Dynamic knowledge graph construction method and system based on streaming data driving

Abstract

The invention relates to the technical field of intelligent operation and maintenance and graph construction, and discloses a dynamic knowledge graph construction method and system based on streaming data driving. The method comprises the following steps: acquiring multi-source streaming monitoring data; windowing processing is carried out by adopting a time sequence sensitive graph structure volume accumulation mechanism; mining a candidate edge pair set by adopting a causal attention guidance mechanism; executing edge set screening; executing an incremental consensus evolution task; and outputting the fusion version dynamic knowledge graph. In the prior art, a graph method which depends on off-line data and static graph model construction cannot realize abnormal perception and graph structure linkage evolution especially in a real-time operation and maintenance monitoring scene oriented to power equipment. Due to the fact that multiple stages of streaming data aggregation, graph embedding updating, structural consistency evolution and the like are introduced, closed-loop construction of node embedding, edge set screening and graph structure increment reconstruction is achieved, and the structural adaptability of the knowledge graph in an intelligent operation and maintenance scene is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance and knowledge graph construction technology, and in particular to a dynamic knowledge graph construction method and system based on streaming data. Background Technology

[0002] Currently, with the widespread application of intelligent sensing, the Internet of Things, and edge computing in power systems, the data generated during the operation of power equipment exhibits a trend of high frequency, multi-dimensionality, and heterogeneity. Especially in typical scenarios such as substations, distribution networks, and switching stations, state variables from power terminals (such as temperature, current, voltage, insulation resistance, and noise) are continuously collected and uploaded in a streaming manner, forming a large-scale equipment operation data stream with time-series characteristics. Building dynamic knowledge graphs based on this data has become an important means to support intelligent diagnosis, fault prediction, and operation optimization of power equipment.

[0003] However, most existing knowledge graph construction methods rely on static data sources and offline batch processing strategies, typically including steps such as entity extraction, relationship mining, graph structure construction, and attribute fusion. While these methods are suitable for scenarios with stable structures and periodically updated data, they exhibit the following significant shortcomings in practical applications: Weak incremental perception capability: When new equipment is added, sensors are expanded, or boundary conditions change, the graph cannot achieve local incremental construction and fusion based on the known graph structure, requiring overall reconstruction and resulting in huge system overhead. Insufficient heterogeneous multi-source processing: Power monitoring data sources are complex, including structured historical operation and maintenance records, semi-structured alarm logs, and unstructured images and sound signals. Traditional graph systems face problems such as inconsistent data standards, difficulties in time alignment, and semantic gaps when fusing these heterogeneous data. Difficulty in decoupling structural conflicts: During graph merging or updating, when multiple edge relationships point to the same entity pair simultaneously, or when the same entity has different semantic interpretations, traditional methods struggle to automatically disambiguate based on credibility, relying solely on manual intervention or prior rules, thus reducing automation capabilities. Unable to support graph evolution and iteration: Existing systems typically treat graphs as static products, lacking version control, graph evolution and backtracking mechanisms, making it difficult to support the knowledge reconstruction needs of device states evolving over time.

[0004] Therefore, there is an urgent need for a dynamic knowledge graph construction method based on streaming data to improve the graph quality and reasoning efficiency in practical power equipment management tasks such as anomaly identification, latent fault location, and complex state modeling. Summary of the Invention

[0005] To address the aforementioned technical shortcomings, the purpose of this invention is to propose a dynamic knowledge graph construction method based on streaming data. This method aims to solve the technical problem that existing graph methods, which rely on offline data and static graph models, cannot achieve the linkage and evolution of anomaly perception and graph structure, especially in the context of real-time operation and maintenance monitoring of power equipment.

[0006] To solve the above technical problems, the present invention adopts the following technical solution: The present invention provides a dynamic knowledge graph construction method based on streaming data. The dynamic knowledge graph construction method based on streaming data includes: Step S10: At time t, acquire multi-source streaming monitoring data from the power equipment monitoring system. Based on the multi-source streaming monitoring data, perform streaming data windowing processing using a time-sensitive graph structure convolutional aggregation mechanism, and output a set of abnormal indicators. and heterogeneous graph structure ; Step S20: Based on the set of abnormal indicators The incremental graph node construction task is performed using a neighborhood-aware incremental graph embedding update mechanism, and the embedded node set is output. ; Step S30: Based on the set of embedded nodes A causal attention-guided relation credibility evaluation mechanism is used to perform the candidate edge set mining task, and outputs a set of candidate edge pairs. ; Step S40: Based on the candidate edge pair set The graph structure incremental update task is performed using a structure consistency-driven edge set selection mechanism, outputting a graph change operation set. ; Step S50: For heterogeneous graph structures Based on the map change operation set An incremental consensus iteration mechanism is used to execute the fusion version evolution and reconstruction task, and output the fusion version dynamic knowledge graph.

