178 results about "Network analytics" patented technology

The invention discloses a cross-border e-commerce compliance intelligent auditing platform and a multi-language contract analysis method, and relates to the field of cross-border contract compliance auditing. In the multi-modal data access step, customs codes, laws and regulations and other multi-source data are collected, and 18 kinds of language contract texts are analyzed; in the cross-language semantic alignment step, a knowledge graph is constructed, and multi-language legal concept mapping is achieved; in the compliance risk reasoning step, a rule engine and an agent cooperatively check a contract, and the compliance conclusion confidence is calculated; the dynamic risk assessment step adopts an LSTM network to analyze historical data and predict a risk trend; in the multi-language report generation step, a multi-format bilingual or multilingual report is generated based on a template engine, encrypted and archived. According to the invention, cross-border contracts are audited efficiently and intelligently, dynamic adaptation laws and regulations are analyzed in multiple languages, compliance risks are identified accurately, and a multi-language report is generated quickly; therefore, the checking efficiency is improved, the manual workload is reduced, the compliance risk is reduced, and the enterprise cross-border business competitiveness and the risk response capability are enhanced.

The invention provides a BIM intelligent management platform oriented to a whole life cycle of project construction. A building information model, Internet of Things sensing data and a block chain evidence storage mechanism are integrated through a multi-source data fusion technology, and a whole-process data chain of association planning, design, construction, operation and maintenance is associated. The platform adopts space optimization Huffman coding to realize model lightweight, combines a constraint genetic algorithm to optimize a construction path, and applies a bidirectional long-short-term memory network to analyze an equipment state. A three-chain block chain system is reconstructed on the architecture, intelligent association of engineering quantity and payment nodes is realized through cooperation of a main chain, a calculation quantity side chain and an auditing side chain, and mobile terminal offline interaction is supported based on a digital-analog separation technology. The platform covers an intelligent design management unit, a block chain investment management unit, a dynamic correction management unit, a quality safety responsibility tracing unit, an NLP risk management unit and a digital twin operation and maintenance unit. The units achieve cross-system cooperation through a unified data bus, and a closed-loop management architecture covering the whole life cycle of project construction is formed.

The invention relates to a data processing method and device based on big data and advertisement pushing, and the method comprises the following steps: obtaining the historical behavior data of a user on a multi-channel platform, constructing a dynamic interest label map according to the historical behavior data, and depicting a user interest evolution process. Combining with a social relation network to analyze an interest propagation path, forming a user social interest diffusion trajectory, and introducing a time decay weighting mechanism to generate a dynamic interest decay curve. According to the method, user interests and advertisement materials are subjected to semantic similarity matching, a personalized advertisement recommendation list is generated, an optimal advertisement putting strategy is determined through multi-target optimization configuration and comprehensive consideration of display positions, opportunities and forms, accurate and efficient advertisement pushing is achieved, and the problems that a traditional user portrait method often depends on a static label system, and the user experience is poor are solved. The dynamic characteristic that the user interest changes along with time is difficult to reflect, so that the advertisement recommendation content lags behind the real intention of the user.

The invention relates to the technical field of environmental monitoring, in particular to an ecological environment and environmental pollution monitoring system and method, and a multi-source heterogeneous data acquisition module comprehensively acquires water quality, weather and space monitoring data. The system comprises a topological mapping data preprocessing module which is used for abnormal data detection and data vacancy filling; the state evolution analysis module is used for constructing an environment state transition network and analyzing an environment state evolution rule; the environment comprehensive evaluation module is used for calculating a water safety level index and a water pollution condition index and generating an environment quality evaluation report; the pollution diffusion prediction module is used for predicting a pollutant diffusion path and concentration distribution and identifying a high-risk area; the intelligent early warning and visualization module is used for generating graded early warning information and visually displaying the environment state and the pollution diffusion condition through multi-dimensional data, so that the coverage and the data integrity of environment monitoring are remarkably improved, the environmental pollution problem can be timely found and dealt with, and the ecological environment safety is guaranteed.

