211 results about "Topological sorting" patented technology

The invention belongs to the field of natural language processing, and provides a task complexity driven graph semantic multi-agent collaborative decision-making method and system.The task complexity driven graph semantic multi-agent collaborative decision-making method comprises the steps that a task text is obtained and subjected to semantic coding to obtain a task semantic vector, evaluation is conducted based on the task semantic vector to obtain a complexity vector, and a task complexity score of the complexity vector is calculated; the task semantic vector and the complexity vector are fused to obtain a task representation vector, an agent capability relation graph is constructed, the participation probability of each agent node is obtained according to the task representation vector and the agent capability relation graph, and a dynamic agent combination scheme is formed; and performing task decomposition according to the agent combination scheme, constructing a sub-task dependency graph, scheduling the execution sequence of the sub-tasks through topological sorting, realizing cooperative execution of the agents, and generating a task result. According to the method, precise matching and efficient cooperation of the agent combination are realized, and the capability of processing complex tasks and the resource utilization efficiency of the multi-agent system are remarkably improved.

The invention relates to the technical field of robot scheduling, and discloses a robot task management framework system based on asynchronous communication. The system comprises a task asynchronous analysis module for performing time sequence decoupling on an original task instruction stream of a heterogeneous robot terminal by means of a multi-channel buffer queue to separate out independent executable task units; the resource dynamic mapping module is used for establishing a variable granularity resource allocation mapping table according to cluster real-time load and task resource requirements, and recording various resource dynamic binding relationships; the task priority reconstruction module is used for constructing a directed acyclic graph scheduling topology based on task logic dependence and deadline constraint, and generating a weighted execution sequence through topological sorting; the exception isolation processing module captures exceptions during operation and triggers an isolation strategy; and the cross-node cooperation module guarantees the multi-robot cooperation task state consensus through a distributed consistency protocol. The system is suitable for heterogeneous robot cluster complex task scenes.

The invention discloses a vehicle-mounted controller firmware upgrading method, which relates to the technical field of intelligent networked automobiles, and comprises the following steps of: accurately describing a dependency relationship between versions by introducing a directed acyclic graph, providing a clear topological sorting basis for an upgrading process, and improving the upgrading efficiency. The system function conflict and disorder caused by disordered upgrading sequence are fundamentally avoided; a distributed consensus protocol is utilized to ensure that each controller achieves state coordination before upgrading, so that the overall robustness of a multi-controller cluster is effectively improved when the multi-controller cluster faces network delay or individual node abnormity, and upgrading failure or version bifurcation caused by the fact that part of nodes are not ready is avoided; through the combination of the group-level binding token and the atomic switching mechanism, the controller group with a strong function dependency relationship can be ensured to be effective and updated as a unified transaction unit, and the safety and reliability of cross-domain function linkage are remarkably enhanced.

The invention discloses a multi-scene multiplexing-oriented VUE dynamic component interface configuration and rendering method, which comprises the following steps of: receiving a multi-scene configuration file, and extracting and coding to generate a scene feature vector; component layout parameters, resource priorities and rendering strategies are automatically generated through the two-stage generative adversarial network, and application parameters are output after the generated parameters are optimized; on the basis of application parameters, asynchronous component loading and priority queue scheduling are adopted, and on-demand loading of components is achieved in combination with dependency topology sorting; dynamically dividing a rendering area by using an adaptive Gaussian algorithm, and rejecting invalid rendering through a boundary box mechanism; the DOM operation frequency is reduced by adopting an optimization Diff algorithm; and constructing a component cache pool through an elimination strategy. According to the method, intelligent configuration and efficient rendering of the interface component under multiple scenes are realized, the development efficiency is improved, and the performance is optimized.

The invention relates to the technical field of constructional engineering informatization and data security, and particularly discloses a BIM and block chain-based whole-process engineering cost data collaborative management system, which comprises a data analysis and dependency graph construction module, which is used for analyzing BIM model component attributes and association relationships and constructing a directed dependency graph; the loop dependency detection and resolution module is used for identifying and resolving a loop dependency structure in the graph; the topological sorting processing module is used for generating an ordered calculation sequence of the components; the cryptographic abstract calculation module is used for generating a unique abstract value of each component in sequence; the global abstract generation module is used for aggregating and generating a global cryptographic abstract representing the overall logic state of the model; and the block chain evidence storage module is used for combining the global abstract and the related metadata to perform uplink evidence storage.

