A knowledge graph-based mine safety hidden danger intelligent investigation method and system

Abstract

The application discloses a kind of based on knowledge graph's mine safety hidden danger intelligent investigation method and system, specifically related to knowledge graph and mine safety information technology field, for solving the problem that the existing hidden danger investigation method based on static knowledge graph cannot follow underground space topology dynamic change and lead to inaccurate reasoning.By obtaining the mine multi-source dynamic data containing sensor data and mining plan, analyzing and identifying the cooperative state change event representing space topology evolution, and comparing the dynamic event subgraph based on this with the static knowledge subgraph extracted from the preset knowledge graph to identify the conflict relationship, then combining the mining plan to verify and correct the conflict relationship to generate the space topology update relationship, finally updating the mine safety knowledge graph in real time according to the update relationship, and based on the updated graph, the safety hidden danger reasoning of real-time monitoring data is carried out, so as to realize the synchronization of knowledge graph and dynamic production environment and the accurate investigation of hidden danger.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph and mine safety information technology, and more specifically, to a method and system for intelligent investigation of mine safety hazards based on knowledge graph. Background Technology

[0002] In the field of mine safety production, information technology can be used to achieve intelligent identification and early warning of potential hazards. Among these technologies, knowledge graph technology, due to its powerful knowledge structure representation and associative reasoning capabilities, can be introduced into this field to construct mine safety knowledge models. Existing technologies typically construct physical entities such as mine roadways, equipment, and hazard sources, along with their attributes, safety regulations, and hazard assessment rules, as nodes and relationships within a knowledge graph. Based on this graph, the system can perform logical reasoning on input safety monitoring data, thereby assisting in the discovery of potential safety hazards. This type of method can effectively improve the efficiency of information association and hazard analysis in application scenarios where entities and relationships are relatively fixed.

[0003] However, the actual production environment of mines is characterized by continuous dynamic engineering changes. In particular, the advancement of mining faces and the expansion and closure of roadway systems cause the underground spatial topology to change rapidly and regularly. Existing knowledge graph-based intelligent hazard identification methods are essentially static snapshots of the system state at a certain point in time. They cannot effectively represent and follow the dynamic evolution of these physical spatial relationships. This leads to a rapid disconnect between the spatial associations between entities in the graph and the real environment. Consequently, the results of hazard location and propagation path reasoning based on such relationships become biased, seriously affecting the accuracy and reliability of hazard identification. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method and system for intelligent investigation of mine safety hazards based on knowledge graphs to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A knowledge graph-based intelligent method for identifying potential safety hazards in mines includes the following steps:

[0007] S1. Acquire multi-source dynamic data of the mine, including real-time monitoring sensor data and mining succession plan data;

[0008] S2. Analyze the temporal trend and spatial distribution of multi-source dynamic data. When a trend change in the state parameters of a downhole spatial entity is detected and a series of causal temporal changes are triggered in its associated neighborhood, the trend change and causal temporal changes are aggregated and identified as a coordinated state change event.

[0009] S3. Construct a dynamic event subgraph based on collaborative state change events, and extract corresponding entities and relationships from the preset mine safety knowledge graph to form a static knowledge subgraph. Identify conflicting entity relationship pairs by comparing the dynamic event subgraph and the static knowledge subgraph.

[0010] S4. Combine the mining succession plan data to verify and correct conflicting entity relationships, and generate spatial topology update relationships.

[0011] S5. Update the mine safety knowledge graph based on the spatial topology update relationship, and perform safety hazard inference on the real-time monitored sensor data.

[0012] Furthermore, acquire multi-source dynamic data from the mine, including real-time monitoring sensor data and mining succession plan data, including:

[0013] Real-time sensor data is collected from various safety monitoring sensors deployed underground.

[0014] Read mining succession plan data from the mine production management system.

[0015] Furthermore, the mining succession plan data read from the mine production management system includes the spatial position change sequence of the planned mining face advancement and the corresponding planned time nodes.

[0016] Furthermore, by analyzing the temporal trends and spatial distribution of multi-source dynamic data, when a trend change in the state parameters of a downhole spatial entity is detected and triggers a series of causal temporal changes within its associated neighborhood, the trend change and causal temporal changes are aggregated and identified as a co-current state change event, including:

[0017] By performing time-series analysis on real-time monitored sensor data, the initial change events that indicate a trend change in the state parameters of the downhole space entity can be identified.

[0018] Based on the spatial neighborhood relationships between entities recorded in the mine safety knowledge graph, starting from the initial change event, we search for derivative changes with causal and temporal relationships within the associated neighborhood.

[0019] Logically associate and encapsulate the initial change event with all the derivative changes it triggers to form a collaborative state change event.

[0020] Furthermore, when searching based on the spatial neighborhood relationships between entities recorded in the mine safety knowledge graph, the criteria for determining the causal temporal relationship are: subsequent derivative changes occur later in time than the initial change event, and spatially they conform to the physical propagation law of radiation from the location of the initial change event to the associated neighborhood.

[0021] Furthermore, a dynamic event subgraph is constructed based on collaborative state change events, and corresponding entities and relationships are extracted from a pre-defined mine safety knowledge graph to form a static knowledge subgraph. Conflicting entity-relationship pairs are identified by comparing the dynamic event subgraph and the static knowledge subgraph, including:

[0022] Based on the changing nodes reflected in the dynamic event subgraph, extract the relationship paths between entities;

[0023] In the static knowledge subgraph, search for whether there is a relationship path that is consistent with or equivalent to the relationship path between the entity nodes corresponding to the changed node;

[0024] When there is no relation path in the static knowledge subgraph that is consistent with or equivalent to the relation path, or when the relation type represented by the existing path contradicts the relation type in the dynamic event subgraph, the graph relations between the entity nodes in this group are determined to be conflicting entity relation pairs.

