs forward propagation on its feature fragments, and the communication between computing nodes is only performed in the input and output stages; finally, the output representations of all computing nodes are transmitted to the master node for splicing, and the output representations of each block of GCN are uniformly adjusted through slice coding, ultimately improving the training efficiency of graph neural networks on large-scale graph data.

The invention relates to the technical field of neural network acceleration, and relates to a CPU-FPGA-oriented graph neural network training acceleration method and system. The method comprises the steps that a CPU stores complete graph feature data and distributes different feature data to storage modules of different FPGAs through data channels according to a load balancing algorithm; partitioning the GNN network model by the CPU, and matching a partitioned model operator with a corresponding FPGA (Field Programmable Gate Array); the CPU completes issuing and sampling of a partition model according to a sampler algorithm, and sends the partition model to the FPGA; reading feature data after the FPGA is started, completing a calculation task by a load execution module, and writing a result back to a CPU memory; the CPU completes data updating and weight updating; and resending the updated weight data to the corresponding FPGA, and carrying out iterative training. According to the invention, operation of the graph neural network on a CPU-FPGA platform is successfully realized, and the overall acceleration effect is excellent.

The invention relates to a distribution sequence optimization method and device for intelligent sending trolleys in a three-dimensional space, and the method comprises the steps: S1, obtaining the three-dimensional map information of a target distribution region, and generating an expression graph of the target region; s2, obtaining delivery order data, and based on a destination of the delivery order data, in combination with three-dimensional map information of a target delivery area, amplifying virtual nodes; s3, based on the generated virtual nodes, generating a distance completion graph of the target area in combination with the expression graph of the target area; s4, based on the distance complete graph, establishing a mathematical optimization model of a sending service sequence corresponding to the park; s5, performing linearization and simplification processing on the mathematical optimization model; and S6, solving the mathematical optimization model by introducing an improved ant colony algorithm of space-time clustering characteristics to obtain a distribution sequence. Compared with the prior art, the method has the advantages that the height information is combined, the distance complete graph is generated, modeling is carried out based on the distance complete graph, a real three-dimensional distribution scene is more conformed, and the optimization effect is improved.

The invention discloses a power distribution network load flow calculation method based on graph synchronization parallel and forward-backward substitution, and the method comprises the steps: building a graph model in a graph database based on a power network, and enabling the graph model to be divided into a plurality of sub-graph models; setting an initial state for nodes in the sub-graph model, traversing the sub-graph model in a graph node parallel computing mode, and recording a node layer number and a node path of the sub-graph model; and in combination with the sub-graph model, completing graph calculation through forward tracking current calculation and backward tracking voltage calculation in sequence by utilizing a forward-backward substitution method, and judging whether an absolute value of a voltage difference value of nodes in the sub-graph model is smaller than the maximum iteration convergence precision or not. According to the method provided by the invention, the graph model is established, and graph calculation is carried out in a graph parallel mode in combination with a forward-backward substitution method, so that the calculation speed can be obviously improved.

The application discloses a water body saliency target detection method based on multi-graph learning, relates to the technical field of saliency target detection, and comprises the following steps: step one: data acquisition: water body information is acquired through unmanned aerial vehicle images or remote sensing images; step two: data preprocessing: an adjacency matrix is generated based on SLIC superpixel segmentation of the acquired images, and a complete graph is constructed; step three: multi-feature extraction: (1) color feature map extraction; (2) contrast feature map extraction; step four: saliency detection by a manifold ordering method; step five: combination of multiple saliency maps; step six: saliency optimization: the images are processed through a saliency optimization framework, and a new saliency effect drawing is obtained; the color feature map, the contrast feature map and the saliency detection map obtained by the manifold ordering method are weighted and fused with different weights, a new saliency map is obtained, and finally saliency optimization is performed, so that a new saliency target detection method based on a multi-graph learning model is realized.

This invention discloses an adaptive graph clustering method based on feature adversarial and graph Transformer, belonging to the field of image processing technology. The original graph is optimized by a feature adversarial autoencoder module to match prior distributions of representation features, obtaining attribute-level features. The feature adversarial autoencoder module includes an autoencoder and a feature adversarial module. The original graph is then processed by a graph Transformer autoencoder module to perform the convergence and fusion of local features and global structure, and information propagation, obtaining structural-level features. Adaptive fusion of attribute-level and structural-level dual-source features from the feature adversarial autoencoder module and the graph Transformer autoencoder module completes graph clustering. The training of the feature adversarial autoencoder module and the graph Transformer autoencoder module employs a self-supervised learning approach. A joint optimization loss function guides the joint optimization of graph representation learning and cluster assignment. The joint optimization loss function includes a feature adversarial loss function, an optimization loss function, and a self-supervised learning loss function. The feature adversarial loss function includes the feature reconstruction loss function of the autoencoder and the minimum cross-entropy loss optimization function of the feature adversarial module.