Microseismic early warning multi-level explainable analysis method and system based on knowledge graph

Abstract

The application discloses a microseismic early warning multi-level explainable analysis method and system based on a knowledge graph. The method comprises the following steps: constructing a microseismic space-time graph based on the space-time and energy relationship between microseismic events, constructing a dynamic knowledge graph, embedding and multi-hop propagation to obtain a vector set, inputting a feature set, a space-time graph and a vector set into a knowledge-enhanced graph neural network model to obtain an early warning result, and performing multi-level explainable analysis on the space-time graph and the knowledge graph to obtain an explanation result. Since the microseismic space-time graph and the dynamic knowledge graph are fused, data driving and geological mechanism constraint are combined in depth, the risk evolution characteristics under complex geological conditions are accurately captured, and the multi-level explanation mechanism is used to realize the tracing from a triggering event to a geological cause-effect chain, so that the accuracy, generalization ability and engineering decision credibility of the microseismic early warning are significantly improved.

Description

Technical Field

[0001] This application relates to the field of microseismic monitoring and early warning, and in particular to a multi-level interpretable analysis method and system for microseismic early warning based on knowledge graphs. Background Technology

[0002] Microseismic monitoring and early warning technology is widely used in mining, underground engineering, shale gas development, and geological disaster prevention. Its core objective is to identify and issue graded warnings of high-risk destructive events in their early stages by analyzing source parameters, energy evolution, and spatiotemporal distribution characteristics. With the improvement of monitoring equipment accuracy, microseismic data exhibits characteristics of high frequency, multi-dimensionality, and strong spatiotemporal correlation, making it difficult for traditional methods based on statistical features or shallow machine learning to fully explore the complex nonlinear relationships between events. Furthermore, the "black box" nature of deep learning-based early warning systems makes it difficult for on-site engineers and decision-makers to fully trust the warning results. It cannot clearly identify the key microseismic events leading to the warning and their action paths, resulting in a lack of traceability and interpretability. This reduces the credibility and usability of the model in actual safety decision-making, limiting its engineering application in high-risk scenarios. Therefore, in mining operations with extremely high safety requirements, improving the interpretability of microseismic early warning results has become a key technical requirement to ensure the reliability of early warning results and the feasibility of engineering applications.

[0003] Existing interpretable machine learning methods, such as ensemble learning models like XGBoost, typically rely on feature aggregation of microseismic event windows, compressing multiple events into fixed-dimensional statistical vectors (e.g., mean, maximum, and energy accumulation). While this improves predictive ability to some extent, it inevitably loses information about the order and spatial topology of the events. Furthermore, at the interpretive level, these methods can only output "feature importance" and cannot pinpoint specific key microseismic events and their interaction mechanisms. Summary of the Invention

[0004] The main purpose of this application is to provide a multi-level interpretable analysis method and system for microseismic early warning based on knowledge graphs, which aims to solve the technical problems of low accuracy and reliability of microseismic early warning due to existing technologies ignoring the structured relationship between microseismic events, lacking physical mechanism constraints and multi-source geological knowledge fusion, and having coarse model interpretation granularity.

[0005] To achieve the above objectives, this application provides a multi-level interpretable analysis method for microseismic early warning based on knowledge graphs, the method comprising the following steps: The raw microseismic monitoring data and geological prior data collected in the target monitoring area are subjected to structured preprocessing to obtain a microseismic event feature set and geological knowledge source data. The geological prior data includes fault data, rock strata data, stress zone data, goaf data, historical earthquake data, and mining disturbance data. A microseismic spatiotemporal map is constructed based on the spatial proximity, temporal decay, and energy propagation relationships among the microseismic events in the microseismic event feature set. A dynamic knowledge graph is constructed based on the geological knowledge source data, and knowledge embedding and multi-hop relationship propagation are performed on the dynamic knowledge graph to obtain a knowledge graph embedding vector set. The microseismic event feature set, the microseismic spatiotemporal map, and the knowledge graph embedding vector set are input into a pre-constructed risk propagation model to obtain microseismic risk warning results. The risk propagation model is a graph neural network model constructed based on a knowledge-enhanced attention mechanism. The microseismic risk warning results include the probability of disaster and the risk level. Based on the microseismic risk warning results, a multi-level interpretable analysis is performed on the microseismic spatiotemporal map and the dynamic knowledge graph to obtain multi-level warning interpretation results.

[0006] Furthermore, to achieve the above objectives, this application also proposes a knowledge graph-based multi-level interpretable analysis system for microseismic early warning, which applies the knowledge graph-based multi-level interpretable analysis method for microseismic early warning described above. The knowledge graph-based multi-level interpretable analysis system for microseismic early warning includes: The data processing module is used to perform structured preprocessing on the raw microseismic monitoring data and geological prior data collected in the target monitoring area to obtain microseismic event feature sets and geological knowledge source data. The geological prior data includes fault data, rock strata data, stress zone data, goaf data, historical earthquake data, and mining disturbance data. The spatiotemporal map construction module is used to construct a microseismic spatiotemporal map based on the spatial proximity, time decay, and energy propagation relationships among the microseismic events in the microseismic event feature set. The knowledge graph construction module is used to construct a dynamic knowledge graph based on the geological knowledge source data, and to perform knowledge embedding and multi-hop relationship propagation on the dynamic knowledge graph to obtain a knowledge graph embedding vector set; The microseismic risk prediction module is used to input the microseismic event feature set, the microseismic spatiotemporal map, and the knowledge graph embedding vector set into a pre-constructed risk propagation model to obtain microseismic risk warning results. The risk propagation model is a graph neural network model constructed based on a knowledge-enhanced attention mechanism. The microseismic risk warning results include the probability of disaster and the risk level. The multi-level analysis module is used to perform multi-level interpretable analysis on the microseismic spatiotemporal map and the dynamic knowledge graph based on the microseismic risk warning results, and obtain multi-level warning interpretation results.

[0007] Furthermore, to achieve the above objectives, this application also proposes a multi-level interpretable analysis device for microseismic early warning, the device comprising: a memory, a processor, and a multi-level interpretable analysis program for microseismic early warning stored in the memory, the processor being used to run the multi-level interpretable analysis program for microseismic early warning, the computer program being configured to implement the steps of the knowledge graph-based multi-level interpretable analysis method for microseismic early warning as described above.

[0008] This application constructs a microseismic spatiotemporal map, preserving the temporal sequence, spatial topology, and energy propagation relationships between microseismic events, thus avoiding information loss caused by traditional window feature aggregation. Simultaneously, it introduces a knowledge-enhanced graph neural network, enabling the model to more fully learn the complex nonlinear correlations between microseismic events, thereby improving the accuracy and stability of early warning. By integrating prior knowledge such as faults, strata, stress zones, goaf areas, and historical earthquake cases through a dynamic knowledge graph, the model not only relies on observational data but is also constrained by geological structures and stress evolution laws, thereby improving its adaptability to different mining areas, different mining stages, and complex geological conditions. Based on geological knowledge, a dynamic knowledge graph is constructed and multi-hop propagation is performed, transforming geological structures and dynamic mining disturbances into computable mechanistic constraint vectors. A knowledge-enhanced attention mechanism is used for risk propagation prediction, achieving a deep integration of observational data-driven and geological mechanism-constrained approaches. This significantly improves the accuracy and cross-regional generalization ability of disaster prediction in complex environments. Combined with early warning results, multi-level interpretable analysis is performed to accurately trace back the triggering events, propagation subgraphs, and geological causal chains. This application enables traceable analysis of key microseismic events, key propagation subgraphs, cluster links, fault propagation paths, and geological causal chains, significantly improving the credibility and engineering usability of early warning results in mine and underground engineering safety decision-making. Attached Figure Description

[0009] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0010] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1This is a flowchart illustrating the first embodiment of the knowledge graph-based multi-level interpretable analysis method for microseismic early warning in this application. Figure 2 This is a flowchart illustrating the second embodiment of the knowledge graph-based multi-level interpretable analysis method for microseismic early warning in this application. Figure 3 This is a flowchart illustrating the third embodiment of the knowledge graph-based multi-level interpretable analysis method for microseismic early warning in this application. Figure 4 This is a schematic diagram of the module structure of the knowledge graph-based multi-level interpretable analysis system for microseismic early warning in this application embodiment; Figure 5 This is a schematic diagram of the hardware operating environment involved in the knowledge graph-based multi-level interpretable analysis method for microseismic early warning in the embodiments of this application.

[0012] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0013] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0014] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0015] It should be noted that the executing entity of this application embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a computer or server, or an electronic device capable of performing the above functions. The following description uses a multi-level interpretable analysis device for microseismic early warning (hereinafter referred to as microseismic early warning analysis device) as an example to illustrate this embodiment and the following embodiments.

[0016] It should be noted that the technical problems to be solved by the embodiments of this application include: 1. Ignore the structured relationships between events: Existing methods based on XGBoost, Random Forest, or other traditional machine learning typically aggregate multiple microseismic events within a certain time window to form fixed-dimensional feature vectors such as mean, maximum, cumulative energy, and event frequency. While these methods facilitate modeling, they weaken or even lose the order of occurrence, spatial proximity, energy transfer relationships, and cluster expansion characteristics among microseismic events, making it difficult to characterize the propagation process of microseismic events from local disturbances to regional ruptures.

