Project risk intelligent early warning method and system based on knowledge graph

Abstract

This invention proposes a knowledge graph-based intelligent early warning method and system for project risks. The method includes: acquiring a structured data table; obtaining the activated nodes and corresponding relational edges in each stage based on the structured data table; inputting multiple stage subgraphs into a graph neural network to obtain a risk trend score; defining high-risk nodes based on the risk trend score; constructing a propagation path set from the upstream nodes of high-risk nodes; selecting a candidate strategy set from a preset strategy set; obtaining a path coverage coefficient and a response delay penalty term based on the candidate strategy set and the propagation path set; constructing a strategy structure influence weight based on the path coverage coefficient and the response delay penalty term; obtaining a strategy benefit score for each candidate strategy based on the strategy structure influence weight, risk suppression consistency term, strategy execution cost, and strategy saturation penalty term; and selecting a recommended execution strategy based on the strategy benefit score. This method can accurately identify project risks and intervene in project risks.

Description

Technical Field

[0001] This invention belongs to the field of project risk, and in particular relates to a method and system for intelligent early warning of project risks based on knowledge graphs. Background Technology

[0002] In managing complex projects, risk warning and response remain core challenges for managers. Existing project management systems mostly record data such as task plans, contract terms, personnel assignments, and organizational structures. However, this data often exists in isolated tables, lacking unified structured modeling capabilities and failing to reveal the complex relationships between tasks, contracts, and personnel as a whole. Traditional risk identification methods rely primarily on single-point indicators, such as schedule delays or contract defaults, lacking the ability to analyze multi-factor coupled risks and structural propagation paths. This leads to inaccurate risk identification and fails to meet the differentiated warning needs at different stages of the project lifecycle. Furthermore, current systems typically only provide risk warnings, lacking the ability to link risks to specific intervention strategies, and cannot offer practical action suggestions before risks materialize. Therefore, accurately identifying and intervening in project risks has become a pressing technical problem. Summary of the Invention

[0003] The purpose of this invention is to design a project risk intelligent early warning method and system based on knowledge graphs, which can accurately identify project risks and intervene in them.

[0004] To achieve the above objectives, a first aspect of the present invention provides a project risk intelligent early warning method based on knowledge graphs, the method comprising:

[0005] Obtain a structured data table, construct a knowledge graph based on the structured data table, and filter out all activated nodes and corresponding relation edges in each stage from the knowledge graph to obtain multiple stage subgraphs;

[0006] The stage subgraph is input into a preset graph neural network to obtain a risk representation vector. The risk representation vector is then linearly transformed to obtain a risk trend score. Nodes with risk trend scores greater than a risk threshold are defined as high-risk nodes. A propagation path set is constructed based on the upstream node paths of the high-risk nodes.

[0007] Candidate strategies whose scope covers at least one node on the propagation path set are selected from the preset strategy set to obtain a candidate strategy set;

[0008] Coverage is calculated based on the candidate policy set and the propagation path set to obtain the path coverage coefficient, the position and timing of the policy action node in the path are obtained, the response delay penalty term is obtained, and the policy structure influence weight is constructed based on the path coverage coefficient and the response delay penalty term.

[0009] Each candidate strategy is scored based on the strategy structure influence weight, the preset risk suppression consistency term, the strategy execution cost, and the strategy saturation penalty term to obtain a strategy benefit score. The corresponding candidate strategy is selected as the recommended execution strategy based on the strategy benefit score.

[0010] Furthermore, the step of obtaining a structured data table and constructing a knowledge graph based on the structured data table includes:

[0011] Obtain a structured data table, and select a set of nodes and a set of relation edges from the structured data table;

[0012] The stage edge activation marker of each relation edge is obtained based on the node set, and the stage position code of each task node is obtained based on the node set.

[0013] The knowledge graph is constructed based on the set of nodes, the set of relation edges, the stage edge activation markers, and the stage position codes.

[0014] Furthermore, the node set includes multiple task nodes, each task node including a current stage number, a start stage number, and an end stage number. Obtaining the stage position code for each task node based on the node set includes:

[0015] Obtain the difference between the current stage number and the starting stage number to get the first difference;

[0016] Obtain the difference between the termination stage number and the starting stage number to get the second difference;

[0017] The ratio of the first difference to the second difference is obtained to obtain the stage position code.