The invention provides a stacked cell hierarchy judgment method based on contour concavity and convexity. The method comprises the following steps: performing contour extraction on a cell binary image; calculating a convex hull for the contour point set of each cell, and taking the convex hull as an ideal form reference of the cell in an unshielded state; detecting the convexity defect between the contour point set of each cell and the corresponding convex hull, and quantifying the characteristic parameter of each convexity defect; screening the convexity defects according to the characteristic parameters to determine significant recesses, marking cells with at least one significant recess as concave cells, and marking cells without significant recesses as convex cells; screening candidate cell pairs according to the spatial position relationship between the cells; according to the convex-concave attributes of the two cells in the candidate cell pair and the position relation between the coordinates of the center of mass and the remarkable concave area, the shielding relation between the two cells is judged; and constructing an attribute graph by taking each cell as a node and taking the occlusion relationship as an edge, and generating a cell stacking hierarchy sequence through topological sorting.

The invention provides a Hadoop cluster incremental backup recovery method based on data block optimization, and relates to the technical field of data management, and the method comprises the following steps: obtaining distribution state information and historical change records of data blocks in a distributed file system, extracting space-time correlation characteristics, and constructing a data block dependency graph containing dependency direction and dependency intensity; identifying the changed data blocks, performing topological sorting, and identifying key path data blocks and independent data block clusters; calculating a distribution entropy value according to a copy distribution position, and determining a backup granularity type; a backup execution sequence is generated according to a topological sorting result, key path data blocks are backed up preferentially, independent data block clusters are backed up in parallel, and backup metadata containing dependency relationship identifiers and distribution entropy values are generated; and performing dependency consistency verification after recovery based on the dependency relationship identifier. According to the method, the backup recovery efficiency is improved, the data consistency is guaranteed, and the storage overhead is reduced.

The invention relates to the technical field of embedded system construction, in particular to a software dependency relationship automatic integration method and system based on meta packaging, and the method comprises the steps: S1, creating a meta packaging framework in an embedded construction system, and defining and constructing a stage hook through a Makefile; s2, automatically deploying the pre-compiled software package and the shared dependency library to a target directory of a construction system through a resource injection script; s3, analyzing a metadata file of the pre-compiled software package, extracting dependency relationship information, and generating a global dependency relationship list based on a directed acyclic graph topological sorting algorithm; and S4, inversely writing the global dependency relationship list into the dependency declaration configuration of the meta-packaging by utilizing the dependency aggregation script. By implementing the scheme, the problems of dependency sequence disorder and metadata auditing failure during embedded system integration are solved, the reliability of system integration is improved, the system construction efficiency is optimized, the safety of system construction is enhanced, and the maintainability and expansibility of system construction are improved.

The invention relates to the technical field of distributed data processing systems, and discloses a data processing system supporting full localization deployment and a compliance guarantee method thereof.The data processing system supporting full localization deployment comprises an audit event collection module used for collecting and storing processing event hash; the event organization module is used for realizing logic time sequence management of events through topological sorting and causal relationship marking; the hash chain construction module is used for constructing a local-to-global double-layer hash chain by utilizing the constructed directed acyclic graph and event hash; the increment synchronization module is used for executing increment synchronization based on Hash tree difference identification; the conflict solving module is used for detecting and solving event conflicts in the distributed environment by applying a version vector technology aiming at data differences found in the incremental synchronization process; according to the method, the lightweight linear hash chain is used for replacing a traditional binary tree structure, high-efficiency and low-cost audit chain maintenance is realized, and the data integrity is guaranteed in an unstable network environment.

The invention relates to the technical field of derivative variable calculation, and discloses a derivative variable calculation method and system based on a graph algorithm, and the method comprises the steps: analyzing derivative variable definition and calculation logic through employing a variable SQL syntax tree analysis technology, taking an analyzed variable as a node, determining a variable dependency relationship as an edge according to a processing task, and carrying out the calculation of the derivative variable; constructing a directed acyclic graph and storing the directed acyclic graph in a graph database; the nodes store variable metadata, execution priorities and calculation information, and store calculation task metadata and calculation information; according to node priorities, finding out high-priority variable nodes, querying all in-degree edges of the high-priority variable nodes, forming one or more subtrees, forming high-priority execution subtrees, performing topological sorting on the execution subtrees, and performing execution by an execution engine; the method aims to solve the problems of low calculation efficiency and insufficient utilization of calculation resources in the prior art.