[0017] 2. The model has limited interpretability and coarse granularity: Traditional interpretable methods often remain at the level of feature importance, such as only indicating that indicators like magnitude, energy, and stress drop contribute significantly to prediction results. However, they cannot further answer engineering questions such as "which micro-seismic events triggered the warning," "what paths did the events propagate through," and "whether there are cluster propagation links or fault propagation channels." Therefore, existing interpretation results are insufficient to meet the needs of refined identification of key triggering events, key propagation edges, and key risk areas in on-site safety decision-making.

[0018] 3. Lack of physical mechanism constraints, resulting in insufficient generalization ability: While existing deep learning or graph neural network methods can learn the nonlinear relationships between microseismic events, most are still primarily driven by observational data and lack explicit modeling of geological mechanisms such as fault structure, rock strata distribution, stress transmission, mining disturbance, and energy propagation. When geological conditions, mining stages, or monitoring environments change in the mining area, the models are prone to distribution shifts, resulting in insufficient stability and cross-regional adaptability of early warning results.

[0019] 4. Lack of integration and utilization of multi-source knowledge: The information required for microseismic early warning includes not only monitoring data such as focal location, magnitude, energy, and stress drop, but also multi-source knowledge such as faults, rock strata, goaf areas, stress concentration zones, historical earthquake cases, and expert experience rules. Existing methods typically only utilize numerical observation data and fail to transform the aforementioned structured geological knowledge into computable, inferable, and fusionable model inputs, resulting in insufficient model cognition of complex geological backgrounds.

[0020] To address the shortcomings of the existing technology, the embodiments of this application aim to solve the following key technical problems: 1. How to construct a structured modeling method that can preserve the temporal, spatial, and energy propagation relationships of microseismic events. This application's embodiments aim to avoid information loss caused by traditional window-level feature aggregation. Individual microseismic events are treated as graph nodes, and a weighted spatiotemporal graph is constructed based on spatial distance, temporal decay, and energy coupling relationships to fully express the triggering, propagation, and correlation mechanisms between microseismic events.

[0021] 2. How to transform prior knowledge such as geological structure, stress evolution, and fault information into computable and reasonable model inputs. This application's embodiments require constructing a knowledge graph for microseismic early warning. Objects such as faults, rock strata, stress zones, goaf areas, and historical earthquake cases are modeled as entities, and the relationships of stress transmission, induction, rupture, and attribution are modeled as relations. Through knowledge graph embedding, geological knowledge is transformed into vector representations that can be integrated with microseismic observation features. Furthermore, a time-dependent relationship evolution function is introduced, enabling the relationship vectors to dynamically update with mining time, stress disturbances, and changes in geological structure, thereby more realistically depicting the evolution process of microseismic risk propagation channels under the influence of mining.

[0022] 3. How to construct a knowledge-enhanced graph neural network model that combines data-driven and knowledge-driven approaches. This application embodiment requires fusing microseismic physical features with knowledge graph embeddings to construct a unified node representation. Through a knowledge-enhanced attention mechanism, data similarity and geological knowledge constraints are considered simultaneously during the graph neural network propagation process, enabling the model to learn complex nonlinear correlations between microseismic events under conditions consistent with geological mechanisms.

[0023] 4. How to achieve multi-level fine-grained interpretation of microseismic early warning results. The embodiments of this application need to not only output the probability of disaster or risk level, but also identify key microseismic events through SHAP, extract microseismic propagation subgraphs, cluster links and fault propagation paths through GNNExplainer, and interpret the geological causal chain through knowledge graph path probabilities, thereby forming a three-level interpretation system of "event-structure-mechanism".

[0024] Based on this, the first embodiment of this application provides a multi-level interpretable analysis method for microseismic early warning based on knowledge graphs, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the knowledge graph-based multi-level interpretable analysis method for microseismic early warning in this application.

[0025] In this embodiment, the knowledge graph-based multi-level interpretable analysis method for microseismic early warning includes: Step S10: Perform structured preprocessing on the raw microseismic monitoring data and geological prior data collected in the target monitoring area to obtain microseismic event feature sets and geological knowledge source data.

[0026] It should be noted that geological prior data refers to data obtained in advance related to geological structure and mining environment. Geological prior data includes fault data, rock strata data, stress zone data, goaf data, historical earthquake data, and mining disturbance data.

[0027] It should be noted that the target monitoring area refers to the spatial area where microseismic risk monitoring and early warning analysis are required. This area can be a mining area, working face, surrounding rock area of ​​a roadway, adjacent area of ​​a goaf, or area affected by a fault. The raw microseismic monitoring data can be event records collected by microseismic sensors, geophones, or underground monitoring stations, including data such as event occurrence time, source location, magnitude, energy, waveform parameters, stress drop, and source radius.

[0028] Understandably, since microseismic events are discrete-time and spatially distributed event streams, this step transforms the raw microseismic monitoring data from discrete records into a structured representation, providing a unified input foundation for subsequent graph structure modeling and knowledge graph fusion. Specifically: Let the set of microseismic events be: in, This is the complete set of microseismic events (all observed microseismic records). This is the i-th microseismic event; This represents the total number of observed microseismic events.

[0029] Each event characteristic is defined as: in, Let x, y, and z be the three-dimensional spatial locations of the i-th microseismic event, respectively, and represent the location of the seismic source. This is the timestamp of the microseismic event. This is the magnitude information; The remaining physical characteristic vectors include P-wave energy, S-wave energy, total energy, energy ratio (S / P), P-wave volume change, S-wave volume change, volume change magnitude, P-wave source radius, S-wave source radius, source radius, static stress drop, apparent stress, apparent volume, average displacement, dynamic stress drop, slip velocity, etc.

[0030] Step S20: Construct a microseismic spatiotemporal map based on the spatial proximity, time decay, and energy propagation relationships among the microseismic events in the microseismic event feature set.

[0031] It should be noted that spatial proximity refers to the closeness of two microseismic events in three-dimensional space. Temporal decay refers to the stronger the likelihood of triggering or correlation between two microseismic events as their occurrence times are closer. Energy propagation refers to the stronger influence that high-energy microseismic events may have on subsequent nearby events. A microseismic spatiotemporal map is a graph-structured data structure built with microseismic events as nodes and spatiotemporal and energy correlations between events as edges.

[0032] It is understandable that microseismic events are essentially complex coupled systems that spread spatially, are triggered temporally, and transfer energy. This step constructs a weighted spatiotemporal graph that integrates spatial distance, temporal decay, and energy propagation mechanisms to achieve a structured expression of the physical relationships between microseismic events, providing a basic topological support for subsequent risk prediction and interpretable analysis based on graph neural networks.

[0033] Step S30: Construct a dynamic knowledge graph based on the geological knowledge source data, and perform knowledge embedding and multi-hop relationship propagation on the dynamic knowledge graph to obtain a knowledge graph embedding vector set.

[0034] It should be noted that a dynamic knowledge graph refers to a knowledge network that can express geological objects, geological relationships, and their updates as they change due to mining disturbances. Geological objects can be faults, rock strata, stress concentration zones, goafs, historical earthquake records, or mining-disturbed areas. Geological relationships can include those related to adjacency, crossing, induction, stress transfer, attribution, and influence.

[0035] It should be noted that knowledge embedding refers to converting entities and relationships in a knowledge graph into vector representations. Multi-hop relationship propagation refers to transmitting indirect influences between distant entities to a target entity or target region through one or more layers of relationship connections.

[0036] Understandably, this step involves constructing a dynamic knowledge graph to explicitly introduce geological mechanisms into the model, enabling the model to understand geological structures.

[0037] Step S40: Input the microseismic event feature set, the microseismic spatiotemporal map, and the knowledge graph embedding vector set into the pre-constructed risk propagation model to obtain the microseismic risk warning result.

[0038] It should be noted that the risk propagation model is a graph neural network model built on a knowledge-enhanced attention mechanism. The knowledge-enhanced attention mechanism refers to an attention calculation method that considers both the similarity of event data and the correlation of geological knowledge when calculating the influence weights between microseismic events.

[0039] It should be noted that the microseismic risk early warning results include both the probability of disaster and the risk level. The probability of disaster refers to the probabilistic result of high-risk events such as rockbursts, rock bursts, and local rupture propagation occurring in the target monitoring area. The risk level refers to the result of low risk, medium risk, high risk, or other levels obtained according to preset classification rules.

[0040] Understandably, the purpose of this step is to combine the characteristics of microseismic events themselves, the propagation structure between events, and geological knowledge constraints for risk assessment, so that the model can both learn the statistical correlation of microseismic activity and conduct risk propagation analysis in conjunction with the geological structure background.

