An event-driven power distribution network time sequence knowledge graph construction method and system

Abstract

The application discloses a power distribution network time sequence knowledge graph construction method and system based on event driving, and the method comprises the following steps: extracting target multi-source heterogeneous data according to a preset event triggering mechanism to obtain target events, and performing standard vectorization and labeling on the target events to obtain feature vectors and event labels corresponding to the target events; mapping the feature vectors of each target event into standardized graph node representations, determining the node types of various graph nodes according to the event labels of each target event, generating initial relationship edges between nodes according to the time sequence, spatial proximity relationship and causal correlation characteristics of various target events, and forming a graph structure; and embedding and updating a time sequence knowledge graph snapshot according to a preset node updating rule to generate graph embedding vectors of each node at the current time, that is, obtaining a power distribution network time sequence knowledge graph, which can improve fault diagnosis efficiency and assist in scheduling optimization decision-making.

Description

Technical Field

[0001] This invention belongs to the field of knowledge graph construction technology, and in particular relates to a method and system for constructing a time-series knowledge graph for power distribution networks based on event-driven principles. Background Technology

[0002] With the rapid development of new power systems, distribution networks are undertaking increasingly complex tasks, especially with the significantly increased pressure on regulation and fault handling of distributed and diversified loads connected to new energy sources. Traditional distribution network operation and maintenance mainly rely on manual judgment and rule-based knowledge bases, lacking the ability to dynamically model and intelligently reason about complex events.

[0003] Existing knowledge graph technology has been initially applied in the power industry, but it is mostly focused on static equipment information management. There is a lack of dynamic knowledge graph construction schemes for event time-series evolution and causal reasoning, which makes it difficult to meet the needs of intelligent fault diagnosis, risk warning and decision support. Summary of the Invention

[0004] This invention provides a method and system for constructing a distribution network time-series knowledge graph based on event-driven principles, which addresses the technical problem that existing knowledge graph construction schemes are insufficient for intelligent fault diagnosis and risk warning.

[0005] In a first aspect, the present invention provides a method for constructing a distribution network time-series knowledge graph based on event-driven principles, comprising:

[0006] Acquire multi-source heterogeneous data in the power distribution network and preprocess the multi-source heterogeneous data to obtain target multi-source heterogeneous data;

[0007] According to the preset event triggering mechanism, the target multi-source heterogeneous data is extracted to obtain the target event, and the target event is standardized and labeled to obtain the feature vector and event label corresponding to the target event.

[0008] The feature vector of each target event is mapped to a standardized graph node representation, and the node type of each graph node is determined according to the event label of each target event.

[0009] Based on the temporal sequence, spatial proximity, and causal relationships of each target event, initial relationship edges are generated between nodes to form a graph structure. Whenever a new target event occurs, a local incremental update of the graph structure is triggered to obtain a snapshot of the temporal knowledge graph. , This is a snapshot of the time-series knowledge graph at time t. Let be the set of nodes in the graph structure at time t. Let be the set of edges in the graph structure at time t, where local incremental updates include adding nodes, adding relational edges, and updating node types;

[0010] The snapshot of the time-series knowledge graph is updated according to the preset node update rules. The embedding is updated to generate the graph embedding vector for each node at the current time, thus obtaining the distribution network time-series knowledge graph.

[0011] Secondly, the present invention provides an event-driven distribution network time-series knowledge graph construction system, comprising:

[0012] The acquisition module is configured to acquire multi-source heterogeneous data in the power distribution network and preprocess the multi-source heterogeneous data to obtain target multi-source heterogeneous data.

[0013] The extraction module is configured to extract the target multi-source heterogeneous data according to a preset event triggering mechanism to obtain target events, and to standardize and label the target events to obtain feature vectors and event labels corresponding to the target events.

