Geological disaster knowledge graph construction method and system based on artificial intelligence

Abstract

The present application relates to the technical field of knowledge graph construction, in particular to a geological disaster knowledge graph construction method and system based on artificial intelligence; the method comprises the following steps: using a semantic physical collaborative preprocessing method to extract a geological semantic triple set and construct a three-layer semantic ontology graph; using a double-chain spatiotemporal graph construction method to generate a spatial profile structure chain and a time sequence state chain and construct a double-chain coupled graph structure; using a disturbance perception graph neural network mechanism to perform node state propagation calculation and obtain risk state prediction results of a landslide structure unit layer at each time node; using a reverse causal path chain reconstruction method to generate a landslide risk propagation path chain, and migrating propagation parameters and response rules in a historical landslide graph structure to the double-chain coupled graph structure to construct a double-chain coupled spatiotemporal geological disaster knowledge graph. The present application realizes landslide state prediction and causal path reasoning through a double-chain graph structure and a disturbance perception graph neural network.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of knowledge graph construction, in particular to a geological disaster knowledge graph construction method and system based on artificial intelligence. BACKGROUND

[0002] As one of the core natural risks affecting human safety and infrastructure stability, geological disasters have the characteristics of wide distribution, complex causes and diverse evolution mechanisms. In order to improve the intelligent expression, structured management and reasoning analysis capabilities of geological disaster data, in recent years, semantic modeling technologies such as knowledge graph have been gradually introduced to explicitly express information such as structural elements, inducing factors and evolution paths in geological disaster events. The information of geological disasters has the characteristics of diverse sources, complex structure and dynamic evolution, which brings great challenges to information integration and risk prediction.

[0003] The existing technology focuses on the recognition and annotation modeling of named entities in the corpus, and the graph structure is still a flat triple. A static geological disaster chain ontology model is constructed, which cannot realize physical prediction modeling and dynamic updating, as well as joint modeling of spatial topological relationship and temporal evolution state, lacks dynamic analysis and prediction ability of geological disaster evolution mechanism, and may cause the problem of explainable prediction of disaster risk state of geological disaster knowledge graph. SUMMARY

[0004] The present application relates to the technical field of knowledge graph construction, in particular to a geological disaster knowledge graph construction method and system based on artificial intelligence.

[0005] To achieve the above-mentioned purpose, the technical scheme of the present application is: a geological disaster knowledge graph construction method based on artificial intelligence, comprising:

[0006] S1, using a semantic physical collaborative preprocessing method based on a geological disaster survey report to extract a set of geological semantic triples, and constructing a three-layer semantic ontology graph containing a landslide structure unit layer, a physical and mechanical parameter layer and a logical causal relationship layer;

[0007] S2, based on the three-layer semantic ontology graph, using a double-chain spatiotemporal graph construction method to generate a spatial profile structure chain and a time sequence state chain respectively, and using a graph neural network graph topology modeling mechanism to integrate the spatial profile structure chain and the time sequence state chain in the form of a heterogeneous graph, and constructing a double-chain coupled graph structure;

[0008] S3, the node state propagation calculation of the node in the double-chain coupled graph structure is carried out by using a perturbation-aware graph neural network mechanism, and the risk state prediction result of the landslide structure unit layer at each time node is obtained;

[0009] S4, in combination with the logic causal relationship layer, a reverse causal path chain reconstruction method is used to generate a landslide risk propagation path chain, and the propagation parameters and response rules in the historical landslide graph structure are migrated to the double-chain coupled graph structure, and finally a double-chain coupled spatiotemporal geological disaster knowledge graph is constructed.

[0010] Preferably, the set of geological semantic triples includes structure configuration triples, physical parameter triples and disaster causal triples;

[0011] The structure configuration triple is a triple composed of the spatial relationship between the geological structure unit entities, and is used to construct the node and structure topology connection edge in the landslide structure unit layer.

[0012] The structure connection type semantic template is constructed, and the structure configuration triple is extracted and generated by traversing the sentence containing the geological structure entity.

[0013] The physical parameter triple is a triple composed of the geological structure unit entity, the physical parameter attribute and the physical parameter value, and is used to construct the binding edge between the attribute node and the structure node in the physical and mechanical parameter layer.

[0014] The disaster causal triple is a triple composed of the causal relationship between the external disturbance source, the physical state parameter and the landslide structure response, and is used to construct the causal edge path of the logic causal relationship layer.