[0025] Furthermore, by combining mining succession plan data, conflicting entity relationships are verified and corrected to generate spatial topology update relationships, including:

[0026] Analyze mining succession plan data to extract the expected spatial topological changes involved in the plan;

[0027] Based on the anticipated spatial topological changes, assess the rationality of the relationship changes reflected by the dynamic event subgraph in mutually conflicting entity relationship pairs.

[0028] When the assessment is deemed reasonable, the corresponding relationships in the static knowledge subgraph are modified based on the dynamic event subgraph;

[0029] When an assessment is deemed unreasonable, conflicting entity relationships are corrected based on anticipated spatial topology changes, thereby generating updated spatial topology relationships.

[0030] Furthermore, the mine safety knowledge graph is updated based on spatial topology update relationships, and safety hazard inferences are performed on real-time monitored sensor data, including:

[0031] The entities, relationships, and safety rules in the updated mine safety knowledge graph are matched with real-time monitored sensor data.

[0032] When the real-time monitored sensor data matches the hazard triggering conditions recorded in the updated mine safety knowledge graph, logical reasoning is performed based on the entity relationships and safety rules in the updated mine safety knowledge graph to generate safety hazard warning information corresponding to the current spatial topology.

[0033] Furthermore, logical reasoning based on the entity relationships and safety rules in the updated mine safety knowledge graph includes: determining the scope of the hazard's impact and possible propagation paths based on the updated spatial topological relationships between entities, and conducting a comprehensive risk assessment in conjunction with the threshold conditions in the safety rules.

[0034] On the other hand, the present invention provides a knowledge graph-based intelligent system for identifying potential safety hazards in mines, comprising the following modules:

[0035] The data acquisition module is used to acquire multi-source dynamic data of the mine, including real-time monitoring sensor data and mining succession plan data.

[0036] The event recognition module is used to analyze the temporal trends and spatial distribution of multi-source dynamic data. When a trend change in the state parameters of a downhole spatial entity is detected and triggers a series of causal temporal changes in its associated neighborhood, the trend change and causal temporal changes are aggregated and identified as a coordinated state change event.

[0037] The subgraph comparison module is used to construct dynamic event subgraphs based on collaborative state change events, and extract corresponding entities and relationships from the preset mine safety knowledge graph to form static knowledge subgraphs. By comparing the dynamic event subgraphs and the static knowledge subgraphs, conflicting entity relationship pairs are identified.

[0038] The relationship correction module is used to verify and correct conflicting entity relationship pairs by combining mining succession plan data, and generate spatial topology update relationships.

[0039] The graph update module is used to update the mine safety knowledge graph based on spatial topology update relationships and to perform safety hazard inference on real-time monitored sensor data.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] 1. By automatically identifying spatiotemporally correlated collaborative state change events from multi-source dynamic data and using these events to drive the updating of the knowledge graph, the synchronous evolution of the mine safety knowledge model and the underground dynamic engineering environment is realized. The continuous changes in the physical world are deconstructed into discrete, computable knowledge conflict events. Through cross-validation with mining plans, it is ensured that the updates to the knowledge graph are both responsive to real-time data and in line with production planning. This transforms the knowledge graph from a static knowledge base into a knowledge evolution system that can adapt to the dynamic changes in the underground spatial topology, thus solving the problem of the disconnect between the static graph model and the dynamic production environment.

[0042] 2. Safety hazard reasoning based on dynamically updated knowledge graphs ensures that the spatial relationships and propagation paths relied upon for hazard identification remain consistent with the actual underground conditions. This enables the system to accurately track the impact range and diffusion path of hazards in the latest spatial topology. Combined with real-time sensor data, it can generate precise early warnings for specific engineering scenarios, significantly improving the accuracy of hazard location and the reliability of risk assessment, and providing up-to-date intelligent decision support for mine safety production. Attached Figure Description

[0043] Figure 1 This is a flowchart of a knowledge graph-based intelligent method for identifying potential safety hazards in mines, according to the present invention.

[0044] Figure 2 This is a schematic diagram of the structure of a knowledge graph-based intelligent system for identifying potential safety hazards in mines, as described in this invention. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0046] Example 1: Figure 1 This invention presents an intelligent method for identifying potential safety hazards in mines based on knowledge graphs, comprising the following steps:

[0047] S1. Acquire multi-source dynamic data of the mine, including real-time monitoring sensor data and mining succession plan data;

[0048] S2. Analyze the temporal trend and spatial distribution of multi-source dynamic data. When a trend change in the state parameters of a downhole spatial entity is detected and a series of causal temporal changes are triggered in its associated neighborhood, the trend change and causal temporal changes are aggregated and identified as a coordinated state change event.

[0049] S3. Construct a dynamic event subgraph based on collaborative state change events, and extract corresponding entities and relationships from the preset mine safety knowledge graph to form a static knowledge subgraph. Identify conflicting entity relationship pairs by comparing the dynamic event subgraph and the static knowledge subgraph.

[0050] S4. Combine the mining succession plan data to verify and correct conflicting entity relationships, and generate spatial topology update relationships.

[0051] S5. Update the mine safety knowledge graph based on the spatial topology update relationship, and perform safety hazard inference on the real-time monitored sensor data.