[0018] Further, the step of selecting the corresponding candidate strategy as the recommended execution strategy based on the strategy effectiveness score includes:

[0019] Sort the strategy benefit scores from largest to smallest, and select the candidate strategies corresponding to the top k strategy benefit scores as the recommended execution strategies; or select the candidate strategies corresponding to strategy benefit scores greater than the score threshold as the recommended execution strategies.

[0020] Further, the step of calculating the path coverage coefficient based on the candidate strategy set and the propagation path set includes:

[0021] Obtain the set of nodes that affect all candidate strategies, and obtain the set of path nodes in the propagation path set;

[0022] The number of nodes that intersect in the set of active nodes and the set of path nodes is obtained to get the number of intersecting nodes;

[0023] The path coverage coefficient is obtained by calculating the ratio between the number of intersection nodes and the number of nodes in the path node set.

[0024] Furthermore, the location and timing of the policy-implementing node in the path are used to obtain a response delay penalty term, including:

[0025] Obtain the distance between each intersection node and the high-risk node, and add them together to obtain the first distance;

[0026] Obtain the distance between each path node and the high-risk node, and add them together to obtain the second distance;

[0027] The response delay penalty term is obtained by calculating the ratio between the first distance and the second distance.

[0028] Furthermore, the method further includes: obtaining the policy saturation penalty term, specifically including:

[0029] The number of times the candidate strategy has been selected in all the high-risk nodes is obtained to obtain the number of candidate strategies;

[0030] Obtain the maximum number of times the candidate strategy can be executed;

[0031] The strategy saturation penalty term is obtained by calculating the ratio of the number of times the candidate strategy is executed to the maximum number of times.

[0032] A second aspect of the present invention provides a knowledge graph-based intelligent early warning system for project risks, the system comprising:

[0033] The acquisition unit is used to acquire a structured data table, construct a knowledge graph based on the structured data table, and filter out all activated nodes and corresponding relation edges in each stage from the knowledge graph to obtain multiple stage subgraphs.

[0034] The construction unit is used to input the stage subgraph into a preset graph neural network to obtain a risk representation vector, and to perform a linear transformation on the risk representation vector to obtain a risk trend score. Nodes with risk trend scores greater than a risk threshold are defined as high-risk nodes. A propagation path set is constructed based on the upstream node paths of the high-risk nodes.

[0035] The filtering unit is used to filter out candidate strategies whose scope covers at least one node on the propagation path set from the preset strategy set, thereby obtaining a candidate strategy set.

[0036] The calculation unit is used to calculate the coverage based on the candidate policy set and the propagation path set, obtain the path coverage coefficient, obtain the position and timing of the policy action node in the path, obtain the response delay penalty term, and construct the policy structure influence weight based on the path coverage coefficient and the response delay penalty term.

[0037] The scoring unit is used to score each candidate strategy based on the strategy structure influence weight, a preset risk suppression consistency term, strategy execution cost, and strategy saturation penalty term to obtain a strategy benefit score, and select the corresponding candidate strategy as the recommended execution strategy based on the strategy benefit score.

[0038] In a third aspect of the invention, an electronic device is provided, the electronic device including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the method described in the first aspect above.

[0039] In a fourth aspect of the invention, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect above.

[0040] The beneficial technical effects of the present invention are at least as follows:

[0041] To address the aforementioned issues, this invention provides a knowledge graph-based intelligent project risk early warning method and system. Its core lies in constructing a multi-entity heterogeneous knowledge graph by introducing stage semantics and location encoding mechanisms, overcoming the problems of traditional data fragmentation and inability to express temporal evolution relationships, thus providing a unified semantic structure for subsequent analysis. Secondly, in the risk identification process, a graph neural network model incorporating path density adjustment factors is designed to achieve stage-aware risk trend prediction and structural propagation path reasoning, enabling the system to identify potential high-risk nodes and trace their causes. Thirdly, in the strategy modeling stage, quantitative indicators of path coverage coefficient and response delay penalty term are introduced to establish a precise mapping between strategies and risk paths, solving the problem that traditional methods cannot structurally evaluate the scope of strategy effectiveness. Finally, in the strategy optimization stage, the invention constructs a strategy benefit scoring function that integrates structural influence, risk consistency, execution cost, and strategy saturation, forming a ranking and recommendation mechanism to ensure that the output strategies are both structurally targeted and conform to project resource constraints, thereby achieving a closed loop of accurate risk identification and response decision-making. The overall solution of this invention closely integrates risk trend modeling and strategy execution, forming a reproducible, interpretable, and implementable intelligent project risk early warning and intervention system. Attached Figure Description

[0042] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0043] Figure 1 This is a flowchart of the project risk intelligent early warning method based on knowledge graph provided in the embodiments of this application.