[0015] The physical parameter triple composed of the geological structure entity and its corresponding physical parameter entity and value is extracted through semantic template matching and numerical attribution recognition.

[0016] Preferably, the three-layer semantic ontology graph is constructed by the landslide structure unit layer, the physical and mechanical parameter layer and the logic causal relationship layer, and serves as the basis for constructing the double-chain coupled graph structure.

[0017] The landslide structure unit layer is used to represent the spatial connection relationship between the geological structure entities in the landslide profile.

[0018] The physical and mechanical parameter layer is used to express the binding relationship between the landslide geological structure unit entity and the physical and mechanical parameter.

[0019] The logic causal relationship layer is used to express the causal chain path relationship between the external disturbance source, the physical state parameter and the landslide structure response.

[0020] Preferably, the spatial profile structure chain is a directed graph structure composed of geological structure entities and their spatial connection relationships, used to represent the spatial topological connection between the geological structure entities in the landslide profile.

[0021] The time series state chain is a time series graph structure composed of physical parameter attributes of geological structure entities at different time points and their time evolution paths, used to express the evolution path of the physical parameters of the geological structure entities over time.

[0022] Preferably, in S2, the double-chain spatiotemporal graph construction method is based on the single-layer topological graph construction method and combines the landslide structure unit layer and the physical and mechanical parameter layer to construct the spatial profile structure chain and the time series state chain, as follows:

[0023] S2.1, traverse all structure configuration triples, take the geological structure unit entities in the structure configuration triples as graph nodes, generate directed edges between the graph nodes according to the spatial connection semantics of the graph nodes, and obtain the spatial profile structure chain;

[0024] S2.2, extract the physical parameter attributes of the geological structure unit entities in the physical parameter triples in the time forward order, and construct the physical parameter attributes of the geological structure unit entities as graph nodes. The adjacent graph nodes are connected by directed edges in the time flow direction, and the difference between the physical parameter values in the physical parameter triples is taken as the edge weight of the directed edges. Finally, the time series state chain is constructed;

[0025] S2.3, based on all geological structure unit entities in the spatial profile structure chain and the time series state chain, a unified node index database with cross-chain index is established.

[0026] Preferably, the double-chain coupled graph structure is a heterogeneous graph structure composed of the spatial profile structure chain and the time series state chain integrated through the unified node index database, used to express the spatial connection relationship of the geological structure unit entities and the time state evolution path simultaneously in the graph neural network propagation.

[0027] Preferably, in S3, the disturbance perception graph neural network mechanism is obtained by introducing a geological disturbance factor as a dynamic adjustment quantity on the basis of the graph neural network propagation model, used to predict the risk state of the landslide structure unit layer under external disturbance;

[0028] The node state propagation calculation is performed on the nodes in the double-chain coupled graph structure using the disturbance perception graph neural network mechanism, and the risk state prediction results of the landslide structure unit layer at each time node are obtained, as follows:

[0029] S3.1, set the geological disturbance factor and add it to the graph nodes of the spatial profile structure chain and the time series state chain;

[0030] S3.2, define the propagation path of the graph neural network, including the spatial structure propagation path and the time series propagation path;

[0031] S3.3, in the propagation process of the graph neural network, updating the state of each node according to the change of the geological disturbance factor, obtaining the node state embedding vector of each round of propagation;

[0032] S3.4, applying an activation function and a linear mapping to the node state embedding vector to obtain a prediction value vector;

[0033] S3.5, the prediction value vector is used for prediction value and classification output, and is mapped to each node in the landslide structure unit layer to obtain a risk state prediction result.

[0034] Preferably, in S4, the reverse causal path chain reconstruction method is a logical causal graph traversal method combined with a prediction state driven path backtracking mechanism, which is used to identify the triggering process of physical parameter change and external disturbance source in reverse from the node as the starting point, and reconstruct the causal propagation path chain of geological disaster event;

[0035] The reverse causal path chain reconstruction method is used to generate the landslide risk propagation path chain, and the specific method steps are as follows:

[0036] S4.1.1, set a risk state threshold, and take the node with a prediction value greater than the risk state threshold in the risk state prediction result as the starting point of the landslide risk propagation path chain;

[0037] S4.1.2, extract the disaster causal triplets in the logical causal relationship layer, and establish a directed causal graph;

[0038] The nodes of the directed causal graph are geological structure entities containing geological disturbance factors and physical state parameters, and the edges of the directed causal graph are external disturbance sources in the disaster causal triplets;

[0039] S4.1.3, for each node of the directed causal graph, perform reverse depth-first traversal in the directed causal graph, and record all causal path chains;

[0040] S4.1.4, remove redundant, conflicting and incomplete causal path chains, and the remaining causal path chain set constitutes the landslide risk propagation path chain.