[0014] The determination module is configured to map the feature vector of each target event to a standardized graph node representation, and determine the node type of each graph node based on the event label of each target event;

[0015] The generation module is configured to generate initial relationship edges between nodes based on the temporal sequence, spatial proximity, and causal relationship characteristics of each target event, forming a graph structure. Whenever a new target event occurs, it triggers a local incremental update of the graph structure to obtain a snapshot of the temporal knowledge graph. , This is a snapshot of the time-series knowledge graph at time t. Let be the set of nodes in the graph structure at time t. Let be the set of edges in the graph structure at time t, where local incremental updates include adding nodes, adding relational edges, and updating node types;

[0016] The update module is configured to update the snapshot of the time-series knowledge graph according to preset node update rules. The embedding is updated to generate the graph embedding vector for each node at the current time, thus obtaining the distribution network time-series knowledge graph.

[0017] Thirdly, an electronic device is provided, comprising: 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, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the event-driven distribution network timing knowledge graph construction method according to any embodiment of the present invention.

[0018] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the steps of the event-driven distribution network timing knowledge graph construction method according to any embodiment of the present invention.

[0019] This application presents an event-driven method and system for constructing a distribution network time-series knowledge graph. It extracts target multi-source heterogeneous data according to a preset event triggering mechanism to obtain target events. These events are then standardized and labeled to obtain corresponding feature vectors and event labels. The feature vector of each target event is mapped to a standardized graph node representation. The node type of each graph node is determined based on its event label. Initial relationship edges between nodes are generated based on the temporal sequence, spatial proximity, and causal relationships of each target event, forming a graph structure. This method enables rapid and accurate identification of distribution network events and timely construction of event association chains. Furthermore, it embeds and updates the time-series knowledge graph snapshot according to preset node update rules, generating a graph embedding vector for each node at the current time, thus obtaining the distribution network time-series knowledge graph. This improves fault diagnosis efficiency and assists in scheduling optimization decisions. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart illustrating an event-driven method for constructing a distribution network time-series knowledge graph, as provided in an embodiment of the present invention;

[0022] Figure 2 The following is a structural block diagram of an event-driven distribution network time-series knowledge graph construction system provided in an embodiment of the present invention;

[0023] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0025] Please see Figure 1 The diagram shows a flowchart of an event-driven method for constructing a distribution network timing knowledge graph according to this application.

[0026] like Figure 1 As shown, the event-driven method for constructing a distribution network time-series knowledge graph includes the following steps:

[0027] Step S101: Obtain multi-source heterogeneous data in the distribution network and preprocess the multi-source heterogeneous data to obtain target multi-source heterogeneous data.

[0028] Step S102: Extract the target multi-source heterogeneous data according to the preset event triggering mechanism to obtain the target event, and perform standard vectorization and labeling on the target event to obtain the feature vector and event label corresponding to the target event.

[0029] In this step, events are determined based on the operational logic of the power distribution network, and structured data is extracted based on the features of context graph embedding, text data is extracted based on knowledge-guided graph-text collaboration, and event detection is performed on image data based on multi-scale perception and structural semantic mapping to obtain the target event.

[0030] It should be noted that the extracted target events are uniformly converted into standard vector form to obtain feature vectors. ,in, For timestamps, For spatial location, This is the event feature vector;

[0031] Each target event is assigned a multidimensional label, which includes fault type, device type, and severity level.

[0032] Step S103: Map the feature vector of each target event to a standardized graph node representation, and determine the node type of each graph node based on the event label of each target event.

[0033] Step S104: Based on the temporal sequence, spatial proximity, and causal relationship characteristics of each target event, initial relationship edges are generated between nodes to form a graph structure. Whenever a new target event occurs, a local incremental update of the graph structure is triggered to obtain a snapshot of the temporal knowledge graph. .

[0034] In this step, This is a snapshot of the time-series knowledge graph at time t. Let be the set of nodes in the graph structure at time t. Let be the set of edges in the graph structure at time t, where local incremental updates include adding nodes, adding relational edges, and updating node types.

