Power grid fault detection method and device, storage medium and electronic equipment

Abstract

The invention discloses a power grid fault detection method and device, a storage medium and electronic equipment. The method comprises the following steps: acquiring a current alarm information text of a target power grid; constructing a target knowledge graph of the target power grid based on the current alarm information text; determining time sequence information based on the target knowledge graph; determining an attenuation factor of the protection action entity based on the time sequence information; based on the attenuation factor, the target knowledge graph and an entity feature matrix, a target fault detection model is adopted to carry out fault detection on the target power grid, a target fault detection result of the target power grid is obtained, and the entity feature matrix comprises feature vectors corresponding to multiple entities respectively; the target fault detection result comprises a target fault category and a target fault position. According to the method and the device, the technical problems of inaccurate fault detection result and poor interpretability of the fault detection result of the power grid in the prior art are solved.

Description

Technical Field

[0001] This application relates to the field of power systems, and more specifically, to a method, apparatus, storage medium, and electronic device for detecting faults in a power grid. Background Technology

[0002] With the continuous expansion of the power grid and the high proportion of renewable energy integration, the complexity of power grid operation has increased dramatically, leading to a dual challenge for power grid fault detection: the timeliness and accuracy of processing massive amounts of alarm information. Current technologies for power grid fault detection rely on expert rules and artificial neural networks, which have several problems. First, the generalization ability of neural network models is insufficient, and the construction of rule bases is difficult to adapt to complex and ever-changing power grid topologies. Second, the interpretability of fault detection results is lacking; data-driven methods cannot intuitively display the fault logic chain, hindering dispatchers' rapid decision-making. Finally, the accuracy of fault detection in key power grid scenarios is insufficient, with low accuracy in detecting complex faults such as protection failure to operate and switch failure to operate. Therefore, current technologies suffer from inaccurate power grid fault detection results and poor interpretability.

[0003] There is currently no effective solution to the above problems. Summary of the Invention

[0004] This application provides a method, apparatus, storage medium, and electronic device for detecting faults in a power grid, in order to at least solve the technical problems of inaccurate fault detection results and poor interpretability of fault detection results in the related art.

[0005] According to one aspect of the embodiments of this application, a fault detection method for a power grid is provided, comprising: acquiring current alarm information text of a target power grid; constructing a target knowledge graph of the target power grid based on the current alarm information text, wherein the target knowledge graph includes an entity set, a relation set, and a relation type set, the entity set including multiple entities within the target power grid, the multiple entities including plant entities, equipment entities, protection action entities, and protection attribute entities, the relation set including the relationships between the multiple entities, and the relation type set including multiple relation types existing between the multiple entities, and the occurrence frequency corresponding to each of the multiple relation types; determining time-series information based on the target knowledge graph, wherein the time-series information is used to describe the time-series characteristics of the state changes of the protection action entities; determining an attenuation factor of the protection action entities based on the time-series information, wherein the attenuation factor is used to quantify the degree of influence of the protection action entities on fault detection; and performing fault detection on the target power grid using a target fault detection model based on the attenuation factor, the target knowledge graph, and the entity feature matrix to obtain a target fault detection result for the target power grid, wherein the entity feature matrix includes feature vectors corresponding to multiple entities, and the target fault detection result includes a target fault category and a target fault location.

[0006] According to another aspect of the embodiments of this application, a fault detection device for a power grid is provided, comprising: a text acquisition module, used to acquire current alarm information text of a target power grid; and a target knowledge graph construction module, used to construct a target knowledge graph of the target power grid based on the current alarm information text, wherein the target knowledge graph includes an entity set, a relationship set, and a relationship type set, the entity set including multiple entities within the target power grid, the multiple entities including plant entities, equipment entities, protection action entities, and protection attribute entities, the relationship set including the relationships between the multiple entities, and the relationship type set including multiple relationship types existing between the multiple entities, and the occurrence frequency corresponding to each of the multiple relationship types; first The first module determines time-series information based on the target knowledge graph, where the time-series information describes the time-series characteristics of the state changes of the protection action entity. The second module determines the attenuation factor of the protection action entity based on the time-series information, where the attenuation factor quantifies the degree of influence of the protection action entity on fault detection. The third module performs fault detection on the target power grid using a target fault detection model based on the attenuation factor, the target knowledge graph, and the entity feature matrix, and obtains the target fault detection result of the target power grid. The entity feature matrix includes feature vectors corresponding to multiple entities, and the target fault detection result includes the target fault category and the target fault location.

[0007] According to another aspect of the embodiments of this application, a non-volatile storage medium is provided, which stores multiple instructions adapted for a power grid fault detection method, any one of which can be loaded and executed by a processor.

[0008] According to another aspect of the embodiments of this application, an electronic device is provided, including: one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement any one of the following power grid fault detection methods.

[0009] According to another aspect of the embodiments of this application, a computer program product is provided, which, when executed on a data processing device, is a program adapted to perform the steps of a power grid fault detection method.

[0010] In this embodiment, the current alarm information text of the target power grid is obtained; based on the current alarm information text, a target knowledge graph of the target power grid is constructed, wherein the target knowledge graph includes an entity set, a relation set, and a relation type set. The entity set includes multiple entities within the target power grid, including plant entities, equipment entities, protection action entities, and protection attribute entities. The relation set includes the relationships between multiple entities, and the relation type set includes various relation types existing between multiple entities, as well as the frequency of occurrence of each relation type. Based on the target knowledge graph, time-series information is determined, wherein the time-series information is used to describe the time-series characteristics of the state changes of the protection action entities. Based on the time-series information, an attenuation factor for the protection action entities is determined, wherein the attenuation factor is used to quantify the degree of influence of the protection action entities on fault detection. Based on the attenuation factor, the target knowledge graph, and the entity feature matrix, a target fault detection model is used to perform fault detection on the target power grid, obtaining the target fault detection result of the target power grid. The entity feature matrix includes feature vectors corresponding to multiple entities, and the target fault detection result includes the target fault category and the target fault location. The goal is to construct a target knowledge graph based on the current alarm information text of the target power grid, and then use a target fault detection model to obtain the target detection results of the target power grid. This aims to improve the accuracy and interpretability of the target detection results of the target power grid, thereby solving the technical problems of inaccurate fault detection results and poor interpretability of fault detection results in related technologies. Attached Figure Description

[0011] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0012] Figure 1 This is a flowchart of a power grid fault detection method according to an embodiment of this application;

[0013] Figure 2This is a flowchart of an optional power grid fault detection method provided according to an embodiment of this application;

[0014] Figure 3 This is a schematic diagram of the detection results of an optional power grid fault detection method provided according to an embodiment of this application;

[0015] Figure 4 This is a schematic diagram of an optional power grid fault detection device provided according to an embodiment of this application. Detailed Implementation

[0016] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0017] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] For ease of description, the following explains some of the nouns or terms used in the embodiments of this application:

[0019] The Viterbi algorithm is a dynamic programming algorithm used to find the most likely sequence. In fields such as natural language processing and speech recognition, the Viterbi algorithm is widely used in tasks such as part-of-speech tagging, syntactic analysis, and sequence labeling. Especially in the sequence labeling problem, when the model needs to determine word segmentation or part-of-speech based on context, the Viterbi algorithm can effectively find the optimal label sequence.