[0041] Preferably, the propagation parameters and response rules in the historical landslide graph structure are migrated to the double-chain coupled graph structure, and finally a double-chain coupled spatio-temporal geological disaster knowledge graph is constructed, and the specific operation includes:

[0042] S4.2.1, a graph embedding representation is performed on the double-chain coupled graph structure, the structural similarity between the historical landslide graph structure is calculated, when the similarity meets the preset threshold, the propagation parameter and the response rule in the selected historical landslide graph structure are migrated and injected into the double-chain coupled graph structure;

[0043] S4.2.2, the landslide risk propagation path chain is written into the double-chain coupled graph structure at the same time, and finally the double-chain coupled spatiotemporal geological disaster knowledge graph is constructed.

[0044] On the other hand, the present application provides an artificial intelligence-based geological disaster knowledge graph construction system, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to realize the artificial intelligence-based geological disaster knowledge graph construction method described above.

[0045] Compared with the prior art, the above technical scheme of the present application has the following beneficial technical effects:

[0046] 1. In the present application, based on the semantic physical collaborative preprocessing method and the three-layer semantic ontology graph construction mechanism, the structural configuration information, physical parameter data and causal mechanism semantics in the original geological disaster text can be uniformly mapped to the structure layer, parameter layer and causal layer, solving the problem of flat graph structure and inability to be used for modeling calculation;

[0047] 2. In the present application, by constructing a double-chain coupled graph structure of spatial profile structure chain and time sequence state chain, and introducing a disturbance perception graph neural network mechanism and a reverse causal path chain reconstruction method, dynamic state propagation, risk score prediction and causal path reasoning and tracing of geological structure units under disturbance conditions are realized. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 The flowchart of the embodiment of the present application is shown. DETAILED DESCRIPTION

[0049] Embodiment one, as shown in the figure, the artificial intelligence-based geological disaster knowledge graph construction method provided by the present application has the following specific implementation steps: Figure 1

[0050] S1, based on the geological disaster investigation report, a semantic physical collaborative preprocessing method is used to extract a set of geological semantic triples, and a three-layer semantic ontology graph comprising a landslide structure unit layer, a physical and mechanical parameter layer and a logical causal relationship layer is constructed;

[0051] In this embodiment, the set of geological semantic triples includes structure configuration triples, physical parameter triples and disaster causal triples;

[0052] ​Wherein, the structure configuration triple is a triple composed of the spatial relationship between the geological structure unit entities, used to build the node and structure topology connection edge in the landslide structure unit layer, and the extraction of the structure configuration triple is as follows:

[0053] The structure connection class semantic template is constructed, the sentences containing the geological structure entities are traversed, and the structure configuration triple is extracted and generated;

[0054] The physical parameter triple is a triple composed of the geological structure unit entity, the physical parameter attribute and the physical parameter value, used to build the binding edge between the attribute node and the structure node in the physical and mechanical parameter layer;

[0055] The disaster causal triple is a triple composed of the causal relationship between the external disturbance source, the physical state parameter and the landslide structure response, used to build the causal edge path in the logical causal relationship layer, and the extraction of the disaster causal triple is as follows:

[0056] The physical parameter triple composed of the geological structure entity and the corresponding physical parameter entity and the value is extracted through semantic template matching and numerical attribution recognition.

[0057] In this embodiment, the format of the structure configuration triple is, the format of the physical parameter triple is, and the format of the disaster causal triple is ;

[0058] The semantic physical collaborative preprocessing method is as follows:

[0059] The natural language sentence processing is performed on the text data, the naming entity recognition model based on BERT is used to mark the geological structure nouns, parameter terms and disturbance trigger factors appearing in the text, and the marked entities are classified according to the semantic types;

[0060] A plurality of sentence templates based on the field language are constructed, including:

[0061] The structure connection relationship sentence, for example, the slide belt is located above the slide bed, and the crack penetrates the slide body;

[0062] The parameter attribution sentence, for example, the internal friction angle of the slide belt is 12°, and the water content at the crack is 23%;

[0063] The causal logic sentence, for example, the water content increases due to the rainfall enhancement, and the slide belt damage is caused by the pore pressure rise;

[0064] The rule engine is used for template matching on the sentences, the sentences meeting the rules are extracted, and the geological semantic triple set is constructed.