[0044] Figure 2 yes Figure 1 The flowchart of step S101.

[0045] Figure 3 This is a schematic diagram of the structure of the knowledge graph-based intelligent early warning system for project risks provided in this application embodiment. Detailed Implementation

[0046] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0047] Please refer to Figure 1 , Figure 1 This is a flowchart of the knowledge graph-based intelligent early warning method for project risks provided in an embodiment of this application. Figure 1 The method may include, but is not limited to, steps S101 to S105.

[0048] Step S101: Obtain a structured data table, construct a knowledge graph based on the structured data table, and filter out all activated nodes and corresponding relation edges in each stage from the knowledge graph to obtain multiple stage subgraphs.

[0049] Step S102: Input the stage subgraph into a preset graph neural network to obtain a risk representation vector, and perform a linear transformation on the risk representation vector to obtain a risk trend score. Define nodes with risk trend scores greater than the risk threshold as high-risk nodes; wherein, construct a propagation path set based on the upstream node paths of high-risk nodes.

[0050] Step S103: Select candidate strategies from the preset strategy set that cover at least one node on the propagation path set to obtain the candidate strategy set;

[0051] Step S104: Calculate the coverage based on the candidate policy set and the propagation path set to obtain the path coverage coefficient, obtain the position and timing of the policy action node in the path, obtain the response delay penalty term, and construct the policy structure influence weight based on the path coverage coefficient and the response delay penalty term.

[0052] Step S105: Each candidate strategy is scored according to the strategy structure influence weight, the preset risk suppression consistency term, the strategy execution cost, and the strategy saturation penalty term to obtain the strategy benefit score. The corresponding candidate strategy is selected as the recommended execution strategy based on the strategy benefit score.

[0053] Please see Figure 2 In some embodiments, step S101 may include, but is not limited to, steps S201 to S203:

[0054] Step S201: Obtain the structured data table, and select the node set and the relation edge set from the structured data table;

[0055] Step S202: Obtain the stage edge activation marker of each relation edge based on the node set, and obtain the stage position code of each task node based on the node set.

[0056] Step S203: Construct a knowledge graph based on the node set, relation edge set, stage edge activation marker, and stage position encoding.

[0057] In step S201 of some embodiments, the input data originates from structured data tables already existing in the project management platform (such as an engineering project management system, an information delivery platform, etc.), mainly including the following five categories:

[0058] Task Information Table: Includes fields such as Task Number, Task Name, Planned Start Time, Planned End Time, Responsible Person Number, Prerequisite Task Number, and Constraint Contract Number;

[0059] Contract Information Sheet: Includes contract number, signing date, payment terms, scope of tasks to be performed, and liability for breach of contract number, etc.

[0060] Personnel and Organization Table: Includes personnel ID, name, department ID, project role, and participation time range;

[0061] Project phase definition table: includes phase number, phase name, phase start and end time;

[0062] A table showing the correspondence between tasks and contracts, tasks and personnel, and personnel and organizations is used to generate connections between entities.

[0063] The above data was collected in a standard structured format through platform interfaces (such as structured query statements, file export, etc.). The recommended format is a comma-separated text file or a data table returned through a database query interface.

[0064] In the graph modeling process, five types of entities—tasks, contracts, personnel, organizations, and phases—are first mapped to nodes in the graph. Each type of node retains information fields with engineering semantics from the original data. For example, the task node includes fields such as: task number, planned start and end time, responsible person number, phase number, and current progress status. All nodes establish connections with phase nodes through the "phase" field.

[0065] Subsequently, based on the entity relationships defined in the structured data table, the relational edges of the graph are constructed, mainly including the following types: dependency relationships between tasks (e.g., task A starts after task B); execution relationships between tasks and personnel (e.g., a task is the responsibility of a certain person); constraint relationships between tasks and contracts (e.g., a task is subject to the delivery conditions of a certain contract); affiliation relationships between personnel and organizations (e.g., a person belongs to a certain department); and affiliation relationships between entities and stages (e.g., a task occurs in the design stage).