The invention discloses a detection method for printing effect verification, and the method comprises the steps: collecting scanning image information and environment sensing information, employing an improved OTSU algorithm and an illumination compensation module to process the scanning image information, and extracting a standardized image feature matrix containing the characteristics of character integrity, edge sharpness and the like; identifying a defect mode by using an EAST text detection algorithm and morphological operation, and generating a quantitative evaluation vector; the vector and environment sensing information are subjected to space-time alignment, a printing quality degradation model is constructed through a density peak value clustering algorithm, and a multi-dimensional quality index set is established; and finally, based on the graph attention network and the time sequence convolutional network, analyzing the incidence relation between the printing quality and the equipment state, and outputting a verification report containing a quality score, a defect positioning graph and a life prediction curve. According to the method, accurate quantitative evaluation of the printing quality is realized, a dynamic association model of the quality, the environmental parameters and the equipment aging is established, and an intelligent decision basis is provided for printing quality maintenance.

The invention relates to the technical field of data processing, and discloses a child autism behavior recognition method and system based on data analysis and a medium. The method comprises the steps of collecting and preprocessing child multi-modal behavior data; extracting nonlinear features to obtain a time sequence feature matrix; identifying a repeated behavior mode through a three-dimensional convolutional network; analyzing behavior time sequence change by using a time convolutional network; fusing a plurality of feature representations to obtain a comprehensive feature vector; a multi-classifier system is applied to identify autism behavior types and evaluate severity. By extracting the nonlinear time sequence features, constructing the deep spatial-temporal feature extraction network and designing a feature fusion mechanism and a multi-classifier integration system, the method can overcome the limitations of a single mode, a linear feature and a single algorithm in the prior art, and improves the accuracy and interpretability of autism behavior recognition.

The invention discloses a content recommendation method and system based on an industry knowledge graph and reinforcement learning, and relates to the technical field of advertisement recommendation, and the method comprises the steps: collecting original advertisement data and user behavior data, carrying out the entity recognition, relation extraction and attribute extraction, and constructing the industry knowledge graph; a graph attention network is adopted to analyze association strength among different modal entity nodes in the industry knowledge graph, advertisement elements are dynamically recombined according to the association strength, and a personalized advertisement material package is generated; mapping the personalized advertisement material package into a reinforcement learning action space, extracting a user historical behavior path from the industry knowledge graph to construct a state space, and obtaining a reinforcement learning environment; in a reinforcement learning environment, mapping the user behavior nodes into a relation chain of an industry knowledge graph, allocating reward values to intermediate nodes, and constructing an RL strategy model; deep modeling and strategy optimization of user behavior intentions are realized, and relevance of recommended content and intelligence of decision making are enhanced.

The invention discloses a public opinion event multi-mode semantic fusion modeling and abstract generation method and system, and relates to the field of natural language processing and social network analysis. Through the multi-mode semantic fusion technology, the short text understanding ability is improved, and the problems of semantic fuzziness and network language diversification are solved. Meanwhile, through a cross-window event cluster matching technology, an event evolution path with time continuity is constructed, and comprehensive capture of event dynamic characteristics is realized. Besides, the structured event abstract is automatically generated by utilizing the generative model, so that the consistency and the information density of the abstract are improved, and the actual application requirements are met. Through the innovations, the defects in the aspects of semantic comprehension, dynamic modeling and abstract generation in the prior art can be effectively overcome, a more efficient and accurate solution is provided for monitoring and analysis of public opinion events, and the method has wide application prospects in the fields of public opinion monitoring, emergency early warning, social media data analysis and the like.

The invention provides a knowledge graph-based public security multi-source risk factor association identification analysis method, which relates to the technical field of risk identification, and comprises the steps of obtaining multi-source risk factor data, extracting information from unstructured data, constructing an initial association network, performing feature analysis and calculating a similarity matrix; and the close association subgroups are identified through community discovery, a multi-level association network is constructed, a conduction path is analyzed, a weight is calculated, and finally risk early warning information is generated. According to the invention, the complex association between public security risk factors can be effectively identified, and the risk prediction accuracy is improved.

The invention relates to the technical field of data processing, and provides an unmanned aerial vehicle-mounted LiDAR geological modeling and intelligent blasting parameter optimization system, which is characterized in that a laser point cloud and a multispectral image are deeply correlated through a data acquisition module based on a tight coupling algorithm, the sampling quality of a point cloud data set is evaluated, and a supplementary flight instruction is judged and generated; a complete data source is provided for subsequent geological modeling; the geological modeling module divides the point cloud data set through an adaptive threshold algorithm to identify rock mass structural surfaces, analyzes a spatial topological relation of the rock mass structural surfaces based on a graph neural network, performs clustering to form a control structural surface group, and associates the control structural surface group with spectral features through a convolutional neural network to cooperatively discriminate rock mass properties; geological indexes of the three-dimensional geological model are extracted through a parameter optimization module, blasting parameters are generated after genetic algorithm processing, the blasting effect is judged through laser point cloud obtained again after blasting and change characteristics are extracted, the parameters of the genetic algorithm are reversely adjusted, and the three-dimensional geological model is updated.