[0065] In the embodiment, the three-layer semantic ontology graph is constructed by a landslide structure unit layer, a physical and mechanical parameter layer and a logical causal relationship layer, serving as a basis for constructing the double-chain coupled graph structure.

[0066] The landslide structure unit layer is used to represent the spatial connection relationship between geological structure entities in a landslide profile.

[0067] The physical and mechanical parameter layer is used to represent the binding relationship between landslide geological structure unit entities and physical and mechanical parameters.

[0068] The logical causal relationship layer is used to represent the causal chain path relationship between external disturbance sources, physical state parameters and landslide structure responses.

[0069] In the embodiment, the landslide structure unit layer is constructed as follows: traversing the structure configuration triplets, geological structure unit entity A and geological structure unit entity B are mapped to structure nodes in the graph, the spatial relationship R is taken as a directed edge label, an edge A->B is generated, the spatial topology structure of the profile structure graph is formed, hierarchical connection is supported, and multiple connection relationships are supported, for example, A is in contact with B, A overlies C.

[0070] The physical and mechanical parameter layer is constructed as follows: traversing the physical parameter triplets, geological structure unit entity A is taken as a structure node and connected to the structure unit layer, the physical parameter attribute is taken as a parameter node and introduced into the graph, and the physical parameter value P is taken as a parameter node attribute, and a relationship edge A-physical parameter attribute-> is created.

[0071] The logical causal relationship layer is constructed as follows: traversing the disaster causal triplets, the external disturbance source D and the landslide structure response E are taken as nodes, a directed edge D->E is established, and the edge label is the causal relationship type; for the logical causal relationship between multiple consecutive nodes, a consecutive causal edge connection is generated.

[0072] S2, based on the three-layer semantic ontology graph, a double-chain spatiotemporal graph construction method is used to generate a spatial profile structure chain and a time sequence state chain, and a graph neural network graph topology modeling mechanism is used to integrate the spatial profile structure chain and the time sequence state chain in the form of a heterogeneous graph to construct a double-chain coupled graph structure.

[0073] The spatial profile structure chain is a directed graph structure composed of geological structure entities and their spatial connection relationships, and is used to represent the spatial topological connection between the geological structure entities in the landslide profile.

[0074] The time sequence state chain is a time sequence graph structure composed of physical parameter attributes of geological structure entities at different time points and time evolution paths thereof, and is used to express evolution paths of physical parameters of the geological structure entities over time.

[0075] In this embodiment, the spatial profile structure chain is a directed graph chain representing spatial connection relationships between geological structure unit entities in a landslide profile, nodes are geological structure entities, edges represent spatial structure semantic relationships, are used for modeling spatial propagation paths in a graph neural network, support information transmission of perturbations and states in a landslide spatial configuration, and are also a basis for constructing visualization relationships of the landslide profile structure; the time sequence state chain is a time sequence graph representing physical parameter evolution paths of geological structure units at multiple time points, is used for modeling time evolution trends of physical states, is a modeling basis for key dynamic processes of landslide formation, and supports accumulation and prediction of risk factors in the time direction.

[0076] The spatial profile structure chain is graph mapping of spatial relationships of a landslide structure unit layer; and the time sequence state chain is time sequence evolution graph mapping of physical and mechanical parameter layers.

[0077] Fusion of the spatial profile structure chain and the time sequence state chain is a projection of a three-layer semantic ontology graph atlas computable substructure, and is used to support propagation calculation of a graph neural network.

[0078] In this embodiment S2, the double-chain spatio-temporal atlas construction method is based on a single-layer topological atlas construction method and combines a landslide structure unit layer and a physical and mechanical parameter layer to construct a spatial profile structure chain and a time sequence state chain, and details are as follows.