[0035] Step S105: Update the snapshot of the time-series knowledge graph according to the preset node update rules. The embedding is updated to generate the graph embedding vector for each node at the current time, thus obtaining the distribution network time-series knowledge graph.

[0036] In this step, the expression for the node update rule is:

[0037] ,

[0038] In the formula, Let v be the hidden state of node v at time t. : Represents the set of neighboring nodes of node v. To represent the embedding state of neighbor node u at time t-1, , For trainable weight matrix, The original features of the node at the current moment. This is the activation function.

[0039] In summary, the method of this application can achieve the following beneficial effects:

[0040] Event-driven dynamic mapping: Breaking through the traditional static mapping, it reflects the temporal evolution of events in the distribution network in real time.

[0041] Automatic causal relationship identification: Constructing accurate causal chains of events using multiple causal reasoning techniques.

[0042] Multimodal fusion: Integrating heterogeneous information from multiple sources such as structured data, text logs, and images to achieve multidimensional representation of knowledge.

[0043] Temporal graph neural network inference: Combining the Transformer attention mechanism to improve the ability to identify and predict complex event associations.

[0044] Dynamic adaptive update: Supports online updates and incremental learning of the knowledge graph, ensuring the timeliness and accuracy of the model.

[0045] In one specific embodiment, the method for constructing a distribution network time-series knowledge graph includes the following steps:

[0046] 1. Multi-source heterogeneous data acquisition and preprocessing

[0047] 1.1 Collect information on distribution network equipment, network topology, telemetry and telecontrol data, dispatch logs, alarm records, environmental and meteorological data, and user load information;

[0048] 1.2 Unify timestamps and device codes to complete data cleaning, format conversion, and multimodal alignment;

[0049] 1.3 A unified data pool is formed to support the subsequent construction of knowledge graphs.

[0050] 2. Event-driven perception and extraction

[0051] 2.1 Event Trigger Mechanism Design

[0052] 2.1.1 Multi-dimensional Business Logic Perception Model Design. Event triggering conditions are not simply based on single-node measurement data, but rather combine key logical judgment mechanisms in distribution network operation, including the following steps:

[0053] (1) Upstream and downstream linkage logic

[0054] When a downstream feeder experiences a voltage drop of ≥10% for 0.5 seconds and the corresponding upstream switch does not operate, an "upstream protection failure to operate event" is triggered; when a main feeder trips and the power of more than two branch feeders in the linkage area suddenly drops by ≥30%, a "regional power supply interruption event" is triggered.

[0055] (2) Timely perception of topological changes

[0056] The system periodically analyzes the topology change diagram (such as adding switching points or temporary access points). If there is a sudden change in circuit breaker status that does not match the actual power flow, a "topology identification anomaly event" is triggered. It determines whether the voltage / current base value of the node's region has changed due to reconfiguration. If it exceeds ±15%, a "reconfiguration configuration conflict warning" is triggered.

[0057] (3) Judgment of behavior-driven events

[0058] When a voltage surge, power flow reversal, or abnormal temperature rise occurs within 30 seconds of receiving a control command (such as switching, merging, or isolation), it is triggered as a "post-control instability event". By constructing a control response map, the operation action is bound to the abnormal event, and the "control response window" is dynamically set.

[0059] 2.1.2 Design a dynamic threshold adaptive algorithm to automatically adjust the threshold based on historical running data, thereby improving the sensitivity and accuracy of event triggering.

[0060]

[0061] in, , These are the historical mean and standard deviation, respectively. This is the sensitivity adjustment factor.

[0062] 2.2 Event Information Extraction

[0063] This includes using natural language processing technology to extract structured events from scheduling and operation logs.