[0020] Neo4j is an open-source graph database management system that is particularly well-suited for handling highly interconnected data, such as applications involving social networks, recommendation systems, and other complex relational networks.

[0021] According to an embodiment of this application, a method embodiment for fault detection of a power grid is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0022] Figure 1 This is a flowchart of a power grid fault detection method according to an embodiment of this application, such as... Figure 1 As shown, the method includes the following steps:

[0023] Step S102: Obtain the current alarm information text of the target power grid;

[0024] It is understandable that the current alarm information text of the target power grid can be obtained by connecting to the SCADA (Supervisory Control and Data Acquisition) system of the target power grid. By obtaining the current alarm information text of the target power grid in real time, fault detection of the target power grid can be performed in a timely manner, ensuring the safe and reliable operation of the target power grid.

[0025] Step S104: Based on the current alarm information text, construct a target knowledge graph of the target power grid. The target knowledge graph includes an entity set, a relationship set, and a relationship type set. The entity set includes multiple entities within the target power grid, including plant entities, equipment entities, protection action entities, and protection attribute entities. The relationship set includes the relationships between multiple entities. The relationship type set includes multiple relationship types between multiple entities and the frequency of occurrence of each relationship type.

[0026] It is understandable that, based on the current alarm information text of the target power grid, a target knowledge graph of the target power grid is constructed. This target knowledge graph includes a set of entities, a set of relationships, and a set of relationship types. The set of entities includes multiple entities within the target power grid, such as substation entities, equipment entities, protection action entities, and protection attribute entities. By constructing a multi-dimensional knowledge graph that includes a set of relationships and a set of relationship types, the spatiotemporal correlation between entities and the dynamic topology of the target power grid are comprehensively considered, thereby improving the target fault detection model's comprehensive understanding of the target power grid's operating status and ultimately improving the accuracy of the target fault detection results.

[0027] Optionally, the plant entity encoding includes basic information such as the type, name, and geographical location of the plant (e.g., power plants, substations, and converter stations within the target power grid). Equipment entities include equipment type (transformers, circuit breakers, lines, etc.), equipment status (operating, under maintenance, faulty), and equipment specifications (rated voltage, rated current, etc.). Protection action entities encode attributes such as action type (tripping, reclosing, etc.) and action result (success, failure). Protection attribute entities include information such as protection type (overcurrent protection, differential protection, etc.), protection level, and the type and location of equipment associated with that protection.

[0028] Optionally, entity extraction mainly refers to using the word segmentation and related descriptions of power equipment fault event text (such as current alarm information text) as entities in the target knowledge graph, and numbering these entities. First, a conventional dictionary and the Viterbi algorithm are used to perform initial word segmentation processing (i.e., initialization) on the current alarm information text of the target power grid. Then, a custom dictionary is used for manual intervention to correct the word segmentation results. Next, the corrected word segmentation results are deduplicated to construct a dedicated fault event description word segmentation dictionary. Finally, based on the above fault event description word segmentation dictionary, all word segments are numbered according to their order of appearance in the dictionary, where each word segment corresponds to an entity in the target knowledge graph.

[0029] Step S106: Based on the target knowledge graph, determine the temporal information, wherein the temporal information is used to describe the time sequence characteristics of the state changes of the protective action entity;

[0030] It is understandable that, based on the constructed target knowledge graph, time-series features are determined to describe the state changes of protective action entities within the entity. These time-series features enrich the input of the target fault detection model from a time-dependency perspective, enabling the model to comprehensively consider both dynamic and static factors of power grid operation, thereby improving the accuracy of target fault detection results.

[0031] In one optional embodiment, determining the timing information based on the target knowledge graph includes: arranging the protection action entities in ascending order according to their occurrence time to obtain sequential protection action entities; and determining the timing information based on the occurrence time of the sequential protection action entities.

[0032] It can be understood that by arranging the protection action entities among multiple entities in the target knowledge graph in ascending order of time, a sorted sequence of protection action entities is obtained. The timing information is determined based on the occurrence time of these sequential protection action entities. The extraction and utilization of this timing information provides a temporal interpretive path for fault detection. By analyzing the temporal patterns of the protection action entities, the target fault detection model can generate time-related fault logic chains, thereby generating a visual topology mapping and enhancing the interpretability of the target fault detection results.

[0033] Optionally, during the relation extraction process, a word segmentation relevance matrix can be constructed by statistically analyzing the co-occurrence frequency of adjacent words in the current alarm information text, and this co-occurrence frequency can be used as a direct basis for quantifying the relationships between entities. The word segmentation relevance matrix can be constructed in the following way:

[0034]

[0035] in, This is the word segmentation relevance matrix (i.e., the relation matrix), where n is the total number of word segments. This indicates the co-occurrence frequency of the word segment numbered i immediately following the word segment numbered j. The larger the value, the stronger the connection between the two word segments.

[0036] Optionally, the timing relationship is mainly used to distinguish the order in which faults occur. Arranging the protection action entities in ascending order of their occurrence time yields the sequence of sequential protection action entities. ,in, This represents the k-th protection action entity in the sequential protection action entity. The time interval between each protection action entity in the sequential protection action entity is calculated. The time interval between the k-th protection action entity and the (k-1)-th protection action entity. The following methods can be used to determine this:

[0037]

[0038] in, This indicates the time when the k-th protective action entity occurs. This indicates the time when the (k-1)th protective action entity occurs.

[0039] Optionally, the valid information triplet (i.e., time sequence information) of the faulty device. And the time-series decay factor (i.e., decay factor) The following methods can be used to determine this:

[0040]

[0041]

[0042] in, Let be the timing decay factor of the k-th protection action entity. This is the attenuation coefficient.

[0043] Alternatively, the target knowledge graph can be defined as a triple. Triplet The following methods can be used to determine this:

[0044]

[0045] in, This represents a set of nodes (i.e., a set of entities), specifically the entity objects in the current alarm message text. The set of edges (i.e., the set of relationships) representing the connection relationships between device entities. A label representing a set of relation types, primarily used to describe the semantics of edges.