The invention discloses an intelligent medical image diagnosis system and method based on hierarchical cross-modal conversion and dynamic feature tracking. The system adopts three-step cross-modal conversion: a first-layer small model for converting user questions to realize medical ontology matching; the second-layer multi-modal model extracts image features, and outputs text states such as JSON data with focus coordinates, density and other features; and the third-layer large model fuses the medical history and the image features to generate diagnosis suggestions, and credibility verification is carried out. A dynamic focus tracking engine is introduced, a focus evolution rule of multiple scanning is analyzed through a convolutional network, and an optical flow field is adopted to compensate artifacts. The system also integrates a multi-expert voting mechanism to simulate a clinical consultation process, and outputs consensus diagnosis and objection viewpoints. A hierarchical routing algorithm is designed for emergency treatment scenes, so that the recognition response time of emergencies such as pneumothorax is shortened. Further, the system automatically generates a full chain of evidence report that conforms to medical regulations, including a model version, a guide reference, and a data hash value.

The invention discloses a city management AI dispatch algorithm and system based on historical mining and responsibility matching, and the method comprises the steps: building and dynamically updating a city management element evolution graph through obtaining the multi-mode description information of a city management case and the real-time state data of disposal resources; calculating potential disposal effects of different candidate dispatching schemes by using a causal inference engine, and generating a comprehensive efficiency estimation vector; on the basis, a multi-target reinforcement learning strategy is adopted to generate an optimal dispatch instruction, and system parameters are continuously optimized through online element learning during execution; cooperative processing network analysis is activated for sudden complex events, and responsibility atlas reconstruction is triggered when the matching efficiency is low. According to the method, the accuracy and efficiency of case disposal are remarkably improved, disposal timeliness optimization, resource load balancing and improvement of the first solution rate are realized, and meanwhile, the adaptive capacity and continuous optimization efficiency of the system to complex scenes are enhanced.

According to the privacy enhancement type (k, d)-truss community search method for the social network, a novel KTG tree structure is constructed by fusing k-truss and G-tree indexes. According to the index structure, hierarchical community decomposition of a social graph and social distance information are fused, a refined boundary vector coding scheme is designed to support efficient distance calculation, and a double-cloud-server non-collusion architecture integrating improved homomorphic encryption and matrix encryption technologies is constructed. Through a two-stage security query process of first structure filtering and then distance verification, on the premise of protecting full-process privacy of a graph structure, a query intention, a distance matrix and an intermediate calculation result, efficient and accurate search of a close community in a large-scale social network is realized. The method is suitable for various scenes such as social recommendation, risk control, public opinion analysis and anti-fraud, and the problems of privacy disclosure and calculation efficiency in social network analysis are effectively solved.

The invention discloses a role representation-based self-excitation multi-agent cooperation method, which comprises the steps of collecting observation data of agents, generating trajectory vector dynamic role allocation, calculating a weighted sum of an environment reward rt and an internal reward based on the similarity of the agents and each role and optimizing role representation, and aggregating Q values of each agent through a value decomposition network. Calculating a joint Q value Qtot (rho t, at); analyzing the observation data based on the optimized strategy network, and generating a coordination action instruction in real time; the intelligent agents obtain local observation of the current behavior through interaction with the environment, the intelligent agents can generate more diversified strategies, and the cooperation efficiency between the intelligent agents is improved.

A system for event-driven distributed data network analysis, the system includes: a multitude of domain data nodes, each comprising a stream storage, a multitude of transformation pods, and a product API, with each domain data node ingesting heterogeneous event streams, validating schema compliance, performing enrichments, and making data products available under versioned schema contracts; each domain data node further comprises a hardware-based edge intelligence appliance comprising an enclosure, a compute module with heterogeneous processing units selected from CPUs, GPUs, and TPUs, a secure enclave module for confidential execution of transformation code on encrypted data, a protocol-structured stream storage medium supporting multi-level retention, and industrial I / O interfaces configured for direct interaction with sensors, programmable logic controllers, and cloud-native services; an event mesh backbone processor configured to connect the multitude of domain data nodes via a publish-subscribe architecture, with the backbone guaranteeing exactly one-time delivery, partition-level sorting, and geo-replicated durability; a Contextual Intelligence Engine coupled with the Event Mesh Backbone processor, the engine comprising a semantic graph memory and a Streaming Graph Reasoner configured to map raw events into an ontology, perform graph joins in real time, and recognize composite causal patterns across multiple domain data nodes; a governance and policy control unit integrated into both the data and control layers. This layer enforces zero-trust security, role-based access control, real-time provenance tracking, and regulatory compliance; and An actuation interface coupled with the contextual intelligence engine and the governance layer. The actuation interface is configured to generate control commands for operational technology devices and digital workflows, thus closing the loop from data acquisition to autonomous decision-making.