[0066] In summary, all nodes and relation edges are selected from the structured data table. All nodes form a node set, and all relation edges form a relation edge set. For each relation edge... Whether an edge is valid within a certain stage depends on the time field of the connected entities. If the valid time intervals of both nodes include a certain stage, then the edge is "valid" in that stage. This state is represented by the stage edge activation function as follows:

[0067] ;

[0068] in, For relation edges The phase edge activation marker indicates the relation edge. Is the stage t in an active state? and These are the two nodes connected by the edge; t is the number or identifier of a certain stage in the lifecycle, which comes from the stage number provided in the project stage table. The purpose of this function is to allow the selection of stage subgraphs in the subsequent graph model, using only the currently active relationships for information dissemination and risk reasoning.

[0069] In step S202 of some embodiments, the task node further includes a current stage number, a start stage number, and an end stage number. To further enhance the temporal semantics of the graph, a stage position code is introduced for all task nodes to express the stage position of a task within its complete lifecycle. Specifically, the calculation method is as follows: obtain the difference between the current stage number and the start stage number to obtain a first difference; obtain the difference between the end stage number and the start stage number to obtain a second difference; obtain the ratio of the first difference to the second difference to obtain the stage position code. As shown in the following formula:

[0070]

[0071] in, Represents task node The stage position code in the current stage t; t represents the current stage number; Represents task node The lifecycle start stage number; Represents a task node. The lifecycle termination stage number. If a task spans multiple stages, this value increments with each stage, representing the task's evolution over time. For example, a task planned for execution between stages 2 and 4 would have a stage position code of 0.5 at stage 3, indicating it is in the middle of its lifecycle.

[0072] knowledge graph The final structure includes a set of nodes, a set of relation edges, a stage edge activation flag for each edge, and a stage position code for each task node, all stored in a graph database format. It is recommended to use a graph database platform that supports heterogeneous entities. All nodes and edges contain a stage label field, supporting subsequent extraction of stage subgraphs based on stages, thus enabling stage-specific risk reasoning.

[0073] In summary, this step implements a multi-entity knowledge graph construction method centered on lifecycle stages, which can unify the originally scattered structured data in a project into a reasonable and analyzable graph structure. The innovation of this solution lies in: not only adding a time-activated mechanism to the graph layer, but also converting task nodes into stage-sensitive inputs through stage-normalized positions, laying the foundation for subsequent stage-based propagation of the model and risk identification. All data used is derived from commonly available fields in the project management system, ensuring feasibility for engineering implementation and reproducibility of the solution.

[0074] Through steps S201 to S203, a knowledge graph with "structural expressiveness, time-phase attributes, and usability for graph model computation" is constructed, serving as the input foundation for subsequent risk identification and strategy evaluation modules of this invention. Considering the typical characteristics of project management data—containing multiple entity categories such as tasks, contracts, organizations, and personnel—and the different semantic roles and risk values ​​exhibited by each entity at different stages of the project lifecycle, this step proposes a modeling method that organizes the graph around the "project lifecycle phase." This method integrates the entity layer (tasks, personnel, contracts), the relationship layer (dependency, affiliation, binding), and the phase time attributes to construct a graph structure with semantic computability and temporal schedulability. This step implements a multi-entity knowledge graph construction method centered on lifecycle phases, enabling the unified incorporation of previously scattered structured data from projects into a reasonable and analyzable graph structure. The innovation of this solution lies in: not only adding a time activation mechanism to the graph layer but also converting task nodes into phase-sensitive inputs through phase normalization positions, laying the foundation for subsequent phased propagation and risk identification of the model. All the data used comes from the regularly available fields in the project management system, ensuring the feasibility of project implementation and the reproducibility of the solution.

[0075] Furthermore, at each stage t, from the knowledge graph The process involves selecting all active nodes and their edges within the current stage to form a stage subgraph. ,Right now:

[0076]

[0077] This stage subgraph This is a dynamic reflection of the current stage structure. Considering that the project structure changes with each stage, all active nodes and their edges within each stage are selected separately, forming an independent stage subgraph for analysis. To identify risk trends in these structures, five-dimensional input features are constructed for each node, as follows:

[0078] Stage position coding ;

[0079] Node type encoding (task / contract / personnel);

[0080] Number of edges associated with a node (representing the connectivity of that node);

[0081] The number of contract terms associated with this node (reflecting risk exposure);

[0082] Stage number.

[0083] These features all originate from the structure fields and require no additional information processing.