[0064] 2.2.1 Structured Data: Feature Enhancement Extraction Based on Context Graph Embedding

[0065] Traditional methods define univariate change rules based solely on device measurements (such as current and voltage). The method proposed in this invention introduces graph embedding constraint information, using the device's "historical state trajectory + changes in topological adjacent nodes" in the graph as the extraction context, thus enhancing the accuracy of state change identification. Specific steps include:

[0066] Constructing device status time series

[0067]

[0068] in, Let represent the state sequence of device i at the current time t. It is an ordered set containing the state data of the device at the most recent k+1 times.

[0069] Calculate its spectral location embedding vector This forms a structured extraction joint representation:

[0070]

[0071] in, This represents the current event characteristics of the device and serves as input for subsequent tasks such as classification, prediction, and clustering. This is a feature fusion function used to encode three types of input features into a high-dimensional vector. Let i be the state time series of device i, representing the state data (such as voltage, current, circuit open state, etc.) of the most recent k moments, reflecting the historical behavior trajectory of the device. This represents the device's current structural embedding vector in the knowledge graph. This represents the rate of change of the structural state of neighboring nodes.

[0072] This method can effectively identify the boundaries of "sudden-non-sudden" and "disturbance-non-disturbance" state changes.

[0073] 2.2.2 Text Data: Knowledge-Guided Collaborative Extraction of Text and Images

[0074] Traditional approaches (BERT+BiLSTM) rely on semantic similarity. This invention proposes an event extraction framework that combines "graph context + action template guidance," comprising the following steps:

[0075] (1) Introduce graph constraint semantic templates, for example:

[0076] "After the XX circuit breaker trips, the YY device becomes overloaded" → triggers the "trip-overload" causal chain template, strengthening the semantic mapping;

[0077] Constructing an event template embedding representation:

[0078]

[0079] in, The original text of the scheduling log. This represents the embedding of a local subgraph of the graph corresponding to the event, serving as a priori context to assist in the extraction of actions and results;

[0080] (2) Introduce a self-supervised training mechanism: use the real labels of historical events as the comparative learning target to guide the system to automatically learn the event structure from a large number of unlabeled texts.

[0081] Furthermore, this method can support subject-verb-object recognition of events under few-shot learning conditions, and non-standard semantic extraction such as disordered word order, abbreviations, and industry terms.

[0082] 2.2.3 Image Data: Event Detection Based on Multi-Scale Perception and Structural Semantic Mapping

[0083] Traditional CNN methods struggle to distinguish abnormal hotspots in complex background regions when extracting event information. This invention innovatively employs a three-stage strategy of "image region segmentation + semantic mapping + map localization" for image data information extraction.

[0084] Step 1: Use Attention UNet to perform region-level hotspot segmentation on infrared images;

[0085] Step 2: Using the trained “structure-location-anomaly” image semantic encoder, thermal anomalies in the image are mapped to “device + location + anomaly type” triples;

[0086] Step 3: Call the equipment topology relationship in the map to automatically infer whether the anomaly belongs to "conduction anomaly" or "single point anomaly", and extract the event information that meets the conditions through anomaly confidence detection, such as "transformer temperature rise > 90℃, contact overheating alarm".

[0087] 2.3 Event Vectorization and Tagging

[0088] 2.3.1 Convert the extracted events into a standardized vector form. ,in, It's a timestamp. It is a spatial location (device ID or network topology coordinates). It is an event feature vector (including multimodal information, such as current value, operation type, fault category, etc.).

[0089] 2.3.2 Each event is automatically assigned a multi-dimensional label (fault type, equipment type, severity level) to facilitate subsequent correlation analysis.

[0090] 3. Construction of Time-Series Knowledge Graph for Distribution Network

[0091] 3.1 Event-Driven Graph Construction Framework

[0092] Trigger update mechanism: Whenever a new event occurs When generated, it triggers a local incremental update of the graph, including adding nodes, adding relation edges, and updating node attributes;

[0093] Multi-granular temporal snapshots: Constructing a knowledge graph snapshot sequence based on a sliding time window It supports multi-timescale modeling from seconds to minutes.