[0079] S2.1, all structure configuration triples are traversed, geological structure unit entities in the structure configuration triples are taken as graph nodes, directed edges between the graph nodes are generated according to spatial connection semantics of the graph nodes, and a spatial profile structure chain is obtained;

[0080] S2.2, physical parameter attributes of the geological structure unit entities in the physical parameter triples are extracted in a time forward order, the physical parameter attributes of the geological structure unit entities are constructed as graph nodes, adjacent graph nodes are established as directed edges in a time flow direction, and a difference between physical parameter values in the physical parameter triples is taken as an edge weight of the directed edges, and finally a time sequence state chain is constructed;

[0081] S2.3, based on all the geological structure unit entities in the spatial profile structure chain and the time sequence state chain, a unified node index database of cross-chain indexes is established.

[0082] In this embodiment, the uniform node index database of cross-chain index is a unique mapping mechanism for spatial profile structure chain and time sequence state chain of geological structure unit nodes, which is used to ensure that the same structure unit in the spatial profile structure chain and the time sequence state chain has a globally unique identifier, and the nodes have locatable, citable and traceable numbers in different atlas substructures, supporting the joint propagation operation of subsequent graph neural networks in heterogeneous graph structures.

[0083] A uniform node index database of cross-chain index is established, and the specific method is as follows:

[0084] A set of geological structure entities is extracted from all triples, each structure unit generates a unique primary key value in the form of structure name + profile number + position code, a UID number system is generated to assign a uniform node identifier, a mapping table is established, and all nodes in the spatial chain and the time chain are traversed. If the node is derived from the same structure entity, it is replaced with a UID node, and the time attribute / physical attribute is kept as an edge or node attribute, but the UID is kept uniform, and finally a database and a mapping structure are established. The uniform node index database stores the unique UID of the node, the semantic name of the node, the type of the chain to which the node belongs, the time label and the attached physical attribute, and realizes the node consistency management and state tracking function in the geological atlas.

[0085] In this embodiment, the double-chain coupled atlas structure is a heterogeneous graph structure integrated by spatial profile structure chain and time sequence state chain through a uniform node index database, which is used to express the spatial connection relationship of geological structure unit entities and the time state evolution path at the same time in the graph neural network propagation.

[0086] In this embodiment, the heterogeneous graph structure refers to a graph structure composed of multiple types of nodes (such as geological structure units, parameter state nodes) and multiple types of edges (such as spatial connection edges, time evolution edges, attribute binding edges). The heterogeneous graph structure integrates the spatial profile structure chain and the time sequence state chain in a uniform graph structure through a uniform node indexing mechanism, forming a graph neural network input structure with spatio-temporal heterogeneous connection relationship.

[0087] S3, the node state propagation calculation of the node in the double-chain coupled atlas structure is carried out by using the perturbation perception graph neural network mechanism, and the risk state prediction result of the landslide structure unit layer at each time node is obtained.

[0088] In this embodiment S3, the perturbation perception graph neural network mechanism is obtained by introducing a geological disturbance factor as a dynamic adjustment amount on the basis of the graph neural network propagation model, which is used to predict the risk state of the landslide structure unit layer under external disturbance.

[0089] In this embodiment, the disturbance perception graph neural network mechanism extends the structure propagation in traditional graph neural networks to structure-disturbance coupled propagation by introducing disturbance perception modeling capability, realizes the learnable modeling of the evolution of risk state in the graph structure with disturbance, and can not only predict the current state of the node, but also respond to the dynamic changes under different external disturbance scenarios, and is suitable for complex disaster scenarios such as landslides induced by multiple factors.

[0090] In this embodiment, the disturbance perception graph neural network mechanism is used to perform node state propagation calculation on the nodes in the double-chain coupled graph structure to obtain the risk state prediction results of the landslide structure unit layer at each time node, as follows:

[0091] S3.1, set the geological disturbance factor, and add the geological disturbance factor to the graph nodes of the spatial profile structure chain and the time sequence state chain;

[0092] S3.2, define the graph neural network propagation path, including the spatial structure propagation path and the time sequence propagation path;

[0093] S3.3, in the graph neural network propagation process, update the state of each node according to the change of the geological disturbance factor, and obtain the node state embedding vector of each round of propagation;

[0094] S3.4, apply an activation function and a linear mapping to the node state embedding vector to obtain a prediction value vector;

[0095] S3.5, perform prediction value and classification output on the prediction value vector, and map it to each node in the landslide structure unit layer to obtain the risk state prediction result.