[0052] In this embodiment of the invention, step S1, acquiring multi-source dynamic data of the mine, is specifically implemented in the following way:

[0053] Multi-source dynamic data includes real-time monitoring sensor data and mining succession plan data. Real-time monitoring sensor data is acquired by collecting raw monitoring data in real time from various safety monitoring sensors deployed underground in the mine. These safety monitoring sensors include gas concentration sensors, carbon monoxide concentration sensors, temperature sensors, wind speed sensors, wind pressure sensors, roof pressure sensors, displacement sensors, and smoke sensors. Each sensor is connected to the mine safety monitoring system according to its design specifications and collects data at a predetermined sampling frequency; for example, a gas concentration sensor can be set to collect data once every 1 second, and a roof pressure sensor can be set to collect data once every 10 minutes. The collected raw monitoring data is transmitted through the underground industrial ring network or a dedicated communication substation, using the Modbus protocol. After transmission to the ground monitoring center server, the raw monitoring data undergoes preliminary processing. This preliminary processing includes adding a precise timestamp to each data record, with timestamp accuracy down to the second, and time alignment of data from different sensors to ensure that data used within the same processing cycle has a consistent time reference. The preliminary processing also includes the standardization of data formats, converting various types of sensor data into standard structured data containing sensor number, type of monitored physical quantity, value, unit and timestamp. This standard structured data after preliminary processing is the real-time monitoring sensor data obtained.

[0054] The mining succession plan data is obtained by reading pre-defined production plan files or accessing the production plan database from the mine production management system. The mine production management system is an information system used by mining enterprises to plan and schedule production tasks. The mining succession plan data stored within it exists in the form of database table records or structured documents. The reading process is achieved by calling the application programming interface provided by the mine production management system or by directly accessing its database. The read mining succession plan data includes the spatial position change sequence of the planned mining face advancement and the corresponding planned time nodes. The spatial position change sequence refers to a set of ordered information describing the position and contour changes of the mining face in the underground space according to the planned time sequence. This sequence includes the three-dimensional coordinates of a series of key points, such as the coordinates of the starting point and ending point of the working face, as well as the coordinates of multiple boundary points describing the contour of the working face. This sequence also includes the coordinate sequence of the roadway centerline points and the design cross-sectional dimensions of the roadway, used to describe the trajectory and morphological changes of the roadway excavation. The corresponding planned time nodes refer to the planned completion time associated with each key change point in the spatial location change sequence. For example, for a fully mechanized mining face, the spatial location change sequence may include the planned location coordinates of the face on the 1st, 5th, and 10th of the month, respectively, with the corresponding planned time nodes being the 1st, 5th, and 10th of the month. When reading mining succession plan data, it is necessary to ensure that the latest version and approved official plan are obtained, and to verify the completeness and rationality of the data. For example, check whether the spatial coordinates are within the valid range of the mine's geographical coordinate system and whether the planned time nodes conform to the chronological logic. The real-time monitoring sensor data and mining succession plan data obtained through the above methods together constitute the multi-source dynamic data foundation required for subsequent analysis and reasoning.

[0055] In this embodiment of the invention, step S2, analyzing the temporal trend and spatial distribution of multi-source dynamic data and identifying cooperative state change events, is specifically implemented in the following ways:

[0056] Time series analysis is performed on real-time monitored sensor data to identify initial change events indicating a trend in the state parameters of downhole entities. The time series analysis is performed independently for each monitored physical quantity time series in the standard structured data acquired and processed in step S1. For a given downhole entity and its corresponding sensor-monitored physical quantity time series, a sliding time window method is used to calculate the statistical characteristic value of that physical quantity within the current time window. The sliding time window method involves setting a fixed-length time interval, for example, 5 minutes, which covers the period backward from the current moment. The arithmetic mean of all monitored values ​​within this 5-minute time window is calculated as the current statistical characteristic value. The time window is then slid forward by a fixed step, for example, 1 minute, to obtain a new time window and its statistical characteristic value is calculated. This new time window overlaps with the previous window by 4 minutes. The statistical characteristic value of the new window is compared with that of the previous window. If the difference between the statistical characteristic values ​​of the two consecutive time windows exceeds a preset trend change threshold for that physical quantity, it is determined that a trend change has occurred in that physical quantity for the downhole entity. The threshold for determining a trend change is set based on the historical statistical variance of the physical quantity within the normal production fluctuation range. This historical statistical variance is calculated by analyzing historical monitoring data from the monitoring point over a past period, such as the past 30 days, under normal production conditions. The calculated historical variance value is multiplied by a factor, such as 2, and the resulting product serves as the threshold for determining a trend change. Once a trend change is determined, an initial change event is generated. This initial change event is recorded as a data structure with the following information: a unique identifier of the downhole spatial entity where the trend change occurred, the corresponding type of monitored physical quantity, whether the trend change is upward or downward, the starting time of the trend change, the time when the trend change was identified, and the magnitude of the trend change.