Described are examples for providing radio access network (RAN) analytics for a virtualized base station. An analytics engine includes a memory storing one or more parameters or instructions for operating the virtualized RAN and at least one processor coupled to the memory. The analytics engine is configured to perform multiple protocol layers of RAN processing for at least one cell at the virtualized base station. The analytics engine is configured to determine a time series of real-time metrics at two or more layers of the multiple protocol layers for the at least one cell or a user equipment (UE) connected to the at least one cell. The analytics engine is configured to correlate a time series for each of the two or more layers to detect a network condition. The analytics engine is configured to modify a configuration of the at least one cell based on the detected network condition.

The invention relates to the technical field of computer technologies, and discloses an old people social network analysis and recommendation system based on a graph neural network, which comprises a data acquisition module, a graph construction module, a graph neural network model module, a social network analysis module, a personalized recommendation module and a user interface module, the data acquisition module is used for acquiring and cleaning social related data of old people; and the graph construction module is used for constructing static and dynamic social network graphs. According to the old people social network analysis and recommendation system based on the graph neural network, a dynamic social network graph is constructed through the graph construction module, sliding time window updating is adopted, dynamic changes of the social relation of old people along with time can be captured, meanwhile, edge weight calculation is combined with the interaction frequency and the interaction depth, and the recommendation efficiency is improved. The interaction depth fuses semantic similarity and emotion scores, the interaction depth is accurately quantified, and a social network analysis module introduces a time decay factor prediction relation.

The invention discloses a family education content recommendation method and system based on big data, and relates to the technical field of big data, and the method comprises the steps: collecting family member multi-source data for preprocessing, carrying out the real-time analysis of the preprocessed data, updating a user file, setting a personalized recommendation target according to an analysis result, and carrying out the recommendation of family education content through the updated user file. The learning progress of a user is predicted by using an LSTM model, and a social relation and an interaction mode among family members are established through a social network analysis algorithm in combination with a robust algorithm and a Pearch ranking algorithm. According to the method, the family member social relation graph is constructed and optimized through social network analysis, the robust algorithm and the Pearch ranking algorithm, the social sub-groups are identified, family education content recommendation is dynamically adjusted in combination with the reinforcement learning algorithm, and the social adaptability and interest matching ability of personalized content recommendation are improved.

The invention discloses an intelligent team forming system and method, relates to the technical field of artificial intelligence and big data technology cross application, and aims to solve the problems of low team forming efficiency and poor team forming effect of a traditional team forming system. The system is provided with a user input module, a large language model support layer, a dialogue type information acquisition module, a multi-dimensional user portrait construction module, a self-adaptive intelligent matching recommendation module and an interpretable matching decision module to form a complete intelligent interaction closed-loop system. When a team formation demand exists, the system performs deep semantic understanding, multi-strategy fusion and graph network analysis to calculate an optimal team combination, and provides clear matching decision interpretation; team forming result feedback is collected, matching strategies and user portraits are continuously optimized, and the accuracy of future recommendation is continuously improved; the team forming quality and the user experience are continuously improved; the problems that a traditional team forming system is low in team forming efficiency and poor in team forming effect are solved.

The invention relates to the technical field of data management, in particular to a project data management system for research and development of small and medium-sized enterprises, which comprises a data acquisition module for acquiring research and development project data of the small and medium-sized enterprises through a multi-source data fusion technology; the quality diagnosis module analyzes research and development data by applying a deep learning network, and automatically identifies and predicts data quality abnormity and a potential risk mode; the process collaboration module is combined with artificial intelligence and a process engine technology, and automatically adjusts research and development task processes and authority distribution according to project progress and member collaboration requirements; the resource interaction module coordinates resource allocation, sharing and recovery between the research and development project and the enterprise internal resource pool; and the decision optimization module coordinates data of each module and a research and development project target to realize collaborative optimization of project planning, resource allocation and risk control, so that the problems of data fragmentation, low flow efficiency and unreasonable resource allocation in the research and development process of small and medium-sized enterprises are solved, and the research and development efficiency and the achievement transformation capability are improved.