[0096] In this embodiment, the geological disturbance factor refers to an external inducing condition or force term that triggers or aggravates the change of the geological structure state, leading to the occurrence of disasters such as landslides, and has a clear physical triggering mechanism and spatio-temporal action characteristics. As a regulating input variable in the disturbance perception graph neural network, it is used to dynamically adjust the risk state propagation path and update amplitude; the geological disturbance factor includes hydrological disturbance, seismic disturbance, engineering disturbance, water storage disturbance, and meteorological / environmental disturbance;

[0097] In this embodiment, the geological disturbance factor can be used as: a component of the node feature vector, a regulating factor of the edge propagation weight, and a risk state classification gating condition;

[0098] The geological disturbance factor is added to the graph nodes of the spatial profile structure chain and the time sequence state chain, in the following specific ways:

[0099] The disturbance factor is taken as a component of the node feature vector: initialize the node attribute, extract the structure and physical parameter feature; call external disturbance monitoring data to obtain the geological disturbance factor of each node; splice the geological disturbance factor into the feature vector of the node and store it in the graph neural network initialization input.

[0100] S4, in combination with the logical causal relationship layer, a reverse causal path chain reconstruction method is used to generate a landslide risk propagation path chain, and the propagation parameters and response rules in the historical landslide graph structure are migrated to the double-chain coupled graph structure, and finally a double-chain coupled spatiotemporal geological disaster knowledge graph is constructed;

[0101] In the present embodiment S4, the reverse causal path chain reconstruction method is realized by fusing the logical causal graph traversal method and the predicted state driven path backtracking mechanism, which is used to identify the triggering process of physical parameter changes and external disturbance sources in reverse from the node as the starting point, and reconstruct the causal propagation path chain of geological disaster events;

[0102] In the present embodiment, the reverse causal path chain reconstruction method is used to generate a landslide risk propagation path chain, and the specific method steps are as follows:

[0103] S4.1.1, set the risk state threshold, and take the node with a predicted value greater than the risk state threshold in the risk state prediction result as the starting point of the landslide risk propagation path chain;

[0104] S4.1.2, extract the disaster causal triplets in the logical causal relationship layer, and establish a directed causal graph;

[0105] The nodes of the directed causal graph are geological structure entities containing geological disturbance factors and physical state parameters, and the edges of the directed causal graph are external disturbance sources in the disaster causal triplets;

[0106] S4.1.3, for each node of the directed causal graph, perform reverse depth-first traversal in the directed causal graph, and record all causal path chains;

[0107] S4.1.4, remove redundant, conflicting and incomplete causal path chains, and the remaining causal path chain set constitutes the landslide risk propagation path chain.

[0108] Wherein, the specific method for removing redundant, conflicting and incomplete causal path chains is as follows:

[0109] Set the minimum path length threshold; set the causal connectivity verification logic for identifying physical mechanism conflict paths; set the path uniqueness standard for determining whether there is a structural redundant path;

[0110] All path chain sets obtained by reverse tracing of the causal diagram are traversed one by one to determine whether the path length is the minimum path length threshold, and if the path is missing an intermediate parameter node, it is determined to be an incomplete structure path and is removed;

[0111] The physical and logical continuity between the nodes in the path is analyzed, and if there is inconsistent logic, it is considered to be a physical mechanism conflict, or if there is a contradictory edge in the path, it is considered to be an anti-logic path and is removed;

[0112] The paths in the path set that are identical in content or have repeated structure nodes but only different lengths are compared, and if there is a long path that contains all the causal segments of a short path, the short path is retained, and if the starting and ending points of two paths are the same but only the number of parameters in the middle is different, the one with more information is retained according to the information entropy priority.

[0113] In this embodiment, the propagation parameters and response rules in the historical landslide graph structure are migrated to the double-chain coupled graph structure, and a double-chain coupled spatiotemporal geological disaster knowledge graph is finally constructed, and the specific operations include:

[0114] S4.2.1, graph embedding representation is performed on the double-chain coupled graph structure, and the structural similarity between the double-chain coupled graph structure and the historical landslide graph structure is calculated, and when the similarity meets a preset threshold, the propagation parameters and response rules in the selected historical landslide graph structure are migrated and injected into the double-chain coupled graph structure;

[0115] S4.2.2, the landslide risk propagation path chain is written into the double-chain coupled graph structure at the same time, and a double-chain coupled spatiotemporal geological disaster knowledge graph is finally constructed.