[0094] 3.2 Atlas Structure Design

[0095] 3.2.1 Node types are further subdivided into device entities, status entities, event entities, and environment entities;

[0096] 3.2.2 Relationship type design includes:

[0097] Temporal sequence: the order in which events occur;

[0098] Causal relationship: automatically determined through causal reasoning algorithms;

[0099] Topology: Electrical connections between devices;

[0100] Control relationship: The impact of instructions on equipment control.

[0101] 3.3 Determination and Completion of Causal Relationships

[0102] To achieve accurate correlation modeling between power distribution network fault events and control behaviors, this invention designs a causal reasoning mechanism that integrates time, equipment characteristics, and topology, while combining rule-based reasoning to achieve automated knowledge graph completion.

[0103] 3.3.1 Joint Edge Construction Based on Time and Features

[0104] For any two events ei and ej, the system calculates the correlation strength based on the following two core indicators:

[0105] Temporal proximity: The closer the times of the events, the higher the likelihood of a connection;

[0106] Feature similarity: Events with similar patterns of change, such as current or voltage, are more likely to form a causal relationship. The association weight is calculated as follows:

[0107] Wi,j = Time weight × Feature similarity

[0108] This weight serves as the basis for edge construction, determining whether to add an "event → event" edge to the graph.

[0109] 3.3.2 Rule-Driven Causal Edge Completion

[0110] (1) Pre-set typical relationship templates (such as "upper-level tripping → lower-level power outage", "remote control closing → current increase");

[0111] (2) Automatically perform rule matching for specific event types;

[0112] (3) Supports the generation of multiple relationship types such as “event → device” and “event → status”.

[0113] This mechanism automatically fills in the edges when the original graph is incomplete, improving the structural coherence and causal reasoning ability of the graph.

[0114] 3.4 Multimodal Data Fusion

[0115] 3.4.1 Multi-source feature unified coding mechanism

[0116] For each perceived or extracted event node ei, the system obtains features from the following three modalities:

[0117] Structured modal features ( ): Extracted from telemetry and teleindication information (such as voltage, current, switch status, etc.);

[0118] Text modal features ( ): Extracting action semantics, device information, alarm status, etc. from scheduling or fault logs using natural language processing technology;

[0119] Image modal features ( ): Abnormal patterns such as temperature rise and discharge identified from infrared or visual images using convolutional neural networks.

[0120] After standardization, the above features are fused into a unified event vector representation using a linear weighting method:

[0121] ,

[0122] in, The fusion weights, automatically learned during system training, reflect the importance of different modalities in event description. This representation possesses good scalability and expressive power, adapting to the modeling needs of various event types in power distribution network scenarios.

[0123] 3.4.2 Integration of Graph Input and Subsequent Inference

[0124] The fused event vector ei is used as the initial feature input for "event nodes" in the knowledge graph, participating in tasks such as embedding computation, causal edge construction, and event evolution reasoning in the graph neural network. This fusion process not only breaks down the barriers between structured and unstructured data but also significantly improves the identifiability and intelligent reasoning capabilities of graph nodes.

[0125] 4. Graph Embedding and Intelligent Reasoning Module

[0126] Based on the weighted time-series knowledge graph constructed in step three, a simplified layout neural network model is used for event node embedding calculation and intelligent reasoning analysis to achieve prediction and auxiliary decision-making for the fault evolution process of the distribution network.

[0127] 4.1 Model Architecture Design

[0128] A graph embedding method combining "mean aggregation + node self-feature update" is adopted to balance information propagation capability and resource consumption.

[0129] By implementing hourly updates to node states, the model retains aggregated information from neighboring nodes while also considering changes in the node's own current characteristics, resulting in good scalability and stability. The node state update formula is as follows:

[0130] ,

[0131] 4.2 Embedded Learning Process

[0132] 4.2.1 Graph Input: Read the snapshot of the time-series knowledge graph constructed in step 3.