[0046] Node set The following methods can be used to determine this:

[0047]

[0048] in, This represents a plant node (i.e., a plant entity). This represents a device node (i.e., a device entity). This represents a protected node (i.e., a protected attribute entity). These are action nodes (i.e., protection action entities). Through the above four types of nodes, an interconnected network of the target power grid is formed.

[0049] Relation set The following methods can be used to determine this:

[0050]

[0051] in Representing node i, This represents node j.

[0052] Relationship Type Set The following methods can be used to determine this:

[0053]

[0054] in, This indicates quantitative relationships, specifically the co-occurrence frequency between adjacent entities. It represents temporal relationships and is used to describe the time-series dependencies of protection action entities.

[0055] Optionally, taking node i as an example, a node insertion operation is defined. The insertion operation for node i can be implemented as follows:

[0056]

[0057] in, This represents the insertion operation on node i. This indicates a creation command.

[0058] Step S108: Based on timing information, determine the attenuation factor of the protection action entity, wherein the attenuation factor is used to quantify the degree of influence of the protection action entity on fault detection.

[0059] It is understandable that an attenuation factor is determined based on time-series information to quantify the impact of protective actions on fault detection. By introducing the attenuation factor, the target fault detection model can dynamically adjust the weights of protective actions in fault detection, focusing on those actions that respond immediately after a fault occurs or trigger consecutively within a short period, thus improving the accuracy of the target fault detection results. Simultaneously, the attenuation factor provides a temporal interpretation of the target fault detection model's output, indicating which protective actions play a more significant role in fault detection, thereby enhancing the interpretability of the target fault detection results.

[0060] Step S110: Based on the attenuation factor, target knowledge graph, and entity feature matrix, a target fault detection model is used to detect faults in the target power grid, and the target fault detection result of the target power grid is obtained. The entity feature matrix includes feature vectors corresponding to multiple entities, and the target fault detection result includes the target fault category and the target fault location.

[0061] It is understandable that by running the target fault detection model, accurate detection and location of faults in the target power grid can be achieved, ensuring that the target fault detection model can fully understand the topology, entity characteristics and temporal relationships of the target power grid. This not only improves the accuracy of fault detection, but also enhances the interpretability of the detection results.

[0062] In one optional embodiment, based on an attenuation factor, a target knowledge graph, and an entity feature matrix, a target fault detection model is used to detect faults in the target power grid, obtaining a target fault detection result for the target power grid. This includes: for any entity among multiple entities, determining an initial relation vector for that entity based on the target knowledge graph, where the initial relation vector describes the relationship type and frequency of occurrence between that entity and other entities, as well as the temporal relationship between that entity and other entities; based on the attenuation factor, using the dynamic relation embedding layer of the target fault detection model, fusing the initial relation vector and historical relation vectors corresponding to multiple historical moments to obtain a target relation vector for that entity, where the historical relation vector describes the relationship type and frequency of occurrence between that entity and other entities at the corresponding historical moment, as well as the temporal relationship between that entity and other entities; fusing the target relation vector with the neighbor feature vectors of the neighboring entities of that entity to obtain a neighbor aggregation feature for that entity; determining the neighbor aggregation features corresponding to multiple entities using the same method as determining the neighbor aggregation features of that entity; and determining the target fault detection result based on the neighbor aggregation features corresponding to multiple entities and the entity feature matrix.

[0063] It is understandable that, for any entity among multiple entities, an initial relation vector for that entity is determined based on the target knowledge graph. Based on a decay factor, a dynamic relation embedding layer of the target fault detection model is used to fuse the initial relation vector of any entity with the historical relation vectors corresponding to multiple historical moments, obtaining the target relation vector of that entity. The target relation vector of any entity is then fused with the neighbor feature vectors of its neighboring entities to obtain the neighbor aggregation feature of that entity. By determining the neighbor aggregation feature of any entity, the neighbor aggregation features corresponding to multiple entities are determined, and based on these neighbor aggregation features, the target fault detection result for the target power grid is determined. The introduction of historical relation vectors compensates for the limitations of single-time information, enabling the target fault detection model to identify long-term stable relation patterns, significantly improving the comprehensiveness and accuracy of fault detection.

[0064] Optionally, the initial relation vector The following methods can be used to determine this:

[0065]

[0066] in, A vector representing the quantitative relationships between entities. Represents a time-series relation vector.

[0067] Alternatively, based on the historical knowledge graph of the target power grid, a GRU (Gated Recurrent Unit) network can be used to extract the historical relationship vectors of nodes. . The following methods can be used to determine this:

[0068]

[0069] in, This represents the historical relation vector extracted through the GRU network. Let t be the historical relationship vector at time t.

[0070] Optionally, the target relation vector is determined based on the initial relation vector and historical relation vectors described above. . The following methods can be used to determine this:

[0071]

[0072] Alternatively, the target relation vector can be fused with the neighbor feature vectors of neighboring entities to generate composite features.

[0073]

[0074]

[0075] in, To fuse the neighbor feature vectors at neighbor entity u Relationship vector with target The composite function, Indicates adaptive weights, For activation function, This represents the gating weight matrix.

[0076] In summary, the neighbor aggregation characteristics of node v The following methods can be used to determine this:

[0077]

[0078] in, The first learnable weight matrix, Let f(v) be the set of neighboring entities of node v.

[0079] In one optional embodiment, the target fault detection result is determined based on the neighbor aggregation features corresponding to multiple entities and the entity feature matrix, including: for any entity among the multiple entities, a gate value is determined based on the neighbor aggregation features of any entity, the feature vector of any entity, and the attenuation factor of the protection action entities included in the neighbor entities, wherein the gate value is used to control the degree of fusion between the feature vector of any entity and the neighbor aggregation features of any entity, and the feature vector of any entity is determined based on the entity feature matrix; based on the gate value, the neighbor aggregation features of any entity and the feature vector of any entity are fused to obtain the fused feature vector of any entity; based on the fused feature vector of any entity, a target feature vector of any entity is determined using a multi-layer cascaded structure of the target fault detection model, wherein the multi-layer cascaded structure is implemented in the form of a residual network; the target feature vectors corresponding to multiple entities are determined by using the method of determining the target feature vector of any entity; and the target fault detection result is determined based on the target feature vectors corresponding to multiple entities.