[0057] Based on the identified initial change event, the spatial neighborhood relationships between entities are pre-recorded in the mine safety knowledge graph. Starting from the underground spatial entity corresponding to the initial change event, derivative changes with causal temporal relationships are searched within its associated neighborhood. The spatial neighborhood relationships between entities recorded in the mine safety knowledge graph exist in the form of graph edges, and the relationship types include direct adjacency, connection through roadways, and location in the same geological unit. Based on these relationships, a set of other underground spatial entities that have direct or indirect spatial connections with the entity of the initial change event is determined, and this set constitutes the associated neighborhood. For each other entity within the associated neighborhood, the real-time monitoring sensor data of that entity within a certain period after the occurrence of the initial change event is retrieved, and the same temporal analysis method as used to identify the initial change event is applied to check whether each monitored physical quantity has also undergone a trend change. The determination of causal temporal relationships includes two conditions that must be met simultaneously. The first condition is the temporal order, that is, the time point when the candidate derivative change is identified must be later than the time point when the initial change event is identified. The second condition is compliance with spatial propagation laws. This means that from the location of the initial change event to the location of the candidate derived change entity, there should exist a path in the mine safety knowledge graph consisting of edges representing spatial neighborhood relationships. Furthermore, the physical connectivity reflected by this path should support the spread of the physical process represented by the initial change. For changes involving gas diffusion, such as an increase in gas concentration, the spatial propagation path must conform to the connectivity of the ventilation network or coal seam fractures. For changes involving stress transmission, such as an increase in roof pressure, the spatial propagation path must conform to the continuity of the rock strata structure and the mechanical conduction relationship. Only candidate changes that simultaneously satisfy both temporal lag and spatial compliance with physical propagation laws are identified as derived changes triggered by the initial change event.

[0058] The initial change event and all its derived changes are logically associated and encapsulated to form a coordinated state change event. Logical association refers to establishing a directional link between the initial change event and each derived change, indicating the relationship of initiation and being initiated. Encapsulation refers to creating a new data structure to uniformly represent this group of interrelated changes. The data structure of the coordinated state change event includes a unique event identifier, embedding the initial change event as its core event, and a list of derived changes, each item in the list recording detailed information about a derived change. This data structure also records the spatial range covered by the entire coordinated state change event and the event's temporal span. Through the above process, step S2 aggregates discrete sensor data anomalies into composite event units with inherent causal relationships in the spatiotemporal dimensions, namely, coordinated state change events.

[0059] In this embodiment of the invention, step S3 constructs a dynamic event subgraph based on collaborative state change events and extracts corresponding entities and relationships from a preset mine safety knowledge graph to form a static knowledge subgraph. Then, it identifies conflicting entity relationships through comparison. This is specifically achieved in the following ways:

[0060] The process of constructing a dynamic event subgraph first involves parsing the collaborative state change events generated in step S2. The data structure of these collaborative state change events includes a core initial change event and a list of derived changes. Unique identifiers for all underground spatial entities recorded in this data structure are extracted and used as nodes in the dynamic event subgraph. Each node is assigned a node label, which represents the type of the corresponding underground spatial entity, such as a roadway type or sensor type. Next, edges in the dynamic event subgraph are constructed based on the triggering relationships between changes recorded in the collaborative state change events. A directed edge is established between the initial change event node and the entity node corresponding to each of its derived changes, with the direction of the directed edge pointing from the initial change event node to the derived change node. The type of the directed edge is defined as the triggering type. Furthermore, if the entities corresponding to two derived changes already have a spatial neighborhood relationship in the mine safety knowledge graph, an undirected edge can also be established between the two corresponding nodes. The type of the undirected edge follows the relationship types defined in the mine safety knowledge graph, such as the adjacent type. The resulting graph structure is the dynamic event subgraph.

[0061] The process of extracting corresponding entities and relationships from a pre-defined mine safety knowledge graph to construct a static knowledge subgraph is performed sequentially with the construction of the dynamic event subgraph. The pre-defined mine safety knowledge graph is a large-scale semantic network stored in a graph database. The extraction operation uses the unique identifiers of all underground spatial entities corresponding to all nodes in the dynamic event subgraph as query conditions. A query request is submitted to the graph database, requesting the return of all entity nodes corresponding to these unique identifiers and all relational edges directly connecting these entity nodes. The query result is a subgraph composed of these nodes and edges; this subgraph is called the static knowledge subgraph. The edge types in the static knowledge subgraph are derived from the definition set of the pre-defined graph, such as being located in a type or connected to a type.

[0062] The core of step S3 is identifying conflicting entity relationship pairs by comparing the dynamic event subgraph and the static knowledge subgraph. Relationship paths between entities are extracted based on the changing nodes reflected in the dynamic event subgraph. A relationship path is a sequence of edges connecting two specific nodes in the dynamic event subgraph. The extraction process is performed on each pair of nodes in the dynamic event subgraph that are directly or indirectly connected by edges. For a pair of nodes, all possible edge sequences connecting them are enumerated; each edge sequence constitutes a relationship path. Each edge in a relationship path carries its type information. For example, in the dynamic event subgraph, there might be a relationship path from node A to node B consisting of triggering edges.

[0063] The static knowledge subgraph searches for a consistent or equivalent relational path between the entity nodes corresponding to the changing node. This search is performed for each relational path extracted from the dynamic event subgraph. The starting and ending nodes of this relational path are found in the static knowledge subgraph, corresponding to the entity nodes in the static knowledge subgraph. All paths connecting these two corresponding entity nodes are searched in the static knowledge subgraph. It is then determined whether the relational path in the dynamic event subgraph is consistent or equivalent to a path found in the static knowledge subgraph. Consistency means that the edge sequences of the two paths are completely isomorphic and the edge types at corresponding positions are exactly the same. Equivalence means that although the edge sequences or specific edge types of the two paths are not completely identical, they semantically represent the same inter-entity association. The determination of equivalence requires adherence to predefined semantic equivalence rules. These rules are defined based on knowledge from the mining engineering field. For example, in a predefined knowledge graph, a directly connected relationship and a relationship connected through a short tunnel (e.g., less than 10 meters in length) can be defined as semantically equivalent; similarly, a triggering relationship in a dynamic event subgraph may be semantically equivalent to a path in a static knowledge subgraph composed of connecting edges and influencing edges.