[0007] Preferably, in step S10, multi-source streaming monitoring data from the power equipment monitoring system is acquired at time t. Based on the multi-source streaming monitoring data, a time-sensitive graph structure convolutional aggregation mechanism is used to perform streaming data windowing processing, and a set of abnormal indicators is output. and heterogeneous graph structure The steps specifically include: Step S101: Preset sampling time window ΔT, and acquire multi-source flow monitoring data from the power equipment monitoring system at time t, which includes current state data, voltage state data, temperature rise state data, power factor, and circuit breaker state data. Obtain the equipment topology connection relationship, and construct a heterogeneous graph structure based on the multi-source flow monitoring data and the equipment topology connection relationship using a causal constraint-based heterogeneous graphing method. , ,in, Indicates monitoring indicator nodes, Indicates a causal path edge; Step S102: For heterogeneous graph structures A time-weighted multi-head graph convolution aggregation method is used to perform one round of time-sensitive graph convolution aggregation operation, outputting a context embedding vector. ; Step S103: Based on context embedding vector A multi-factor anomaly detection task is performed using a saliency scoring mechanism driven by a graph Transformer architecture, outputting node saliency scores. When the node saliency score Greater than the preset significance score threshold At that time, the corresponding heterogeneous graph structure Included in the abnormal indicator set The final output is a set of abnormal indicators. .

[0008] Preferably, in step S20, based on the set of abnormal indicators... The incremental graph node construction task is performed using a neighborhood-aware incremental graph embedding update mechanism, and the embedded node set is output. The steps specifically include: Step S201: Based on the set of abnormal indicators A structure dependency traversal method based on time decay weights is used to extract heterogeneous graph structure-aware neighbors, resulting in a first-order neighbor set. ; Step S202: For the first-order neighbor set A multi-channel attention fusion method based on edge type is used to perform the neighbor context representation aggregation task, and the output is a neighbor context embedding vector. Step S203: Perform fusion processing on all neighbor context embedding vectors using an attention-weighted embedding compression mechanism to output a set of embedded nodes. .

[0009] Preferably, in step S30, based on the set of embedded nodes... A causal attention-guided relation credibility evaluation mechanism is used to perform the candidate edge set mining task, and outputs a set of candidate edge pairs. The steps specifically include: Step S301: Based on the embedded node set Constructing embedded node sequences ,in, This is the embedding vector of the first newly added abnormal indicator node in the set of embedded nodes; This is the embedding vector for the second newly added abnormal event node in the set of embedded nodes; This is the embedding vector of the nth newly added abnormal indicator node in the set of embedded nodes; The embedding vector for the second newly added abnormal event node in the embedded node set; the embedding vector for the i-th node in the embedded node sequence. and the j-th embedding vector Any pair of nodes formed A causal prior guidance matrix is ​​constructed by combining a pre-set graph prior knowledge base. ; Step S302: Based on the causal prior guidance matrix For any pair of nodes The task of evaluating relation credibility weights is performed using a causal-guided Transformer attention mechanism, and the relation credibility weights are output. ; Step S303: Based on the relationship credibility weight Execution threshold determination: When the relation credibility weight When the confidence weight exceeds the preset threshold, any pair of nodes is judged. For a strong association, any pair of nodes that satisfy the strong association relationship... Included in the candidate edge pair set The final output is a set of candidate edge pairs. .

[0010] Preferably, in step S30, the relationship credibility weight... The formula is expressed as: ;in, This is the query vector projection matrix, used to map node embeddings to query vectors; This is the key vector projection matrix, used to map node embeddings to key vectors; For embedding dimensions, used for normalized scaling; It serves as a causal guiding factor, used to control the degree to which causal priors enhance the attention outcome; This is the standard normalization function, used for normalization processing.

[0011] Preferably, in step S40, based on the candidate edge pair set... The graph structure incremental update task is performed using a structure consistency-driven edge set selection mechanism, outputting a graph change operation set. The steps specifically include: Step S401: For the candidate edge pair set Any pair of nodes in The intersection-union ratio of the neighbor set is calculated using the neighborhood overlap calculation method. And based on the intersection and union ratio of the neighbor sets Relationship credibility weight A structural consistency weighted score is constructed using linear interpolation. , ,in, Assign factors to the weights. ; Step S402: Weight the structural consistency score Greater than the preset structural consistency threshold any pair of nodes Edge pairs are included in the effective graph increment edge set; the effective graph increment edge set is used to update the graph structure; Step S403: Obtain the heterogeneous graph structure Based on the effective graph increment edge set, an incremental graph merging method based on conflict resolution is used to process the heterogeneous graph structure. Perform structural fusion and version update processing, and output a set of map change operations. .