[0080] It can be understood that for any entity among multiple entities, a feature vector of that entity is determined based on the entity feature matrix. Then, based on the neighbor aggregation features and feature vector of that entity, as well as the attenuation factors of the protection action entities included in the multiple neighboring entities, a gating mechanism is used to determine a gating value for that entity. This gating value controls the degree of fusion between the feature vector of that entity and the neighbor aggregation features of that entity. Based on the gating value, the neighbor aggregation features and feature vector of that entity are fused to obtain a fused feature vector for that entity. Based on the fused feature vector of that entity, a target feature vector for that entity is determined using a multi-layer cascaded structure implemented in the form of a residual network in the target fault detection model. By determining the target feature vector of that entity, target feature vectors corresponding to multiple entities are determined. Based on the target feature vectors corresponding to multiple entities, fault detection is performed on the target power grid to obtain the target fault detection result of the target power grid. Through the gating mechanism, the target fault detection model can adaptively adjust the feature fusion ratio, enabling more accurate identification of key information when handling different types of faults, thus improving the accuracy of the target fault detection result. Meanwhile, the calculation process of the gating value intuitively demonstrates the weight assigned to the corresponding entity in fault detection, providing a basis for understanding the target fault detection results.

[0081] Optionally, the gate value of node v The following methods can be used to determine this:

[0082]

[0083] in, This is the learnable second weight matrix.

[0084] Alternatively, the fused feature vector of node v can be determined in the following manner. :

[0085]

[0086] Here, ⊙ represents element-level multiplication.

[0087] Optionally, the multi-layered cascaded structure can be implemented using a residual network to avoid degradation in deep networks. The multi-layered cascaded structure can be modeled as follows:

[0088]

[0089] in, Indicates the first The feature vector output by layer node v, assuming there are L layers in total, then Let V represent the feature vector of the output of node v in the Lth layer, which is also the target feature vector of node v. This represents the residual connection weight matrix. Presentation layer normalization operation.

[0090] In one optional embodiment, determining the target fault detection result based on the target feature vectors corresponding to multiple entities includes: determining the relationship strength corresponding to multiple relationship types based on the target relationship vectors corresponding to multiple entities; determining the output feature vector based on the relationship strength corresponding to multiple relationship types and the target feature vectors corresponding to multiple entities; and determining the target fault detection result based on the output feature vector.

[0091] It can be understood that the relationship strength corresponding to various relationship types is determined based on the target relationship vectors corresponding to multiple entities. Based on the relationship strengths corresponding to various relationship types and the target relationship vectors corresponding to multiple entities, an output feature vector is determined, and the target fault detection result of the target power grid is determined based on this output feature vector. By calculating relationship strength, the target fault detection model can effectively integrate entity relationships at different levels, avoiding information redundancy or omission, and improving the comprehensiveness and accuracy of the target fault detection results.

[0092] Alternatively, the strength of the relationship can be determined in the following ways. :

[0093]

[0094] in, The importance weight of relation type r is represented. This is an important vector for interpretable relationships.

[0095] Optionally, output feature vector The following methods can be used to determine this:

[0096]

[0097] in, For mean pooling operation, Let r be a set of nodes with relation r. Represents a set of relation types.

[0098] In one optional embodiment, determining the target fault detection result based on the output feature vector includes: determining the fault probabilities corresponding to multiple preset fault categories based on the output feature vector, and determining the preset fault category corresponding to the highest fault probability as the target fault category; determining the importance of the equipment entity based on the current alarm information text; determining the target fault location based on the importance and the frequency matrix of the equipment entity, wherein the frequency matrix is ​​obtained based on the target knowledge graph and is used to quantify the correlation between the equipment entity and the current fault event of the target power grid; and determining the target fault detection result based on the target fault category and the target fault location.

[0099] It is understandable that, based on the output feature vector, the fault probabilities corresponding to various preset fault categories are determined, and the preset fault category corresponding to the highest fault probability is identified as the target fault category of the target power grid. Text parsing of the current alarm information text determines the importance of equipment entities, and based on the importance and frequency matrix of these entities, the target fault location in the target power grid—that is, the specific faulty equipment within the target power grid—is determined. The aforementioned target fault category and target fault location are used as the target fault detection result for the target power grid. The output feature vector integrates the relationship patterns and temporal characteristics of all entities in the target power grid, providing a rich source of information for determining the target fault category, thus ensuring the accuracy of the target fault category determination result. The equipment entity importance assessment allows dispatchers to understand which equipment plays a key role in the faults of the target power grid, enhancing the rationality and interpretability of the target fault location determination result.

[0100] Optionally, the failure probability of the preset failure category c can be determined in the following manner. :

[0101]

[0102] in, and These represent the first weight vector and the second weight vector, respectively. and Let represent the first bias term and the second bias term, respectively; m is the total number of fault categories; q is the index variable used to traverse all possible fault categories; and y is a random variable representing the predicted fault category.

[0103] Alternatively, the target fault location can be determined in the following manner. This refers to the specific faulty equipment in the target power grid.

[0104]

[0105] in, The frequency matrix representing device entity s; The importance of device entity s is determined based on the frequency of its appearance in the current alarm message text. This is the equipment type normalization coefficient.

[0106] The role of Fore(x) is to further perform equipment-level target fault location based on the target fault category results of the target power grid. This process, combined with the target fault category results, constitutes a two-dimensional fault detection result decision output system for the target power grid, namely "fault category-location".

[0107] Through the above steps S102 to S110, the goal of constructing a target knowledge graph based on the current alarm information text of the target power grid and obtaining the target detection results of the target power grid using the target fault detection model can be achieved. This will improve the accuracy and interpretability of the target detection results of the target power grid, thereby solving the technical problems of inaccurate fault detection results and poor interpretability of fault detection results in related technologies.

[0108] Based on the above embodiments and optional embodiments, this application proposes an optional implementation method for power grid fault detection. This implementation method can be understood as an intelligent equipment detection method based on a power dispatch knowledge graph (i.e., a target knowledge graph). By describing the fault events in the current alarm information text of the target power grid, and utilizing word segmentation and time-series fusion technology, a multi-dimensional relationship network of "equipment-action-time sequence" is constructed. Combined with a dynamic relationship graph convolutional network (i.e., a target fault detection model), the accuracy of the target detection results is improved. This method transforms the target detection results into an interpretable fault logic chain through visual topology mapping, promoting the leap from "experience-driven" to "knowledge-data dual-driven" fault handling of the target power grid, and improving the accuracy and interpretability of the target power grid's target detection results.

[0109] First, entity extraction and relationship mining are performed on the current alarm information text to construct a multi-dimensional target knowledge graph containing quantitative relationships and decay factors. Then, based on this target knowledge graph, a Dynamic Relational Graph Convolutional Network (DRGCN) is input. A dynamic relationship embedding layer fuses time-dependent weights (i.e., decay factors) with relationship vectors to enhance the representation of dynamic relationships between entities. Next, temporal gating aggregation technology is used to fuse the neighbor aggregation features of entities with feature vectors. Finally, relationship attention pooling generates the fault probability of a preset fault category, and combined with the importance of device entities, achieves precise device-level localization, forming an interpretable "fault type-location" decision loop.