The invention discloses an open source component analysis method based on a static program, and relates to the technical field of software analysis. Project source codes are input and analyzed, a semantic calling graph is obtained, and hierarchical management is conducted on the semantic calling graph; carrying out topological sorting on the layered semantic calling atlas to obtain a component check sequence; and comparing the semantic call maps corresponding to the new version and the old version of the to-be-upgraded component, performing software detection operation on the to-be-upgraded component, and outputting a conflict report. According to the method, a layered semantic calling graph is constructed and contains rich context information, so that analysis can go deep into the level of whether behaviors and semantics of the method are changed or not, topological sorting is carried out on component dependency relationships, a component checking sequence is provided, the problem that a conflict is repaired in reality is solved, and the method is more practical. By means of the method, it is guaranteed that before any component is checked, all the components on which the component depends are analyzed and confirmed, and orderliness and certainty of the analysis process are guaranteed.

Provided are a text generation method, apparatus and system, a computing device, and a readable storage medium. The method comprises: acquiring target data to be analyzed, and on the basis of the target data, generating a text generation template on which a text needs to be based; on the basis of each piece of configuration information in the text generation template, determining a plurality of tasks to be executed for the target data; performing topological sorting on the plurality of tasks on the basis of a dependency relationship between the plurality of tasks, and determining, from among the plurality of tasks, target tasks supporting parallel execution; and executing the target tasks in parallel for the target data, and integrating the execution results of the target tasks to generate a text conforming to the text generation template. The method can improve the text generation efficiency.

The invention relates to the technical field of software, in particular to an automatic construction method based on dependency topology analysis and dynamic orchestration, which comprises the following steps of: reading and analyzing a predefined configuration file to construct a service dependency relationship graph; wherein nodes in the service dependency relationship graph represent services, and directed edges represent dependency relationships among the services. According to the method, the predefined configuration file is read and analyzed to construct the service dependency relationship graph, so that automation of dependency management is realized, and manual identification errors are avoided; through dynamic mapping of services and construction tasks, the problem of conversion from a logic architecture to physical execution is solved; based on collaborative optimization of topological sorting and a priority strategy, the construction efficiency and the resource utilization rate are improved, and time consumed by a critical path is shortened. Therefore, full-link automation from dependency identification and task mapping to intelligent arrangement is completed, and the efficiency bottleneck and error risk caused by manual intervention are effectively solved.

The invention provides a low-code and component intelligent collaborative application construction method based on a dual-drive architecture, and belongs to the technical field of computer programming. The method comprises the following steps: establishing a component dependency digraph, optimizing a component collaborative relationship by adopting topological sorting, graph coloring and a minimum spanning tree algorithm, collecting multi-dimensional performance indexes to establish a monitoring matrix, processing monitoring data by utilizing a collaborative optimization model containing a gating information propagation mechanism, and generating a dynamic scheduling decision vector; the load change of the system is judged through a change rate threshold equation, quick or slow response adjustment is executed, and finally, a low-code component library and a visual configuration interface are established to realize intelligent collaboration and dynamic resource allocation among components. The technical problems that the collaboration efficiency among low-code platform components is low and intelligent dynamic scheduling cannot be achieved according to system load changes are solved.

The invention provides a dynamic reasoning chain construction and dependency task arrangement method and device based on an MCP agent, and relates to the field of artificial intelligence. The method comprises the following steps: acquiring context information input by a user, and acquiring a current task environment of a target agent based on the context information; based on a task environment, combining context information and taking a task target as a guide to construct a target reasoning chain; performing task dependency analysis on the target reasoning chain, and constructing a target dependency graph based on element constraint conditions; performing topological sorting on the target dependency graph through a graph isomorphic coding sorting model, and outputting an ordered task sequence with priority labels; and constructing a dynamic scheduling framework based on the ordered task sequence, and outputting a task processing result corresponding to the task target according to the dynamic scheduling framework. According to the method, the problem that the task execution efficiency of an existing multi-agent task arrangement method is remarkably reduced when the task targets are diversified and a complex dependency relationship exists among the task nodes is solved.