[0116] In embodiment two, the geological disaster knowledge graph construction system based on artificial intelligence proposed by the present application is applied to the geological disaster knowledge graph construction method based on artificial intelligence proposed in embodiment one, and includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to realize the geological disaster knowledge graph construction method based on artificial intelligence in embodiment one.

[0117] The embodiments of the present application are described in detail above in combination with the drawings, but the present application is not limited thereto, and various changes can be made within the knowledge of those skilled in the art without departing from the purpose of the present application.

Claims

1. A method for constructing a geological disaster knowledge graph based on artificial intelligence, characterized in that, Includes the following steps: S1. Based on the geological disaster investigation report, a semantic-physical collaborative preprocessing method is used to extract the geological semantic triple set and construct a three-layer semantic ontology map containing a landslide structural unit layer, a physical and mechanical parameter layer, and a logical causal relationship layer. S2. Based on the three-layer semantic ontology graph, a dual-chain spatiotemporal graph construction method is used to generate spatial profile structure chains and time series state chains respectively. Then, a graph neural network graph topology modeling mechanism is adopted to integrate the spatial profile structure chains and time series state chains in a heterogeneous graph form to construct a dual-chain coupled graph structure. S3. The node state propagation calculation is performed on the nodes in the double-chain coupled graph structure using the perturbation-sensing graph neural network mechanism to obtain the risk state prediction results of the downslope structural unit layer at each time node. S4. Combining the logical causal relationship layer, the reverse causal path chain reconstruction method is used to generate the landslide risk propagation path chain, and the propagation parameters and response rules in the historical landslide map structure are transferred to the dual-chain coupled map structure, and finally a dual-chain coupled spatiotemporal geological disaster knowledge graph is constructed.

2. The method for constructing a geological disaster knowledge graph based on artificial intelligence according to claim 1, characterized in that: The geological semantic triplet set includes structural configuration triplets, physical parameter triplets, and disaster causality triplets; Among them, the structural configuration triplet is a triplet composed of the spatial relationships between geological structural unit entities. It is used to construct the nodes and structural topological connection edges in the landslide structural unit layer. The specific extraction of the structural configuration triplet is as follows: Construct a semantic template for structural connection, traverse sentences containing geological structure entities, and extract and generate structural configuration triples; The physical parameter triplet is a triplet consisting of a geological structural unit entity, a physical parameter attribute, and a physical parameter value. It is used to construct the binding edges between attribute nodes and structural nodes in the physical and mechanical parameter layer. The disaster causal triple is a triple consisting of the causal relationship between the external disturbance source, physical state parameters, and landslide structural response. It is used to construct the causal edge paths of the logical causal relationship layer. The specific extraction of the disaster causal triple is as follows: By using semantic template matching and numerical attribution identification, physical parameter triplets consisting of geological structural entities, their corresponding physical parameter entities, and values ​​are extracted.

3. The method for constructing a geological disaster knowledge graph based on artificial intelligence according to claim 2, characterized in that: The three-layer semantic ontology graph is constructed from the landslide structural unit layer, the physical and mechanical parameter layer, and the logical causal relationship layer, serving as the basis for constructing the dual-chain coupled graph structure. Among them, the landslide structural unit layer is used to characterize the spatial connection relationship between geological structural entities in the landslide profile; The physical and mechanical parameter layer is used to express the binding relationship between the landslide geological structural unit entity and the physical and mechanical parameters; The logical causal relationship layer is used to express the causal chain path relationship between external disturbance sources, physical state parameters and landslide structural response.

4. The method for constructing a geological disaster knowledge graph based on artificial intelligence according to claim 3, characterized in that: The spatial profile structure chain is a directed graph structure composed of geological structural entities and their spatial connection relationships, used to represent the spatial topological connections between geological structural entities in the landslide profile. A time series state chain is a time series diagram structure composed of the physical parameter attributes of a geological structural entity at different time points and its temporal evolution path. It is used to express the evolution path of the physical parameters of a geological structural entity over time.

5. The method for constructing a geological disaster knowledge graph based on artificial intelligence according to claim 4, characterized in that: In S2, the dual-chain spatiotemporal map construction method is based on the single-layer topological map construction method and combines the landslide structural unit layer and the physical and mechanical parameter layer to construct a spatial profile structural chain and a time series state chain, as detailed below: S2.

1. Traverse all structural configuration triples, take the geological structural unit entities in the structural configuration triples as graph nodes, generate directed edges between graph nodes according to the spatial connection semantics of graph nodes, and obtain the spatial profile structure chain. S2.

