A multi-diagram collaborative reasoning method and system for engineering scenarios

Abstract

The present application relates to a kind of multi-diagram collaborative reasoning method and system for engineering scene, wherein the method is mapped to the same entity shared semantic space by semantic embedding algorithm to three kinds of engineering subgraph nodes, construct bidirectional semantic association path and be associated as initial engineering knowledge graph for multi-diagram collaborative reasoning, execute multi-diagram collaborative forward reasoning, traverse from process node to risk node along bidirectional semantic association path, process node is used as condition variable, and condition risk probability reasoning propagation is carried out to risk node, and multi-diagram collaborative reverse update reasoning is executed to the optimal process node screened out, realize the reasoning reconstruction of process topology graph, re-perform condition risk probability reasoning propagation, cyclic iteration to realize the adaptive linkage update of process topology graph and risk propagation graph, complete collaborative reasoning evolution to initial engineering knowledge graph, output final engineering knowledge graph. Thus, the present application realizes multi-diagram collaborative reasoning for engineering scene.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a multi-graph collaborative reasoning method and system for engineering scenarios. Background Technology

[0002] With the continuous improvement of digitalization in engineering construction, the review of construction plans, risk assessment, and process decisions are gradually shifting from reliance on manual experience to data-driven approaches. In recent years, knowledge graph technology has been gradually introduced into the engineering field for scenarios such as construction process management, safety risk identification, and process solution recommendation.

[0003] However, existing knowledge graph technologies in the engineering field have the following shortcomings: (1) The existing solution constructs three independent graph structures for the three types of information: process, risk and technology. It only performs simple result summarization at the output layer of the graph structure and fails to achieve unified representation and fusion of the three heterogeneous graph structures at the data level. It cannot support multi-graph collaborative reasoning for engineering scenarios.

[0004] (2) In existing solutions, once the graph structure is constructed, its node set, edge set, and topological connection relationship are fixed. Subsequent processing only performs one-way path queries on the fixed graph structure, and the processing results only output the query conclusions without feedback to the graph structure itself. For example, the AI-driven construction drawing compliance review system and method with announcement number CN120671228B constructs a construction process specification map and uses a simulation engine to compare the component behavior with the specification map in real time to mark violations. Its processing direction is a one-way comparison from construction drawings to review results, and the processing results are only used to output review suggestions and cannot reverse the graph structure. Also, the multimodal fusion-driven interactive visualization briefing system for construction processes with announcement number CN120428886A introduces reinforcement learning algorithms to optimize the construction script generation strategy, but its optimization only applies to the strategy parameter level, and the node composition, topological structure, and associated edges of the graph remain static. None of the above solutions have the ability to iteratively update the graph structure itself based on the processing results. Once the graph structure is constructed, it is fixed and cannot adaptively adjust with the dynamic changes of the application scenario.

[0005] (3) In existing risk propagation graphs or risk knowledge graphs, the risk propagation probability is a predefined static rule and is decoupled from the construction process. Although some methods can establish the relationship between the process and the technology, such as CN120671228B mentioned above, the risk propagation probability itself does not change with the choice of technology, and it cannot adapt to the differentiated risk evolution under different technologies.

[0006] Therefore, a novel multi-graph collaborative reasoning scheme for engineering scenarios is proposed, which can integrate three types of heterogeneous engineering graph structures into a unified graph data structure at the data level. It supports graph traversal and dynamic calculation of edge weights with condition variables as parameters during the reasoning process, and realizes iterative feedback and dynamic updates of the graph structure itself by the output of graph structure traversal results. Summary of the Invention

[0007] The technical problem to be solved by this invention is: This invention provides a multi-graph collaborative reasoning method and system for engineering scenarios, which supports conditional risk probability reasoning propagation and dynamic calculation of edge weights with conditional variables as parameters in the process of multi-graph collaborative forward reasoning, and realizes closed-loop iteration of multi-graph collaborative forward reasoning and multi-graph collaborative reverse update reasoning, thus completing the collaborative reasoning evolution of engineering knowledge graphs.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides a multi-graph collaborative reasoning method for engineering scenarios, comprising: Multi-source heterogeneous data from engineering scenarios are collected, and semantic parsing is performed on the multi-source heterogeneous data to generate three types of engineering subgraphs. The nodes of the three types of engineering subgraphs are uniformly mapped to the same entity shared semantic space through a semantic embedding algorithm. Bidirectional semantic association paths are constructed between nodes of different types of engineering subgraphs and nodes of the same entity, so that the three types of engineering subgraphs are associated based on bidirectional semantic association paths, generating an initial engineering knowledge graph. The three types of engineering subgraphs are process topology graph, risk propagation graph, and process relationship graph. A forward graph structure traversal is performed on the engineering knowledge graph. Starting from the process node of the process topology graph, the traversal is performed along the bidirectional semantic association path to the risk node of the risk propagation graph. The process nodes in the process relationship graph are used as condition variables to perform conditional risk probability reasoning propagation on the risk nodes to obtain the risk reasoning feature vector corresponding to each process node. The risk reasoning feature vector corresponding to each process node is then matched with the risk suppression feature vector pre-configured for the corresponding process node to select the optimal process node. A reverse graph structure update is performed on the optimal process node. The process decomposition template corresponding to the optimal process node is obtained from the process relationship graph. The process topology graph is reconstructed based on the process decomposition template to obtain the reconstructed process topology graph. The reconstructed process topology graph is then re-inferred and traversed to the risk node of the risk propagation graph through the bidirectional semantic association path. Conditional risk probability inference propagation is performed again and conditional risk inference verification is executed. The forward cross-graph inference and the reverse cross-graph update are iterated repeatedly until the optimal process node passes the conditional risk inference verification, thus completing the graph structure iteration. This achieves adaptive linkage update of the process topology graph and the risk propagation graph, generating the final engineering knowledge graph.

[0009] The beneficial effects of this invention are as follows: It maps nodes of three types of engineering subgraphs to a shared semantic space of the same entity, constructs bidirectional semantic association paths between nodes of different types of subgraphs, and achieves the fusion and unified representation of process topology graphs, risk propagation graphs, and process relationship graphs at the data level. This transforms the three types of graph structures from a mutually isolated state into a joint graph structure that can be traversed across graphs via bidirectional association paths, providing a unified data foundation for multi-graph collaborative reasoning in engineering scenarios. In multi-graph collaborative forward reasoning, process nodes are used as switchable condition variables, allowing the risk propagation graph to dynamically calculate the conditional risk propagation probability of each link based on the current condition variable during each graph reasoning traversal. The same risk propagation graph exhibits differentiated edge weight distributions under different process conditions, overcoming the limitations of traditional static edge weights. The process template corresponding to the optimal process node selected by the multi-graph collaborative forward reasoning traversal is used as a feedback signal to adjust the process topology graph in reverse. The reconstructed process topology graph is then input again for forward reasoning, achieving closed-loop iteration of multi-graph collaborative reasoning. This enables the graph structure of the knowledge graph to dynamically evolve with the results of multi-graph collaborative reasoning, completing the collaborative reasoning evolution of the initial engineering knowledge graph.