[0110] Figure 2 This is a flowchart of an optional power grid fault detection method provided according to an embodiment of this application, such as... Figure 2 As shown, the steps of the intelligent equipment detection method based on power dispatch knowledge graph include:

[0111] Step S1: Entity extraction from the knowledge graph for fault detection of the target power grid.

[0112] Entity extraction mainly refers to using the word segmentation and related descriptions of power equipment fault event text (such as current alarm information text) as entities in the target knowledge graph, and assigning numbers to these entities.

[0113] Step S11: First, the current alarm information text of the target power grid is initially segmented using a conventional dictionary and the Viterbi algorithm (i.e., initialization processing).

[0114] Step S12: Then, manual intervention is performed using a custom dictionary to correct the word segmentation results.

[0115] Step S13: Then, perform deduplication filtering on the corrected word segmentation results to construct a dedicated word segmentation dictionary for describing fault events.

[0116] Step S14: Finally, based on the above fault event description word segmentation dictionary, all word segments are numbered according to the order in which they appear in the dictionary, where each word segment corresponds to an entity in the target knowledge graph.

[0117] Step S2: Extracting relationships from the knowledge graph for fault detection of the target power grid.

[0118] During the relation extraction process, a word segmentation relevance matrix is ​​constructed by statistically analyzing the co-occurrence frequency of adjacent words in the current alarm information text, and the co-occurrence frequency is used as a direct basis for quantifying the relationship between entities.

[0119] Step S21, the word segmentation relevance matrix is ​​constructed in the following way:

[0120]

[0121] in, This is the word segmentation relevance matrix (i.e., the relation matrix), where n is the total number of word segments. This indicates the co-occurrence frequency of the word segment numbered i immediately following the word segment numbered j. The larger the value, the stronger the connection between the two word segments.

[0122] Step S22: Dynamic modeling of temporal relationships.

[0123] The timing relationship is mainly used to distinguish the order in which faults occur. Arranging the protection action entities in ascending order of their occurrence time yields the sequence of sequential protection action entities. ,in, This represents the k-th protection action entity in the sequential protection action entity.

[0124] Calculate the time interval between each protection action entity in the sequential protection action entity. The time interval between the k-th protection action entity and the (k-1)-th protection action entity. The following method is used to determine:

[0125]

[0126] in, This indicates the time when the k-th protective action entity occurs. This indicates the time when the (k-1)th protective action entity occurs.

[0127] The valid information triplet (i.e., time sequence information) of the faulty device. And the time-series decay factor (i.e., decay factor) The following method is used to determine:

[0128]

[0129]

[0130] in, Let be the timing decay factor of the k-th protection action entity. This is the attenuation coefficient.

[0131] Step S3: Structure the storage of the power equipment knowledge graph (i.e., the target knowledge graph).

[0132] Step S31: Construction of the target knowledge graph structure.

[0133] Define the target knowledge graph as a triple. Triplet The following method is used to determine:

[0134]

[0135] in, This represents a set of nodes (i.e., a set of entities), specifically the entity objects in the current alarm message text. The set of edges (i.e., the set of relationships) representing the connection relationships between device entities. A label representing a set of relation types, primarily used to describe the semantics of edges.

[0136] Node set The following method is used to determine:

[0137]

[0138] in, This represents a plant node (i.e., a plant entity). This represents a device node (i.e., a device entity). This represents a protected node (i.e., a protected attribute entity). These are action nodes (i.e., protection action entities). Through the above four types of nodes, an interconnected network of the target power grid is formed.

[0139] Relation set The following method is used to determine:

[0140]

[0141] in Representing node i, This represents node j.

[0142] Relationship Type Set The following method is used to determine:

[0143]

[0144] in, This indicates quantitative relationships, specifically the co-occurrence frequency between adjacent entities. It represents temporal relationships and is used to describe the time-series dependencies of protection action entities.

[0145] Step S32, storage mapping function based on Neo4j database.

[0146] Taking node i as an example, the node insertion operation is defined. The insertion operation of node i is implemented in the following way:

[0147]

[0148] in, This represents the insertion operation on node i. This indicates a creation command.

[0149] Step S4: Dynamic Relationship Graph Convolutional Network (DRGCN) detection model.

[0150] The dynamic relation graph convolutional network consists of several parts, including a dynamic relation embedding layer, a temporal gated aggregation layer, a multi-layer cascaded structure, a device fault output layer, and a loss function. These will be introduced separately below.

[0151] Step S41, Dynamic Relationship Embedding Layer.

[0152] Step S411: First, define the input and obtain the four-dimensional tensor based on the target knowledge graph. . The following method is used to determine:

[0153]

[0154] in, Represents a weighted adjacency matrix. The node feature matrix represents four categories: plant nodes, equipment nodes, protection attribute nodes, and protection action nodes. The time interval Δt represents the edge of the temporal relationship. For the pre-calculated attenuation factor .

[0155] Step S412: Dynamic relation vector generation.

[0156] The first step is relation type embedding (i.e., initial relation vector). . The following method is used to determine:

[0157]

[0158] in, A vector representing the quantitative relationships between entities. Represents a time-series relation vector.

[0159] Secondly, historical relationship vectors are extracted, primarily based on the historical knowledge graph of the target power grid. These vectors utilize a GRU (Gated Recurrent Unit) network to extract the historical relationship vectors of the nodes. . The following method is used to determine:

[0160]

[0161] in, This represents the historical relation vector extracted through the GRU network. Let t be the historical relationship vector at time t.

[0162] Based on the initial relation vector and historical relation vectors mentioned above, determine the target relation vector. . The following method is used to determine:

[0163]

[0164] Step S42, timing-gated aggregation layer.

[0165] Step S421, timing-gated aggregation.

[0166] The target relation vector is fused with the neighbor feature vectors of neighboring entities to generate composite features using the following method.

[0167]

[0168]

[0169] in, To fuse the neighbor feature vectors at neighbor entity u Relationship vector with target The composite function, Indicates adaptive weights, For activation function, This represents the gating weight matrix.

[0170] Neighbor aggregation features of node v The following method is used to determine:

[0171]

[0172] in, The first learnable weight matrix, Let f(v) be the set of neighboring entities of node v.

[0173] Step S422, dynamic gating mechanism.

[0174] First, a gating value is calculated. The gating value controls the fusion ratio between the entity's feature vector and the aggregated features of its neighbors. The gating value for node v... The following method is used to determine:

[0175]

[0176] in, This is the learnable second weight matrix.