[0084] In step S102 of some embodiments, then, in the stage subgraph A three-layer graph neural network is constructed for propagation and representation learning. To enhance the "causal path saliency" in risk structure identification, a new "path density adjustment factor" is introduced into the original graph convolutional propagation mechanism. This factor assigns higher weights to intermediate nodes with lower degrees (i.e., more likely to form bottlenecks) in the propagation path, thereby making the model focus on nodes that may become risk "bottlenecks." The propagation of each layer of the graph neural network is defined as follows:

[0085] ;

[0086] in, Represents a node In the Layer representation vector; Represents a node In the Layer representation vector; It is the set of its neighboring nodes; and They are nodes and The degree; For the first Layered learnable weights; For bias terms; This is the path density adjustment factor designed in this step, defined as:

[0087]

[0088] The initial design principle of this factor is that, in risk propagation scenarios, the smaller the degree of a node, the more likely it is to be a bottleneck or control node in the path, thus requiring greater attention weight in graph propagation. This design differs from the average propagation strategy of general graph neural networks and is more in line with the project risk characteristic that "local weak connections can easily lead to global failure."

[0089] After propagation through a three-layer graph, the final risk representation vector of the node is obtained. Then, a risk trend score is output through a linear transformation. :

[0090] ;

[0091] in, Let be the weight vector. This is the transpose of the weight vector. For bias terms, Represents a node The risk trend score at the current stage t is typically normalized to [0,1]. The higher the score, the more likely the node is to evolve into a high-risk node at the current stage due to structural coupling, time pressure, or dependency failure.

[0092] Nodes with risk trend scores greater than the risk threshold are defined as high-risk nodes. After identifying high-risk nodes, the upstream node paths are searched backwards from the graph structure, using the high-risk node as the endpoint, to generate a set of propagation paths that triggered the high-risk node. The path search algorithm uses a reverse shortest path plus structural density constraint method to retain only the set of paths with a length not exceeding L and an average node degree not exceeding a set threshold, ensuring that the propagation path has a clear structure and strong semantic interpretability.

[0093] The lifecycle knowledge graph constructed in the previous step Based on this, we identify key points in the project structure at the current stage t that may evolve into risk nodes, and deduce the structural path of risk formation. Compared with traditional risk identification methods that only score and predict based on single events such as task delays and contract defaults, this step emphasizes the joint driving force of structural propagation and time stages, and proposes a risk trend identification method that integrates stage-based graph convolutional propagation and structural attribution mechanisms, which has stronger structural interpretability and scenario adaptability.

[0094] In step S103 of some embodiments, a preset strategy set is introduced. Each strategy item *s* corresponds to a type of executable intervention action, such as "replacing the person responsible for the task," "adjusting the plan cycle," or "terminating the contract clause." These strategies are predefined by project managers and entered into the system in the form of a configuration table during the deployment phase. Each strategy has the following attributes: the type of node affected, a list of node numbers that can be affected, the expected intervention stage, the execution cost, and the maximum number of operations that can be performed. For example, strategy... It can be applied to all planned task nodes, and its scope of influence is the task number within the specified range. to The maximum number of times the strategy can be executed is 2. All strategy data is structured configuration data, requiring no real-time modeling.

[0095] To map the policy to the propagation path, policy pruning is first performed, that is, selecting a set of policies from the preset policy set S whose scope covers the propagation path. The candidate policy set is obtained by considering all candidate policies of at least one node. The path pruning process strictly adheres to structural connections and policy activation conditions, without introducing reasoning or generalization, ensuring a clear execution coupling between the policy and the propagation path.

[0096] In step S104 of some embodiments, two response capability indicators are introduced, namely the path coverage coefficient and the response delay penalty term, to quantify the coverage breadth and responsiveness of the strategy on the structural path.

[0097] The first metric is the path coverage coefficient. Measure the effect of strategy s on the propagation path The coverage degree of the upper node. Obtain the set of nodes acting on all candidate strategies, and the set of path nodes in the propagation path set; obtain the number of nodes that intersect between the set of acting nodes and the set of path nodes, thus obtaining the number of intersection nodes; calculate the path coverage coefficient by comparing the number of intersection nodes with the number of nodes in the path node set. The definition is as follows:

[0098] ;

[0099] in, This represents the set of nodes where candidate strategy s applies. Represents a node In the set of propagation paths at stage t, This represents the number of nodes in the set. express and There are nodes that intersect. The larger the value, the stronger the coverage of the strategy on the path, the more intervention it can have on more causal nodes, reflecting the structural coupling capability.

