Abnormal equipment identification method and device based on unified data base modeling of power equipment, terminal equipment and storage medium

Abstract

The invention discloses an abnormal equipment identification method and device based on unified data base modeling of power equipment, terminal equipment and a storage medium, and belongs to the technical field of term equipment anomaly identification. The method comprises the following steps: acquiring each piece of power equipment in a power system and original heterogeneous label data, and constructing an initial power system mapping model; then matching the original heterogeneous tag data according to a preset dictionary tree to obtain candidate tags; inputting each candidate tag into a preset power professional word vector model to obtain a selected tag, and replacing the corresponding original heterogeneous tag data according to the selected tag to obtain a final power system mapping model; and finally, according to the final power system mapping model, performing anomaly identification on each power device. By implementing the method and the device, the problem of inaccurate anomaly recognition due to non-uniform data semantics of the power equipment in the prior art can be solved.

Description

Technical Field

[0001] This invention relates to the field of equipment anomaly identification technology, and in particular to an anomaly identification method, device, terminal equipment and storage medium based on a unified data base model for power equipment. Background Technology

[0002] Equipment in power systems, such as protection devices, automation systems, and stability control devices, are the "nerve center" for ensuring the safe and stable operation of the power grid. With the development of smart grids, these devices generate massive amounts of multi-source, heterogeneous data. However, currently, various professional systems, such as information security, security automation, and communication power supply, are built independently, resulting in serious data silos.

[0003] In existing technologies, power system mapping models are typically used to detect equipment anomalies. However, since various systems often use proprietary protocols or different international standards, the semantics of the data in the mapping models are inconsistent. For example, the same protection signal may be named differently in different systems. It is difficult to associate "zero-sequence overcurrent protection stage 1" with "zero current stage 1". Since anomaly detection depends on the regularity of equipment operation characteristics, semantic confusion can lead to incomplete feature extraction or repeated calculations, interfering with the normal anomaly judgment logic. Therefore, there is a problem of inaccurate anomaly identification. Summary of the Invention

[0004] This invention provides a method, device, terminal device, and storage medium for identifying abnormal equipment based on a unified data foundation model for power equipment. This invention can solve the problem of inaccurate anomaly identification in the prior art due to the lack of unified data semantics of power equipment.

[0005] An embodiment of the present invention provides a method for identifying abnormal equipment based on a unified data base model for power equipment, comprising: Obtain the original heterogeneous label data of each power device in the power system and the aforementioned power devices, and construct an initial power system mapping model; Based on the preset dictionary tree constructed using power industry terminology, the above-mentioned original heterogeneous label data is matched to obtain the candidate label that is most similar to each of the original heterogeneous label data in the preset dictionary tree. Each candidate label is input into a preset power professional word vector model so that the preset power professional word vector model can calculate the similarity between each candidate label and each preset standard label. The standard label with the highest similarity is selected as the selected label, and the corresponding original heterogeneous label data is replaced according to the selected label to obtain the final power system mapping model. Based on the final power system mapping model, anomalies are identified in each power device.

[0006] Furthermore, the training of the aforementioned pre-defined power industry term vector model includes: Obtain several heterogeneous data label sample groups with real labels; wherein, each group of the above heterogeneous data label sample includes a heterogeneous label sample, a positive example heterogeneous label sample whose real semantic similarity with the above heterogeneous label sample is not less than a preset similarity threshold, and a negative example heterogeneous label sample whose real semantic similarity with the above heterogeneous label sample is less than the above preset similarity threshold; the above real label is used to represent the real similarity between the heterogeneous label sample and the positive example heterogeneous label sample and the negative example heterogeneous label sample, respectively; The above heterogeneous data label sample group is input into the power professional word vector model to be trained for iterative training until the loss function converges, and the trained preset power professional word vector model is obtained. In each iteration of training, the first predicted semantic similarity between the current heterogeneous label sample and the corresponding positive heterogeneous label sample, and the second predicted semantic similarity between the current heterogeneous label sample and the corresponding negative heterogeneous label sample are calculated based on the current heterogeneous data label sample group. The current loss function is calculated based on the first predicted semantic similarity, the second predicted semantic similarity, and the corresponding real labels. If the current loss function converges, the current power industry word vector model is used as the aforementioned preset power industry word vector model; otherwise, the model parameters in the current power industry word vector model are adjusted, and training continues.