[0041] In some embodiments, the physical characteristics of each microseismic event can be fused with the corresponding knowledge graph embedding vector to form a node input representation. Then, the information propagation range between nodes can be determined using the microseismic spatiotemporal graph. During processing, the risk propagation model can assign different attention weights based on the similarity of event features, the strength of spatial connectivity, and the strength of geological relationships. For example, if two microseismic events are close in time and space and located within the same fault influence zone, the model can assign them higher information propagation weights; if two events are spatially close but geologically separated by stable rock layers, the model can reduce the propagation impact between them. After multi-level graph propagation and graph-level aggregation, the model outputs the catastrophic probability of the target monitoring area and converts it into the corresponding risk level based on the catastrophic probability or a preset risk threshold.

[0042] Step S50: Based on the microseismic risk warning results, perform multi-level interpretable analysis on the microseismic spatiotemporal map and the dynamic knowledge graph to obtain multi-level warning interpretation results.

[0043] It should be noted that multi-level interpretable analysis refers to the process of interpreting risk warning results from different levels, which can include event-level interpretation, structural-level interpretation, and mechanism-level interpretation. Event-level interpretation explains which microseismic events contribute significantly to risk assessment. Structural-level interpretation explains which events may connect to or propagate through structures along which risks may extend. Mechanism-level interpretation explains the geological causes of risk propagation related to faults, stress zones, goafs, or rock structures. The results of multi-level warning interpretation can include key microseismic events, key propagation relationships, risk propagation sub-regions, geological correlation paths, and textual explanatory information.

[0044] In some embodiments, the microseismic early warning analysis device can combine SHAP to achieve event-level contribution analysis, use GNNExplainer to extract key propagation subgraphs, and achieve geological mechanism-level causal explanation based on knowledge graph path probability. This enables a unified mapping from microseismic prediction results to a three-layer interpretable system of "event-structure-mechanism", significantly improving the credibility and engineering usability of the model in high-risk geological environments.

[0045] This embodiment improves the availability of multi-source data through structured preprocessing, preserves the spatial, temporal, and energy correlations between events through microseismic spatiotemporal maps, introduces geological structure and mining disturbance constraints through dynamic knowledge graphs, fuses data propagation relationships and geological knowledge relationships through knowledge-enhanced graph neural networks, and outputs risk sources, propagation structures, and geological causes through multi-level interpretable analysis. Therefore, this embodiment combines data perception, structural modeling, knowledge constraints, risk prediction, and result interpretation into a continuous microseismic early warning analysis process, which can effectively mine the propagation characteristics between microseismic events, improve the stability of risk identification under complex geological conditions, and enhance the interpretability and credibility of early warning results in engineering applications.

[0046] Based on the first embodiment described above, in the second embodiment of this application, the same or similar content as the first embodiment can be referred to the above description, and will not be repeated hereafter.

[0047] Based on this, please refer to Figure 2 , Figure 2 This is a flowchart illustrating the second embodiment of the knowledge graph-based multi-level interpretable analysis method for microseismic early warning. In this embodiment, step S20 further includes: Step S201: Extract the spatial proximity relationship, time decay relationship and energy propagation relationship of each microseismic event pair based on the microseismic event feature set.

[0048] It should be noted that a microseismic event pair refers to an analysis object consisting of two microseismic events, used to determine whether there is a spatial, temporal, or energy correlation between the two microseismic events.

[0049] It should be understood that this embodiment expands a single microseismic event into an event pair analysis object, so that subsequent graph construction no longer only deals with isolated events, but can obtain a candidate correlation basis between events.

[0050] For example, in the case of multiple microseismic events that occur consecutively in a short period of time within the same mining area, two adjacent events can be considered as a pair of microseismic events, or two events located within the same fault influence zone can be considered as a pair of microseismic events.

[0051] Step S202: Determine the spatial proximity weights of each microseismic event pair based on the spatial proximity relationship, and generate a spatial proximity weight set.

[0052] It should be noted that spatial proximity weight refers to a numerical value used to characterize the strength of spatial correlation between two microseismic events. A spatial proximity weight set is a dataset consisting of the spatial proximity weights of multiple microseismic event pairs.

[0053] It should be understood that this embodiment converts three-dimensional spatial distance into weighted data that can participate in the calculation of graph edge weights, so that events close to the same rupture area, fault zone or mining disturbance area can have a higher degree of correlation in the graph structure.

[0054] Step S203: Determine the time decay weight of each microseismic event pair based on the time decay relationship, and generate a time decay weight set.

[0055] It should be noted that time decay weights are numerical values ​​used to characterize the strength of the correlation between two microseismic events in terms of their occurrence time. A time decay weight set is a dataset consisting of the time decay weights of multiple microseismic event pairs.

[0056] It should be understood that this embodiment incorporates the order of occurrence and time interval between microseismic events into the graph edge weight calculation, enabling the microseismic spatiotemporal graph to express the event triggering and continuous evolution relationship.

[0057] Step S204: Determine the energy propagation weights of each microseismic event pair based on the energy propagation relationship, and generate an energy propagation weight set.

[0058] It should be noted that energy propagation weights are numerical values ​​used to characterize the degree of energy influence between two microseismic events. An energy propagation weight set is a dataset consisting of the energy propagation weights of multiple microseismic event pairs.

[0059] It should be understood that this embodiment incorporates the impact of high-energy microseismic events on subsequent events into the graph structure modeling, enabling the microseismic spatiotemporal graph to express the characteristics of energy release and energy transfer.

[0060] Step S205: Merge the spatial proximity weight set, the time decay weight set, and the energy propagation weight set to obtain the event association weight set.

[0061] It should be noted that event correlation weights refer to the correlation strength of microseismic event pairs obtained after considering spatial, temporal, and energy factors. The event correlation weight set is a dataset composed of the event correlation weights of multiple microseismic event pairs.

[0062] It should be understood that this embodiment unifies the three types of heterogeneous relationships into graph edge weights, thus avoiding insufficient association judgment caused by relying solely on a single spatial distance or a single time window.

[0063] Step S206: Generate a microseismic spatiotemporal map with microseismic events as nodes based on the microseismic event feature set and the event association weight set.

[0064] In practical implementation, the microseismic spatiotemporal diagram is defined as: in, This is a spatiotemporal diagram of microseismic events. This is the set of nodes, i.e., the set of microseismic events; The set of edges represents the relationships between events; It is an adjacency matrix.

[0065] Each node in the microseismic spatiotemporal diagram represents a microseismic event: in, The i-th node in the graph corresponds to the microseismic event. .

[0066] In the microseismic spatiotemporal graph, edges represent the existence of influence relationships between events, and edge weights represent the degree of influence. Therefore, edge weights, combined with time, space, and energy, construct a physically constrained event influence network. Specifically: in, , Indicate the spatial location of events i and j; Indicates spatial distance; This is the squared term of the spatial influence scale parameter; This is a spatial decay term; the closer the event space is, the higher its weight. , Let i and j be the times when events occur. Indicates time difference; This is the squared term of the time decay coefficient; This is a time decay term; the closer the time, the stronger the correlation. , Let represent the energies of events i and j, indicating that higher-energy events have a greater impact; This represents a small constant that prevents division by zero. This is an energy transfer term.

[0067] This embodiment extracts three basic associations—space, time, and energy—by event pair relationship extraction. It then quantifies spatial proximity, temporal continuity, and energy propagation factors through three types of weight calculations. By fusing these weights, a unified event association weight is formed, and a microseismic spatiotemporal map is generated using these event association weights. This transforms discrete microseismic events into graph data with propagation structures, preserving the spatiotemporal coupling and energy associations between events, and providing structured input for subsequent risk propagation models.

[0068] Based on the above embodiments, please refer to Figure 3 , Figure 3 This is a flowchart illustrating the third embodiment of the knowledge graph-based multi-level interpretable analysis method for microseismic early warning in this application. In this third embodiment, step S30 further includes: Step S301: Extract the set of geological entities and the set of geological relationships from the geological knowledge source data.

[0069] It should be noted that the set of geological entities includes fault entities, rock strata entities, stress zone entities, goaf entities, historical earthquake entities, and mining disturbance entities, while the set of geological relationships includes stress transmission relationships, induced relationships, fracture relationships, attribution relationships, and disturbance evolution relationships.

[0070] Step S302: Generate a set of geological knowledge triples based on the set of geological entities and the set of geological relationships.

[0071] It should be noted that each geological knowledge triple in the geological knowledge triple set includes a head entity, a geological relation, and a tail entity.

[0072] It should be noted that relational state evolution refers to updating the state or intensity of geological relations based on time, mining disturbances, and stress changes.

[0073] In its specific implementation, a knowledge graph is defined as: in, For dynamic knowledge graphs; It is a collection of entities that represent geological objects, such as faults, rock strata, stress zones, and goaf areas; This is a set of relationships, representing geological processes such as stress transfer, induction, and fracturing. It is a set of triples, namely, structured knowledge of entity-relation-entity.

[0074] The triplet form is as follows: in, The head entity (such as fault A) represents the initial geological object; For relations (such as "inducing"), it indicates the type of action; The tail entity (such as microseismic zone B) represents the target object.