[0010] Optionally, the multi-source heterogeneous data includes construction procedure data, construction risk data, and construction technology data. The step of semantically parsing the multi-source heterogeneous data to generate three types of engineering sub-graphs includes: Semantic parsing is performed on the construction process data to identify the process set and the temporal dependencies between different processes in the process set. Each process in the process set is treated as an independent process node, and the temporal dependencies between different processes are treated as directed dependency edges between different process nodes to construct a process topology graph. Semantic parsing is performed on construction risk data to identify the set of risk factors, the triggering conditions of each risk factor in the set of risk factors, and the propagation relationship between different risk factors. Each risk factor in the set of risk factors is treated as an independent risk node. Based on the triggering conditions of each risk factor and the propagation relationship between different risk factors, a risk causal topology link and a risk propagation link between risk factors are generated. The conditional probability with process as the conditional variable is used as the risk propagation probability of the risk propagation link to construct a risk propagation graph. Semantic parsing is performed on the construction process data to identify the process set and the substitution, inclusion, and constraint relationships between different processes within the process set. Each process in the process set is treated as an independent process node, and a unique process decomposition template and risk suppression feature vector are configured for each process node. The substitution, inclusion, and constraint relationships between different processes are used as process relationship edges between different processes to construct a process relationship graph. The process decomposition template is the set of processes that should be included when using the corresponding process and the temporal dependencies between processes in the process set. The risk suppression feature vector is the ability to suppress each risk factor in the risk factor set when using the corresponding process.

[0011] As described above, each process is treated as an independent process node, and the temporal dependencies between different processes are used as directed dependency edges to ensure that the constructed process topology graph can accurately reproduce the topological relationships of the construction process. Each risk factor is treated as an independent risk node, and risk causal topology links and risk propagation links are automatically generated based on the triggering conditions and propagation relationships of the risk factors. At the same time, the conditional probability with the process as a conditional variable is introduced as the risk propagation probability of the risk propagation link, so that the risk propagation path and risk propagation probability can dynamically change with the process selection, adapting to differentiated process selections and improving the authenticity and accuracy of dynamic construction risk propagation in engineering scenarios. Each process is treated as an independent process node, and each process node is configured with a dedicated process decomposition template and risk suppression feature vector, so that the process node not only has the correlation between processes, but also carries process adaptation and risk adaptation information, providing support for subsequent process selection, process topology graph reasoning and reconstruction, and conditional risk reasoning verification, ensuring the professionalism and adaptability of subsequent multi-graph collaborative forward reasoning and multi-graph collaborative reverse update reasoning.

[0012] Optionally, the step of uniformly mapping the nodes of the three types of engineering subgraphs to the same entity shared semantic space through a semantic embedding algorithm, constructing bidirectional semantic association paths between nodes of different types of engineering subgraphs and shared nodes of the same entity, so that the three types of engineering subgraphs are associated based on bidirectional semantic association paths, and generating an initial engineering knowledge graph includes: The nodes of the three types of engineering subgraphs are uniformly mapped to the same entity shared semantic space through a semantic embedding algorithm. In the same entity shared semantic space, nodes from different types of engineering subgraphs that point to the same engineering object are anchored to the same entity shared node, and multi-dimensional semantic information is given to the entity shared node to obtain a set of entity shared nodes carrying multi-dimensional semantic information. The multi-dimensional semantic information is the semantic information corresponding to the nodes of different types of engineering subgraphs. The entity shared nodes in the entity shared node set are assigned roles. If the multi-dimensional semantic information carried by the entity shared node is the semantic information of the process node from the process topology diagram and the semantic information of the risk node from the risk propagation diagram, then the entity shared node carrying the multi-dimensional semantic information is marked as a first role shared node. If the multi-dimensional semantic information carried is the semantic information of the process node from the process topology diagram and the semantic information of the process node from the process relationship diagram, then the entity shared node carrying the multi-dimensional semantic information is marked as a second role shared node. If the multi-dimensional semantic information carried is the semantic information of the risk node from the risk propagation diagram and the semantic information of the process node from the process relationship diagram, then the entity shared node carrying the multi-dimensional semantic information is marked as a third role shared node. Based on the first role-shared node, a first bidirectional semantic association path is established between the process topology diagram and the risk propagation diagram, and a first inference traversal weight of the first bidirectional semantic association path is generated through a first association strength matrix. Based on the second role-shared node, a second bidirectional semantic association path is established between the process topology diagram and the process relationship diagram, and a second inference traversal weight of the second bidirectional semantic association path is generated through a second association strength matrix. Based on the third role-shared node, a third bidirectional semantic association path is established between the risk propagation diagram and the process relationship diagram, and a third inference traversal weight of the third bidirectional cross-graph mapping is generated through a third association strength matrix. The process topology graph, the risk propagation graph, and the process relationship graph are associated through the first bidirectional semantic association path, the second bidirectional semantic association path, the third bidirectional semantic association path, the first reasoning traversal weight, the second reasoning traversal weight, and the third reasoning traversal weight to generate an initial engineering knowledge graph.

[0013] As described above, a semantic embedding algorithm maps nodes from different types of engineering subgraphs to the same entity's shared semantic space, anchoring nodes pointing to the same engineering object to the same entity's shared node. This enables unified representation of different engineering subgraphs in the vector space, overcoming the problem of semantic fragmentation in multi-source heterogeneous data. By assigning three differentiated role labels to entity-shared nodes, the cross-graph semantic attributes of different entity-shared nodes are accurately distinguished. Based on these role labels, three sets of bidirectional semantic association paths are constructed, and corresponding inference traversal weights are generated using association strength matrices. This elevates cross-graph associations from static logical pointing to weighted path construction at the graph data structure level. Thus, the process topology diagram, risk propagation diagram, and process relationships are... Figure 3 Subgraphs are associated as a unified engineering knowledge graph in the same semantic space, providing a complete data foundation for subsequent graph structure traversal and iterative updates.