[0100] The second metric is the response delay penalty. This measures the position and timing of the policy-implementing nodes within the path. The distance between each intersection node and the high-risk node is obtained and summed to obtain the first distance; the distance between each path node and the high-risk node is obtained and summed to obtain the second distance; the ratio of the first and second distances is calculated to obtain the response latency penalty term. The definition is as follows:

[0101] ;

[0102] in, Represents nodes on the path With high-risk nodes The distance (i.e. from) arrive (shortest path length) Represents the path nodes in the propagation path set. With high-risk nodes distance. To avoid small constants with a denominator of zero (e.g., 0.01). The smaller the value, the closer the strategy's impact point is to the source of the path, allowing for earlier intervention in the risk formation process, reflecting proactive response and high value.

[0103] In obtaining and After considering the two indicators, further construct the strategy structure to influence the weights. A comprehensive reflection of the strategy's impact on the set of propagation paths The intensity of its effect. As shown in the following formula:

[0104] ;

[0105] in, This is an adjustment factor (typically 1-3) used to adjust the sensitivity of "path response delay" in the scoring. This formula fulfills the engineering design requirement that "optimization of response strategy must consider both structural coverage and time anteness," and also differs from traditional scoring mechanisms that rely solely on strategy effectiveness.

[0106] All strategies s establish response edges with their target nodes, forming a response subgraph with the propagation path as the main branch and the response strategies as the extensions. The edge attributes contain The value is used as an input variable in the subsequent optimization and sorting stage. The subgraph structure maintains consistency with the original graph structure and can be directly embedded into the graph query module to achieve graph structure maintenance.

[0107] In step S105 of some embodiments, a strategy benefit scoring function is constructed. This function scores each candidate strategy based on the strategy structure influence weight, a preset risk suppression consistency term, strategy execution cost, and strategy saturation penalty term, thus obtaining a strategy benefit score. This score reflects the impact of candidate strategy s on the node. The overall response value of the risk path is defined as follows:

[0108] ;

[0109] Indicates candidate strategy s for nodes The strategy effectiveness score. Indicated in the response subgraph The policy structure influence weights calculated from the edge attributes have taken into account the path coverage coefficient. With response delay item The combined effects of these factors. The weighting coefficient for strategy saturation penalty is adjustable based on project resource management constraints, and is typically set between 0.5 and 2.0. To avoid division by zero, use a small constant, such as 0.01.

[0110] This is a risk suppression consistency term, reflecting the risk score of the nodes covered by the strategy and the target node. The similarity is defined as follows:

[0111] ;

[0112] in, Let s be the set of nodes that affect strategy s. Represents a node The set of propagation paths at the current stage t. For nodes Risk trend score at the current stage t For nodes Risk trend score at stage t. The design idea of ​​this item is: if the risk score of the policy action node is related to... The convergence of values ​​indicates that they can work together to suppress the current risk trend; the closer the value is to 1, the more consistent the risk suppression targets are.

[0113] The cost of strategy execution can be expressed in units such as budget amount, time duration, or resource man-days, derived from predefined fields in the project management system. This is a strategy saturation penalty, representing the frequency with which the strategy has been called by the current system. It prevents the same strategy from being repeatedly called, which could lead to a decrease in marginal returns. It is calculated by obtaining the number of times a candidate strategy has been selected across all high-risk nodes; obtaining the maximum number of times a candidate strategy has been executed; and then calculating the strategy saturation penalty by dividing the number of candidate strategy executions by the maximum number of executions. Specifically:

[0114] ;

[0115] in, This represents the number of times candidate strategy s has been selected among all high-risk paths. This specifies the maximum number of times the strategy is allowed to execute (given in the strategy configuration table). This field reflects the limitations of project resource allocation and the concurrency constraints of the response strategy.

[0116] The design goal of this scoring function is to comprehensively evaluate the "return on investment" of a strategy by considering factors such as high structural coverage, high pre-response capability, high risk consistency, low cost, and low repeatability. The most innovative aspects include:

[0117] Integrating risk structure scoring with response timing;

[0118] Introduced as a policy saturation penalty term This reflects the project's true "resource exclusivity";

[0119] use Describe risk matching and enhance the coordination of objectives between paths and strategies;

[0120] All parameters can be directly obtained from the previous stage's structure or project configuration, ensuring a stable implementation path.

[0121] In all candidate strategies Finish After calculation, the strategies are sorted from highest to lowest effectiveness score. The candidate strategies corresponding to the top k strategy effectiveness scores are selected as recommended execution strategies, or the candidate strategies corresponding to strategy effectiveness scores greater than the score threshold are selected as recommended execution strategies. This forms a sorted sequence. The sorting results can be directly input into the project management and control system for human-machine collaborative decision-making or automated execution processes.