[0133] 4.2.2 Feature Initialization: Assign initial features to each node. This includes device attributes, event characteristics, status parameters, etc.

[0134] 4.2.3 Node aggregation calculation: The embedding is updated according to the above formula to generate the graph embedding vector of each node at the current time;

[0135] 4.2.4 Output Results: The embedded vectors are used for subsequent tasks such as event prediction, anomaly detection, and risk identification.

[0136] 4.3 Intelligent Reasoning Task

[0137] After obtaining the node embedding vectors, the following graph inference capabilities are performed:

[0138] (1) Event evolution prediction

[0139] Predict whether a specific device may experience similar or derivative failures in the near future; for example, after a current switch trips, the probability of another feeder in the same area becoming overloaded within 5 minutes is estimated to be 0.7.

[0140] (2) Anomaly source identification

[0141] Based on the event propagation path and embedded spatial distance, potential fault sources are identified in reverse; an "event reverse chain" is realized to assist operation and maintenance in locating the initial problem node.

[0142] (3) Regulatory decision support

[0143] By combining historical event graphs and the current state, the system simulates the impact of different control actions (such as reclosing and load shedding); it recommends the optimal or suboptimal control strategies to achieve intelligent auxiliary decision-making.

[0144] 4.4 Online Model Learning and Adaptive Adjustment

[0145] Online learning mechanism: As new events and data are continuously added, the model supports dynamic incremental training, eliminating the need for training from scratch;

[0146] Feedback mechanism optimization: The annotations of maintenance personnel and the actual actions of the system are used as feedback samples to automatically update the graph structure and weights;

[0147] Continuously enhance intelligence: Through real-time data stream training, the model continuously adapts to changes in network topology and operating status, improving stability and timeliness.

[0148] Please see Figure 2 The diagram shows a structural block diagram of an event-driven distribution network timing knowledge graph construction system according to this application.

[0149] like Figure 2 As shown, the power distribution network time-series knowledge graph construction system 200 includes an acquisition module 210, an extraction module 220, a determination module 230, a generation module 240, and an update module 250.

[0150] The module 210 is configured to acquire multi-source heterogeneous data in the power distribution network and preprocess the multi-source heterogeneous data to obtain target multi-source heterogeneous data. The extraction module 220 is configured to extract the target multi-source heterogeneous data according to a preset event triggering mechanism to obtain target events, and to standardize and label the target events to obtain feature vectors and event labels corresponding to the target events. The determination module 230 is configured to map the feature vector of each target event to a standardized graph node representation, and to determine the node type of each graph node according to the event label of each target event. The generation module 240 is configured to generate initial relationship edges between nodes based on the temporal sequence, spatial proximity, and causal relationship features of each target event to form a graph structure, and to trigger a local incremental update of the graph structure whenever a new target event is generated to obtain a snapshot of the temporal knowledge graph. , This is a snapshot of the time-series knowledge graph at time t. Let be the set of nodes in the graph structure at time t. Let be the set of edges in the graph structure at time t, wherein local incremental updates include adding nodes, adding relation edges, and updating node types; update module 250 is configured to update the snapshot of the temporal knowledge graph according to preset node update rules. The embedding is updated to generate the graph embedding vector for each node at the current time, thus obtaining the distribution network time-series knowledge graph.

[0151] It should be understood that Figure 2 The modules and references described in the document Figure 1 The steps described in the text correspond to those in the method described above. Therefore, the operations, features, and corresponding technical effects described above also apply to the method described in the text. Figure 2 The various modules in the document will not be described in detail here.

[0152] In other embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the event-driven distribution network timing knowledge graph construction method in any of the above method embodiments.