[0177] The fusion feature vector of node v is determined as follows: :

[0178]

[0179] Here, ⊙ represents element-level multiplication.

[0180] Step S43, multi-layer cascade structure.

[0181] The multi-layered cascaded structure is implemented using residual networks to avoid degradation in deep networks. The multi-layered cascaded structure is modeled as follows:

[0182]

[0183] in, Indicates the first The feature vector output by layer node v, assuming there are L layers in total, then Let V represent the feature vector of the output of node v in the Lth layer, which is also the target feature vector of node v. This represents the residual connection weight matrix. Presentation layer normalization operation.

[0184] Step S44, Equipment Fault Output Layer.

[0185] Step S441, relational attention pooling.

[0186] The strength of the relationship is determined as follows: :

[0187]

[0188] in, The importance weight of relation type r is represented. This is an important vector for interpretable relationships.

[0189] Output feature vector The following method is used to determine:

[0190]

[0191] in, For mean pooling operation, Let r be a set of nodes with relation r. Represents a set of relation types.

[0192] Step S442, equipment fault classification.

[0193] The failure probability of the preset failure category c is determined using the following method. :

[0194]

[0195] in, and These represent the first weight vector and the second weight vector, respectively. and Let represent the first bias term and the second bias term, respectively; m is the total number of fault categories; q is the index variable used to traverse all possible fault categories; and y is a random variable representing the predicted fault category.

[0196] Step S443, equipment fault classification.

[0197] The target fault location is determined using the following method. This refers to the specific faulty equipment in the target power grid.

[0198]

[0199] in, The frequency matrix representing device entity s; The importance of device entity s is determined based on the frequency of its appearance in the current alarm message text. This is the equipment type normalization coefficient.

[0200] The role of Fore(x) is to further perform equipment-level target fault location based on the target fault category results of the target power grid. This process, combined with the target fault category results, constitutes a two-dimensional fault detection result decision output system for the target power grid, namely "fault category-location".

[0201] Step S45, loss function design.

[0202] The loss function is the guiding principle for training the DRGCN model. By coordinating classification accuracy, temporal prediction, and regularization objectives, it improves the accuracy of fault detection in the DRGCN model.

[0203] Step S451, total loss for multiple tasks.

[0204] The loss function of the DRGCN model is determined as follows:

[0205]

[0206] in, This represents the total loss from multi-task collaboration. Represents classification loss, For time-series loss weighting coefficients, For time series prediction loss, These are the regularization loss weight coefficients. This represents the loss of sparse regularization.

[0207] Step S452, classification loss.

[0208] Classification loss The following method is used to determine:

[0209]

[0210] in, This represents the label smoothing coefficient.

[0211] Step S453, timing loss.

[0212] Timing loss The following method is used to determine:

[0213]

[0214] in, Represents a set of relationships based on a target knowledge graph. For the relationship between node i and node j, Indicates the predicted action delay. This indicates the actual delay in the action.

[0215] Step S454, relation sparsity regularization loss.

[0216] Relation sparsity regularization loss The following method is used to determine:

[0217]

[0218] in, This represents the Frobenius norm, which is mainly used to control the magnitude of weights.

[0219] In summary, the process of the intelligent equipment detection method based on power dispatch knowledge graph is as follows: First, based on entity extraction, relation extraction, and target knowledge graph storage structuring, a target knowledge graph for target power grid fault detection is constructed; second, based on the target knowledge graph, a dynamic relation graph convolutional network (DRGCN) detection model is proposed to improve the detection accuracy of complex faults.

[0220] The above-mentioned optional implementation can be applied to an intelligent equipment detection system based on power dispatch knowledge graph, which mainly includes the following parts:

[0221] The entity extraction part of the target knowledge graph is used to achieve word segmentation standardization and node identification of the current alarm information text of the target power grid;

[0222] The target knowledge graph relation extraction part is used to quantify the strength of associations and temporal dependencies between entities;

[0223] The target knowledge graph stores the structured portion, which is used to realize the physical storage and dynamic query of the device relationship network;

[0224] The Dynamic Relationship Graph Convolutional Network (DRGCN) detection component is used to achieve accurate fault detection.

[0225] The entity extraction part of the target knowledge graph includes: text segmentation unit, dictionary correction unit, and node encoding unit.

[0226] The text segmentation unit performs initial word segmentation processing on the current alarm information text of the target power grid using the Viterbi algorithm, and identifies four types of entity boundary dictionaries: power plants, equipment, protection attributes, and protection actions.

[0227] The correction unit, in conjunction with an electrical engineering dictionary, uses manual intervention to correct word segmentation results (such as distinguishing the operating subject and action of "circuit breaker tripping"), and eliminates ambiguity in terminology.

[0228] The node encoding unit numbers all the words according to the order in which they appear in the dictionary, thus constructing the entity base library for device detection.

[0229] The knowledge graph relationship extraction part includes: frequency matrix unit and time series modeling unit.

[0230] The frequency matrix unit is used to construct a word segmentation relevance matrix based on the co-occurrence frequency of adjacent words in the current alarm information text (such as the high-frequency association between "line overload" and "protection trip");

[0231] The timing modeling unit is used to extract the occurrence time of protection action entities, calculate the time interval of protection actions, and generate timing decay factors (such as assigning high weight to short-delay action chains from tripping to reclosing) to form effective information triplet.

[0232] The structured storage component of a knowledge graph includes: graph construction units and database mapping units.

[0233] The graph construction unit is used to define four types of nodes (plant nodes / equipment nodes / protection attribute nodes / protection action nodes) and two types of relationship edges (quantitative relationships / temporal relationships) to construct a multidimensional knowledge network.

[0234] The database mapping unit uses Neo4i commands to physically store nodes and relationships, supporting real-time queries such as "device fault tracing".

[0235] The detection part of the Dynamic Relationship Graph Convolutional Network (DRGCN) includes: a dynamic relationship embedding layer, a temporal gated aggregation layer, and a fault decision output layer.

[0236] The dynamic relation embedding layer is used to fuse the decay factor and relation vector to generate a dynamic relation vector (target relation vector).

[0237] The timing-gated aggregation layer filters key neighbor information through an adaptive gating mechanism, strengthens the propagation of short-delay action chains (such as the "trip-lock" fast sequence), and suppresses long-delay noise.

[0238] The fault decision output layer generates fault classification probabilities (i.e., fault probabilities of preset fault categories, such as line faults / bus faults) through relational attention pooling, and combines the importance of equipment entities determined by frequency-text weights to locate specific faulty equipment (such as "the #1 main transformer of a certain substation"), forming a two-dimensional fault detection decision closed loop of "fault type-location".