[0064] When there is no relation path in the static knowledge subgraph that is consistent with or equivalent to the relation path, or when the relation type represented by the existing path contradicts the relation type in the dynamic event subgraph, the graph relations between the entity nodes in that group are determined to be conflicting entity relation pairs. A contradiction refers to the semantic exclusivity of relation types. For example, the dynamic event subgraph represents a triggering relation between two entities, while the static knowledge subgraph represents an isolating relation; these two relations constitute a contradiction. Each identified contradiction instance is recorded as a conflicting entity relation pair. A conflicting entity relation pair contains at least the following information: unique identifiers of the two entity nodes involved in the conflict, a relation path description extracted from the dynamic event subgraph, a corresponding path description found in the static knowledge subgraph, or a declaration of a non-existent path, and an identifier of the conflict type. Through the comparative analysis of the above system, step S3 outputs a series of knowledge conflict points that require further verification and correction.

[0065] In this embodiment of the invention, step S4, which combines mining succession plan data to verify and correct conflicting entity relationship pairs and generate spatial topology update relationships, is specifically implemented in the following way:

[0066] The analysis process extracts expected spatial topology changes from the mining succession plan data. This analysis focuses on the spatial position change sequence of the mining face advancement and its corresponding planned time nodes within the mining succession plan data obtained in step S1. The extraction of expected spatial topology changes involves identifying key spatial events described in the sequence that will fundamentally alter the relationships between entities in the mine safety knowledge graph. The analysis algorithm sequentially reads each record in the spatial position change sequence. Each record contains a planned time node and a set of spatial coordinates or geometric descriptions describing the mining status at that moment. The algorithm infers the spatial topology changes that occur between two consecutive planned time nodes by comparing the differences in spatial states. For example, if a significant shift in the boundary coordinates of a working face is observed between two consecutive planned time nodes, the algorithm infers that the working face has advanced, which will cause changes in the spatial relationships between the working face entity and the entities in the goaf behind, the entities in the mining area ahead, and the entities in adjacent roadways. Each extracted expected spatial topology change is represented as a structured expectation entry. An expected entry must contain at least the following elements: a unique identifier for one or more mining entities involved in the expected change; the type of spatial relationship change expected to occur, such as from adjacent to contained, or the creation of a new connection; the planned time point or time interval for the expected change to take effect; and the location index of the source planning data on which the change is based.

[0067] The rationality of relationship changes reflected in the dynamic event subgraph of conflicting entity relationship pairs is evaluated based on the expected spatial topology changes. For each conflicting entity relationship pair output in step S3, the evaluation process proceeds as follows: Key information is extracted from the conflicting entity relationship pair, including the unique identifiers of the two entity nodes involved in the conflict and the relationship path description extracted from the dynamic event subgraph representing the newly observed dynamic relationship between the two entities. A search is performed on all extracted expected spatial topology change entries to find those expected entries whose entity set involved in the expected change covers the two entities involved in the current conflicting entity relationship pair. If such an expected entry is found, a detailed matching evaluation is initiated. The detailed matching evaluation is performed from three dimensions: temporal consistency, spatial consistency, and conformity to the direction of change. The temporal consistency evaluation checks whether the time point at which the relationship change is observed in the dynamic event subgraph falls within a reasonable time window near the expected time interval recorded in the relevant expected spatial topology change entry. The size of the reasonable time window can be set according to the actual engineering situation; for example, for a daily plan, the reasonable time window can be set to 6 hours before and after the planned time point. Spatial consistency assessment checks whether the types of relationship changes reflected in the dynamic event subgraph semantically match or are compatible with the types of relationship changes expected in the spatial topology change expectation entries. Change direction conformity assessment checks whether the change trends reflected in the dynamic event subgraph are consistent with the engineering progress direction guided by the spatial topology change expectations. The assessment results from these three dimensions are combined, and predefined rules are used to determine reasonableness. The rules may involve assigning weighted scores to each dimension and calculating a weighted total score; a reasonableness judgment is made when the weighted total score exceeds a reasonableness threshold. The weighted scores are allocated based on the importance of each dimension in historical assessments; for example, time conformity has a weight of 0.3, spatial consistency has a weight of 0.5, and change direction conformity has a weight of 0.2.

[0068] The reasonableness judgment threshold is obtained through statistical learning by analyzing historical correct and incorrect cases. Specifically, samples of verified reasonable and unreasonable conflict relationships in historical cases are collected, and quantitative scores for each sample are extracted in three dimensions: temporal consistency, spatial consistency, and conformity of change direction. The model is trained using a logistic regression model, and the 0.5 probability cutoff point output by the model is used as the initial reasonableness judgment threshold. It is then fine-tuned based on feedback from field applications, for example, adjusted to 0.7, to achieve accurate screening of the reasonableness of dynamic events.

[0069] When deemed reasonable, the corresponding relationships in the static knowledge subgraph are corrected based on the dynamic event subgraph. The correction operation targets specific relationships in the static knowledge subgraph that are linked to by mutually conflicting entity relationships deemed reasonable. The logic of the correction is to adopt new relationships reflected in the dynamic event subgraph and verified as reasonable by the planned data to cover or supplement outdated or missing relationships in the static knowledge subgraph. Specific operations include: if no corresponding relationship exists in the static knowledge subgraph, a new edge is added between the corresponding two entity nodes in the static knowledge subgraph, with the edge type set according to the relationship path type in the dynamic event subgraph; if a relationship path exists in the static knowledge subgraph but contradicts the dynamic event subgraph, the type or attribute of the key edge in that relationship path is modified to align with the dynamic event subgraph. Figure 1 To.