The invention provides a multi-format document analysis and verification method and system and a program product, and relates to the technical field of document processing and intelligent verification. The method adopts a layered modular architecture, and comprises the following steps: a document analysis layer selects an analysis strategy according to document types and complexity, and uniformly converts various formats of documents such as PDF, Word, Excel and the like into a Markdown intermediate format; the index extraction layer uses an abstract base class and a dynamic Prompt optimization mechanism to extract structured data from an intermediate document and integrate the structured data into a unified JSON format to realize complete decoupling of data extraction and rule verification; and the rule verification layer dynamically loads a business rule by scanning an independent file, constructs a rule-dependent directed acyclic graph, and determines an execution sequence by utilizing topological sorting. According to the method, the problems of high data and rule coupling degree, high maintenance cost and poor expansibility in an existing system are effectively solved, hot plugging of the rules, dependency on automatic management and multi-document concurrent processing are realized, and the reusability and execution efficiency of the system are remarkably improved.

The invention relates to the technical field of field bridge scheduling, in particular to a partial order and multi-index evaluation-based field bridge scheduling method, device and equipment and a medium, and the method comprises the steps of obtaining a task queue which comprises a plurality of container management tasks; analyzing a dependency relationship among all the container management tasks in the task queue, generating a partial order relationship based on the dependency relationship, and traversing all the tasks according to the partial order relationship to generate a directed acyclic graph; generating a plurality of scheduling sequences based on the directed acyclic graph by using a topological sorting algorithm; calculating a plurality of core indexes of each scheduling sequence according to a moving route planned by a field bridge for completing the task queue in each scheduling sequence; and on the basis of each core index and the preset weight corresponding to the core index, the comprehensive evaluation score of each scheduling sequence is calculated, and the scheduling sequence with the lowest comprehensive evaluation score is a pre-selected scheduling sequence. The method can optimize the dispatching decision of the yard bridge, and improves the operation efficiency of the yard equipment.

The invention is applicable to the technical field of artificial intelligence, and provides a large-model multi-agent-based method for automatically generating a complex-structure document, which comprises the following steps of: mapping an original structured document template into a compressed structure mark sequence with a reserved layout, and performing zero-sample semantic understanding and planning through a structure planning agent to obtain a complex-structure document; generating an atomic operation execution plan; according to the semantic logic relationship among the atomic operations in the structured execution plan, constructing a field-dependent directed acyclic graph, carrying out topological sorting on the field-dependent directed acyclic graph, and determining an execution sequence of content generation; through collaborative iteration of the content generation agent and the symbol verification agent, each field content is generated under the condition that a hard constraint condition is met; and finally synthesizing a complete document. According to the method, accurate understanding, reliable generation and format guarantee of the complex document under the zero sample condition are realized.

The invention belongs to the technical field of intelligent education and personalized learning, and provides a personalized learning path recommendation method based on a course knowledge point sequence. The method aims at solving the problems that existing learning path recommendation logic association is insufficient, the personalized matching degree is low, and the optimization efficiency is not high. The method comprises the following steps: constructing a knowledge point sequence model by adopting a topological sorting algorithm on the basis of a curriculum knowledge point pre-modification and post-modification relationship, a difficulty level and dependence intensity, and generating a structured knowledge relationship graph; a learner feature model and a knowledge resource model are established, learner features are described from the dimensions of learning targets, cognitive levels, learning styles and the like, and the matching degree with knowledge points is calculated; and constructing an optimization model taking learning cost minimization, mastering degree maximization and path continuity optimization as targets, performing iterative optimization by using an improved intelligent optimization algorithm, and generating a learning path with logic rationality and personalized adaptability. According to the method, the scientificity and learning efficiency of learning path recommendation can be effectively improved.

A method for adaptive repair of a freeform surface includes: performing data discretization on a worn workpiece to obtain a worn body to be repaired, and performing adaptive segmentation on the worn body to be repaired to obtain initial reference surfaces, wherein the initial reference surfaces include a worn model bottom surface and a standard model top surface; constructing a freeform surface slice based on each of the initial reference surfaces; calculating surface curvature distribution data of the freeform surface slice, and constructing a curvature-overlapping pitch optimization model to perform adaptive trajectory filling on the freeform surface slice to obtain freeform surface scanning trajectories; and performing topological sorting on scanning trajectory points of the freeform surface scanning trajectories to obtain processing trajectories, fitting the processing trajectories, and performing adaptive repair on the freeform surface. This method eliminates a severe “staircase effect” in a traditional slicing method.