[0075] Step S303: Based on the mining disturbance data, the establishment time of the geological relationship, and the stress zone data, perform relationship state evolution on the geological relationships in the geological knowledge triplet set to obtain a dynamic geological knowledge triplet set.

[0076] It should be noted that relational state evolution refers to updating the state or intensity of geological relations based on time, mining disturbances, and stress changes.

[0077] Understandably, since traditional knowledge graph relationships are static and cannot reflect geological evolution, this method introduces a time-dependent relationship evolution mechanism to characterize the dynamic characteristics of geological structures as they evolve with mining and stress, thus expanding the static knowledge graph into a dynamic one. The relationship vector is defined as follows: in, for The time-relation vector represents The state of geological relationships at any given time; This is the initial relation vector, representing the initial geological structure and stress relationship; This is the attenuation coefficient, used to control the "aging rate" of the relationship; The current time; This refers to the initial time, i.e., the moment the relationship is established. The external disturbance term represents the impact of mining / stress disturbance. This formula shows that geological relationships are not fixed structures, but rather update over time and with external disturbances, such as the gradual reduction of the influence of old faults and stress redistribution caused by mining.

[0078] Based on this, the triplet is expanded to This enables knowledge graphs to characterize the time-varying properties of geological relationships and further influence the propagation process of graph neural networks.

[0079] Step S304: Based on the dynamic geological knowledge triplet set, perform vector initialization on the geological entity set and the geological relation set to obtain the initial entity vector set and the initial relation vector set.

[0080] Understandably, in order for the neural network to understand entities such as faults, stresses, and rock strata, they need to be transformed into points in a numerical vector space. Therefore, the initialization vector for each entity is: in, Let i be the initial vector of the i-th entity; It follows a Gaussian distribution; Let Variance be the variance.

[0081] Step S305: Based on the dynamic geological knowledge triplet set, the initial entity vector set, and the initial relation vector set, perform relation constraint learning to obtain relation constraint embedding data.

[0082] It should be noted that relation constraint learning refers to the process of training and updating entity vectors and relation vectors based on the matching relationships between the head entity, relation, and tail entity in the geological knowledge triplet. Relation constraint embedded data refers to the entity vectors and relation vectors obtained after relation constraint learning.

[0083] Furthermore, to improve the consistency of geological relationships in the knowledge embedding results, step S305 above may include: Step S3051: Extract the target geological knowledge triple from the dynamic geological knowledge triple set to obtain the target head entity, target dynamic relationship and target tail entity; Step S3052: Extract the head entity vector corresponding to the target head entity and the tail entity vector corresponding to the target tail entity from the initial entity vector set to obtain the target entity vector group; Step S3053: Extract the relation vector corresponding to the target dynamic relation from the initial relation vector set to obtain the target relation vector; Step S3054: Calculate the triplet matching deviation based on the target entity vector group and the target relation vector to obtain relation constraint deviation data; Step S3055: Update the initial entity vector set and the initial relation vector set based on the relation constraint deviation data, and use the updated entity vector set and the updated relation vector set as relation constraint embedding data.

[0084] Understandably, microseismic early warning analysis equipment calculates triplet matching bias based on the distance or similarity between the head entity vector, the target relation vector, and the tail entity vector. For triplets based on real geological knowledge, the equipment tends to minimize the matching bias; for negative samples constructed by replacing the head or tail entities, the equipment tends to maintain a larger matching bias. Relationship constraint bias data can include positive sample bias, negative sample bias, and positive-negative sample interval bias.

[0085] In its implementation, the microseismic early warning analysis equipment uses the TransE core formula for relational modeling: in, This is the head entity vector, representing the starting fault, rock mass, etc.; As a relation vector, it represents Stress transmission and induction at any given moment; This is the tail entity vector, i.e., the endpoint structure.

[0086] The loss function for this part is: in, For knowledge graph training loss; These are the head entity vector, relation vector, and tail entity vector of the triplet, respectively. It is the squared L2 norm (Euclidean distance).

[0087] Step S306: Perform multi-hop relation propagation based on the relation constraint embedding data, and extract the embedding vectors corresponding to the microseismic event nodes according to the mapping relationship between the microseismic event nodes and geological entities to obtain the knowledge graph embedding vector set.

[0088] It should be noted that the mapping relationship refers to the correspondence between microseismic event nodes and geological entities, which can be determined based on the location of the microseismic event, the fault to which it belongs, the stress zone it is located in, or the adjacent goaf. The knowledge graph embedding vector set refers to the set of geological knowledge vectors corresponding to the microseismic event nodes.

[0089] It should be understood that this embodiment transmits multi-hop geological influences from the geological knowledge graph to microseismic event nodes, enabling subsequent risk propagation models to utilize knowledge information such as fault connectivity, stress transmission, and mining disturbances.

[0090] In the specific implementation, when node i updates its state, it does not only look at itself, but also considers the influence of all neighbors and geological relationships. The specific relationship propagation is as follows: in, For node i, the first Layered entity representation; For node j, the first Layered entity representation; For the set of neighboring nodes; This is the normalization factor (sum of neighbor weights); For activation functions; For a moment In this context, the relationship vector from i to j represents the geological interaction between i and j. This multi-hop propagation formula, through joint modeling of the states of neighboring nodes and the relationship vector, realizes the recursive information propagation of microseismic events within the spatiotemporal graph structure, enabling the model to capture the energy transfer and catastrophic evolution process of microseismic events under the constraints of geological structures. From a physical perspective, this can be intuitively understood as the degree of danger of a microseismic event being jointly determined by all surrounding events and geological structures; node updates are equivalent to "the rupture risk at a certain location = the combined effect of all surrounding microseismic activities + geological channels + propagation paths".

[0091] After K layers of propagation, the output is: in, Embed the knowledge graph of the i-th microseismic node; The entity vector after propagation through the K layers; This represents the number of propagation layers.

[0092] By performing triplet structure modeling, entity vector initialization, relation constraint learning, and multi-hop propagation reasoning on the knowledge graph, an entity representation reflecting geological semantics and structural information is finally obtained. This representation can then be used to construct a unified node representation space after being fused with microseismic observation features, thereby achieving microseismic risk modeling that combines data-driven and mechanism-driven approaches.

[0093] This embodiment forms a knowledge graph foundation by extracting geological entities and geological relationships, generates a unified geological knowledge expression through triples, introduces mining disturbances and stress changes through relational state evolution, obtains computable knowledge representations through vector initialization and relational constraint learning, and transmits indirect geological influences to microseismic event nodes through multi-hop relational propagation. Therefore, it can transform geological prior knowledge into embedded vectors that can be fused with microseismic event characteristics, providing geological mechanism constraints for risk propagation models.

[0094] Based on the above embodiments, in the fourth embodiment of the knowledge graph-based multi-level interpretable analysis method for microseismic early warning in this application, step S40 may include: Step S401: Based on the microseismic event node identifiers, perform node alignment and vector fusion on the microseismic event feature set and the knowledge graph embedding vector set to obtain a node fusion input set; Step S402: Based on the microseismic spatiotemporal diagram, determine the set of neighboring nodes for each microseismic event node to obtain the node adjacency relationship set; Step S403: Input the node fusion input set and the node adjacency relationship set into the pre-constructed risk propagation model to obtain the microseismic risk warning result.

[0095] It should be noted that the risk propagation model is a graph neural network built on a knowledge-enhanced attention mechanism, including a knowledge-enhanced attention layer, a graph propagation layer, and a graph-level prediction layer; The knowledge-enhanced attention layer is configured to calculate attention weights based on the data similarity between microseismic event nodes and the geological relationship scores in the dynamic knowledge graph, thereby obtaining a knowledge-enhanced attention weight set. The graph propagation layer is configured to perform neighborhood-weighted propagation based on the node fusion input set and the knowledge-enhanced attention weight set, and output a high-level node representation set. The graph-level prediction layer is configured to perform graph-level aggregation and risk mapping on the high-level representation set of the nodes, and output microseismic risk warning results.

[0096] Understandably, the graph-level risk prediction of the risk propagation model is represented as follows: in, This is a graph-level feature vector (state encoding of the entire microseismic region). This represents the number of nodes in the graph (total number of microseismic events). Let i be the node of the i-th microseismic event; The representation of the i-th microseismic event node after processing by the graph neural network in the L-th layer contains information such as geological knowledge, physical observation data, and graph neural network propagation. This represents the number of propagation layers in the graph neural network. This formula ultimately yields a feature representation of the overall state of the region.

[0097] The probability prediction formula is: in, To predict probabilities; Microseismic image; Catastrophic events (such as rock bursts); These are graph-level feature vectors; This is a risk prediction weight matrix; For bias terms; This indicates the use of the sigmoid function for probability mapping, outputting the catastrophe probability. This formula maps complex states to a risk space, representing a high-dimensional graph. Convert it into a risk scalar.

[0098] This embodiment constructs a regional overall state vector by performing graph-level aggregation of the high-level semantic representations of all microseismic nodes, and outputs the probability of disaster occurrence through a nonlinear mapping function, thereby achieving a holistic and probabilistic prediction of microseismic risks in complex geological environments and providing a quantitative basis for engineering safety decisions.