[0012] Preferably, in step S403, the heterogeneous graph structure is processed using an incremental graph merging method based on conflict resolution according to the effective graph increment edge set. Perform structural fusion and version update processing, and output a set of map change operations. The steps specifically include: Step S4031: Candidate edge conflict detection stage: Obtain heterogeneous graph structure , ,in, Indicates monitoring indicator nodes, Represents causal path edges; for each edge to be merged in the effective graph increment edge set. and any pair of nodes In heterogeneous graph structure The following conflict detection operations are performed in the middle: like Then define the edges to be incrementally merged. For structural conflict edges, output the conflict marking function. , =1; like Then define the edges to be incrementally merged. For newly added edges, output the collision flag function. , =0; like and Furthermore, a single node still exists in the effective graph increment edge set. Make any pair of nodes With a single node If a closed-loop structure is formed, then define the edges to be incrementally merged. Output a conflict marking function for redundant triangular closed conflict edges. , =1; Step S4032: Conflict Edge Calculation and Structure Fusion Stage: Targeting the conflict marking function Edges to be merged with a weight of 1 are updated using the principle of edge weight steady state preservation, and conflict resolution results are generated. Step S4033: Finally, based on the conflict resolution operation results, output the map change operation set using a version difference tracking method. .

[0013] This invention also provides a dynamic knowledge graph construction system based on streaming data, comprising: The multi-source streaming sensing module is used to acquire multi-source streaming monitoring data from the power equipment monitoring system at time t. Based on the multi-source streaming monitoring data, it uses a time-sensitive graph structure convolutional aggregation mechanism to perform streaming data windowing processing tasks and outputs a set of abnormal indicators. and heterogeneous graph structure ; The graph node construction module is used for constructing graph nodes based on a set of anomaly indicators. The incremental graph node construction task is performed using a neighborhood-aware incremental graph embedding update mechanism, and the embedded node set is output. ; The relation edge mining module is used to mine relationships based on the set of embedded nodes. A causal attention-guided relation credibility evaluation mechanism is used to perform the candidate edge set mining task, and outputs a set of candidate edge pairs. ; The graph structure evolution module is used to determine the evolution of the graph structure based on the set of candidate edge pairs. The graph structure incremental update task is performed using a structure consistency-driven edge set selection mechanism, outputting a graph change operation set. ; The fusion evolution and reconstruction module is used for heterogeneous graph structures. Based on the map change operation set An incremental consensus iteration mechanism is used to execute the fusion version evolution and reconstruction task, and output the fusion version dynamic knowledge graph.

[0014] The present invention also provides a dynamic knowledge graph construction device based on streaming data, comprising: a memory, a processor, and a dynamic knowledge graph construction program based on streaming data stored in the memory and executable on the processor. When the dynamic knowledge graph construction program based on streaming data is executed by the processor, it implements a dynamic knowledge graph construction method based on streaming data.

[0015] The present invention also provides a computer program product, including a dynamic knowledge graph construction program based on streaming data, wherein the dynamic knowledge graph construction program based on streaming data is executed by a processor to implement the dynamic knowledge graph construction method based on streaming data.

[0016] The beneficial effects of this invention are as follows: By introducing a streaming data processing mechanism based on time-sensitive graph structure convolutional aggregation, this invention can dynamically extract key anomaly indicators during the real-time arrival of monitoring data, and combine graph embedding update and causal attention mechanism to achieve adaptive evolution of graph structure, which significantly improves the response speed of graph to changes in the state of power equipment and the ability to express anomaly structures.

[0017] This invention employs a structural consistency-driven edge set selection mechanism and an incremental consensus evolution strategy to maintain the stability and evolutionary continuity of the graph topology during candidate edge pair conflict resolution and heterogeneous graph fusion and reconstruction. This effectively avoids the structural disorder and version drift problems existing in traditional static graph updates, thereby improving the accuracy of dynamic graph fusion for complex power operation and maintenance scenarios. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the first embodiment of a dynamic knowledge graph construction method based on streaming data according to the present invention.

[0020] Figure 2 This is a heatmap diagram of the consistency factor scoring of a first embodiment of a dynamic knowledge graph construction method based on streaming data according to the present invention.

[0021] Figure 3 This is a schematic diagram of a device for a dynamic knowledge graph construction method based on streaming data according to the present invention. Detailed Implementation

[0022] 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Example 1: As Figure 1 The diagram shown is a flowchart of the first embodiment of the dynamic knowledge graph construction method based on streaming data driven by the present invention, which presents the first embodiment of the dynamic knowledge graph construction method based on streaming data driven by the present invention.

[0024] In the first embodiment, the dynamic knowledge graph construction method based on streaming data includes: Step S10: At time t, acquire multi-source streaming monitoring data from the power equipment monitoring system. Based on the multi-source streaming monitoring data, perform streaming data windowing processing using a time-sensitive graph structure convolutional aggregation mechanism, and output a set of abnormal indicators. and heterogeneous graph structure ; It should be noted that the "time-sensitive graph structure convolutional aggregation mechanism" in this step refers to the following: when processing multi-source streaming monitoring data, using a time sliding window as the basic unit, real-time monitoring values ​​from different modes such as temperature, current, voltage, partial discharge, vibration, and switching status are aligned according to a unified timestamp and constructed into data blocks with graph structure constraints. This mechanism includes: dynamic time-series encoding based on edge weights using a time decay function; normalized feature construction based on heterogeneous mode alignment; graph convolutional aggregation based on node neighborhood correlation; and a cross-window residual smoothing strategy for suppressing short-term spike noise.