[0153] In one embodiment, the computer-readable storage medium of the present invention stores computer-executable instructions, which are configured as follows:

[0154] Acquire multi-source heterogeneous data in the power distribution network and preprocess the multi-source heterogeneous data to obtain target multi-source heterogeneous data;

[0155] According to the preset event triggering mechanism, the target multi-source heterogeneous data is extracted to obtain the target event, and the target event is standardized and labeled to obtain the feature vector and event label corresponding to the target event.

[0156] The feature vector of each target event is mapped to a standardized graph node representation, and the node type of each graph node is determined according to the event label of each target event.

[0157] Based on the temporal sequence, spatial proximity, and causal relationships of each target event, initial relationship edges are generated between nodes to form a graph structure. Whenever a new target event occurs, a local incremental update of the graph structure is triggered to obtain a snapshot of the temporal knowledge graph. , This is a snapshot of the time-series knowledge graph at time t. Let be the set of nodes in the graph structure at time t. Let be the set of edges in the graph structure at time t, where local incremental updates include adding nodes, adding relational edges, and updating node types;

[0158] The snapshot of the time-series knowledge graph is updated according to the preset node update rules. The embedding is updated to generate the graph embedding vector for each node at the current time, thus obtaining the distribution network time-series knowledge graph.

[0159] Computer-readable storage media may include a stored program area and a stored data area, wherein the stored program area may store an operating system and an application program required for at least one function; the stored data area may store data created based on the use of the event-driven distribution network timing knowledge graph construction system, etc. Furthermore, the computer-readable storage medium may include high-speed random access memory, and may also include memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include memory remotely configured relative to a processor, which can be connected to the event-driven distribution network timing knowledge graph construction system via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0160] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiment of the present invention, such as... Figure 3 As shown, the device includes a processor 310 and a memory 320. The electronic device may also include an input device 330 and an output device 340. The processor 310, memory 320, input device 330, and output device 340 can be connected via a bus or other means. Figure 3Taking a bus connection as an example, the memory 320 is the computer-readable storage medium described above. The processor 310 executes various server functions and data processing by running non-volatile software programs, instructions, and modules stored in the memory 320, thereby realizing the event-driven distribution network time-series knowledge graph construction method described in the above embodiment. The input device 330 can receive input digital or character information and generate key signal inputs related to user settings and function control of the event-driven distribution network time-series knowledge graph construction system. The output device 340 may include a display screen or other display device.

[0161] The aforementioned electronic device can execute the method provided in the embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in the embodiments of the present invention.

[0162] In one implementation, the above-described electronic device is applied to an event-driven power distribution network time-series knowledge graph construction system for a client, comprising: 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, the instructions being executed by the at least one processor to enable the at least one processor to:

[0163] Acquire multi-source heterogeneous data in the power distribution network and preprocess the multi-source heterogeneous data to obtain target multi-source heterogeneous data;

[0164] According to the preset event triggering mechanism, the target multi-source heterogeneous data is extracted to obtain the target event, and the target event is standardized and labeled to obtain the feature vector and event label corresponding to the target event.

[0165] The feature vector of each target event is mapped to a standardized graph node representation, and the node type of each graph node is determined according to the event label of each target event.

[0166] Based on the temporal sequence, spatial proximity, and causal relationships of each target event, initial relationship edges are generated between nodes to form a graph structure. Whenever a new target event occurs, a local incremental update of the graph structure is triggered to obtain a snapshot of the temporal knowledge graph. , This is a snapshot of the time-series knowledge graph at time t. Let be the set of nodes in the graph structure at time t. Let be the set of edges in the graph structure at time t, where local incremental updates include adding nodes, adding relational edges, and updating node types;

[0167] The snapshot of the time-series knowledge graph is updated according to the preset node update rules. The embedding is updated to generate the graph embedding vector for each node at the current time, thus obtaining the distribution network time-series knowledge graph.