The invention belongs to the technical field of early warning and alarming, and particularly discloses an Internet of Things fire-fighting facility early warning method and system based on data processing, and the method comprises the steps: collecting real-time temperature and smoke concentration data from each floor and a fireproof region of a high-rise building through a sensor network; performing initial spatial modeling on the acquired data by adopting a graph neural network to obtain an incidence matrix representing internal relation of adjacent regions; performing space segmentation processing on the acquired data according to the incidence matrix, and meanwhile, integrating a time segmentation mechanism to determine cross-regional boundary information reserved in the segmented data; if the temperature difference of adjacent areas in the segmented data exceeds a preset threshold value, analyzing time sequence change through a time sequence convolutional network, and judging a potential fire diffusion direction; the invention aims to solve the problems of inaccurate prediction diffusion direction and slow cross-floor linkage response caused by difficulty in capturing space correlation and time dynamic change in real-time fire monitoring in a high-rise building in the prior art.

The invention belongs to the technical field of bioinformatics and medical data analysis, and discloses a neurodegenerative disease comprehensive analysis platform based on multi-source data fusion, and the platform comprises a data integration module; a gene name conversion module; a core analysis module; a network analysis module; a drug screening module; the omics visualization module systematically catalogs data of genes related to at least 10 main neurodegenerative diseases, 486 drug-derived 3, 957 bioactive components and 18 lifestyle factors through the data integration module, and the data source is wide; according to the method, artificially sorted literature evidence, reanalyzed batch transcriptome data, single-cell RNA sequencing data and standardized data from a public database are covered, new disease targets and potential treatment strategies can be found easily, and the neurodegenerative disease research efficiency and comprehensiveness are improved.

The invention discloses a graph neural network fairness improvement method and system, and relates to the technical field of graph neural networks. The method comprises the following steps: analyzing graph data to construct a node set and dividing sensitive attribute subgroups; estimating probability distribution of subgroup prediction results by adopting a Bayesian smoothing technology; calculating a mutual information difference between a node prediction result and the sensitive subgroup as a node-level prejudice value, and further generating a global average prejudice; global prejudice is fused into a loss function, task loss and fairness constraint are balanced through a dynamic weight strategy, and model parameters are optimized. The system comprises a node set construction module, a sensitive attribute subgroup analysis module, a node prejudice calculation module, a global constraint generation module and a joint optimization training module. According to the invention, the limitation of macroscopic statistics is broken through, and node-level prejudice accurate positioning is realized; fusing graph structure information to improve fairness optimization efficiency; and the dynamic weight strategy balances the performance and fairness. The method is suitable for scenes such as social network analysis and financial risk control.

Systems, methods, apparatuses, and computer program products for network data analytics profiling for a user equipment (UE) are provided. One method may include obtaining data related to a UE and generating analytics for the UE based on the obtained data, generating a UE analytics profile for the UE using at least one of the obtained data and / or the analytics generated for the UE, and storing the UE analytics profile in a repository. The method may also include receiving, from a network node in the visited network, a request to retrieve the UE analytics profile for the visited network. The method may then include generating the visited network UE analytics profile, and providing the visited network UE analytics profile to the network node in the visited network.

The invention relates to the field of financial risk protection, in particular to a financial risk protection method and system based on data mining, and the method comprises the following steps: collecting enterprise multi-dimensional financial logs, carrying out the deep mining of transaction behaviors, and constructing an enterprise financial behavior map; performing multi-signal correlation degree calculation and liability structure change perception based on the enterprise financial behavior map, and marking potential fund chain vulnerabilities; upstream and downstream supply chain information of an enterprise is identified, supply chain ecological topology network analysis and risk effect calculation are carried out, and a key risk critical link is identified; carrying out multi-scene financial risk simulation on the potential fund chain vulnerability points and the critical links of the key risks, and constructing a financial risk simulation sand table; and carrying out impact bearing capability evaluation based on the financial risk simulation sand table. According to the method, the potential financial risk of the enterprise is accurately identified in advance, a protection decision is made in a targeted manner, the risk protection capability of the enterprise is improved, and enterprise assets are effectively protected.