[0099] Furthermore, to improve the risk prediction performance of the model, the following may be included before step S40: Step S41: Construct a model training sample set based on the historical microseismic event feature set, historical microseismic spatiotemporal map, historical knowledge graph embedding vector set, and historical risk labels.

[0100] It should be noted that historical risk labels refer to risk outcome markers obtained based on historical microseismic disaster records, on-site response records, or expert annotations, and may include whether a disaster has occurred, the risk level, or the category of high-risk events. The model training sample set refers to the dataset used to train the risk propagation model.

[0101] In some embodiments, the microseismic early warning analysis device divides training samples according to historical time windows, historical mining areas, or historical working faces, and stores the historical microseismic event feature set, historical microseismic spatiotemporal map, historical knowledge graph embedded vector set, and historical risk label corresponding to each sample in association. For historical samples with incomplete risk labels, supplementary annotations can be made based on field records, energy anomaly levels, or expert review results. After completing the data association, the model training sample set is obtained.

[0102] Step S42: Construct node fusion training samples based on the historical microseismic event feature set and the historical knowledge graph embedding vector set to obtain a node fusion training sample set.

[0103] It should be noted that the node fusion training samples refer to the data samples used to train the input layer of the model nodes, which are obtained by fusing the feature vectors of historical microseismic events and the embedding vectors of historical knowledge graphs.

[0104] In some embodiments, the microseismic early warning analysis device performs node matching on the historical microseismic event feature set and the historical knowledge graph embedded vector set based on the historical microseismic event node identifiers. The matched vectors are then concatenated, normalized, and their dimensions transformed to generate node fusion training samples. For cases where the feature dimensions differ between different historical mining areas, standardization can be performed first, followed by vector fusion. After all historical samples have been processed, a node fusion training sample set is obtained.

[0105] Step S43: Based on the historical microseismic spatiotemporal map and the historical knowledge graph embedding vector set, configure a knowledge-enhanced attention layer to obtain the initial attention propagation network.

[0106] It should be noted that the initial attention propagation network refers to a network structure that contains knowledge-enhanced attention computation rules but has not yet been trained.

[0107] In some embodiments, the microseismic early warning analysis device reads historical node adjacency relationships and historical edge weights based on historical microseismic spatiotemporal maps, and calculates geological relationship scores between adjacent nodes based on the embedded vector set of the historical knowledge graph. Subsequently, the device sets the input dimension, attention parameters, and normalization method of the knowledge-enhanced attention layer, and combines the data similarity calculation branch and the geological relationship scoring branch into an initial attention propagation network. This initial attention propagation network can adopt a multi-head attention structure or a single-head attention structure.

[0108] Step S44: Construct an initial risk propagation model based on the graph propagation layer, the graph-level prediction layer, and the initial attention propagation network; Step S45: Train the initial risk propagation model based on the model training sample set; Step S46: If the model loss of the initial risk propagation model after training satisfies the preset convergence condition, the initial risk propagation model after training is determined as the risk propagation model.

[0109] Understandably, the goal of this embodiment is to construct a risk propagation model that integrates data-driven and knowledge-driven approaches, based on a "microseismic spatiotemporal map + knowledge graph," to predict the probability of microseismic disasters. The overall input is: microseismic event characteristics. Knowledge graph embedding Spatiotemporal diagram structure .

[0110] Initial node representation of a graph neural network: in, The 0th layer node represents the i-th microseismic event, which is the input feature of the graph neural network; This represents the physical characteristic vector of the microseismic event; Embed vectors (geological semantics) into the knowledge graph to represent prior geological information; This represents a vector concatenation operation that merges two types of information.

[0111] The knowledge-enhanced attention mechanism of the knowledge-enhanced attention layer is as follows: in, This represents the weight of the influence of node j on node i; A data-driven similarity function; For the knowledge graph relationship scoring function; Let i be the geological relationship vector from node i to j; This refers to the knowledge weighting coefficient (the intensity of the influence of the control mechanism). This is a normalization function that makes the weights sum to 1. This part is used to calculate how important the influence of node j is on node i.

[0112] The node update of a graph neural network is represented as: in, The neighboring nodes of the microseismic node i; The weight matrix is ​​a learnable matrix; It is a non-linear activation function; Let j be the weight of the influence of node j on node i; For node i, the first Layer node representation; For node j, the first Layer node representation. This formula shows that when node i updates its own state, it integrates the influence of all its neighbors and weights them according to their importance.

[0113] This embodiment constructs a graph neural network that integrates "data-driven similarity" and "geological knowledge constraints," and achieves interpretable risk propagation modeling of microseismic events in spatiotemporal structure through knowledge-enhanced attention mechanism and multi-hop propagation mechanism.

[0114] Based on the above embodiments, in the fifth embodiment of the knowledge graph-based multi-level interpretable analysis method for microseismic early warning in this application, step S50 may include: Step S501: Extract the disaster probability and risk level from the microseismic risk early warning results to obtain the early warning objects to be interpreted.

[0115] It should be noted that the early warning object to be interpreted refers to the current early warning result or current early warning sample that needs to be interpreted and analyzed, which may include the corresponding target monitoring area, monitoring time window, disaster probability and risk level.

[0116] In some embodiments, the microseismic early warning analysis device can read the probability of disaster and the risk level from the microseismic risk early warning results, and bind the corresponding target monitoring area, monitoring time window, input microseismic spatiotemporal map, and dynamic knowledge graph to form an early warning object to be interpreted. For example, when a working face is determined to be at a high risk level within the current time window, the device uses the high risk level and the corresponding probability of disaster as the core fields of the early warning object to be interpreted.

[0117] Step S502: Based on the early warning object to be explained, perform event-level contribution analysis on each microseismic event node in the microseismic spatiotemporal map to obtain an event contribution score set.

[0118] It should be noted that event-level contribution analysis refers to the process of assessing the impact of each microseismic event node on the current early warning result. The event contribution score set refers to the data set composed of the contribution scores corresponding to each microseismic event node.

[0119] It is understood that the multi-level interpretable analysis in this embodiment includes three layers of interpretive analysis: the first layer is event-level interpretive analysis, the second layer is structural-level interpretive analysis, and the third layer is mechanistic-level interpretive analysis. This embodiment completes event-level interpretive analysis by combining SHAP to achieve event-level contribution analysis; it completes structural-level interpretive analysis by using GNNExplainer to extract key propagation subgraphs; and it completes mechanistic-level interpretive analysis by achieving geological mechanism-level causal interpretation based on knowledge graph path probabilities.

[0120] Among them, SHAP can be an event contribution interpretation method based on Shapley values, used to calculate the marginal contribution of microseismic event nodes or node features to the probability of catastrophe and risk level. GNNExplainer can be a structural interpretation method based on graph masking learning, used to screen key nodes, key propagation edges, and key propagation subgraphs from the microseismic spatiotemporal map while maintaining the stability of the risk propagation model prediction results. KGPath can be a mechanistic interpretation method based on knowledge graph path probability, used to calculate the probability of the success of candidate geological causal paths based on the probability of relationships between geological entities in the dynamic knowledge graph, and output a geological causal chain to explain the causes of microseismic risk formation.

[0121] In practical implementation, the event hierarchy interpretation analysis interprets the importance contribution of each microseismic event node, referring to the following formula: in, This represents the importance contribution value of the i-th microseismic event; The node currently being interpreted; For the set of all microseismic nodes; A subset that does not contain node i; The total number of nodes; For subset Prediction results after inputting the data into the model; The subset after adding node i; This represents the factorial, used for combining weight allocation. This step is used to explain how much a particular microseismic event i contributes to the final catastrophic prediction, thereby identifying key triggering events.

[0122] Step S503: Based on the event contribution score set, filter micro-seismic events whose event contribution scores are greater than the preset contribution score threshold, use the filtered micro-seismic events as key triggering events, and construct a key triggering event set.

[0123] It should be noted that the preset contribution scoring threshold refers to the contribution scoring criteria set in advance for filtering key triggering events. The key triggering event set refers to the data set consisting of microseismic events that meet the contribution scoring criteria.

[0124] In some embodiments, the microseismic early warning analysis device can sort events from highest to lowest according to their event contribution scores and identify microseismic events with event contribution scores greater than a preset contribution score threshold as key triggering events. If the number of events meeting the threshold condition is too small, additional filtering can be performed according to a preset key event quantity threshold (e.g., the top 10 events). After filtering, the device stores the event identifier, occurrence time, spatial location, energy parameters, and contribution score of the key triggering events into a key triggering event set.

[0125] Step S504: Under the constraint of keeping the prediction results of the object to be explained and warned stable, perform edge mask learning on the propagation edges in the microseismic spatiotemporal map based on the risk propagation model and the microseismic spatiotemporal map to obtain the edge mask matrix.

[0126] It should be noted that the edge mask matrix refers to matrix data composed of the importance weights of each propagation edge. The constraint of stable prediction results means that after edge filtering, the risk propagation model still maintains an output close to the original warning result, for example, the change in the probability of disaster is less than a preset probability change threshold.