[0168] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0169] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An event-driven power distribution network timing knowledge graph construction method, characterized in that, The method comprises the following steps: acquiring multi-source heterogeneous data in a power distribution network, and preprocessing the multi-source heterogeneous data to obtain target multi-source heterogeneous data; extracting target events from the target multi-source heterogeneous data according to a preset event triggering mechanism, and performing standard vectorization and labeling on the target events to obtain feature vectors and event labels corresponding to the target events; mapping the feature vectors of each target event into standardized graph node representations, and determining node types of each graph node according to the event labels of each target event; According to the time sequence, the spatial proximity relationship and the cause-effect correlation characteristics of each target event, an initial relationship edge between nodes is generated to form a graph structure, and each time a new target event occurs, a local incremental update of the graph structure is triggered to obtain a time sequence knowledge graph snapshot , is a time sequence knowledge graph snapshot at t time, is a node set in the graph structure at t time, is an edge set in the graph structure at t time, wherein the local incremental update includes a new node, an added relationship edge and an updated node type; According to a preset node updating rule, the time sequence knowledge graph snapshot is updated The embedding is updated to generate a graph embedding vector of each node at the current time, that is, a power distribution network time sequence knowledge graph is obtained, wherein an expression of the node updating rule is: ， wherein, is the hidden state of node v at time t, : denotes the set of neighbor nodes of node v, is the embedding state of neighbor node u at time t-1, , is a trainable weight matrix, is the original feature of the node at the current time, is an activation function.

2. The power distribution network timing knowledge graph construction method based on event driving according to claim 1, characterized in that, the extracting target events from the target multi-source heterogeneous data according to a preset event triggering mechanism comprises: determining events according to operation logic in power distribution network operation, extracting structured data based on context graph embedded features, extracting text data based on knowledge guided graph-text collaboration, and detecting events based on multi-scale perception and structure semantic mapping to obtain target events.

3. The power distribution network timing knowledge graph construction method based on event driving according to claim 1, characterized in that, the performing standard vectorization and labeling on the target events to obtain feature vectors and event labels corresponding to the target events comprises: The extracted target event is uniformly converted into a standard vector form to obtain a feature vector wherein, is a timestamp, is a spatial position, is an event feature vector; assigning multi-dimensional labels to each target event, wherein the multi-dimensional labels include fault types, device types, and severity levels.

4. An event-driven power distribution network timing knowledge graph construction system, characterized in that, The method comprises the following steps: an acquisition module configured to acquire multi-source heterogeneous data in a power distribution network, and to preprocess the multi-source heterogeneous data to obtain target multi-source heterogeneous data; an extraction module configured to extract target events from the target multi-source heterogeneous data according to a preset event triggering mechanism, and to perform standard vectorization and labeling on the target events to obtain feature vectors and event labels corresponding to the target events; a determination module configured to map the feature vectors of each target event into standardized graph node representations, and to determine node types of each graph node according to the event labels of each target event. The generating module is configured to generate initial relationship edges between nodes according to time sequence, spatial proximity and causal correlation characteristics of each target event, form a graph structure, and trigger local incremental update of the graph structure whenever a new target event occurs to obtain a time sequence knowledge graph snapshot , is a time sequence knowledge graph snapshot at t moment, is a node set in the graph structure at t moment, is an edge set in the graph structure at t moment, wherein the local incremental update includes adding a new node, adding a relationship edge and updating a node type; an updating module, configured to update the time-series knowledge graph snapshot according to a preset node updating rule perform embedding updating to generate a graph embedding vector of each node at the current time, i.e., obtain a power distribution network time-series knowledge graph, wherein an expression of the node updating rule is: ， wherein, is the hidden state of node v at time t, : denotes the set of neighbor nodes of node v, is the embedding state of neighbor node u at time t-1, , is a trainable weight matrix, is the original feature of the node at the current time, is an activation function.

5. An electronic device, comprising: The method comprises the following steps: at least one processor, and a memory connected to the at least one processor in communication, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 3.

6. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the method of any one of claims 1 to 3.

Images