[0025] Understandably, this invention enables feature aggregation and structured representation of multi-source monitoring data within each short-cycle sliding window as streaming data continuously arrives, allowing for the simultaneous identification of abnormal fluctuations, trend shifts, and cross-modal correlation features. The graph-structured convolutional aggregation mechanism automatically learns the correlation strength of different monitoring indicators within the window, thereby accurately outputting a set of abnormal indicators that can represent changes in operating status and generating a local heterogeneous graph structure characterizing the current health status of the equipment. This processing method ensures a response to drastic fluctuations in the status of field equipment with a latency of only seconds, improving the accuracy of subsequent inference chains in the dynamic knowledge graph.

[0026] It should be understood that, unlike traditional methods that rely on static batch data to construct graphs, this step expands the anomaly detection method, which originally relied solely on numerical threshold comparisons, into a streaming structured recognition method that combines "time sensitivity" and "structure awareness" by introducing time decay coding, structural convolution aggregation, and cross-window residual smoothing. Traditional methods cannot identify the collaborative changes between multiple indicators and are prone to misjudgment in scenarios with short-term drastic fluctuations, pattern drift, or structural coupling anomalies; while the mechanism adopted in this invention can simultaneously capture local trend anomalies and cross-modal correlation anomalies, thereby effectively improving the accuracy of anomaly recognition in complex power scenarios.

[0027] For example, during routine monitoring of a 110kV line, the following streaming data was received within a 3-minute sliding window: the current in phase A of the line increased by 18% in a short period of time, the conductor temperature showed a continuous upward trend (6°C higher than the baseline), and the partial discharge pulse count increased from 8 to 31, while phases B and C showed no significant changes during the same period. Traditional threshold methods can only identify current and temperature as "exceeding limits," but cannot understand the coupling relationship between the three. In this step, the multi-source data within the window is first constructed into a set of nodes (temperature nodes, current nodes, and partial discharge nodes), and edge weights are generated based on the changing trends and statistical correlations. Subsequently, a local anomaly subgraph consisting of "current surge—temperature rise—partial discharge jump" is obtained through time-sensitive convolution aggregation, thereby simultaneously incorporating the three monitored quantities into the anomaly index set and outputting a heterogeneous graph structure that reflects the potential risk of poor contact or local insulation degradation, which is used to guide subsequent incremental evolution of the graph.

[0028] Step S20: Based on the set of abnormal indicators The incremental graph node construction task is performed using a neighborhood-aware incremental graph embedding update mechanism, and the embedded node set is output. ; It should be noted that the "neighborhood-aware incremental graph embedding update mechanism" in this step refers to the following: After the anomaly indicator set is generated, candidate neighborhoods are constructed based on the existing nodes and edges in the existing graph structure. Factors such as timestamp differences between nodes, attribute similarity, and edge weight fluctuations are used to dynamically evaluate the influence weight of candidate neighborhoods on the embedding of new nodes. Based on this, a graph embedding algorithm (such as GraphSAGE or GAT) is used to perform incremental embedding calculations on the candidate nodes and their adjacent edges, generating embedding vectors corresponding to the current anomaly state. These vectors are then written into the graph as attributes of the new nodes, achieving uninterrupted incremental construction.

[0029] Understandably, this mechanism enables dynamic responses to sudden anomalies without reconstructing the entire graph. Each newly added graph node derived from anomaly indicators not only possesses independent representation capabilities in the attribute space but also completes semantic alignment and behavioral inference preparation through neighborhood context information, ensuring that subsequent knowledge graph reasoning modules can accurately identify new risk patterns or event chains. The neighborhood awareness mechanism effectively avoids information silos among newly added nodes, improving the connectivity and semantic integrity of the graph structure.

[0030] For example, in a 220kV substation monitoring system, a drastic fluctuation in the A-phase voltage of the busbar (a 9.6% decrease from the baseline) was detected within 20 seconds, along with a simultaneous 5°C rise in the main transformer oil temperature and a sudden change in transient flux leakage. These three monitoring points were identified as high-risk indicators from the set of abnormal indicators. Using these three indicators as the center, their direct neighbors in the current graph were retrieved, including busbar protection devices, ring main unit temperature nodes, and grounding current nodes, constructing a local candidate neighborhood graph. Subsequently, based on a graph attention mechanism, behavioral features in the neighborhood (such as temperature rise rate and state change frequency) were aggregated, generating a 5-dimensional embedding vector for each new node, and automatically establishing edge connections based on the degree of association. Finally, an embedded node set was generated and written into the graph, realizing incremental graph modeling of sudden equipment state anomalies, laying a structural foundation for subsequent fault causal chain mining.