[0127] Understandably, the goal of structural hierarchical interpretation analysis is to extract complete microseismic spatiotemporal maps from a pre-trained knowledge-enhanced graph neural network model. The system automatically selects the most critical nodes, edges, and propagation structures for the early warning results, forming an interpretable microseismic propagation subgraph, cluster links, and fault propagation paths.

[0128] First, let the original microseismic map be: in, Represents the set of nodes for microseismic events; The set of edges between microseismic events; It is an adjacency matrix; This is the node feature matrix.

[0129] Secondly, a learnable edge mask matrix is ​​introduced through GNNExplainer. .set up Indicates the importance of the edges. This indicates that the edge is very important to the prediction result; This indicates that the edge contributes little to the prediction result.

[0130] Filtering the original adjacency matrix using a mask matrix: in, This is the interpreted weighted adjacency matrix; This is Hadamard's element-wise multiplication; This is the edge mask matrix.

[0131] Therefore, we can conclude that: in, The original edge weights; This is the importance mask for the edge; The valid edge weights are retained after interpretation. (Through...) This filter is used to determine which connections between microseismic events truly support the model's early warning judgment.

[0132] The optimization objective of the mask matrix can be written as: in, This is the optimal edge mask matrix; This indicates the search for a mask that minimizes the objective function; The mask matrix (between 0 and 1); Microseismic image; The microseismic submap is filtered by a mask. Indicates element-wise multiplication; Enhance the GNN prediction model with pre-trained knowledge; To explain the loss function; This is a true risk label (whether a risk has occurred); For sparse constraint weights; for The L1 norm is used to minimize the size of the interpretive subgraph; The entropy constraint coefficient; For mask entropy; used to allow Closer to 0 or 1.

[0133] Step S505: Based on the edge mask matrix and the preset edge mask threshold, perform sub-graph filtering on the microseismic spatiotemporal map to obtain the key propagation sub-graph.

[0134] It is understandable that training yields the optimal mask. Then, threshold filtering is performed on the edges: Thus, we obtain the key propagation subgraph: in, The set of key edges that are retained; The set of key nodes that are retained; This is a sub-map of microseismic propagation. The mask threshold; The edge between nodes i and j; Let i be the node representation of the microseismic event i; Explanation of the edge It is a key propagation edge.

[0135] Therefore, the output logic of the microseismic propagation subgraph is: original microseismic map —Learning side masks —Threshold-based key edge filtering—Resulting key subgraphs This key subplot represents the microseismic events that contribute most to the current risk warning and their propagation relationships.

[0136] Step S506: Based on the key propagation subgraph and the dynamic knowledge graph, perform structural hierarchical interpretation to obtain the set of mass transmission links and the fault propagation path.

[0137] It should be noted that a cluster of microseismic events refers to a continuous propagation link or a locally concentrated link formed by multiple high-risk microseismic events. A fault propagation path refers to the geological propagation path determined by the connection between the microseismic propagation structure and the fault entity.

[0138] In some embodiments, the microseismic early warning analysis device can perform structural hierarchy analysis on the critical propagation subgraph based on node connectivity, edge mask weights, and fault entities in the dynamic knowledge graph. The device can first identify high-importance nodes in the critical propagation subgraph, then perform connectivity analysis on these nodes to form a set of propagation links; subsequently, it maps the microseismic event nodes in the critical propagation subgraph to fault entities in the dynamic knowledge graph, and combines this with fault connectivity to filter fault propagation paths.

[0139] Furthermore, to enhance the traceability and engineering reliability of the microseismic early warning model, step S506 above may include: Step S5061: Generate node-fault mapping relationship based on the three-dimensional spatial location of each microseismic event node in the key propagation subgraph and the fault entity in the dynamic knowledge graph; Step S5062: Generate a set of candidate propagation paths based on the propagation edges in the key propagation subgraph and the node fault mapping relationship; Step S5063: Calculate the structural importance score of each microseismic event node based on the edge mask weights and node connection relationships in the key propagation subgraph to obtain the node structural importance score set; Step S5064: Filter high-risk mass-sending nodes according to the node structure importance score set and the preset structure importance threshold, and identify the connected components of the high-risk mass-sending nodes to obtain a mass-sending link set; Step S5065: Based on the fault connectivity in the dynamic knowledge graph, the geological consistency score of the candidate propagation path set is calculated, and the path with the highest score is selected from the candidate propagation path set as the fault propagation path based on the score results.

[0140] Understandably, after obtaining the key propagation subgraph... Then, the importance of the node structure can be further calculated: in, For nodes Structural importance score; For nodes The set of neighbors; The importance of the edge pointing from i to j; Let j represent the importance of the edge pointing from j to i. If some nodes satisfy: This is then defined as a set of high-risk mass-sending nodes: Subsequently Top Perform connected component or community detection algorithms: in, This is a set of high-risk mass-sending nodes; The threshold for node importance; For mass sending links or mass sending clusters; This is a subgraph induced by high-risk nodes; For nodes The structural importance score.

[0141] Therefore, the output logic of the mass transmission link is: key propagation subgraph —Importance of computation nodes — Screening high-risk nodes —Connectivity Components / Community Detection—Output Cluster Links. This section indicates that if multiple high-importance microseismic events are sequentially connected in time and space, forming a high-density local structure, it suggests that a cluster of microseismic events or rupture propagation may exist in the region.

[0142] The fault propagation path requires the interpretation of structural results (microseismograms). Align with the fault entities in the knowledge graph.

[0143] Let the set of entities in the knowledge graph break layer be: Each microseismic node It can be mapped to a specific fault entity based on spatial location or knowledge relationship: in, A collection of fault entities; This is the k-th fault; Number of faults; This is the mapping function from microseismic nodes to the fault entity.

[0144] In key subgraph In the middle, define a candidate propagation path: in, This is a candidate propagation path; This is the l-th node in the path; This represents the path length.

[0145] Path score is defined as: in, Score the importance of the path; The structural importance of adjacent nodes in the path; For nodes The fault to which it belongs; For nodes The fault to which it belongs; The knowledge consistency score for fault lines can be defined as: in, There are two fault entities; This is the reduction factor for cross-fault propagation.

[0146] The path with the highest score is ultimately selected: in, The optimal fault propagation path; The path that maximizes the path score; This indicates that the candidate path originates from the key propagation subgraph.

[0147] Therefore, the output logic of the fault propagation path is: key propagation subgraph —Mapping microseismic nodes to fault entities—Calculating candidate path scores—Selecting the highest-scoring path Only paths that simultaneously satisfy the criteria of "the model deems them important" and "the geological structure allows for their propagation" will be output as fault propagation paths.

[0148] It should be noted that the structural hierarchy interpretation and analysis in this embodiment uses GNNExplainer to learn the side mask matrix on the original microseismic spatiotemporal map. While maintaining the stability of the model's prediction results, the graph structure is sparsified to extract the key subgraphs that contribute most to the early warning results. This module first uses thresholding to optimize the mask. Output microseismic propagation subgraph This is used to reveal key microseismic events and their propagation relationships; subsequently, based on the importance scores of the nodal structures... Community detection methods are used to identify cluster fault links to discover localized microseismic clusters and rupture propagation areas. Finally, by combining fault entities and geological connections in the knowledge graph, geological consistency scores are applied to candidate paths in the key subgraphs to output the most probable fault propagation path. Therefore, this structural explanation mechanism can not only explain "which events are important", but also further answer "how events propagate" and "whether the propagation conforms to the geological tectonic mechanism", thereby significantly enhancing the traceability and engineering credibility of the microseismic early warning model.

[0149] Step S507: Based on the fault propagation path and the dynamic knowledge graph, perform a hierarchical interpretation of the mechanism to obtain the geological causal chain.

[0150] Understandably, this embodiment uses KGPath to identify the geological causal chain of the disaster and find the physical cause, referring to the following formula: in, This is a geological causal path; This represents the probability that the path is valid. The relationship between node i and node j in the knowledge graph; Let be the probability of the relationship from node i to j; ∏ represents the chain product (chain calculation of path probabilities). The above formula explains from a geological perspective why this microseismic disaster occurred. For example, by using the probability of the highest path being true, the disaster path can be found as "fault A—stress concentration zone B—microfracture C—microseismic swarm D".

[0151] Step S508: Summarize the set of key triggering events, the key propagation subgraph, the set of mass transmission links, the fault propagation path, and the geological causal chain to obtain multi-level early warning interpretation results.

[0152] In some embodiments, the microseismic early warning analysis device can integrate a set of key triggering events, a key propagation subgraph, a set of clustered links, fault propagation paths, and geological causal chains into an interpretation report. The report may include risk level, key event number, event location, key edge weight, clustered link number, fault propagation path name, and geological causal chain description. If deployed in a mine safety platform, key events and propagation paths can be marked on a 3D geological model or mining plan, and textual descriptions can be output simultaneously.