[0007] Furthermore, based on the final power system mapping model, the above-mentioned anomaly identification is performed on each power device, including: Using each power device in the final power system mapping model as a node, the topological connection relationship between each power device as an edge, and the selected labels and the original heterogeneous data labels as node information, a device association knowledge graph is constructed. Obtain the power time-series data of each power device within a preset time period; A three-dimensional feature matrix is ​​constructed based on the preset time period, power time series data, and the topological location of the corresponding power equipment in the above-mentioned equipment association knowledge graph. The aforementioned three-dimensional feature matrix is ​​input into a preset anomaly identification model so that the preset anomaly identification model can obtain the anomaly score of each power device based on the aforementioned three-dimensional feature matrix; For each power device, if the above-mentioned abnormal score is greater than the preset score threshold, it is preliminarily determined that the above-mentioned power device has an abnormality; otherwise, it is preliminarily determined that the above-mentioned power device does not have an abnormality.

[0008] Furthermore, based on the preset time period, power time-series data, and the topological location of the corresponding power equipment in the final power system mapping model, a three-dimensional feature matrix is ​​constructed, including: Perform a Fourier transform on the above power time-series data to obtain the power data spectrum; The amplitude of the preset harmonic order in the above power data spectrum is extracted to obtain the frequency feature vector; Based on the final power system mapping model, the associated power equipment is identified, and the phase difference of the power time series data of the aforementioned associated power equipment is calculated. Based on the preset time period, frequency feature vector, and phase difference of all power equipment, the above three-dimensional feature matrix is ​​constructed.

[0009] Furthermore, after initially determining that the aforementioned power equipment is malfunctioning, the following also applies: From the aforementioned knowledge graph of device associations, we extract the associated device information of the associated devices that are associated with the aforementioned power devices; Based on a preset consistency rule built on power knowledge, the information of the aforementioned associated equipment is verified for faults, and the final anomaly identification result of the aforementioned power equipment is obtained.

[0010] Furthermore, it also includes: Several anomaly handling strategies were developed to obtain the final anomaly identification results for power equipment with anomalies. A power system simulation model is constructed based on the above power system mapping model; The above-mentioned anomaly handling strategies are executed in the power system simulation model, and several power system operation data are obtained after execution. Based on the aforementioned power system operation data, security verification is performed. The anomaly handling strategy that passes the security verification is taken as the target anomaly handling strategy, and the target anomaly handling strategy is executed in the aforementioned power system.

[0011] Based on the above method embodiments, the present invention provides corresponding apparatus embodiments; This invention provides an abnormal equipment identification device based on a unified data foundation model for power equipment, comprising: The system includes an initial mapping model construction module, a label matching module, a mapping model optimization module, and an anomaly detection module. The aforementioned initial mapping model construction module is used to obtain the original heterogeneous label data of each power device in the power system and the power device, and to construct the initial power system mapping model. The aforementioned tag matching module is used to match the aforementioned original heterogeneous tag data according to a preset dictionary tree constructed based on power industry terminology, and to obtain the candidate tag that is most similar to each of the original heterogeneous tag data in the preset dictionary tree. The aforementioned mapping model optimization module is used to input each candidate label into a preset power professional word vector model, so that the preset power professional word vector model calculates the similarity between each candidate label and each preset standard label, takes the standard label with the highest similarity as the selected label, and replaces the corresponding original heterogeneous label data according to the selected label to obtain the final power system mapping model. The aforementioned anomaly identification module is used to identify anomalies in each power device based on the final power system mapping model.