[0070] When an assessment deems the relationship unreasonable, conflicting entity pairs are corrected based on anticipated spatial topology changes. The correction logic adjusts the understanding of conflicts revealed in the dynamic event subgraph based on the anticipated spatial topology changes, or directly provides a basis for updating the static knowledge subgraph. Specific operations include: creating or modifying relationships in the static knowledge subgraph based on anticipated spatial topology changes, ensuring the graph state aligns with the planned expectations, while simultaneously recording observations from the dynamic event subgraph as anomalous events awaiting verification.

[0071] A set of explicit spatial topology update relationships is generated through verification and correction operations. The spatial topology update relationship is a structured output set. Each record in the set defines a specific update operation on the predefined mine safety knowledge graph. A spatial topology update relationship record contains the following information: the target entity or entity pair to be updated; the type of operation to be performed, such as adding a relationship; the new relationship type and attribute values; and the source on which this update is based, i.e., dynamic event subgraph verification or expected spatial topology changes. This set serves as the direct input for step S5.

[0072] In this embodiment of the invention, step S5, which updates the mine safety knowledge graph based on spatial topology update relationships and performs safety hazard inference on real-time monitored sensor data, is specifically implemented in the following way:

[0073] Updating the mine safety knowledge graph based on spatial topology update relationships is the primary operation performed. Spatial topology update relationships are the structured output set generated in step S4, where each record defines a specific graph update operation. The update process is completed by traversing each spatial topology update relationship record in the set and executing the corresponding graph operation instructions according to the record content. For a spatial topology update relationship record specifying the operation type as "add relationship," the update system locates the target entity node specified in the record within the preset mine safety knowledge graph, then creates a new relationship edge between these two entity nodes, and assigns the new relationship type and attribute value specified in the record to this new edge. For a spatial topology update relationship record specifying the operation type as "delete relationship," the update system locates the target entity node in the preset mine safety knowledge graph and finds the relationship edge connecting these two nodes that matches the record description, then removes the relationship edge from the graph. For a spatial topology update relationship record specifying the operation type as "modify relationship attribute," after locating the specific relationship edge, the update system updates the attribute value corresponding to that relationship edge to the new value specified in the record. All update operations are executed sequentially within a single, independent transaction of the pre-defined mine safety knowledge graph to ensure data consistency. Upon completion, the pre-defined mine safety knowledge graph evolves into a new version reflecting the latest spatial topological understanding; this new version is the updated mine safety knowledge graph. The updated mine safety knowledge graph inherits all entities, all relationships except those that have been modified or deleted, and all safety rules from the original graph, and incorporates the latest changes carried by the spatial topological update relationships.

[0074] Matching entities, relationships, and safety rules in the updated mine safety knowledge graph with real-time monitoring sensor data is a prerequisite for reasoning. The matching process is a bidirectional association mapping. On one hand, for each newly arrived piece of real-time monitoring sensor data, the corresponding sensor entity node is searched in the updated mine safety knowledge graph based on the sensor number or location code carried in the data. Since sensor entity nodes in the updated mine safety knowledge graph are connected to specific roadway or working face spatial entity nodes through location relationships, this association can determine the specific location of the monitoring point represented by the real-time monitoring sensor data in the current spatial topology and its spatial region. On the other hand, all hazard triggering conditions defined in the updated mine safety knowledge graph are traversed. A hazard triggering condition is part of a safety rule, which is usually expressed as a logical expression. The logical expression of a hazard triggering condition consists of multiple sub-conditions connected by logical AND and logical OR operations. Each sub-condition is associated with a specific graph entity or entity relationship and contains one or more judgment conditions for the physical quantity monitored by the sensor. For example, a potential hazard triggering condition for gas accumulation might be defined as follows: the logical expression for the hazard triggering condition is that conditions A and B are simultaneously true; condition A is that the concentration reading of the associated gas sensor at a specific coal mining face is greater than the gas concentration alarm threshold; condition B is that the wind speed reading of the associated wind speed sensor at a specific coal mining face is less than the minimum allowable wind speed threshold. The matching process involves checking in real-time whether the latest real-time monitored sensor data makes the calculation result of the logical expression of a certain hazard triggering condition true. During calculation, the specific monitored physical quantity values ​​involved in the logical expression of the hazard triggering condition are extracted from the real-time monitored sensor data and compared with the threshold specified in the logical expression of the hazard triggering condition. The Boolean comparison result is then combined and calculated according to the logical expression of the hazard triggering condition.