[0153] This embodiment determines the interpretation scope by identifying the object to be interpreted in the early warning, locates key triggering events by analyzing the contribution of events, extracts key propagation structures by learning edge masks and filtering subgraphs, obtains the cluster propagation links and fault propagation paths by interpreting the structure hierarchy, forms a geological causal chain by interpreting the mechanism hierarchy, and forms a multi-level interpretation output by summarizing the results. Therefore, it can decompose the risk warning results into event sources, propagation structures and geological causes, and enhance the traceability of microseismic early warning results.

[0154] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the multi-level interpretable analysis method for microseismic early warning based on knowledge graphs in this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0155] This application also provides a multi-level interpretable analysis system for microseismic early warning based on knowledge graphs. Please refer to [reference needed]. Figure 4 , Figure 4 This is a schematic diagram of the module structure of a knowledge graph-based multi-level interpretable analysis system for microseismic early warning. The system includes: Data processing module 10 is used to perform structured preprocessing on the raw microseismic monitoring data and geological prior data collected in the target monitoring area to obtain microseismic event feature set and geological knowledge source data. The geological prior data includes fault data, rock strata data, stress zone data, goaf data, historical earthquake data and mining disturbance data. The spatiotemporal diagram construction module 20 is used to construct a microseismic spatiotemporal diagram based on the spatial proximity, time decay, and energy propagation relationships among the microseismic events in the microseismic event feature set. The knowledge graph construction module 30 is used to construct a dynamic knowledge graph based on the geological knowledge source data, and to perform knowledge embedding and multi-hop relationship propagation on the dynamic knowledge graph to obtain a knowledge graph embedding vector set. The microseismic risk prediction module 40 is used to input the microseismic event feature set, the microseismic spatiotemporal map, and the knowledge graph embedding vector set into a pre-constructed risk propagation model to obtain microseismic risk warning results. The risk propagation model is a graph neural network model constructed based on a knowledge-enhanced attention mechanism. The microseismic risk warning results include the probability of disaster and the risk level. The multi-level analysis module 50 is used to perform multi-level interpretable analysis on the microseismic spatiotemporal map and the dynamic knowledge graph based on the microseismic risk warning results, and obtain multi-level warning interpretation results.

[0156] The knowledge graph-based multi-level interpretable analysis system for microseismic early warning provided in this application employs the knowledge graph-based multi-level interpretable analysis method for microseismic early warning in the above embodiments, and can solve the technical problems to be solved by the aforementioned knowledge graph-based multi-level interpretable analysis method for microseismic early warning. Compared with the prior art, the beneficial effects of the knowledge graph-based multi-level interpretable analysis system for microseismic early warning provided in this application are the same as the beneficial effects of the knowledge graph-based multi-level interpretable analysis method for microseismic early warning provided in the above embodiments, and other technical features of the knowledge graph-based multi-level interpretable analysis system for microseismic early warning are the same as the features disclosed in the methods of the above embodiments, and will not be repeated here.

[0157] This application provides a multi-level interpretable analysis device for microseismic early warning, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to execute the knowledge graph-based multi-level interpretable analysis method for microseismic early warning in the first embodiment described above.

[0158] The following is for reference. Figure 5 The diagram illustrates a structural schematic suitable for implementing a multi-level interpretable analysis device for microseismic early warning in the embodiments of this application. The multi-level interpretable analysis device for microseismic early warning in the embodiments of this application may include, but is not limited to, servers, edge computing devices, industrial control computers, mine safety monitoring platform terminals, desktop computers, or other electronic devices with data processing capabilities. Figure 5 The microseismic early warning multi-level interpretable analysis device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0159] like Figure 5As shown, the microseismic early warning multi-level interpretable analysis device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in ROM 1002 (Read Only Memory) or a program loaded from storage device 1003 into RAM 1004 (Random Access Memory). RAM 1004 also stores various programs and data required for the operation of the microseismic early warning multi-level interpretable analysis device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via bus 1005. I / O interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touch screens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, LCDs (Liquid Crystal Displays), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the microseismic early warning multi-level interpretable analysis device to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows a microseismic early warning multi-level interpretable analysis device with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0160] The microseismic early warning multi-level interpretable analysis device provided in this application employs the knowledge graph-based multi-level interpretable analysis method for microseismic early warning in the above embodiments, and can solve the technical problems to be solved by the knowledge graph-based multi-level interpretable analysis method for microseismic early warning. Compared with the prior art, the beneficial effects of the microseismic early warning multi-level interpretable analysis device provided in this application are the same as the beneficial effects of the knowledge graph-based multi-level interpretable analysis method for microseismic early warning provided in the above embodiments, and other technical features in this microseismic early warning multi-level interpretable analysis device are the same as the features disclosed in the previous embodiment method, and will not be repeated here.

[0161] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0162] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0163] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the knowledge graph-based multi-level interpretable analysis method for microseismic early warning in the above embodiments.

[0164] The computer-readable storage medium provided in this application can be a tangible medium capable of storing computer programs, such as a hard disk, random access memory, read-only memory, flash memory, optical storage device, or magnetic storage device. When the computer program is executed by a processor, it implements the aforementioned knowledge graph-based multi-level interpretable analysis method for microseismic early warning.

[0165] The aforementioned computer-readable storage medium may be included in the microseismic early warning multi-level interpretable analysis device; or it may exist independently and not be assembled into the microseismic early warning multi-level interpretable analysis device.

[0166] The aforementioned computer-readable storage medium carries one or more programs. When the aforementioned one or more programs are executed by the microseismic early warning multi-level interpretable analysis device, the microseismic early warning multi-level interpretable analysis device performs the knowledge graph-based microseismic early warning multi-level interpretable analysis method as described in the above embodiments.

[0167] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including LAN (Local Area Network) or WAN (Wide Area Network)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0168] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0169] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the aforementioned knowledge graph-based multi-level interpretable analysis method for microseismic early warning, thereby solving the technical problems to be addressed by the aforementioned knowledge graph-based multi-level interpretable analysis method for microseismic early warning. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the knowledge graph-based multi-level interpretable analysis method for microseismic early warning provided in the above embodiments, and will not be repeated here.

[0170] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A multi-level interpretable analysis method for microseismic early warning based on knowledge graphs, characterized in that, The method includes: The raw microseismic monitoring data and geological prior data collected in the target monitoring area are subjected to structured preprocessing to obtain a microseismic event feature set and geological knowledge source data. The geological prior data includes fault data, rock strata data, stress zone data, goaf data, historical earthquake data, and mining disturbance data. A microseismic spatiotemporal map is constructed based on the spatial proximity, temporal decay, and energy propagation relationships among the microseismic events in the microseismic event feature set. A dynamic knowledge graph is constructed based on the geological knowledge source data, and knowledge embedding and multi-hop relation propagation are performed on the dynamic knowledge graph to obtain a knowledge graph embedding vector set. The relation vector of the dynamic knowledge graph is defined as follows: in, for The time-relation vector represents The state of geological relationships at any given time; This is the initial relation vector, which represents the initial geological structure and stress relationship; The attenuation coefficient; The current time; This is the initial time; External disturbance term; The microseismic event feature set, the microseismic spatiotemporal map, and the knowledge graph embedding vector set are input into a pre-constructed risk propagation model to obtain microseismic risk warning results. The risk propagation model is a graph neural network model constructed based on a knowledge-enhanced attention mechanism. The microseismic risk warning results include the probability of disaster and the risk level. The knowledge-enhanced attention mechanism refers to an attention calculation method that considers both the similarity of event data and the correlation of geological knowledge when calculating the influence weights between microseismic events. Extract the probability of disaster and the risk level from the microseismic risk early warning results to obtain the early warning objects to be interpreted; Based on the aforementioned early warning object, an event-level contribution analysis is performed on each microseismic event node in the microseismic spatiotemporal map to obtain an event contribution score set; Based on the event contribution score set, micro-seismic events with event contribution scores greater than a preset contribution score threshold are selected, and the selected micro-seismic events are used as key triggering events to construct a key triggering event set. Under the constraint of keeping the prediction results of the object to be explained stable, the propagation edges in the microseismic spatiotemporal map are learned by edge masking based on the risk propagation model and the microseismic spatiotemporal map to obtain the edge mask matrix. The constraint of the prediction results being stable means that the change in the probability of disaster of the prediction results output by the risk propagation model after edge filtering is less than a preset probability change threshold. Based on the edge mask matrix and a preset edge mask threshold, the microseismic spatiotemporal map is subjected to sub-graph selection to obtain key propagation sub-graphs. The sub-graph selection is performed according to the following formula: in, The set of key edges that are retained; The set of key nodes that are retained; This is the key propagation subgraph; The mask threshold; Let i be the edge between nodes i and j; Let i be the node representation of the microseismic event i; Explanation of the edge It is a key propagation edge; Based on the key propagation subgraph and the dynamic knowledge graph, a structural hierarchy interpretation is performed to obtain a set of clustered links and fault propagation paths. The structural hierarchy interpretation refers to identifying important nodes in the key propagation subgraph, and then performing connectivity analysis on the important nodes to form a set of clustered links. The microseismic event nodes in the key propagation subgraph are mapped to fault entities in the dynamic knowledge graph, and fault propagation paths are selected by combining fault connection relationships. Based on the fault propagation path and the dynamic knowledge graph, a hierarchical explanation of the mechanism is performed to obtain the geological causal chain; By summarizing the key triggering event set, the key propagation subgraph, the mass transmission link set, the fault propagation path, and the geological causal chain, a multi-level early warning interpretation result is obtained.