[0239] Figure 3 This is a schematic diagram illustrating the detection results of an optional power grid fault detection method provided in an embodiment of this application, as shown below. Figure 3 As shown, using the DRGCN model described above to detect faults in the target power grid can significantly improve the detection accuracy of common simple faults, switch failure to operate faults, and protection failure to operate faults compared to using common knowledge graph detection models.

[0240] The above-mentioned optional implementation methods achieve at least the following effects: By constructing a multi-dimensional knowledge graph containing a set of relationships and a set of relationship types, the spatiotemporal correlation between entities and the dynamic topology of the target power grid are comprehensively considered, improving the target fault detection model's comprehensive understanding of the target power grid's operating state, thereby improving the accuracy of the target fault detection results; the extraction and utilization of time-series information provides a time-dimensional interpretation path for fault detection. The target fault detection model, by analyzing the time-series patterns of protection action entities, can generate time-related fault logic chains, and thus generate a visual topology mapping, enhancing the interpretability of the target fault detection results; by running the target fault detection model, accurate detection and location of faults in the target power grid can be achieved, ensuring that the target fault detection model can fully understand the topology, entity characteristics, and time-series relationships of the target power grid, which not only improves the accuracy of fault detection but also enhances the interpretability of the detection results.

[0241] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0242] This embodiment also provides a power grid fault detection device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the terms "module" and "device" can refer to a combination of software and / or hardware that performs a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0243] According to an embodiment of this application, an apparatus embodiment for implementing a power grid fault detection method is also provided. Figure 4 This is a schematic diagram of a power grid fault detection device according to an embodiment of this application, as shown below. Figure 4 As shown, the above-mentioned power grid fault detection device includes a text acquisition module 402, a target knowledge graph construction module 404, a first determination module 406, a second determination module 408, and a fault detection module 410. The device will be described below.

[0244] The text acquisition module 402 is used to acquire the current alarm information text of the target power grid;

[0245] The target knowledge graph construction module 404 is connected to the text acquisition module 402 and is used to construct a target knowledge graph of the target power grid based on the current alarm information text. The target knowledge graph includes an entity set, a relationship set, and a relationship type set. The entity set includes multiple entities within the target power grid, including plant entities, equipment entities, protection action entities, and protection attribute entities. The relationship set includes the relationships between multiple entities. The relationship type set includes multiple relationship types that exist between multiple entities, as well as the frequency of occurrence of each relationship type.

[0246] The first determining module 406 is connected to the target knowledge graph construction module 404 and is used to determine the temporal information based on the target knowledge graph. The temporal information is used to describe the time sequence characteristics of the state changes of the protection action entity.

[0247] The second determining module 408, connected to the first determining module 406, is used to determine the attenuation factor of the protection action entity based on timing information, wherein the attenuation factor is used to quantify the degree of influence of the protection action entity on fault detection.

[0248] The fault detection module 410, connected to the second determination module 408, is used to perform fault detection on the target power grid based on the attenuation factor, the target knowledge graph, and the entity feature matrix, using a target fault detection model to obtain the target fault detection result of the target power grid. The entity feature matrix includes feature vectors corresponding to multiple entities, and the target fault detection result includes the target fault category and the target fault location.

[0249] In the power grid fault detection device provided in this application embodiment, by setting a text acquisition module 402, a target knowledge graph construction module 404, a first determination module 406, a second determination module 408, and a fault detection module 410, the device aims to obtain the target detection result of the target power grid based on the target knowledge graph constructed from the current alarm information text of the target power grid and using the target fault detection model. This achieves the technical effect of improving the accuracy and interpretability of the target detection result of the target power grid, thereby solving the technical problems of inaccurate fault detection results and poor interpretability of fault detection results in related technologies.

[0250] It should be noted that the above modules can be implemented by software or hardware. For example, for the latter, it can be implemented in the following ways: the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0251] It should be noted that the text acquisition module 402, target knowledge graph construction module 404, first determination module 406, second determination module 408, and fault detection module 410 correspond to steps S102 to S110 in the embodiments. The instances and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in the above embodiments. It should be noted that the above modules, as part of the device, can run in a computer terminal.

[0252] It should be noted that the optional or preferred implementation methods of this embodiment can be found in the relevant descriptions in the embodiments, and will not be repeated here.

[0253] The aforementioned power grid fault detection device may also include a processor and a memory. The text acquisition module 402, the target knowledge graph construction module 404, the first determination module 406, the second determination module 408, the fault detection module 410, etc. are all stored in the memory as program units, and the processor executes the aforementioned program units stored in the memory to realize the corresponding functions.

[0254] The processor contains a core that retrieves the corresponding program unit from memory. One or more cores may be configured. Memory may include non-persistent memory in computer-readable media, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory includes at least one memory chip.

[0255] This application provides a non-volatile storage medium storing a program that, when executed by a processor, implements a power grid fault detection method.

[0256] This application provides an electronic device, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs the following steps: acquiring the current alarm information text of the target power grid; constructing a target knowledge graph of the target power grid based on the current alarm information text, wherein the target knowledge graph includes an entity set, a relationship set, and a relationship type set. The entity set includes multiple entities within the target power grid, including substation entities, equipment entities, protection action entities, and protection attribute entities. The relationship set includes the relationships between multiple entities, and the relationship type set includes various relationship types existing between multiple entities, as well as the frequency of occurrence of each relationship type; determining time-series information based on the target knowledge graph, wherein the time-series information describes the time-series characteristics of the state changes of protection action entities; determining an attenuation factor for protection action entities based on the time-series information, wherein the attenuation factor quantifies the impact of protection action entities on fault detection; and performing fault detection on the target power grid using a target fault detection model based on the attenuation factor, the target knowledge graph, and the entity feature matrix, to obtain the target fault detection result of the target power grid, wherein the entity feature matrix includes feature vectors corresponding to multiple entities, and the target fault detection result includes the target fault category and the target fault location. The devices mentioned in this article can be servers, PCs, etc.

[0257] This application also provides a computer program product, which, when executed on a data processing device, is suitable for executing an initialization program with the following method steps: obtaining the current alarm information text of the target power grid; constructing a target knowledge graph of the target power grid based on the current alarm information text, wherein the target knowledge graph includes an entity set, a relation set, and a relation type set, the entity set including multiple entities within the target power grid, the multiple entities including plant entities, equipment entities, protection action entities, and protection attribute entities, the relation set including the relationships between multiple entities, and the relation type set including multiple relation types existing between multiple entities, and the occurrence frequency corresponding to each of the multiple relation types; determining time-series information based on the target knowledge graph, wherein the time-series information is used to describe the time-series characteristics of the state changes of the protection action entities; determining the attenuation factor of the protection action entities based on the time-series information, wherein the attenuation factor is used to quantify the degree of influence of the protection action entities on fault detection; and performing fault detection on the target power grid using a target fault detection model based on the attenuation factor, the target knowledge graph, and the entity feature matrix to obtain the target fault detection result of the target power grid, wherein the entity feature matrix includes feature vectors corresponding to multiple entities, and the target fault detection result includes the target fault category and the target fault location.