[0014] Optionally, the step of starting from the process node in the process topology graph, traversing along the bidirectional semantic association path to the risk node in the risk propagation graph, using the process nodes in the process relationship graph as condition variables to perform conditional risk probability reasoning propagation on the risk nodes, obtaining the risk reasoning feature vector corresponding to each process node, and performing reasoning matching between the risk reasoning feature vector corresponding to each process node and the pre-configured risk suppression feature vector of the corresponding process node to select the optimal process node includes: The current construction stage of the project scenario is collected, the process node corresponding to the current construction stage is extracted from the process topology diagram as the activated process node, and the current execution status of the activated process node is obtained. The activated process node is reasoned and traversed along the bidirectional semantic association path to the risk node of the risk propagation graph to obtain the activated risk node triggered by the current execution state, and the activated risk propagation link triggered from the activated risk node is obtained based on the risk causal topology link of the risk propagation graph. The process nodes in the process relationship diagram are used as condition variables for the activated risk nodes. Dynamic risk reasoning propagation is performed on the activated risk propagation link to obtain the risk reasoning feature vector corresponding to each process node. Calculate the cosine similarity between the risk inference feature vector corresponding to each process node and the risk suppression feature vector pre-configured for the corresponding process node, and select the process node with the highest cosine similarity as the optimal process node.

[0015] As described above, by extracting activated process nodes and their current execution status based on the current construction stage, forward cross-graph reasoning is bound to the real-time construction scenario, ensuring its real-time performance and accuracy. Based on bidirectional semantic association paths and risk causal topology links, activated risk nodes triggered by the current execution status and activated risk propagation links triggered from these nodes are obtained. This allows subsequent dynamic risk propagation calculations to be based on this, avoiding issues of low global computational efficiency and excessive redundancy. By using process nodes as condition variables for dynamic risk propagation, the different risk reasoning feature vectors under different process nodes are reflected, achieving dynamic coupling between risk propagation and process. The process node with the highest cosine similarity between the risk reasoning feature vector and the risk suppression feature vector is selected as the optimal process node, ensuring the rationality of the optimal process node selection.

[0016] Optionally, the step of reasoning and reconstructing the process topology diagram based on the process decomposition template includes: The process decomposition template is compared with the process topology diagram to determine whether there are process nodes that exist in the process decomposition template but not in the process topology diagram. If there are, the corresponding process node is taken as a missing process node. The temporal dependency relationship of the missing process node is obtained from the process decomposition template. Based on the temporal dependency relationship of the missing process node, the missing process node is added to the process topology diagram to realize the reasoning reconstruction of the process topology diagram. Determine whether there are process nodes that exist in the process topology diagram but not in the process decomposition template. If so, the corresponding process node is regarded as a redundant process node and suspended in the process topology diagram to realize the reasoning reconstruction of the process topology diagram.

[0017] As described above, by comparing the process decomposition template with the process topology diagram, missing process nodes are added to the process topology diagram, enabling the reconstructed process topology diagram to adapt to the process requirements of the optimal process node. Redundant process nodes are suspended rather than deleted, ensuring that the current process topology diagram adapts to the optimal process node, eliminating interference from invalid process nodes, and supporting the restoration and adjustment of the process topology diagram for subsequent iterations of the optimal process node. This avoids the computational overhead caused by repeated deletion and reconstruction, and balances the adaptability and flexibility of iterative reconstruction of the process topology diagram.

[0018] Optionally, the step of re-inferring the reconstructed process topology graph to the risk nodes of the risk propagation graph through the bidirectional semantic association path, re-performing conditional risk probability reasoning propagation and performing conditional risk reasoning verification, and iteratively iterating the forward cross-graph reasoning and the reverse cross-graph update until the optimal process node passes the conditional risk reasoning verification, thereby achieving adaptive linkage update of the process topology graph and the risk propagation graph, includes: The process topology graph after reasoning reconstruction is re-inferred and traversed to the risk node of the risk propagation graph through the bidirectional semantic association path. The optimal process node is used as the condition variable to re-perform conditional risk probability reasoning propagation to obtain the optimal risk reasoning feature vector corresponding to the optimal process node. The optimal risk inference feature vector corresponding to the optimal process node is matched with the pre-configured risk suppression features to obtain the corresponding inference matching degree. The inference matching degree is compared with a preset adaptation threshold to determine whether the inference matching degree is lower than the preset adaptation threshold. If so, the conditional risk inference verification fails, triggering the next round of forward cross-graph inference and reverse cross-graph update until the re-selected optimal process node passes the conditional risk inference verification. If not, the conditional risk inference verification succeeds, and the reconstructed process topology graph is used as the updated process topology graph. At the same time, the risk propagation link of the risk propagation graph is updated synchronously according to the optimal risk inference feature vector corresponding to the optimal process node, realizing adaptive linkage update of the process topology graph and the risk propagation graph.

[0019] As described above, conditional risk probability reasoning propagation is re-performed based on the reconstructed process topology graph, generating the optimal risk reasoning feature vector corresponding to the optimal process node. This achieves real-time linkage between process topology graph reconstruction and risk propagation graph, ensuring that the obtained optimal risk reasoning feature vector always matches the latest optimal process node. The optimal risk reasoning feature vector is then matched with pre-configured risk suppression features. Conditional risk reasoning verification is performed based on the reasoning matching degree. If the conditional risk reasoning verification fails, the next round of forward cross-graph reasoning and reverse cross-graph update is triggered until the optimal process node passes the conditional risk reasoning verification. This achieves adaptive linkage update of the process topology graph and the risk propagation graph, that is, realizing the transformation of the engineering knowledge graph from static modeling to dynamic self-evolution, and improving the adaptability of the engineering knowledge graph.

[0020] Secondly, the present invention provides an engineering knowledge graph construction system for multi-graph fusion bidirectional reasoning, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the multi-graph collaborative reasoning method for engineering scenarios described in the first aspect.

[0021] The technical effects of the engineering knowledge graph construction system with multi-graph fusion and bidirectional reasoning provided in the second aspect are the same as those of the multi-graph collaborative reasoning method for engineering scenarios provided in the first aspect. Attached Figure Description

[0022] Figure 1 This is a flowchart of a multi-graph collaborative reasoning method for engineering scenarios provided in this embodiment; Figure 2 This is a schematic diagram illustrating the construction and fusion process of the three types of engineering sub-graphs involved in this embodiment; Figure 3 This is a schematic diagram illustrating the process of performing forward cross-graph reasoning and reverse cross-graph updating on the engineering knowledge graph involved in this embodiment; Figure 4 This is a schematic diagram of the structure of an engineering knowledge graph construction system for multi-graph fusion and bidirectional reasoning provided in this embodiment.

[0023] [Explanation of Labels in the Attached Image] 1. An engineering knowledge graph construction system with multi-graph fusion and bidirectional reasoning; 2. Processor; 3. Memory. Detailed Implementation

[0024] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present invention can be understood more clearly and thoroughly, and that the scope of the present invention can be fully conveyed to those skilled in the art.