[0122] Steps S101 to S105 of this embodiment involve obtaining a structured data table, constructing a knowledge graph based on the structured data table, and filtering out all activated nodes and corresponding relational edges within each stage from the knowledge graph to obtain multiple stage subgraphs. The stage subgraphs are input into a preset graph neural network to obtain a risk representation vector, and a linear transformation is performed on the risk representation vector to obtain a risk trend score. Nodes with risk trend scores greater than a risk threshold are defined as high-risk nodes. A propagation path set is constructed based on the upstream node paths of high-risk nodes. Candidate strategies whose scope covers at least one node in the propagation path set are selected from a preset strategy set to obtain a candidate strategy set. Coverage is calculated based on the candidate strategy set and the propagation path set to obtain a path coverage coefficient. The position and timing of the strategy's active node in the path are obtained to obtain a response delay penalty term. Each candidate strategy is scored based on the path coverage coefficient, response delay penalty term, preset risk suppression consistency term, strategy execution cost, and strategy saturation penalty term to obtain a strategy benefit score. The corresponding candidate strategy is selected as the recommended execution strategy based on the strategy benefit score, thus achieving accurate identification of project risks and intervention in project risks.

[0123] Please see Figure 3This application also provides a knowledge graph-based intelligent early warning system for project risks, which can implement the above-mentioned knowledge graph-based intelligent early warning method for project risks. The system includes:

[0124] The acquisition unit 301 is used to acquire a structured data table, construct a knowledge graph based on the structured data table, and filter out all activated nodes and corresponding relation edges in each stage from the knowledge graph to obtain multiple stage subgraphs.

[0125] The construction unit 302 is used to input the stage subgraph into a preset graph neural network to obtain a risk representation vector, and to perform a linear transformation on the risk representation vector to obtain a risk trend score. Nodes with a risk trend score greater than a risk threshold are defined as high-risk nodes. A set of propagation paths is constructed based on the upstream node paths of high-risk nodes.

[0126] The filtering unit 303 is used to filter out candidate strategies whose scope covers at least one node on the propagation path set from the preset strategy set, and obtain a candidate strategy set.

[0127] The calculation unit 304 is used to calculate the coverage based on the candidate policy set and the propagation path set, obtain the path coverage coefficient, obtain the position and timing of the policy action node in the path, obtain the response delay penalty term, and construct the policy structure influence weight based on the path coverage coefficient and the response delay penalty term.

[0128] The scoring unit 305 is used to score each candidate strategy based on the strategy structure influence weight, the preset risk suppression consistency item, the strategy execution cost, and the strategy saturation penalty item to obtain the strategy benefit score. Based on the strategy benefit score, the corresponding candidate strategy is selected as the recommended execution strategy.

[0129] The specific implementation of this knowledge graph-based intelligent early warning system for project risks is basically the same as the specific implementation of the knowledge graph-based intelligent early warning method for project risks described above, and will not be repeated here.

[0130] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A knowledge graph-based intelligent early warning method for project risks, characterized in that, The method includes: Obtain a structured data table, construct a knowledge graph based on the structured data table, and filter out all activated nodes and corresponding relation edges in each stage from the knowledge graph to obtain multiple stage subgraphs; The stage subgraph is input into a preset graph neural network to obtain a risk representation vector. The risk representation vector is then linearly transformed to obtain a risk trend score. Nodes with risk trend scores greater than a risk threshold are defined as high-risk nodes. A propagation path set is constructed based on the upstream node paths of the high-risk nodes. Candidate strategies whose scope covers at least one node on the propagation path set are selected from the preset strategy set to obtain a candidate strategy set; Coverage is calculated based on the candidate policy set and the propagation path set to obtain the path coverage coefficient. The position and timing of the policy-implementing nodes in the path are obtained, and a response latency penalty term is derived. The policy structure influence weights are constructed based on the path coverage coefficient and the response latency penalty term. The path coverage coefficient calculation based on the candidate policy set and the propagation path set includes: Obtain the set of nodes that affect all candidate strategies, and obtain the set of path nodes in the propagation path set; The number of nodes that intersect in the set of active nodes and the set of path nodes is obtained to get the number of intersecting nodes; The path coverage coefficient is obtained by calculating the ratio between the number of intersection nodes and the number of nodes in the path node set; the position and timing of the policy-implementing nodes in the path are obtained to obtain the response latency penalty term, including: Obtain the distance between each intersection node and the high-risk node, and add them together to obtain the first distance; Obtain the distance between each path node and the high-risk node, and add them together to obtain the second distance; The response delay penalty term is obtained by calculating the ratio between the first distance and the second distance; Each candidate strategy is scored based on the strategy structure influence weight, the preset risk suppression consistency term, the strategy execution cost, and the strategy saturation penalty term to obtain a strategy benefit score. The corresponding candidate strategy is selected as the recommended execution strategy based on the strategy benefit score.