[0258] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0259] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0260] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0261] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0262] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0263] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0264] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0265] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0266] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0267] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for detecting faults in a power grid, characterized in that, include: Obtain the current alarm information text of the target power grid; Based on the current alarm information text, a target knowledge graph of the target power grid is constructed. The target knowledge graph includes an entity set, a relationship set, and a relationship type set. The entity set includes multiple entities within the target power grid, including plant entities, equipment entities, protection action entities, and protection attribute entities. The relationship set includes the relationships between the multiple entities. The relationship type set includes various relationship types that exist between the multiple entities, as well as the frequency of occurrence of each of the various relationship types. Based on the target knowledge graph, temporal information is determined, wherein the temporal information is used to describe the time sequence characteristics of the state changes of the protection action entity; Based on the timing information, an attenuation factor for the protection action entity is determined, wherein the attenuation factor is used to quantify the degree of influence of the protection action entity on fault detection. Based on the attenuation factor, the target knowledge graph, and the entity feature matrix, a target fault detection model is used to detect faults in the target power grid, and a target fault detection result of the target power grid is obtained. The entity feature matrix includes feature vectors corresponding to the multiple entities, and the target fault detection result includes the target fault category and the target fault location.

2. The method according to claim 1, characterized in that, The determination of temporal information based on the target knowledge graph includes: The protection action entities are arranged in ascending order according to their occurrence time to obtain sequential protection action entities; The timing information is determined based on the occurrence time of the sequential protection action entity.

3. The method according to any one of claims 1 or 2, characterized in that, The method of using a target fault detection model to detect faults in the target power grid based on the attenuation factor, the target knowledge graph, and the entity feature matrix, to obtain the target fault detection result of the target power grid, includes: For any entity among the plurality of entities, an initial relation vector is determined based on the target knowledge graph. The initial relation vector is used to describe the relationship type and frequency of occurrence between the entity and other entities, as well as the temporal relationship between the entity and the other entities. The other entities are entities other than the entity among the plurality of entities. Based on the attenuation factor, the dynamic relationship embedding layer of the target fault detection model is used to fuse the initial relationship vector and the historical relationship vectors corresponding to multiple historical moments to obtain the target relationship vector of any entity. The historical relationship vector is used to describe the relationship type and frequency of occurrence between any entity and other entities at the corresponding historical moment, as well as the temporal relationship between any entity and other entities. The target relation vector is fused with the neighbor feature vector of the neighboring entities of any entity to obtain the neighbor aggregation feature of any entity. The neighbor aggregation features corresponding to multiple entities are determined by using the method of determining the neighbor aggregation features of any one of the entities; Based on the neighbor aggregation features corresponding to the multiple entities and the entity feature matrix, the target fault detection result is determined.

4. The method according to claim 3, characterized in that, The step of determining the target fault detection result based on the neighbor aggregation features corresponding to the multiple entities and the entity feature matrix includes: For any entity among the plurality of entities, a gating value is determined based on the neighbor aggregation features of the entity, the feature vector of the entity, and the attenuation factor of the protection action entities included in the neighbor entities. The gating value is used to control the degree of fusion between the feature vector of the entity and the neighbor aggregation features of the entity. The feature vector of the entity is determined based on the entity feature matrix. Based on the gating value, the neighbor aggregation feature of any entity is fused with the feature vector of any entity to obtain the fused feature vector of any entity; Based on the fused feature vector of any entity, the target feature vector of any entity is determined using the multi-layer cascaded structure of the target fault detection model, wherein the multi-layer cascaded structure is implemented in the form of a residual network. The target feature vectors corresponding to the plurality of entities are determined by using the method of determining the target feature vector of any one of the entities; The target fault detection result is determined based on the target feature vectors corresponding to the multiple entities.

5. The method according to claim 4, characterized in that, The step of determining the target fault detection result based on the target feature vectors corresponding to the multiple entities includes: Based on the target relation vectors corresponding to multiple entities, the relation strengths corresponding to the various relation types are determined. Based on the relationship strength corresponding to the various relationship types and the target feature vectors corresponding to the various entities, the output feature vector is determined. The target fault detection result is determined based on the output feature vector.

6. The method according to claim 5, characterized in that, Determining the target fault detection result based on the output feature vector includes: Based on the output feature vector, the fault probabilities corresponding to various preset fault categories are determined, and the preset fault category corresponding to the highest fault probability is determined as the target fault category. Based on the current alarm information text, the importance of the device entity is determined; Based on the importance and the frequency matrix of the device entity, the target fault location is determined, wherein the frequency matrix is ​​obtained based on the target knowledge graph and is used to quantify the correlation between the device entity and the current fault event of the target power grid; The target fault detection result is determined based on the target fault category and the target fault location.

7. A fault detection device for a power grid, characterized in that, include: The text acquisition module is used to acquire the current alarm information text of the target power grid; The target knowledge graph construction module is used to construct a target knowledge graph of the target power grid based on the current alarm information text. The target knowledge graph includes an entity set, a relationship set, and a relationship type set. The entity set includes multiple entities within the target power grid, including plant entities, equipment entities, protection action entities, and protection attribute entities. The relationship set includes the relationships between the multiple entities. The relationship type set includes multiple relationship types that exist between the multiple entities, as well as the occurrence frequency of each of the multiple relationship types. The first determining module is used to determine time-series information based on the target knowledge graph, wherein the time-series information is used to describe the time-series characteristics of the state changes of the protection action entity; The second determining module is used to determine the attenuation factor of the protection action entity based on the timing information, wherein the attenuation factor is used to quantify the degree of influence of the protection action entity on fault detection. The fault detection module is used to perform fault detection on the target power grid based on the attenuation factor, the target knowledge graph, and the entity feature matrix, using a target fault detection model to obtain the target fault detection result of the target power grid. The entity feature matrix includes feature vectors corresponding to the multiple entities respectively, and the target fault detection result includes the target fault category and the target fault location.

8. A non-volatile storage medium, characterized in that, The non-volatile storage medium stores multiple instructions adapted for loading and execution by a processor of the power grid fault detection method according to any one of claims 1 to 6.

9. An electronic device, characterized in that, include: One or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the power grid fault detection method according to any one of claims 1 to 6.

10. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the power grid fault detection method according to any one of claims 1 to 6.

Images