[0075] When real-time monitored sensor data matches the hazard triggering conditions recorded in the updated mine safety knowledge graph (i.e., the logical expression of a hazard triggering condition is determined to be true), logical reasoning is performed based on the entity relationships and safety rules in the updated mine safety knowledge graph to generate a safety hazard warning information corresponding to the current spatial topology. This logical reasoning process first determines the hazard's impact range and possible propagation paths based on the updated spatial topological relationships between entities. Determining the hazard's impact range involves using the entity triggering the hazard as the core and performing a breadth-first or depth-first traversal along the spatial relationship edges in the updated mine safety knowledge graph. The termination condition of the traversal is determined by the propagation constraints defined in the safety rules. For example, for a gas over-limit hazard, the traversal can proceed along the ventilation airflow direction and is limited to connected roadway entities until it encounters an isolation entity such as an air door or a sealed wall. All traversed entities constitute the initial hazard's impact range. Possible propagation paths are all reachable paths from the core entity to other entities within the hazard's impact range, connected by spatial relationship edges. Based on determining the hazard's impact range and possible propagation paths, a comprehensive risk assessment is conducted in conjunction with threshold conditions in the safety rules. Comprehensive risk assessment quantifies or grades the potential severity of a hazard. Input parameters include the extent to which the core monitored physical quantity triggers the hazard exceeds its limits, the number of personnel or critical equipment involved within the hazard's impact area, and whether there are other coupling factors that exacerbate the risk along possible propagation paths, such as the accumulation of flammable materials. Safety regulations define a method for integrating these multi-dimensional parameters into a single risk assessment value. This method may involve assigning weights to each parameter and calculating a weighted sum. The weights are set based on statistical analysis of the impact of each parameter on accident consequences in historical accident cases. For example, through analysis of 100 past gas accidents, the weight for the extent to which the core parameter exceeds its limits is determined to be 0.6, the weight for the number of personnel within the hazard's impact area to be 0.3, and the weight for coupling risks to be 0.1. The weighted sum is mapped to a predefined risk level range, such as a score range of 0 to 100, with higher scores indicating greater risk. This range is further divided into four levels: blue (general risk), yellow (significant risk), orange (major risk), and red (extremely high risk). The final generated safety hazard warning information is a structured data object, which includes at least the following fields: a unique warning identifier, the type of hazard that triggered the warning, the location of the core entity corresponding to the warning, the timestamp of the warning generation, the determined risk level, a detailed list of entities within the hazard's impact range, a description of the main possible propagation paths, and a preliminary outline of handling suggestions based on safety rules. This safety hazard warning information is output to the monitoring terminal of the mine safety management personnel and can trigger different levels of audible and visual alarm signals according to the risk level, thereby completing intelligent safety hazard investigation and warning based on the latest spatial topology knowledge.

[0076] The gas concentration alarm threshold is set comprehensively based on the mine's absolute gas emission rate, ventilation design capacity, and the lower limit value stipulated in the "Coal Mine Safety Regulations". First, the lower limit of the threshold is determined according to the regulations, for example, 1.0%. Then, it is adjusted upwards by combining the measured data of coal seam gas content in the mining area with the maximum predicted emission rate of the mining face. The final threshold is written into the safety rules of the knowledge graph and is regularly reviewed and updated as the mining depth increases.

[0077] The minimum allowable wind speed threshold is set according to the roadway's purpose, cross-sectional area, and safety regulations for different types of roadways. For example, the minimum wind speed is set at 0.25 m / s for coal mining faces and 0.15 m / s for coal roadways in tunneling. This threshold serves as a key parameter for triggering ventilation-related hazards in the knowledge graph, used to determine whether the ventilation system meets the basic requirements for diluting methane and dust.

[0078] The threshold conditions in comprehensive risk assessment involve combinations and logical relationships among multiple physical quantity thresholds. For example, the gas concentration alarm threshold and the minimum allowable wind speed threshold are logically ANDed to form the triggering condition for gas accumulation hazards. The risk level classification thresholds are set based on the statistical relationship between the values ​​of various parameters and casualties in historical accident data. For example, a weighted risk score of 60 points and 80 points are set as the thresholds for relatively large risk and major risk, respectively.

[0079] Example 2: Figure 2 A schematic diagram of a knowledge graph-based intelligent system for identifying mine safety hazards is provided. This system includes the following modules:

[0080] The data acquisition module is used to acquire multi-source dynamic data of the mine, including real-time monitoring sensor data and mining succession plan data.

[0081] The event recognition module is used to analyze the temporal trends and spatial distribution of multi-source dynamic data. When a trend change in the state parameters of a downhole spatial entity is detected and triggers a series of causal temporal changes in its associated neighborhood, the trend change and causal temporal changes are aggregated and identified as a coordinated state change event.

[0082] The subgraph comparison module is used to construct dynamic event subgraphs based on collaborative state change events, and extract corresponding entities and relationships from the preset mine safety knowledge graph to form static knowledge subgraphs. By comparing the dynamic event subgraphs and the static knowledge subgraphs, conflicting entity relationship pairs are identified.

[0083] The relationship correction module is used to verify and correct conflicting entity relationship pairs by combining mining succession plan data, and generate spatial topology update relationships.

[0084] The graph update module is used to update the mine safety knowledge graph based on spatial topology update relationships and to perform safety hazard inference on real-time monitored sensor data.

[0085] All calculations involved in the embodiments are dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to the actual situation.

[0086] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0087] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0088] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0089] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0090] 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.

[0091] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent investigation of mine safety hazards based on knowledge graphs, characterized in that, Includes the following steps: S1. Acquire multi-source dynamic data of the mine, including real-time monitoring sensor data and mining succession plan data; S2. Analyze the temporal trend and spatial distribution of multi-source dynamic data. When a trend change in the state parameters of a downhole spatial entity is detected and a series of causal temporal changes are triggered in its associated neighborhood, the trend change and causal temporal changes are aggregated and identified as a coordinated state change event. S3. Construct a dynamic event subgraph based on collaborative state change events, and extract corresponding entities and relationships from the preset mine safety knowledge graph to form a static knowledge subgraph. Identify conflicting entity relationship pairs by comparing the dynamic event subgraph and the static knowledge subgraph. S4. Combine the mining succession plan data to verify and correct conflicting entity relationships, and generate spatial topology update relationships. S5. Update the mine safety knowledge graph based on the spatial topology update relationship, and perform safety hazard inference on the real-time monitored sensor data.