[0025] Example 1 Please refer to Figures 1 to 3 This invention provides a multi-graph collaborative reasoning method for engineering scenarios, comprising the following steps: S1. Collect multi-source heterogeneous data of engineering scenarios, perform semantic parsing on the multi-source heterogeneous data, generate three types of engineering subgraphs, map the nodes of the three types of engineering subgraphs to the same entity shared semantic space through a semantic embedding algorithm, construct bidirectional semantic association paths between nodes of different types of engineering subgraphs and shared nodes of the same entity, so that the three types of engineering subgraphs are associated based on bidirectional semantic association paths, and generate an initial engineering knowledge graph. The three types of engineering subgraphs are process topology graph, risk propagation graph and process relationship graph. In this embodiment, as Figure 2As shown, multi-source heterogeneous data from engineering scenarios are collected. Semantic parsing is performed on this data to generate three types of engineering subgraphs. The multi-source heterogeneous data includes construction procedure data, construction risk data, and construction technology data. Construction procedure data originates from construction organization designs, construction plans, and construction schedule documents; construction risk data comes from construction safety risk assessment reports, historical construction accident records, and engineering geological survey reports; and construction technology data comes from construction technical specifications and construction manuals. The nodes of the three types of engineering subgraphs are uniformly mapped to the same entity-shared semantic space using a semantic embedding algorithm. This constructs bidirectional semantic association paths between nodes of different types of engineering subgraphs and shared nodes of the same entity, enabling the three types of engineering subgraphs to be associated based on these bidirectional semantic association paths, generating an initial engineering knowledge graph. The entity-shared semantic space refers to a high-dimensional vector space that uniformly represents nodes from different source graph structures.

[0026] At this point, the multi-source heterogeneous data mentioned in step S1 includes construction procedure data, construction risk data, and construction technology data. The step of semantically parsing the multi-source heterogeneous data to generate three types of engineering sub-graphs includes: S11. Perform semantic parsing on the construction process data to identify the process set and the temporal dependencies between different processes in the process set. Treat each process in the process set as an independent process node and the temporal dependencies between different processes as directed dependency edges between different process nodes to construct a process topology graph. S12. Perform semantic parsing on the construction risk data to identify the risk factor set, the triggering conditions of each risk factor in the risk factor set, and the propagation relationship between different risk factors. Treat each risk factor in the risk factor set as an independent risk node. Generate risk causal topology links and risk propagation links between risk factors based on the triggering conditions of each risk factor and the propagation relationship between different risk factors. Use the conditional probability with process as the conditional variable as the risk propagation probability of the risk propagation link to construct a risk propagation graph. S13. Perform semantic parsing on the construction process data to identify the process set and the substitution, inclusion, and constraint relationships between different processes in the process set. Treat each process in the process set as an independent process node, and configure a unique process decomposition template and risk suppression feature vector for each process node. Treat the substitution, inclusion, and constraint relationships between different processes as process relationship edges between different processes to construct a process relationship graph. The process decomposition template is the set of processes that should be included when using the corresponding process and the temporal dependency relationship between processes in the process set. The risk suppression feature vector is the ability to suppress each risk factor in the risk factor set when using the corresponding process.

[0027] In this embodiment, as Figure 2 As shown, semantic parsing is performed on the construction process data to identify the process set and the temporal dependencies between different processes within the process set. These temporal dependencies include sequential dependencies, such as process A: dewatering construction, which must be completed before process B: earthwork excavation; and parallel dependencies, such as process C: rebar processing and process D: formwork fabrication, which can be performed simultaneously. Each process in the process set is treated as an independent process node, and the temporal dependencies between different processes are represented as directed dependency edges between these process nodes to construct a process topology graph.

[0028] Semantic analysis is performed on construction risk data to identify a set of risk factors, the triggering conditions for each risk factor within the risk set, and the propagation relationships between different risk factors. Risk factors include: pit collapse, trench wall collapse, piping, surface settlement, support structure instability, and pile foundation misalignment. Triggering conditions refer to the direct causes that trigger a risk factor, such as rising groundwater levels triggering piping, or excessive axial force in the support structure triggering instability. Propagation relationships refer to the transmission paths between risk factors. Each risk factor in the risk factor set is treated as an independent risk node. Based on the triggering conditions of each risk factor and the propagation relationships between different risk factors, a causal topological link and a risk propagation link are generated between risk factors. The conditional probability, with the process as the conditional variable, is used as the risk propagation probability of the risk propagation link. Therefore, the risk propagation link is a computable path with a risk propagation probability assigned to it based on the causal topological link. Furthermore, the risk propagation probability of the same risk propagation link differs under different processes, thus constructing a risk propagation graph. For example, the risk propagation chain is: rising groundwater level → piping burst → foundation pit collapse. When the groundwater level rises, the risk propagation probability of piping burst is 'a' when using slope excavation technology, and 'b' when using diaphragm wall technology. That is, the risk propagation probability changes dynamically with the technology used. The conditional probability can be expressed as: ; in, Indicates risk factors That is: risk nodes , Indicates risk factors That is: risk nodes , Indicating process .

[0029] Semantic parsing is performed on construction technology data to identify the process set and the substitution, inclusion, and constraint relationships between different processes within the process set. Each process in the process set is treated as an independent process node, and a unique process decomposition template and risk suppression feature vector are configured for each process node. The process decomposition template consists of the set of processes that should be included when using the corresponding process and the temporal dependencies between processes in the process set. For example, when using the diaphragm wall process, the corresponding process decomposition template is: guide wall construction → trenching construction → rebar cage hoisting → concrete pouring. When using the slope excavation process, the corresponding process decomposition template is: site leveling → layered excavation → slope trimming → drainage construction. The risk suppression feature vector represents the ability of a given process to suppress each risk factor in the risk factor set. The specific suppression capability is assigned based on the physical characteristics of the process, its construction principles, historical engineering data, or expert evaluation. For example, diaphragm walls act as water-cutting curtains, effectively blocking groundwater seepage; therefore, they have a higher suppression capability against piping and sudden inrush than slope excavation. As a rigid support structure, diaphragm walls also have a higher suppression capability against foundation pit collapse than slope excavation. The substitution, inclusion, and constraint relationships between different processes are used as process relationship edges to construct a process relationship diagram.