[0031] Step S30: Based on the set of embedded nodes A causal attention-guided relation credibility evaluation mechanism is used to perform the candidate edge set mining task, and outputs a set of candidate edge pairs. ; It should be noted that the "causal attention-guided relationship credibility assessment mechanism" in this step refers to: for the set of embedded nodes obtained in step S20, based on the time attributes, feature embedding vectors, sources of abnormal indicators and potential causal propagation paths between nodes, the node pairs are combined, and an attention scoring structure with causal bias terms is used to calculate the relationship credibility between nodes.

[0032] It should be understood that, compared to traditional methods that use static indicators such as correlation coefficients, mutual information, and cosine similarity to mine node relationships, the innovation of this step lies in the introduction of a "causal attention" modeling approach. This invention uses causal structures such as temporal sequence, anomaly type, and neighborhood propagation trajectory as attention bias terms, enabling the attention score to reflect the true directionality and strength of the anomaly propagation chain. This makes the relationship credibility assessment more robust and more consistent with the equipment's operating logic. This mechanism is particularly suitable for typical "anomaly chain propagation" phenomena in power monitoring scenarios, such as short circuit → current jump → temperature rise → partial discharge enhancement.

[0033] For example, during the operation of a 110kV distribution ring network, the following abnormal node embedding vectors were detected within one minute: Node N1: The temperature of the cable joint continues to rise, with a sudden increase at timestamp 14:03; At node N2, the distortion coefficient of phase A current fluctuates more significantly starting at 14:02:45. Node N3: The partial discharge pulse count jumps at 14:03:10; Node N4: The voltage of the adjacent bus dropped slightly at 14:04.

[0034] Based on the above nodes, a causal prior matrix is ​​constructed, where: N2 has the earliest timestamp, therefore the highest causal bias; N1 and N3 appear together in adjacent windows, exhibiting coupling gain; N4's fluctuation appears later, placing it in an "affected" position in the initial propagation chain. Based on this prior, the following attention scores are obtained under the multi-head attention mechanism: Att(N2→N1) = 0.86 (consistent with the physical logic of current fluctuation leading to temperature rise) Att(N1→N3) = 0.78 (temperature rise will lead to a decrease in local insulation performance → enhanced partial discharge) Att(N3→N4) = 0.64 (weak correlation between partial discharge and voltage fluctuation) Att(N4→N1) = 0.19 (not conforming to causal directionality). The edge pairs with the highest final credibility fusion calculation scores are: {N2, N1}, {N1, N3}, {N3,N4}, forming a candidate edge pair set for subsequent incremental graph merging and updating, indicating that the "abnormal propagation chain" can be automatically reconstructed and illogical reverse relationships can be filtered out.

[0035] Step S40: Based on the candidate edge pair set The graph structure incremental update task is performed using a structure consistency-driven edge set selection mechanism, outputting a graph change operation set. ; It should be noted that the "structural consistency-driven edge set selection mechanism" in this step refers to introducing the existing adjacency structure, node semantic similarity, and relation semantic stability in the heterogeneous graph structure as consistency constraint factors into the candidate edge pair set, and assigning a confidence score to each candidate edge by constructing a consistency scoring function. This mechanism comprehensively considers three types of structural consistency features: (1) structural adjacency reconstruction degree, that is, whether the newly added edge helps to restore the structurally missing areas in the original graph; (2) node semantic fusion degree, that is, whether the embedding vectors of the nodes at both ends of the edge have high semantic proximity; (3) relation type confidence, that is, whether the proposed new relation conforms to the frequency distribution of existing relation types and contextual logic. Weighted linear fusion and dynamic thresholds are used to select edges based on the multi-factor scoring results, and finally a graph change operation set is formed.

[0036] Understandably, the edge set selection mechanism driven by structural consistency can significantly suppress erroneous connections introduced by noisy data or non-causal events, effectively ensuring the structural stability and semantic consistency of the knowledge graph during incremental evolution.

[0037] It should be understood that, compared to the static edge confidence ranking and manually set thresholds commonly used in traditional graph construction methods, this invention introduces a structural consistency factor as a dynamic constraint. This not only improves the accuracy and adaptability of edge selection but also possesses stronger context awareness and adaptive generalization capabilities, making it particularly suitable for dynamic graph update tasks in scenarios with rapid changes in device status and sudden bursts of indicator noise. For example, as Figure 2 As shown, the heatmap illustrates the score distribution of candidate edges across three structural consistency factors: structural adjacency reconstruction degree, node semantic fusion degree, and relation type confidence degree. Darker colors in the graph indicate higher structural consistency scores, and concentrated hot areas represent edge sets prioritized for retention under multi-factor fusion evaluation. This heatmap provides a clear visual identification of edge pairs with high credibility and consistency, validating the technical advantages of the "structural consistency-driven edge set selection mechanism" in this invention in noise suppression and relation credibility enhancement. It is particularly suitable for high-quality incremental update tasks of graph structures in dynamic scenarios.