The invention relates to the technical field of data processing, in particular to a data information storage method and system for an online Chinese language learning platform. The method comprises the following steps: obtaining platform user interaction data, and carrying out pinyin input track analysis and Chinese character component structure coding to obtain a pinyin alignment input frame; extracting a pinyin-Chinese character mapping relation in the pinyin alignment input frame, and performing Chinese character structure path reconstruction to obtain a user input word structure path set; learning content data are called, multi-modal content node embedding is carried out, and a content linkage fragment cluster is obtained; constructing a single session learning topology track according to the content linkage fragment cluster, and performing bidirectional topology sorting to obtain a learning track topological graph; and establishing a multi-dimensional primary key index and behavior triggering mechanism for the learning track topological graph to obtain a dynamic index distribution block, and transmitting the dynamic index distribution block to an online Chinese language learning platform. According to the invention, the stability and expansibility of the learning platform are obviously improved.

This invention discloses an online editing and parallel execution method and system for vehicle network test cases, relating to the field of vehicle network testing technology. It establishes an online editing environment via a web interface, enabling automatic generation of protocol code and providing testers with a convenient and efficient development experience. Regarding test case execution, it accurately analyzes the dependencies of test cases, uses a topology sorting algorithm to construct a reasonable execution plan, and adaptively switches between sequential and parallel execution modes. Combined with containerization isolation technology, it allocates an independent execution container for each test case, ensuring the isolation and stability of the testing environment. Simultaneously, it initializes a unified test execution context for each test case and establishes a communication connection with the vehicle network bus simulation engine through a standardized interface, achieving standardization and normalization of operations such as message sending and receiving, fault injection, and assertion verification. In variable management, it adopts a copy-on-write mechanism to achieve isolated sharing of project variables, avoiding data crosstalk.

This invention discloses an automatic business application construction method based on AI semantic analysis, belonging to the field of artificial intelligence semantic analysis and automatic software application generation technology. The method includes: acquiring user-input business requirement description text, performing word segmentation and stop word filtering to form a requirement word sequence; inputting the requirement word sequence into a pre-trained semantic encoding model to generate semantic feature vectors; extracting business intent tags, functional entity sets, and dependency triples; performing similarity matching between the business intent tags and functional entities and component semantic vectors in a component knowledge base to determine the main control component and functional components; constructing a functional component call relationship graph and generating a component topology structure according to topological sorting; and calling an application building engine to instantiate each component sequentially according to the component topology structure, outputting a business application package. This invention enables automatic mapping of business requirements to application structures, reduces manual configuration, and improves the intelligence and consistency of business application construction.

This invention relates to a multi-agent orchestration and execution method supporting complex cyclic graph structures, comprising: S1: receiving workflow configuration and constructing a directed graph data structure; S2: performing a depth-first search traversal on the directed graph data structure, detecting and outputting strongly connected components in the directed graph data structure; S3: abstracting strongly connected components containing multiple nodes into super nodes, and keeping single-node strongly connected components as ordinary nodes, thereby constructing a directed acyclic graph containing super nodes; S4: performing topological sorting on the directed acyclic graph to generate a hierarchical execution plan containing multiple execution layers; S5: scheduling execution items according to the hierarchical execution plan, and when an execution item is the super node, starting a loop executor for iterative execution. This method, by constructing a multi-layered technical system, integrates graph algorithm innovation, unified context management, abstraction layer design, metadata-driven techniques, etc., to achieve automated orchestration and execution of multi-agent workflows supporting cyclic graph structures.

The application provides a causal discovery method and device fusing field knowledge of a large language model and a flow model, and relates to the technical fields of artificial intelligence and data mining. The method comprises the following steps: acquiring observation data and variable semantic information of a system to be analyzed; determining a causal topological order between variables by using a large language model; constructing a mask affine autoregressive flow model based on the causal topological order, and training the model by using the observation data; calculating a Jacobian matrix of the flow model with respect to input data, constructing a statistic based on the Jacobian matrix, and performing hypothesis testing; pruning according to a test result to obtain a candidate causal graph; inputting the candidate causal graph into the large language model, and performing secondary verification and pruning on candidate connections by using the variable semantic information to obtain a final sparse causal graph. The application introduces prior knowledge by using semantic information of the large language model, and fits data by using a more flexible function form of the flow model, thereby significantly improving the accuracy and robustness of causal discovery.