[0030] At this point, step S1 involves mapping the nodes of the three types of engineering subgraphs to the same entity's shared semantic space using a semantic embedding algorithm, constructing bidirectional semantic association paths between nodes of different types of engineering subgraphs and nodes shared by the same entity, enabling the three types of engineering subgraphs to be associated based on these bidirectional semantic association paths, and generating an initial engineering knowledge graph, including: S14. Map the nodes of the three types of engineering subgraphs to the same entity shared semantic space through a semantic embedding algorithm. In the same entity shared semantic space, anchor the nodes from different types of engineering subgraphs that point to the same engineering object to the same entity shared node, and assign multi-dimensional semantic information to the entity shared node to obtain a set of entity shared nodes carrying multi-dimensional semantic information. The multi-dimensional semantic information is the semantic information corresponding to the nodes of different types of engineering subgraphs. S15. Role-label the entity shared nodes in the entity shared node set. If the multi-dimensional semantic information carried by the entity shared node is the semantic information of the process node from the process topology diagram and the semantic information of the risk node from the risk propagation diagram, then the entity shared node carrying the multi-dimensional semantic information is labeled as a first role shared node. If the multi-dimensional semantic information carried is the semantic information of the process node from the process topology diagram and the semantic information of the process node from the process relationship diagram, then the entity shared node carrying the multi-dimensional semantic information is labeled as a second role shared node. If the multi-dimensional semantic information carried is the semantic information of the risk node from the risk propagation diagram and the semantic information of the process node from the process relationship diagram, then the entity shared node carrying the multi-dimensional semantic information is labeled as a third role shared node. S16. Based on the first role-shared node, establish a first bidirectional semantic association path between the process topology diagram and the risk propagation diagram, and generate a first inference traversal weight of the first bidirectional semantic association path through a first association strength matrix. Based on the second role-shared node, establish a second bidirectional semantic association path between the process topology diagram and the process relationship diagram, and generate a second inference traversal weight of the second bidirectional semantic association path through a second association strength matrix. Based on the third role-shared node, establish a third bidirectional semantic association path between the risk propagation diagram and the process relationship diagram, and generate a third inference traversal weight of the third bidirectional cross-graph mapping through a third association strength matrix. S17. By using the first bidirectional semantic association path, the second bidirectional semantic association path, the third bidirectional semantic association path, the first reasoning traversal weight, the second reasoning traversal weight, and the third reasoning traversal weight, the process topology graph, the risk propagation graph, and the process relationship graph are associated to generate an initial engineering knowledge graph.

[0031] In this embodiment, as Figure 2As shown, nodes from the three types of engineering subgraphs are uniformly mapped to the same entity shared semantic space using a semantic embedding algorithm. Specifically, process nodes from the process topology graph, risk nodes from the risk propagation graph, and process nodes from the process relationship graph are all mapped to the same entity shared semantic space. Nodes from different types of engineering subgraphs that point to the same engineering object are anchored to the same entity shared node. Here, an engineering object refers to a complete engineering event or construction activity, containing multi-dimensional information about the process attributes, technological attributes, and risk attributes of that event or activity. For example, the process node "diaphragm wall construction" in the process topology graph, the risk node "trench wall collapse" in the risk propagation graph, and the process node "diaphragm wall" in the process relationship graph all point to the same engineering object: diaphragm wall construction activity. Therefore, these nodes are associated with the same entity shared node, allowing the entity shared node to carry the semantic information corresponding to nodes from different types of engineering subgraphs. In other words, multi-dimensional semantic information is assigned to the entity shared node, resulting in a set of entity shared nodes carrying multi-dimensional semantic information.

[0032] The entity-shared nodes in the entity-shared node set are assigned roles. If the multi-dimensional semantic information carried by the entity-shared node comes from the semantic information of the process nodes in the process topology diagram and the semantic information of the risk nodes in the risk propagation diagram, then the entity-shared node carrying the multi-dimensional semantic information is marked as a first-role shared node. If the multi-dimensional semantic information carried comes from the semantic information of the process nodes in the process topology diagram and the semantic information of the process nodes in the process relationship diagram, then the entity-shared node carrying the multi-dimensional semantic information is marked as a second-role shared node. If the multi-dimensional semantic information carried comes from the semantic information of the risk nodes in the risk propagation diagram and the semantic information of the process nodes in the process relationship diagram, then the entity-shared node carrying the multi-dimensional semantic information is marked as a third-role shared node.

[0033] A first semantic association path is established between the process topology diagram and the risk propagation diagram based on the first role shared nodes. Specifically, using the first role shared nodes as anchor points, a bidirectional semantic association path is established between the process nodes in the process topology diagram and the risk nodes in the risk propagation diagram. A first inference traversal weight for the first bidirectional semantic association path is generated through a first association strength matrix. The first association strength matrix is ​​an N×L matrix with process nodes as the main nodes and risk nodes as the columns. The element values ​​in the matrix, i.e., the first weights, are determined based on historical engineering data or pre-assigned manually. Similarly, a second bidirectional semantic association path is established between the process topology diagram and the process relationship diagram based on the second role shared nodes. Specifically, using the second role shared nodes as anchor points, a bidirectional semantic association path is established between the process nodes in the process topology diagram and the process nodes in the process relationship diagram. A second inference traversal weight for the second bidirectional cross-graph mapping is generated through a second association strength matrix. The second association strength matrix is ​​an N×M matrix with process nodes as the main nodes and process nodes as the columns. The element values ​​in the matrix, i.e., the second weights, are determined based on historical engineering data or pre-assigned manually. A third bidirectional semantic association path is established between the risk propagation graph and the process relationship graph based on the third role shared node. That is, with the third role shared node as the anchor point, a bidirectional semantic association path is established between the risk node in the risk propagation graph and the process node in the process relationship graph. The third inference traversal weight of the third bidirectional cross-graph mapping is generated through the third association strength matrix. The third association strength matrix is ​​an L×M matrix of behavioral risk nodes and process nodes. The element values ​​in the matrix, i.e. the third weight, are determined based on historical engineering data or are pre-assigned manually.

[0034] By using the first, second, and third bidirectional semantic association paths, along with the first, second, and third reasoning traversal weights, and the third reasoning traversal weights, the three originally independent engineering subgraphs are linked into an engineering knowledge graph. In this engineering knowledge graph, the nodes of the three types of engineering subgraphs exist uniformly within the same entity's shared semantic space. Furthermore, based on the multi-dimensional semantic information they carry, not only can the originating engineering subgraph be identified, but the node's position within the corresponding engineering subgraph and its connection relationships with other nodes within the same subgraph can also be determined. In other words, the original internal edge structure of the three types of engineering subgraphs is preserved.