The invention discloses an urban public facility layout optimization method and system based on multi-source space-time big data, and relates to the technical field of urban planning. Comprising the following steps: firstly, acquiring and fusing multi-source space-time big data such as mobile equipment signaling data and point-of-interest data; identifying and delimiting an urban public service three-level center system based on the fusion data through a spatial clustering and network analysis algorithm, and determining each center level and a spatial influence range; constructing a differentiated service demand quantification model for each level of service center; constructing a multi-stage collaborative layout optimization model, and completing facility hierarchical assignment and site selection optimization; evaluating the layout scheme in the simulation environment and outputting a grading evaluation report; through dynamic analysis and multi-level collaborative optimization driven by multi-source space-time big data, accurate matching of public facility layout and service requirements is achieved, and meanwhile the service coverage rate and cross-level connection efficiency are improved.

The invention relates to the technical field of user layering methods, in particular to a user layering method and device and a storage medium, and the method specifically comprises the following steps: 1, collecting basic information data of users, including names, ages, genders, contact information and registration time, collecting behavior data of the users, and collecting social relation data of the users, cleaning the basic data and the behavior data according to the social relation data; 2, modeling a user behavior sequence; step 3, analyzing the user social network; 4, establishing a user value evaluation model; 5, layering the users based on deep learning; step 6, carrying out dynamic optimization on a layering result; step 7, applying and feeding back a layering result; according to the method, the problem of single traditional data is solved through multi-source data acquisition and processing, and accurate features are comprehensively extracted; behavior sequence modeling and social network analysis break through the limitation that only individual behaviors are concerned, and users with similar behaviors and social contact are deeply mined.

The invention discloses a social group member dynamic management method and system, and relates to the field of social network analysis, and the method comprises the steps: obtaining online interaction data and offline co-occurrence data of social group members, cleaning the online interaction data and the offline co-occurrence data, and converting the cleaned online interaction data and offline co-occurrence data into a structured data set; importing the structured data set into a graph database to obtain an initial weight, calculating a dynamic attenuation weight, and constructing a basic relation network; and converting an edge weight matrix in the basic relation network into a quantum bit entanglement state, measuring the quantum bit entanglement state to form a quantum probability amplitude, setting a quantum entanglement judgment threshold, and generating a quantum relation thermodynamic diagram through a Grover algorithm. Through the technical scheme of combining quantum calculation and dynamic entropy analysis, accurate management and risk prevention and control of the social group membership are realized, an edge weight matrix of a basic relation network is converted into a quantum bit entanglement state, and a quantum relation thermodynamic diagram is generated by using quantum measurement and a Grover algorithm.

The invention relates to the technical field of transformer substations, and discloses a transformer substation remote intelligent patrol method and system based on big data, and the method comprises the steps: obtaining and preprocessing multi-source heterogeneous data; the data tokens are converted into a unified device state sequence; generating attention bias based on the causal knowledge base, guiding the cross-modal fusion network to analyze the sequence and outputting an abnormal token; constructing a dynamic causal map based on the abnormal token, identifying a fault root cause and forming a diagnosis conclusion; and generating a report according to the diagnosis conclusion, and performing closed-loop updating on the causal knowledge base. The system comprises a data acquisition module; a data processing module; an intelligent analysis module; and a decision support module. According to the method, causal reasoning is introduced, deep fusion of multi-source data and accurate positioning of fault root causes are achieved, the diagnosis conclusion interpretability is high, self-evolution can be achieved through closed-loop updating of the causal knowledge base, and the intelligent level of substation patrol is comprehensively improved.

The invention relates to a node injection attack method based on adaptive target selection, and the method comprises the following steps: S1, target node selection: calculating a comprehensive score of a node based on uncertainty and topology centrality, and dynamically selecting a target node set of a current attack round; s2, feature generation: using an adaptive feature generator to generate node features which are similar to target node distribution and have strong aggressiveness; and S3, disturbance edge construction: selecting an optimal disturbance edge connection mode for the injection node according to strategy network output in reinforcement learning. According to the method, the attack flexibility can be improved through dynamic target selection, the attack performance can be remarkably enhanced through combination of disturbance characteristics and structures, and the method has good concealment, expandability and generalization ability. The method is widely applied to security evaluation and defense research fields related to graph neural networks, such as social network analysis, recommendation systems, knowledge graphs and the like.