[0038] Step S50: For heterogeneous graph structures Based on the map change operation set An incremental consensus iteration mechanism is used to execute the fusion version evolution and reconstruction task, and output the fusion version dynamic knowledge graph.

[0039] It should be noted that the "incremental consensus iteration mechanism" in this step refers to the following: upon receiving the set of graph change operations, instead of directly performing a batch replacement of the original graph, it performs local graph structure reconstruction, semantic consistency verification, and cross-version consistency detection based on each candidate change operation, forming multiple candidate evolutionary branches. Subsequently, a consensus determination mechanism is adopted for multiple evolutionary branches. By comparing the structural matching degree, semantic stability, and reasoning consistency of each branch in the historical version sequence, the optimal evolutionary path is dynamically selected for version merging. This mechanism integrates the temporal evolution trajectory and structural semantic features of the graph, ensuring that logical closed loops and evolutionary controllability are maintained during the continuous evolution of the heterogeneous graph structure.

[0040] Understandably, by introducing an incremental consensus iteration mechanism, this step can effectively improve the stability and accuracy of the knowledge graph version evolution process, avoiding problems such as structural drift, concept conflicts, or broken causal chains in environments with rapidly changing heterogeneous data. Simultaneously, because the mechanism possesses the ability to evaluate multiple branches of the evolution path and constrain global consistency, it can continuously build fused version knowledge graphs, adapting to the deployment needs of online scenarios driven by streaming data.

[0041] It should be understood that, compared with traditional static merging strategies or graph update methods based on single-version voting mechanisms, this invention breaks through the limitations of existing methods in the face of graph heterogeneity, multi-source conflicts and continuous evolution, and improves the stability, adaptability and semantic coherence of the fused version. It is particularly suitable for knowledge reasoning and event modeling in high-frequency changing data scenarios such as industrial equipment and power grid systems.

[0042] For example, consider the following sets of graph change operations received during three consecutive rounds of evolution of a power grid equipment knowledge graph: The first round includes the newly added edge "Device Node A ↔ Alarm Event X"; The second round includes the relationship replacement of "device node A ↔ fault event Y"; The third round includes an update of the upstream and downstream relationships of "Event X ↔ Repair Operation Z".

[0043] In traditional static merging strategies, these three rounds of operations will be directly superimposed on the original graph, which may cause structural redundancy, semantic conflicts (such as inconsistent logic between X and Y) or contextual breaks (Z and Y becoming disconnected).

[0044] The incremental consensus iteration mechanism of this invention is as follows: each round of operation is treated as an independent evolutionary candidate branch; the structural matching degree (e.g., whether the node degree change is reasonable), semantic stability (e.g., whether the node embedding distance change is too large), and reasoning consistency (whether it supports the existing reasoning path) of these branches in the historical versions of the graph are calculated; after multiple rounds of consensus voting and normalized scoring, the optimal evolutionary path is selected, such as retaining "device node A ↔ fault event Y" and "event X ↔ repair operation Z", and constructing auxiliary edges between Y and Z to maintain semantic closed loop; finally, a fusion version dynamic knowledge graph with compact structure, semantic coherence, and evolutionary rationality is formed.

[0045] Example 2: Furthermore, the present invention provides a dynamic knowledge graph construction system based on streaming data, employing a dynamic knowledge graph construction method based on streaming data as described in the above embodiments, which can solve the technical problem of dynamic knowledge graph construction based on streaming data. Compared with the prior art, the beneficial effects of the dynamic knowledge graph construction system based on streaming data provided by the present invention are the same as those of the dynamic knowledge graph construction method based on streaming data provided in the above embodiments, and other technical features of the dynamic knowledge graph construction system based on streaming data are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0046] Example 3: This invention provides a dynamic knowledge graph construction device based on streaming data. Please refer to... Figure 3A dynamic knowledge graph construction device based on streaming data drive 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 dynamic knowledge graph construction method based on streaming data drive described in Embodiment 1 above. The dynamic knowledge graph construction device based on streaming data drive in this embodiment of the invention may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. This dynamic knowledge graph construction device based on streaming data drive is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the invention. A streaming data-driven dynamic knowledge graph construction device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the streaming data-driven dynamic knowledge graph construction device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An I / O interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. Communication device 1009 allows a streaming data-driven dynamic knowledge graph construction device to communicate wirelessly or wiredly with other devices to exchange data. While the figure illustrates a streaming data-driven dynamic knowledge graph construction device with various systems, it should be understood that implementing or having all of the systems shown is not required. More or fewer systems may be implemented alternatively.