[0035] S2. Perform a forward graph structure traversal on the engineering knowledge graph. Starting from the process node of the process topology graph, traverse along the bidirectional semantic association path to the risk node of the risk propagation graph. Use the process nodes in the process relationship graph as condition variables to perform conditional risk probability reasoning propagation on the risk nodes to obtain the risk reasoning feature vector corresponding to each process node. Perform reasoning matching between the risk reasoning feature vector corresponding to each process node and the pre-configured risk suppression feature vector of the corresponding process node to select the optimal process node. At this point, in step S2, starting from the process nodes of the process topology graph, the process reasoning traverses along the bidirectional semantic association path to the risk nodes of the risk propagation graph. The process nodes in the process relationship graph are used as condition variables to perform conditional risk probability reasoning propagation on the risk nodes, obtaining the risk reasoning feature vectors corresponding to each process node. The risk reasoning feature vectors corresponding to each process node are then matched with the pre-configured risk suppression feature vectors of the corresponding process nodes to select the optimal process nodes, including: S21. Collect the current construction stage of the engineering scene, extract the process node corresponding to the current construction stage from the process topology diagram as the activated process node, and obtain the current execution status of the activated process node. S22. The activated process node is reasoned and traversed along the bidirectional semantic association path to the risk node of the risk propagation graph to obtain the activated risk node triggered by the current execution state, and the activated risk propagation link triggered from the activated risk node is obtained based on the risk causal topology link of the risk propagation graph. S23. Using the process nodes of the process relationship diagram as condition variables of the activated risk nodes, perform dynamic risk reasoning propagation on the activated risk propagation link to obtain the risk reasoning feature vector corresponding to each process node. S24. Calculate the cosine similarity between the risk inference feature vector corresponding to each process node and the risk suppression feature vector pre-configured for the corresponding process node, and take the process node with the highest cosine similarity as the optimal process node.

[0036] In this embodiment, as Figure 3As shown, forward cross-graph reasoning is performed on the engineering knowledge graph. The current construction stage of the engineering scenario is collected, and the process nodes corresponding to the current construction stage are extracted from the process topology graph as activated process nodes. The current execution status of the activated process nodes is obtained, including the execution progress. The activated process nodes are passed to the risk nodes in the risk propagation graph along the bidirectional semantic association path. That is, according to the bidirectional semantic association path constructed in step S16, the process is passed to the activated risk nodes triggered by the current execution status. Based on the risk causal topology link of the risk propagation graph, the activated risk propagation link triggered by the activated risk node is obtained. That is, the subset of links triggered by the risk causal topology link starting from the activated risk node. For example, the activated process node: diaphragm wall trenching is mapped to the risk node: trench wall collapse in the risk propagation graph through the first cross-graph mapping relationship. That is, the activated risk node: trench wall collapse. The risk propagation graph shows a causal topological link: abnormal groundwater level → trench wall collapse → foundation pit collapse → surrounding building tilting. Starting from the activated risk node: trench wall collapse, the causal topological link further triggers the risk node: foundation pit collapse, thus forming an activated risk propagation link: trench wall collapse → foundation pit collapse. The risk node: surrounding building tilting, as the end of this causal topological link, is not activated because the risk node: foundation pit collapse has not yet actually occurred; therefore, it does not belong to the activated risk propagation link.

[0037] The process relationship diagram is traversed, and each process node in the diagram is used as a condition variable for an activated risk node. Dynamic risk propagation is then performed on the activated risk propagation path. Specifically, the conditional probability of each process node as a condition variable for an activated risk node is first calculated. Then, based on these conditional probabilities, risk propagation is extrapolated along the activated risk propagation path. The propagation probability of each risk node along the activated risk propagation path and the propagation range of the activated risk propagation path are statistically analyzed, thus obtaining the risk inference feature vector corresponding to different process nodes. For example, in the activated risk propagation path: trench wall collapse → foundation pit collapse, the conditional probabilities of the activated risk node: trench wall collapse are calculated using the process nodes: diaphragm wall and slope excavation as condition variables. The conditional probability in this case is the propagation probability along the activated risk propagation path: trench wall collapse → foundation pit collapse to the next risk node. The propagation range of the activated risk propagation path refers to the number of risk nodes affected, thus obtaining the risk inference feature vector corresponding to different process nodes. Calculate the cosine similarity between the risk inference feature vector corresponding to each process node and the risk suppression feature vector pre-configured for the corresponding process node, and select the process node with the highest cosine similarity as the optimal process node.

[0038] S3. Perform reverse graph structure update on the optimal process node, obtain the process decomposition template corresponding to the optimal process node from the process relationship graph, and perform reasoning reconstruction on the process topology graph according to the process decomposition template to obtain the reconstructed process topology graph. Re-infer and traverse the reasoned and reconstructed process topology graph to the risk node of the risk propagation graph through the bidirectional semantic association path, re-perform conditional risk probability reasoning propagation and perform conditional risk reasoning verification, and iterate the forward cross-graph reasoning and the reverse cross-graph update until the optimal process node passes the conditional risk reasoning verification, complete the graph structure iteration, realize the adaptive linkage update of the process topology graph and the risk propagation graph, and generate the final engineering knowledge graph.

[0039] At this point, step S3, which involves reasoning and reconstructing the process topology diagram based on the process decomposition template, includes: S31. Compare the process decomposition template with the process topology diagram to determine whether there are process nodes that exist in the process decomposition template but not in the process topology diagram. If there are, take the corresponding process node as a missing process node, obtain the temporal dependency relationship of the missing process node from the process decomposition template, and add the missing process node to the process topology diagram based on the temporal dependency relationship of the missing process node to realize the reasoning reconstruction of the process topology diagram. S32. Determine whether there are process nodes that exist in the process topology diagram but not in the process decomposition template. If so, treat the corresponding process node as a redundant process node and suspend the redundant process node in the process topology diagram to realize the reasoning reconstruction of the process topology diagram.

[0040] In this embodiment, as Figure 3 As shown, a reverse cross-graph update is performed on the optimal process node selected in step S2. The process decomposition template corresponding to the optimal process node is obtained from the process relationship graph. The process decomposition template is compared with the process topology graph to identify process nodes that exist in the process decomposition template but not in the process topology graph (i.e., missing process nodes). The temporal dependencies of the process nodes are obtained from the process decomposition module to clarify the position of the missing process nodes in the process topology graph, allowing them to be added and thus reconstructing the process topology graph. Simultaneously, process nodes that exist in the process topology graph but not in the process decomposition template (i.e., redundant process nodes) are identified. These redundant process nodes are suspended in the process topology graph. Suspension means that the redundant process node is not activated under the current optimal process node. If it needs to be restored in subsequent iterations, it can be reactivated without rebuilding, thus reconstructing the process topology graph.