A pipeline generation method, comprising: acquiring a plurality of data cleaning operators for generating a pipeline; acquiring an attribute association relationship of data; extracting an attribute set from the plurality of data cleaning operators; grouping attributes in the attribute set on the basis of the attribute association relationship, so as to obtain a plurality of attribute groups which support concurrent computation, and the sorting of the plurality of attribute groups; on the basis of a topological sorting result of at least one attribute group among the plurality of attribute groups, determining the intra-group sorting of an operator group that corresponds to the at least one attribute group; and on the basis of the sorting of the plurality of attribute groups and the intra-group sorting of the operator group that corresponds to the at least one attribute group, orchestrating the plurality of data cleaning operators, so as to obtain a pipeline. In the method, attributes are first grouped to realize coarse-granularity pipeline recommendation, and the execution sequence of operators in a group is then intelligently orchestrated, so as to greatly reduce the number of operator combinations which require searching, thereby improving the efficiency of pipeline generation; moreover, the execution time is shortened by means of concurrency on the basis of ensuring precision, thereby improving the execution efficiency.

The application provides a method for obtaining power network layout based on directed acyclic graph topology unfolding, comprising the following steps: step 1, separating unconnected components; step 2, adding a virtual node before each source end, and adding the out-degree of the virtual node to the source end to form a directed acyclic graph with a single starting point; step 3, determining the node serial number, and rearranging the node out-degree according to the node serial number; step 4, performing topology sorting on the directed acyclic graph to obtain a topology sorting list; and step 5, obtaining the layout from the topology sorting. The application introduces a deep analysis method based on directed acyclic graph topology, and innovatively adjusts the out-degree of the directed acyclic graph to generate an optimized topology sorting.

The invention relates to a large language model driven NL2SQL intelligent query system in the financial field, and belongs to the technical field of SQL query. The system comprises a domain semantic analysis module, a query planning and disassembling module, an SQL generation and optimization module, a query assembly and verification module and a continuous learning and feedback module. The domain semantic analysis module is used for realizing conversion from natural language query to structured semantic representation, and constructing an explicit calculation dependency relationship graph through domain self-adaptive pre-training and semantic dependency analysis; and the query planning and disassembly module is used for identifying calculation nodes and types by analyzing the dependency graph, extracting atomic calculation units, planning an execution sequence through topological sorting, and generating a structured disassembly plan. According to the method, the financial semantic understanding capability is constructed through field self-adaptive fine tuning, the calculation dependence graph explicit modeling query logic is utilized, reliable sub-queries are generated through modular disassembly and optimization, and finally a complete SQL is formed through combination of a verification mechanism.

The application discloses a kind of based on multi-focal sharpness analysis's stacked cell hierarchy determination method and system.Acquire Z-stack multi-focal image sequence and extract each focal plane cell contour;Based on centroid coordinates and area features, establish the profile correlation data table of the same cell in different focal planes;Ellipse fitting algorithm is used to complete the shape of the profile of the cell that is blocked;Based on complete geometry mask, screen potential stacked candidate cell pair and locate overlapping area profile;With gradient function as evaluation index, calculate the overlapping area profile sharpness of each cell under multiple focal planes, determine the best focus focal length and judge cell hierarchy relationship;Attribute graph is constructed and topological sorting is executed, and global stacked hierarchy sequence is output.The application establishes hierarchy reasoning rule with rigorous physical basis according to the physical law that upper cell best focus focal length is higher, is compatible with existing Z-stack imaging process, adapts to basic laboratory high-throughput analysis demand.

The application provides a causal discovery method and device fusing field knowledge of a large language model and a flow model, and relates to the technical fields of artificial intelligence and data mining. The method comprises the following steps: acquiring observation data and variable semantic information of a system to be analyzed; determining a causal topological order between variables by using a large language model; constructing a mask affine autoregressive flow model based on the causal topological order, and training the model by using the observation data; calculating a Jacobian matrix of the flow model with respect to input data, constructing a statistic based on the Jacobian matrix, and performing hypothesis testing; pruning according to a test result to obtain a candidate causal graph; inputting the candidate causal graph into the large language model, and performing secondary verification and pruning on candidate connections by using the variable semantic information to obtain a final sparse causal graph. The application introduces prior knowledge by using semantic information of the large language model, and fits data by using a more flexible function form of the flow model, thereby significantly improving the accuracy and robustness of causal discovery.