[0047] Example 4: This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method for constructing a dynamic knowledge graph based on streaming data. The computer program product provided by this invention can solve the technical problem of constructing a dynamic knowledge graph based on streaming data. Compared with the prior art, the beneficial effects of the computer program product provided by this invention are the same as those of the method for constructing a dynamic knowledge graph based on streaming data provided in the above embodiments, and will not be repeated here.

[0048] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this invention.

[0049] It should be understood that the various parts disclosed in this invention can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

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

Claims

1. A method for constructing dynamic knowledge graphs based on streaming data, characterized in that, The methods include: Step S10: At time t, acquire multi-source streaming monitoring data from the power equipment monitoring system. Based on the multi-source streaming monitoring data, perform streaming data windowing processing using a time-sensitive graph structure convolutional aggregation mechanism, and output a set of abnormal indicators. and heterogeneous graph structure ; Step S20: Based on the set of abnormal indicators The incremental graph node construction task is performed using a neighborhood-aware incremental graph embedding update mechanism, and the embedded node set is output. ; Step S30: Based on the set of embedded nodes A causal attention-guided relation credibility evaluation mechanism is used to perform the candidate edge set mining task, and outputs a set of candidate edge pairs. ; Step S40: Based on the candidate edge pair set The graph structure incremental update task is performed using a structure consistency-driven edge set selection mechanism, outputting a graph change operation set. ; Step S50: For heterogeneous graph structures Based on the map change operation set An incremental consensus iteration mechanism is used to execute the fusion version evolution and reconstruction task, and output the fusion version dynamic knowledge graph.

2. The method for constructing a dynamic knowledge graph based on streaming data as described in claim 1, characterized in that, In step S10, multi-source streaming monitoring data from the power equipment monitoring system is acquired at time t. Based on the multi-source streaming monitoring data, a time-sensitive graph structure convolutional aggregation mechanism is used to perform streaming data windowing processing, and a set of abnormal indicators is output. and heterogeneous graph structure The steps specifically include: Step S101: Preset sampling time window ΔT, and acquire multi-source flow monitoring data from the power equipment monitoring system at time t, which includes current state data, voltage state data, temperature rise state data, power factor, and circuit breaker state data. Obtain the equipment topology connection relationship, and construct a heterogeneous graph structure based on the multi-source flow monitoring data and the equipment topology connection relationship using a causal constraint-based heterogeneous graphing method. , ,in, Indicates monitoring indicator nodes, Indicates a causal path edge; Step S102: For heterogeneous graph structures A time-weighted multi-head graph convolution aggregation method is used to perform one round of time-sensitive graph convolution aggregation operation, outputting a context embedding vector. ; Step S103: Based on context embedding vector A multi-factor anomaly detection task is performed using a saliency scoring mechanism driven by a graph Transformer architecture, outputting node saliency scores. When the node saliency score Greater than the preset significance score threshold At that time, the corresponding heterogeneous graph structure Included in the abnormal indicator set The final output is a set of abnormal indicators. .

3. The method for constructing a dynamic knowledge graph based on streaming data as described in claim 1, characterized in that, In step S20, based on the set of abnormal indicators The incremental graph node construction task is performed using a neighborhood-aware incremental graph embedding update mechanism, and the embedded node set is output. The steps specifically include: Step S201: Based on the set of abnormal indicators A structure dependency traversal method based on time decay weights is used to extract heterogeneous graph structure-aware neighbors, resulting in a first-order neighbor set. ; Step S202: For the first-order neighbor set A multi-channel attention fusion method based on edge type is used to perform the neighbor context representation aggregation task, and the output is a neighbor context embedding vector. Step S203: Perform fusion processing on all neighbor context embedding vectors using an attention-weighted embedding compression mechanism to output a set of embedded nodes. .

4. The method for constructing a dynamic knowledge graph based on streaming data as described in claim 1, characterized in that, In step S30, based on the set of embedded nodes A causal attention-guided relation credibility evaluation mechanism is used to perform the candidate edge set mining task, and outputs a set of candidate edge pairs. The steps specifically include: Step S301: Based on the set of embedded nodes Constructing embedded node sequences ,in, This is the embedding vector of the first newly added abnormal indicator node in the set of embedded nodes; This is the embedding vector for the second newly added abnormal event node in the set of embedded nodes; This is the embedding vector of the nth newly added abnormal indicator node in the set of embedded nodes; The embedding vector for the second newly added abnormal event node in the embedded node set; the embedding vector for the i-th node in the embedded node sequence. and the j-th embedding vector Any pair of nodes formed A causal prior guidance matrix is ​​constructed by combining a pre-set graph prior knowledge base. ; Step S302: Based on the causal prior guidance matrix For any pair of nodes The task of evaluating relation credibility weights is performed using a causal-guided Transformer attention mechanism, and the relation credibility weights are output. ; Step S303: Based on the relationship credibility weight Execution threshold determination: When the relation credibility weight When the confidence weight exceeds the preset threshold, any pair of nodes is judged. For a strong association, any pair of nodes that satisfy the strong association relationship will be considered. Included in the candidate edge pair set The final output is a set of candidate edge pairs. .