2. The intelligent method for identifying mine safety hazards based on knowledge graphs according to claim 1, characterized in that, Acquire multi-source dynamic data from the mine, including real-time monitoring sensor data and mining succession plan data, including: Real-time sensor data is collected from various safety monitoring sensors deployed underground. Read mining succession plan data from the mine production management system.

3. The intelligent method for identifying mine safety hazards based on knowledge graphs according to claim 2, characterized in that, The mining succession plan data read from the mine production management system includes the spatial position change sequence of the planned mining face advancement and the corresponding planned time nodes.

4. The intelligent method for identifying mine safety hazards based on knowledge graphs according to claim 1, characterized in that, Analyzing the temporal trends and spatial distribution of multi-source dynamic data, when a trend change in the state parameters of a downhole spatial entity is detected and triggers a series of causal temporal changes within its associated neighborhood, the trend change and causal temporal changes are aggregated and identified as a co-current state change event, including: By performing time-series analysis on real-time monitored sensor data, the initial change events that indicate a trend change in the state parameters of the downhole space entity can be identified. Based on the spatial neighborhood relationships between entities recorded in the mine safety knowledge graph, starting from the initial change event, we search for derivative changes with causal and temporal relationships within the associated neighborhood. Logically associate and encapsulate the initial change event with all the derivative changes it triggers to form a collaborative state change event.

5. The intelligent method for identifying mine safety hazards based on knowledge graphs according to claim 4, characterized in that, When searching based on the spatial neighborhood relationships between entities recorded in the mine safety knowledge graph, the criteria for determining causal temporal relationships are: subsequent derivative changes occur later in time than the initial change event, and spatially they conform to the physical propagation law of radiation from the location of the initial change event to the associated neighborhood.

6. The intelligent method for identifying mine safety hazards based on knowledge graphs according to claim 1, characterized in that, A dynamic event subgraph is constructed based on collaborative state change events, and corresponding entities and relationships are extracted from a pre-defined mine safety knowledge graph to form a static knowledge subgraph. Conflicting entity-relation pairs are identified by comparing the dynamic event subgraph and the static knowledge subgraph, including: Based on the changing nodes reflected in the dynamic event subgraph, extract the relationship paths between entities; In the static knowledge subgraph, search for whether there is a relationship path that is consistent with or equivalent to the relationship path between the entity nodes corresponding to the changed node; When there is no relation path in the static knowledge subgraph that is consistent with or equivalent to the relation path, or when the relation type represented by the existing path contradicts the relation type in the dynamic event subgraph, the graph relations between the entity nodes in this group are determined to be conflicting entity relation pairs.

7. The intelligent method for identifying mine safety hazards based on knowledge graphs according to claim 1, characterized in that, By combining mining succession plan data, conflicting entity relationships are verified and corrected to generate spatial topology update relationships, including: Analyze mining succession plan data to extract the expected spatial topological changes involved in the plan; Based on the anticipated spatial topological changes, assess the rationality of the relationship changes reflected by the dynamic event subgraph in mutually conflicting entity relationship pairs. When the assessment is deemed reasonable, the corresponding relationships in the static knowledge subgraph are modified based on the dynamic event subgraph; When an assessment is deemed unreasonable, conflicting entity relationships are corrected based on anticipated spatial topology changes, thereby generating updated spatial topology relationships.

8. The intelligent method for identifying mine safety hazards based on knowledge graphs according to claim 1, characterized in that, The mine safety knowledge graph is updated based on spatial topology updates, and safety hazard inferences are performed on real-time monitored sensor data, including: The entities, relationships, and safety rules in the updated mine safety knowledge graph are matched with real-time monitoring sensor data. When the real-time monitored sensor data matches the hazard triggering conditions recorded in the updated mine safety knowledge graph, logical reasoning is performed based on the entity relationships and safety rules in the updated mine safety knowledge graph to generate safety hazard warning information corresponding to the current spatial topology.

9. The intelligent method for identifying mine safety hazards based on knowledge graphs according to claim 8, characterized in that, Logical reasoning based on the relationships between entities and safety rules in the updated mine safety knowledge graph includes: determining the scope of impact and possible propagation paths of potential hazards based on the updated spatial topological relationships between entities, and conducting a comprehensive risk assessment in conjunction with the threshold conditions in the safety rules.

10. A knowledge graph-based intelligent mine safety hazard investigation system, used to implement the knowledge graph-based intelligent mine safety hazard investigation method according to any one of claims 1-9, characterized in that, Includes the following modules: The data acquisition module is used to acquire multi-source dynamic data of the mine, including real-time monitoring sensor data and mining succession plan data. The event recognition module is used to analyze the temporal trends and spatial distribution of multi-source dynamic data. When a trend change in the state parameters of a downhole spatial entity is detected and triggers a series of causal temporal changes in its associated neighborhood, the trend change and causal temporal changes are aggregated and identified as a coordinated state change event. The subgraph comparison module is used to construct dynamic event subgraphs based on collaborative state change events, and extract corresponding entities and relationships from the preset mine safety knowledge graph to form static knowledge subgraphs. By comparing the dynamic event subgraphs and the static knowledge subgraphs, conflicting entity relationship pairs are identified. The relationship correction module is used to verify and correct conflicting entity relationship pairs by combining mining succession plan data, and generate spatial topology update relationships. The graph update module is used to update the mine safety knowledge graph based on spatial topology update relationships and to perform safety hazard inference on real-time monitored sensor data.

Images