[0041] At this point, step S3, which involves re-inferring the reconstructed process topology graph through the bidirectional semantic association path to the risk nodes of the risk propagation graph, re-performing conditional risk probability inference propagation and performing conditional risk inference verification, and iterating the forward cross-graph inference and the reverse cross-graph update until the optimal process node passes the conditional risk inference verification, thereby achieving adaptive linkage updates of the process topology graph and the risk propagation graph, includes: S33. The process topology graph after reasoning reconstruction is re-inferred and traversed to the risk nodes of the risk propagation graph through the bidirectional semantic association path. The optimal process node is used as a condition variable to re-perform conditional risk probability reasoning propagation to obtain the optimal risk reasoning feature vector corresponding to the optimal process node. S34. The optimal risk reasoning feature vector corresponding to the optimal process node is reasoned and matched with the pre-configured risk suppression features to obtain the corresponding reasoning matching degree. S35. Compare the inference matching degree with a preset adaptation threshold to determine whether the inference matching degree is lower than the preset adaptation threshold. If so, the conditional risk inference verification fails, triggering the next round of forward cross-graph inference and reverse cross-graph update until the re-selected optimal process node passes the conditional risk inference verification. If not, the conditional risk inference verification succeeds, and the reconstructed process topology graph is used as the updated process topology graph. At the same time, the risk propagation link of the risk propagation graph is updated synchronously according to the optimal risk inference feature vector corresponding to the optimal process node, realizing adaptive linkage update of the process topology graph and the risk propagation graph.

[0042] In this embodiment, as Figure 3As shown, the reconstructed process topology graph is traversed again to the risk nodes of the risk propagation graph through a bidirectional semantic association path. At this time, the activated process nodes are re-extracted and the risk nodes are re-mapped. Then, the conditional risk probability inference propagation is carried out again with the optimal process node as the condition variable. That is, the dynamic risk propagation is carried out again in the manner of step S23 to obtain the optimal risk inference feature vector corresponding to the optimal process node. The optimal risk inference feature vector corresponding to the optimal process node is matched with the pre-configured risk suppression features, i.e., the cosine similarity is calculated. The calculated cosine similarity is the inference matching degree. If the inference matching degree is lower than the preset adaptation threshold, the conditional risk inference verification fails, triggering the next round of forward cross-graph inference and reverse cross-graph update, until the re-selected optimal process node passes the conditional risk inference verification. Conversely, if the inference matching degree is not lower than the preset adaptation threshold, the conditional risk inference verification succeeds. The reconstructed process topology graph is used as the updated process topology graph. At the same time, the risk propagation link of the risk propagation graph is updated synchronously according to the optimal risk inference feature vector corresponding to the optimal process node, realizing adaptive linkage update of the process topology graph and the risk propagation graph. Through the above iterative forward cross-graph inference and reverse cross-graph update, the engineering knowledge graph is continuously converged, thereby generating the final engineering knowledge graph.

[0043] Example 2 Please refer to Figure 4 The present invention provides an engineering knowledge graph construction system 1 with multi-graph fusion and bidirectional reasoning, including a memory 3, a processor 2, and a computer program stored on the memory 3 and run on the processor 2. When the processor 2 executes the computer program, it implements the steps in Embodiment 1.

[0044] Since the systems / devices described in the above embodiments of the present invention are systems / devices used to implement the methods of the above embodiments of the present invention, those skilled in the art can understand the specific structure and modifications of the systems / devices based on the methods described in the above embodiments of the present invention, and therefore will not be repeated here. All systems / devices used in the methods of the above embodiments of the present invention fall within the scope of protection of the present invention.

[0045] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0046] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions.

[0047] It should be noted that any reference numerals placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In claims that enumerate several means, several of these means may be embodied by the same hardware. The use of the terms first, second, third, etc., is merely for convenience of expression and does not indicate any order. These terms can be understood as part of the component names.

[0048] Furthermore, it should be noted that in the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0049] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the claims should be interpreted to include both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0050] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, then this invention should also include these modifications and variations.

Claims

1. A multi-graph collaborative reasoning method for engineering scenarios, characterized in that, include: Multi-source heterogeneous data from engineering scenarios are collected, and semantic parsing is performed on the multi-source heterogeneous data to generate three types of engineering subgraphs. The nodes of the three types of engineering subgraphs are uniformly mapped to the same entity shared semantic space through a semantic embedding algorithm. Bidirectional semantic association paths are constructed between nodes of different types of engineering subgraphs and nodes of the same entity, so that the three types of engineering subgraphs can be associated based on bidirectional semantic association paths, generating an initial engineering knowledge graph for multi-graph collaborative reasoning. The three types of engineering subgraphs are process topology graph, risk propagation graph, and process relationship graph. Multi-graph collaborative forward reasoning is performed on the initial engineering knowledge graph. Starting from the process node of the process topology graph, the reasoning traverses along the bidirectional semantic association path to the risk node of the risk propagation graph. The process nodes in the process relationship graph are used as condition variables to perform conditional risk probability reasoning propagation on the risk nodes to obtain the risk reasoning feature vector corresponding to each process node. The risk reasoning feature vector corresponding to each process node is then matched with the risk suppression feature vector pre-configured for the corresponding process node to select the optimal process node. Multi-graph collaborative reverse update reasoning is performed on the optimal process node. The process decomposition template corresponding to the optimal process node is obtained from the process relationship graph. The process topology graph is reconstructed based on the process decomposition template to obtain the reconstructed process topology graph. The reconstructed process topology graph is then re-inferred and traversed to the risk node of the risk propagation graph through the bidirectional semantic association path. Conditional risk probability reasoning propagation is performed again and conditional risk reasoning verification is executed. The multi-graph collaborative forward reasoning and multi-graph collaborative reverse update reasoning are iterated until the optimal process node passes the conditional risk reasoning verification. This achieves adaptive linkage update of the process topology graph and the risk propagation graph, completes the collaborative reasoning evolution of the initial engineering knowledge graph, and outputs the final engineering knowledge graph.

2. The multi-graph collaborative reasoning method for engineering scenarios as described in claim 1, characterized in that, The multi-source heterogeneous data includes construction procedure data, construction risk data, and construction technology data. The semantic parsing of the multi-source heterogeneous data to generate three types of engineering sub-graphs includes: Semantic parsing is performed on the construction process data to identify the process set and the temporal dependencies between different processes in the process set. Each process in the process set is treated as an independent process node, and the temporal dependencies between different processes are treated as directed dependency edges between different process nodes to construct a process topology graph. Semantic parsing is performed on construction risk data to identify the set of risk factors, the triggering conditions of each risk factor in the set of risk factors, and the propagation relationship between different risk factors. Each risk factor in the set of risk factors is treated as an independent risk node. Based on the triggering conditions of each risk factor and the propagation relationship between different risk factors, a risk causal topology link and a risk propagation link between risk factors are generated. The conditional probability with process as the conditional variable is used as the risk propagation probability of the risk propagation link to construct a risk propagation graph. Semantic parsing is performed on the construction process data to identify the process set and the substitution, inclusion, and constraint relationships between different processes within the process set. Each process in the process set is treated as an independent process node, and a unique process decomposition template and risk suppression feature vector are configured for each process node. The substitution, inclusion, and constraint relationships between different processes are used as process relationship edges between different processes to construct a process relationship graph. The process decomposition template is the set of processes that should be included when using the corresponding process and the temporal dependencies between processes in the process set. The risk suppression feature vector is the ability to suppress each risk factor in the risk factor set when using the corresponding process.