2. The intelligent early warning method for project risks based on knowledge graphs according to claim 1, characterized in that, The step of obtaining the structured data table and constructing a knowledge graph based on the structured data table includes: Obtain a structured data table, and select a set of nodes and a set of relation edges from the structured data table; The stage edge activation marker of each relation edge is obtained based on the node set, and the stage position code of each task node is obtained based on the node set. The knowledge graph is constructed based on the set of nodes, the set of relation edges, the stage edge activation markers, and the stage position codes.

3. The intelligent early warning method for project risks based on knowledge graphs according to claim 2, characterized in that, The node set includes multiple task nodes, each task node including a current stage number, a start stage number, and an end stage number. Obtaining the stage position code for each task node based on the node set includes: Obtain the difference between the current stage number and the starting stage number to get the first difference; Obtain the difference between the termination stage number and the starting stage number to get the second difference; The ratio of the first difference to the second difference is obtained to obtain the stage position code.

4. The intelligent early warning method for project risks based on knowledge graphs according to claim 1, characterized in that, The step of selecting the corresponding candidate strategy as the recommended execution strategy based on the strategy effectiveness score includes: Sort the strategy benefit scores from largest to smallest, and select the candidate strategies corresponding to the top k strategy benefit scores as the recommended execution strategies; or select the candidate strategies corresponding to strategy benefit scores greater than the score threshold as the recommended execution strategies.

5. The intelligent early warning method for project risks based on knowledge graphs according to claim 1, characterized in that, The method further includes: obtaining the policy saturation penalty term, specifically including: The number of times the candidate strategy has been selected in all the high-risk nodes is obtained to obtain the number of candidate strategies; Obtain the maximum number of times the candidate strategy can be executed; The strategy saturation penalty term is obtained by calculating the ratio of the number of times the candidate strategy is executed to the maximum number of times.

6. A project risk intelligent early warning system based on knowledge graphs, characterized in that: The system includes: The acquisition unit is used to acquire a structured data table, construct a knowledge graph based on the structured data table, and filter out all activated nodes and corresponding relation edges in each stage from the knowledge graph to obtain multiple stage subgraphs. The construction unit is used to input the stage subgraph into a preset graph neural network to obtain a risk representation vector, and to perform a linear transformation on the risk representation vector to obtain a risk trend score. Nodes with risk trend scores greater than a risk threshold are defined as high-risk nodes. A propagation path set is constructed based on the upstream node paths of the high-risk nodes. The filtering unit is used to filter out candidate strategies whose scope covers at least one node on the propagation path set from the preset strategy set, thereby obtaining a candidate strategy set. The calculation unit is configured to perform coverage calculation based on the candidate policy set and the propagation path set to obtain a path coverage coefficient, acquire the position and timing of the policy-implementing node in the path, obtain a response latency penalty term, and construct a policy structure influence weight based on the path coverage coefficient and the response latency penalty term; wherein, the coverage calculation based on the candidate policy set and the propagation path set to obtain the path coverage coefficient includes: Obtain the set of nodes that affect all candidate strategies, and obtain the set of path nodes in the propagation path set; The number of nodes that intersect in the set of active nodes and the set of path nodes is obtained to get the number of intersecting nodes; The path coverage coefficient is obtained by calculating the ratio between the number of intersection nodes and the number of nodes in the path node set; the position and timing of the policy-implementing nodes in the path are obtained to obtain the response latency penalty term, including: Obtain the distance between each intersection node and the high-risk node, and add them together to obtain the first distance; Obtain the distance between each path node and the high-risk node, and add them together to obtain the second distance; The response delay penalty term is obtained by calculating the ratio between the first distance and the second distance; The scoring unit is used to score each candidate strategy based on the strategy structure influence weight, a preset risk suppression consistency term, strategy execution cost, and strategy saturation penalty term to obtain a strategy benefit score, and select the corresponding candidate strategy as the recommended execution strategy based on the strategy benefit score.

7. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the knowledge graph-based intelligent early warning method for project risks as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the knowledge graph-based intelligent early warning method for project risks as described in any one of claims 1 to 5.

Images