2. Extract the physical parameter attributes of the geological structure unit entities in the physical parameter triplet according to the forward time sequence, and construct the physical parameter attributes of the geological structure unit entities as graph nodes. Establish directed edges between adjacent graph nodes according to the time flow direction, and use the difference of the physical parameter values ​​in the physical parameter triplet as the edge weight of the directed edge to finally construct the time series state chain. S2.

3. Based on all geological structural unit entities in the spatial profile structure chain and time series state chain, establish a unified node index database for cross-chain indexing.

6. The method for constructing a geological disaster knowledge graph based on artificial intelligence according to claim 5, characterized in that: The dual-chain coupled graph structure is a heterogeneous graph structure composed of a spatial profile structure chain and a time series state chain integrated through a unified node index database. It is used to simultaneously express the spatial connection relationship and temporal state evolution path of geological structural unit entities in graph neural network propagation.

7. The method for constructing a geological disaster knowledge graph based on artificial intelligence according to claim 6, characterized in that: In S3, the disturbance perception graph neural network mechanism is obtained by introducing a geological disturbance factor as a dynamic adjustment quantity on the basis of the graph neural network propagation model, which is used to predict the risk status of the external disturbance landslide structural unit layer. A perturbation-aware graph neural network mechanism was used to calculate the node state propagation of nodes in the dual-chain coupled graph structure, and the risk state prediction results of the downslope structural unit layer at each time node were obtained as follows: S3.1 Set geological disturbance factors and attach them to the graph nodes of the spatial profile structure chain and the time series state chain; S3.2 Define the propagation path of the graph neural network, including the spatial structure propagation path and the time series propagation path; S3.3 During the propagation of the graph neural network, the state of each node is updated according to the changes in the geological disturbance factor to obtain the node state embedding vector for each round of propagation. S3.4 Apply activation functions and linear mappings to the node state embedding vectors to obtain the predicted value vectors; S3.

5. Calculate the predicted values ​​and classifications of the predicted value vector and map them to each node in the landslide structural unit layer to obtain the risk state prediction results.

8. The method for constructing a geological disaster knowledge graph based on artificial intelligence according to claim 7, characterized in that: In S4, the reverse causal path chain reconstruction method is achieved by integrating the logical causal graph traversal method with the predicted state-driven path backtracking mechanism. It is used to identify the physical parameter changes and external disturbance source triggering process from the node as the starting point, and reconstruct the causal propagation path chain of geological disaster events. A reverse causal path chain reconstruction method is used to generate a landslide risk propagation path chain. The specific steps are as follows: S4.1.1 Set a risk status threshold and take the nodes whose predicted values ​​are greater than the risk status threshold in the risk status prediction results as the starting point of the landslide risk propagation path chain. S4.1.2 Extract the disaster causal triples from the logical causal relationship layer and establish a directed causal graph; The nodes of the directed causal graph are geological structural entities containing geological disturbance factors and physical state parameters, and the edges of the directed causal graph are external disturbance sources in the disaster causal triplet. S4.1.3 For each node in the directed causal graph, perform a reverse depth-first traversal in the directed causal graph and record all causal path chains; S4.1.

4. Remove redundant, conflicting, and incomplete causal path chains. The set of remaining causal path chains constitutes the landslide risk propagation path chain.

9. The method for constructing a geological disaster knowledge graph based on artificial intelligence according to claim 8, characterized in that: The process of transferring propagation parameters and response rules from the historical landslide map structure to the dual-chain coupled map structure, ultimately constructing a dual-chain coupled spatiotemporal geological hazard knowledge graph, includes the following specific operations: S4.2.

1. Perform graph embedding representation on the double-chain coupled graph structure, calculate the structural similarity between it and the historical landslide graph structure. When the similarity meets the preset threshold, transfer the propagation parameters and response rules in the selected historical landslide graph structure into the double-chain coupled graph structure. S4.2.2 Simultaneously, the landslide risk propagation path chain is written into the dual-chain coupled graph structure, and finally a dual-chain coupled spatiotemporal geological disaster knowledge graph is constructed.

10. A geological disaster knowledge graph construction system based on artificial intelligence, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: The processor executes a computer program to implement the method for constructing a geological disaster knowledge graph based on artificial intelligence as described in any one of claims 1-9.

Images