5. The method for constructing a dynamic knowledge graph based on streaming data as described in claim 4, characterized in that, In step S30, the relationship credibility weight The formula is expressed as: ;in, This is the query vector projection matrix, used to map node embeddings to query vectors; This is the key vector projection matrix, used to map node embeddings to key vectors; For embedding dimensions, used for normalized scaling; It serves as a causal guiding factor, used to control the degree to which causal priors enhance the attention outcome; This is the standard normalization function, used for normalization processing.

6. The method for constructing a dynamic knowledge graph based on streaming data as described in claim 4, characterized in that, In step S40, based on the candidate edge pair set The graph structure incremental update task is performed using a structure consistency-driven edge set selection mechanism, outputting a graph change operation set. The steps specifically include: Step S401: For the candidate edge pair set Any pair of nodes in The intersection-union ratio of the neighbor set is calculated using the neighborhood overlap calculation method. And based on the intersection and union ratio of the neighbor sets Relationship credibility weight A structural consistency weighted score is constructed using linear interpolation. , ,in, Assign factors to the weights. ; Step S402: Weight the structural consistency score Greater than the preset structural consistency threshold any pair of nodes Edge pairs are included in the effective graph increment edge set; the effective graph increment edge set is used to update the graph structure; Step S403: Obtain the heterogeneous graph structure Based on the effective graph increment edge set, an incremental graph merging method based on conflict resolution is used to process the heterogeneous graph structure. Perform structural fusion and version update processing, and output a set of map change operations. .

7. The method for constructing a dynamic knowledge graph based on streaming data as described in claim 6, characterized in that, In step S403, the heterogeneous graph structure is processed using an incremental graph merging method based on conflict resolution, according to the effective graph increment edge set. Perform structural fusion and version update processing, and output a set of map change operations. The steps specifically include: Step S4031: Candidate edge conflict detection stage: Obtain heterogeneous graph structure , ,in, Indicates monitoring indicator nodes, Represents causal path edges; for each edge to be merged in the effective graph increment edge set. and any pair of nodes In heterogeneous graph structure The following conflict detection operations are performed in the middle: like Then define the edges to be incrementally merged. For structural conflict edges, output the conflict marking function. , =1; like Then define the edges to be incrementally merged. For newly added edges, output the collision flag function. , =0; like and Furthermore, a single node still exists in the effective graph increment edge set. Make any pair of nodes With a single node If a closed-loop structure is formed, then define the edges to be incrementally merged. Output a conflict flag function for redundant triangular closed conflict edges. , =1; Step S4032: Conflict Edge Calculation and Structure Fusion Stage: Targeting the conflict marking function Edges to be merged with a weight of 1 are updated using the principle of edge weight steady state preservation, and conflict resolution results are generated. Step S4033: Finally, based on the conflict resolution operation results, output the map change operation set using a version difference tracking method. .

8. A dynamic knowledge graph construction system based on streaming data, applied to the dynamic knowledge graph construction method based on streaming data as described in any one of claims 1 to 7, characterized in that, The dynamic knowledge graph construction system based on streaming data includes: The multi-source streaming sensing module is used to acquire multi-source streaming monitoring data from the power equipment monitoring system at time t. Based on the multi-source streaming monitoring data, it uses a time-sensitive graph structure convolutional aggregation mechanism to perform streaming data windowing processing tasks and outputs a set of abnormal indicators. and heterogeneous graph structure ; The graph node construction module is used for constructing graph nodes based on a set of anomaly indicators. The incremental graph node construction task is performed using a neighborhood-aware incremental graph embedding update mechanism, and the embedded node set is output. ; The relation edge mining module is used to mine relationships based on the set of embedded nodes. A causal attention-guided relation credibility evaluation mechanism is used to perform the candidate edge set mining task, and outputs a set of candidate edge pairs. ; The graph structure evolution module is used to determine the evolution of the graph structure based on the set of candidate edge pairs. The graph structure incremental update task is performed using a structure consistency-driven edge set selection mechanism, outputting a graph change operation set. ; The fusion evolution and reconstruction module is used for heterogeneous graph structures. Based on the map change operation set An incremental consensus iteration mechanism is used to execute the fusion version evolution and reconstruction task, and output the fusion version dynamic knowledge graph.

9. A dynamic knowledge graph construction device based on streaming data, characterized in that, The dynamic knowledge graph construction device based on streaming data includes: a memory, a processor, and a dynamic knowledge graph construction program based on streaming data stored in the memory and executable on the processor. When the dynamic knowledge graph construction program based on streaming data is executed by the processor, it implements a dynamic knowledge graph construction method based on streaming data as described in any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes a dynamic knowledge graph construction program based on streaming data, which, when executed by a processor, implements a dynamic knowledge graph construction method based on streaming data as described in any one of claims 1 to 7.

Images