2. The multi-level interpretable analysis method for microseismic early warning based on knowledge graphs as described in claim 1, characterized in that, The construction of a microseismic spatiotemporal map based on the spatial proximity, temporal decay, and energy propagation relationships among microseismic events in the microseismic event feature set includes: Based on the microseismic event feature set, the spatial proximity relationship, temporal decay relationship, and energy propagation relationship of each microseismic event pair are extracted; Based on the spatial proximity relationship, the spatial proximity weights of each microseismic event pair are determined, and a spatial proximity weight set is generated; Based on the time decay relationship, the time decay weight of each microseismic event pair is determined, and a set of time decay weights is generated; Based on the energy propagation relationship, the energy propagation weights of each microseismic event pair are determined, and an energy propagation weight set is generated. By fusing the spatial proximity weight set, the time decay weight set, and the energy propagation weight set, an event association weight set is obtained; A microseismic spatiotemporal diagram with microseismic events as nodes is generated based on the microseismic event feature set and the event association weight set.

3. The multi-level interpretable analysis method for microseismic early warning based on knowledge graphs as described in claim 1, characterized in that, The process involves constructing a dynamic knowledge graph based on the geological knowledge source data, and then embedding knowledge and performing multi-hop relationship propagation on the dynamic knowledge graph to obtain a knowledge graph embedding vector set, including: Geological entity sets and geological relationship sets are extracted from the geological knowledge source data. The geological entity sets include fault entities, rock strata entities, stress zone entities, goaf entities, historical earthquake case entities, and mining disturbance entities. The geological relationship sets include stress transmission relationships, induced relationships, fracture relationships, attribution relationships, and disturbance evolution relationships. A set of geological knowledge triples is generated based on the set of geological entities and the set of geological relationships. Each geological knowledge triple in the set of geological knowledge triples includes a head entity, a geological relationship, and a tail entity. Based on the mining disturbance data, the establishment time of the geological relationships, and the stress zone data, the geological relationships in the geological knowledge triplet set are subjected to relationship state evolution to obtain a dynamic geological knowledge triplet set. Based on the dynamic geological knowledge triplet set, the geological entity set and the geological relation set are vector-initialized to obtain the initial entity vector set and the initial relation vector set. Based on the dynamic geological knowledge triplet set, the initial entity vector set, and the initial relation vector set, relation constraint learning is performed to obtain relation constraint embedded data; Multi-hop relationship propagation is performed based on the relationship constraint embedding data, and the embedding vectors corresponding to the microseismic event nodes are extracted according to the mapping relationship between microseismic event nodes and geological entities to obtain the knowledge graph embedding vector set.

4. The multi-level interpretable analysis method for microseismic early warning based on knowledge graphs as described in claim 3, characterized in that, The relation constraint learning based on the dynamic geological knowledge triplet set, the initial entity vector set, and the initial relation vector set yields relation constraint embedding data, including: Extract the target geological knowledge triplet from the dynamic geological knowledge triplet set to obtain the target head entity, the target dynamic relationship, and the target tail entity; Extract the head entity vector corresponding to the target head entity and the tail entity vector corresponding to the target tail entity from the initial entity vector set to obtain the target entity vector group; Extract the relation vector corresponding to the target dynamic relation from the initial relation vector set to obtain the target relation vector; Calculate the triplet matching deviation based on the target entity vector group and the target relation vector to obtain relation constraint deviation data; The initial entity vector set and the initial relation vector set are updated based on the relation constraint deviation data, and the updated entity vector set and the updated relation vector set are used as relation constraint embedding data.

5. The multi-level interpretable analysis method for microseismic early warning based on knowledge graphs as described in claim 1, characterized in that, The step of inputting the microseismic event feature set, the microseismic spatiotemporal map, and the knowledge graph embedding vector set into a pre-constructed risk propagation model to obtain microseismic risk warning results includes: Based on the microseismic event node identifiers, the microseismic event feature set and the knowledge graph embedding vector set are aligned and fused to obtain a node fusion input set. Based on the microseismic spatiotemporal diagram, the set of neighboring nodes for each microseismic event node is determined, thus obtaining the node adjacency relationship set; The node fusion input set and the node adjacency relationship set are input into a pre-constructed risk propagation model to obtain microseismic risk early warning results; The risk propagation model includes a knowledge-enhanced attention layer, a graph propagation layer, and a graph-level prediction layer. The knowledge-enhanced attention layer is configured to calculate attention weights based on the data similarity between microseismic event nodes and the geological relationship scores in the dynamic knowledge graph, thereby obtaining a knowledge-enhanced attention weight set. The graph propagation layer is configured to perform neighborhood-weighted propagation based on the node fusion input set and the knowledge-enhanced attention weight set, and output a high-level node representation set. The graph-level prediction layer is configured to perform graph-level aggregation and risk mapping on the high-level representation set of the nodes, and output microseismic risk warning results.

6. The multi-level interpretable analysis method for microseismic early warning based on knowledge graphs as described in claim 5, characterized in that, Before inputting the microseismic event feature set, the microseismic spatiotemporal map, and the knowledge graph embedding vector set into the pre-constructed risk propagation model to obtain the microseismic risk warning result, the method further includes: A model training sample set is constructed based on historical microseismic event feature sets, historical microseismic spatiotemporal maps, historical knowledge graph embedding vector sets, and historical risk labels. Based on the historical microseismic event feature set and the historical knowledge graph embedding vector set, a node fusion training sample set is constructed to obtain the node fusion training sample set. Based on the historical microseismic spatiotemporal map and the historical knowledge graph embedding vector set, a knowledge-enhanced attention layer is configured to obtain the initial attention propagation network; An initial risk propagation model is constructed based on the graph propagation layer, the graph-level prediction layer, and the initial attention propagation network. The initial risk propagation model is trained based on the model training sample set. If the model loss of the initial risk propagation model after training satisfies the preset convergence condition, the initial risk propagation model after training is determined as the risk propagation model.

7. The multi-level interpretable analysis method for microseismic early warning based on knowledge graphs as described in claim 1, characterized in that, The structural hierarchy interpretation based on the key propagation subgraph and the dynamic knowledge graph yields a set of mass transmission links and discontinuous propagation paths, including: Based on the three-dimensional spatial location of each microseismic event node in the key propagation subgraph and the fault entities in the dynamic knowledge graph, a node-fault mapping relationship is generated. A set of candidate propagation paths is generated based on the propagation edges in the key propagation subgraph and the node fault mapping relationship; The structural importance score of each microseismic event node is calculated based on the edge mask weights and node connection relationships in the key propagation subgraph, resulting in a set of node structural importance scores. High-risk mass-sending nodes are selected based on the node structure importance score set and the preset structure importance threshold, and the connected component identification is performed on the high-risk mass-sending nodes to obtain a mass-sending link set; The candidate propagation path set is geologically consistent with the fault connectivity in the dynamic knowledge graph, and the path with the highest score is selected as the fault propagation path based on the score results.

8. A multi-level interpretable analysis system for microseismic early warning based on knowledge graphs, characterized in that, The system employs the knowledge graph-based multi-level interpretable analysis method for microseismic early warning as described in any one of claims 1 to 7, and the system comprises: The data processing module is used to perform structured preprocessing on the raw microseismic monitoring data and geological prior data collected in the target monitoring area to obtain microseismic event feature sets and geological knowledge source data. The geological prior data includes fault data, rock strata data, stress zone data, goaf data, historical earthquake data, and mining disturbance data. The spatiotemporal diagram construction module is used to construct a microseismic spatiotemporal diagram based on the spatial proximity, time decay, and energy propagation relationships among the microseismic events in the microseismic event feature set. The knowledge graph construction module is used to construct a dynamic knowledge graph based on the geological knowledge source data, and to perform knowledge embedding and multi-hop relationship propagation on the dynamic knowledge graph to obtain a knowledge graph embedding vector set; The microseismic risk prediction module is used to input the microseismic event feature set, the microseismic spatiotemporal map, and the knowledge graph embedding vector set into a pre-constructed risk propagation model to obtain microseismic risk warning results. The risk propagation model is a graph neural network model constructed based on a knowledge-enhanced attention mechanism. The microseismic risk warning results include the probability of disaster and the risk level. The multi-level analysis module is used to perform multi-level interpretable analysis on the microseismic spatiotemporal map and the dynamic knowledge graph based on the microseismic risk warning results, and obtain multi-level warning interpretation results.

9. A multi-level interpretable analysis device for microseismic early warning, characterized in that, The microseismic early warning multi-level interpretable analysis device includes: a memory, a processor, and a microseismic early warning multi-level interpretable analysis program stored in the memory. The processor is used to run the microseismic early warning multi-level interpretable analysis program, which is configured to implement the knowledge graph-based microseismic early warning multi-level interpretable analysis method as described in any one of claims 1 to 7.

Images