3. The multi-graph collaborative reasoning method for engineering scenarios as described in claim 1, characterized in that, The process of mapping nodes of the three types of engineering subgraphs to the same entity shared semantic space using a semantic embedding algorithm, constructing bidirectional semantic association paths between nodes of different types of engineering subgraphs and nodes of the same entity, and enabling the three types of engineering subgraphs to be associated based on bidirectional semantic association paths to generate an initial engineering knowledge graph for multi-graph collaborative reasoning includes: The nodes of the three types of engineering subgraphs are uniformly mapped to the same entity shared semantic space through a semantic embedding algorithm. In the same entity shared semantic space, nodes from different types of engineering subgraphs that point to the same engineering object are anchored to the same entity shared node, and multi-dimensional semantic information is given to the entity shared node to obtain a set of entity shared nodes carrying multi-dimensional semantic information. The multi-dimensional semantic information is the semantic information corresponding to the nodes of different types of engineering subgraphs. The entity shared nodes in the entity shared node set are assigned roles. If the multi-dimensional semantic information carried by the entity shared node is the semantic information of the process node from the process topology diagram and the semantic information of the risk node from the risk propagation diagram, then the entity shared node carrying the multi-dimensional semantic information is marked as a first role shared node. If the multi-dimensional semantic information carried is the semantic information of the process node from the process topology diagram and the semantic information of the process node from the process relationship diagram, then the entity shared node carrying the multi-dimensional semantic information is marked as a second role shared node. If the multi-dimensional semantic information carried is the semantic information of the risk node from the risk propagation diagram and the semantic information of the process node from the process relationship diagram, then the entity shared node carrying the multi-dimensional semantic information is marked as a third role shared node. Based on the first role-shared node, a first bidirectional semantic association path is established between the process topology diagram and the risk propagation diagram, and a first inference traversal weight of the first bidirectional semantic association path is generated through a first association strength matrix. Based on the second role-shared node, a second bidirectional semantic association path is established between the process topology diagram and the process relationship diagram, and a second inference traversal weight of the second bidirectional semantic association path is generated through a second association strength matrix. Based on the third role-shared node, a third bidirectional semantic association path is established between the risk propagation diagram and the process relationship diagram, and a third inference traversal weight of the third bidirectional cross-graph mapping is generated through a third association strength matrix. The process topology graph, the risk propagation graph, and the process relationship graph are associated through the first bidirectional semantic association path, the second bidirectional semantic association path, the third bidirectional semantic association path, the first reasoning traversal weight, the second reasoning traversal weight, and the third reasoning traversal weight, thereby generating an initial engineering knowledge graph for multi-graph collaborative reasoning.

4. The multi-graph collaborative reasoning method for engineering scenarios as described in claim 1, characterized in that, Starting from the process nodes in the process topology graph, the process traverses along the bidirectional semantic association path to the risk nodes in the risk propagation graph. The process nodes in the process relationship graph are used as condition variables to perform conditional risk probability reasoning propagation on the risk nodes, resulting in risk reasoning feature vectors corresponding to each process node. These feature vectors are then matched with pre-configured risk suppression feature vectors for the corresponding process nodes to select the optimal process nodes. The current construction stage of the project scenario is collected, the process node corresponding to the current construction stage is extracted from the process topology diagram as the activated process node, and the current execution status of the activated process node is obtained. The activated process node is reasoned and traversed along the bidirectional semantic association path to the risk node of the risk propagation graph to obtain the activated risk node triggered by the current execution state, and the activated risk propagation link triggered from the activated risk node is obtained based on the risk causal topology link of the risk propagation graph. The process nodes in the process relationship diagram are used as condition variables for the activated risk nodes. Dynamic risk reasoning propagation is performed on the activated risk propagation link to obtain the risk reasoning feature vector corresponding to each process node. Calculate the cosine similarity between the risk inference feature vector corresponding to each process node and the risk suppression feature vector pre-configured for the corresponding process node, and select the process node with the highest cosine similarity as the optimal process node.

5. The multi-graph collaborative reasoning method for engineering scenarios as described in claim 1, characterized in that, The step of reasoning and reconstructing the process topology diagram based on the process decomposition template includes: The process decomposition template is compared with the process topology diagram to determine whether there are process nodes that exist in the process decomposition template but not in the process topology diagram. If there are, the corresponding process node is taken as a missing process node. The temporal dependency relationship of the missing process node is obtained from the process decomposition template. Based on the temporal dependency relationship of the missing process node, the missing process node is added to the process topology diagram to realize the reasoning reconstruction of the process topology diagram. Determine whether there are process nodes that exist in the process topology diagram but not in the process decomposition template. If so, the corresponding process node is regarded as a redundant process node and suspended in the process topology diagram to realize the reasoning reconstruction of the process topology diagram.

6. The multi-graph collaborative reasoning method for engineering scenarios as described in claim 1, characterized in that, The process topology graph re-infers and traverses to the risk nodes of the risk propagation graph through the bidirectional semantic association path, performs conditional risk probability reasoning propagation again and executes conditional risk reasoning verification, and iterates the multi-graph collaborative forward reasoning and the multi-graph collaborative reverse update reasoning until the optimal process node passes the conditional risk reasoning verification, thereby achieving adaptive linkage update of the process topology graph and the risk propagation graph, including: The process topology graph after reasoning reconstruction is re-inferred and traversed to the risk node of the risk propagation graph through the bidirectional semantic association path. The optimal process node is used as the condition variable to re-perform conditional risk probability reasoning propagation to obtain the optimal risk reasoning feature vector corresponding to the optimal process node. The optimal risk inference feature vector corresponding to the optimal process node is matched with the pre-configured risk suppression features to obtain the corresponding inference matching degree. The inference matching degree is compared with a preset adaptation threshold to determine whether the matching degree is lower than the preset adaptation threshold. If so, the conditional risk inference verification fails, triggering the next round of multi-graph collaborative forward inference and multi-graph collaborative reverse update inference, until the re-selected optimal process node passes the conditional risk inference verification. If not, the conditional risk inference verification succeeds, and the reconstructed process topology graph is used as the updated process topology graph. At the same time, the risk propagation link of the risk propagation graph is updated synchronously according to the optimal risk inference feature vector corresponding to the optimal process node, realizing adaptive linkage update of the process topology graph and the risk propagation graph.

7. A multi-graph fusion bidirectional reasoning engineering knowledge graph construction system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 6.

Images