[0012] Furthermore, the above-mentioned mapping model optimization module includes: Sample data acquisition unit and model training unit; The aforementioned sample data acquisition unit is used to acquire several heterogeneous data label sample groups with real labels; wherein, each group of the aforementioned heterogeneous data label samples includes a heterogeneous label sample, a positive example heterogeneous label sample whose real semantic similarity with the aforementioned heterogeneous label sample is not less than a preset similarity threshold, and a negative example heterogeneous label sample whose real semantic similarity with the aforementioned heterogeneous label sample is less than the aforementioned preset similarity threshold; the aforementioned real label is used to represent the real similarity between the heterogeneous label sample and the positive example heterogeneous label sample and the negative example heterogeneous label sample, respectively; The aforementioned model training unit is used to input the aforementioned heterogeneous data label sample group into the power industry word vector model to be trained for iterative training until the loss function converges, thereby obtaining the trained preset power industry word vector model. In each iteration of training, the first predicted semantic similarity between the current heterogeneous label sample and the corresponding positive heterogeneous label sample, and the second predicted semantic similarity between the current heterogeneous label sample and the corresponding negative heterogeneous label sample are calculated based on the current heterogeneous data label sample group. The current loss function is calculated based on the first predicted semantic similarity, the second predicted semantic similarity, and the corresponding real labels. If the current loss function converges, the current power industry word vector model is used as the aforementioned preset power industry word vector model; otherwise, the model parameters in the current power industry word vector model are adjusted, and training continues.

[0013] Based on the above method embodiments, the present invention provides a corresponding terminal device embodiment; The present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the abnormal device identification method based on the unified data base modeling of power equipment described in any embodiment of the present invention.

[0014] Based on the above method embodiments, the present invention provides a corresponding storage medium embodiment; The present invention provides a storage medium including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the abnormal device identification method based on the unified data base modeling of power equipment described in any embodiment of the present invention.

[0015] The embodiments of the present invention have the following beneficial effects: This invention provides a method, apparatus, terminal device, and storage medium for identifying abnormal equipment based on a unified data foundation model for power equipment. The method includes: acquiring various power devices in a power system and their original heterogeneous label data, and constructing an initial power system mapping model; then, matching the original heterogeneous label data against a preset dictionary tree constructed based on power industry terminology to obtain candidate labels that are most similar to each original heterogeneous label data in the preset dictionary tree; then, inputting each candidate label into a preset power industry term vector model to calculate the similarity between each candidate label and each preset standard label, selecting the standard label with the highest similarity as the selected label, and replacing the corresponding original heterogeneous label data according to the selected label to obtain the final power system mapping model; finally, identifying anomalies in each power device based on the final power system mapping model. Therefore, after initially constructing the power system mapping model, this invention utilizes a preset dictionary tree and a preset power industry term vector model to unify the terminology of the original heterogeneous label data of each power device, and then performs anomaly identification based on the terminology-unified mapping model, thus improving the accuracy of the anomaly identification results. Attached Figure Description

[0016] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating an abnormal equipment identification method based on a unified data base model for power equipment, provided by an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram of the structure of an abnormal equipment identification device based on a unified data base model for power equipment, provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0021] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0022] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0023] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0024] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0025] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0026] See Figure 1 To address the problem of inaccurate anomaly identification due to inconsistent data semantics in existing power equipment technologies, an embodiment of this invention provides an anomaly equipment identification method based on a unified data foundation model for power equipment, comprising: Step S101: Obtain the original heterogeneous label data of each power device in the power system and the above power devices, and construct the initial power system mapping model; Specifically, using the electrical bays of the substation (such as line bays, transformer bays, bus bays, etc.) as the basic modeling unit and communication bridge, the primary equipment (such as circuit breakers and disconnect switches) and all related secondary equipment (such as protection devices, measurement and control devices, safety and automation devices, fault recorders, network equipment, etc.) within the bay are mapped onto the entity. At the same time, each device has its own original heterogeneous label (such as protection settings, fault record files, alarm information, measurement data, and network security information, etc.), thus constructing an initial power system mapping model.

[0027] Step S102: Based on the preset dictionary tree constructed based on power industry terminology, match the above-mentioned original heterogeneous label data to obtain the candidate label that is most similar to each of the original heterogeneous label data in the preset dictionary tree. Specifically, the aforementioned pre-defined trie is a double-array trie containing standard power terms such as "zero-sequence differential protection." This pre-defined trie breaks down power terms into keyword fragments (such as "zero-sequence differential protection" and "overcurrent protection stage I"). The AC automaton then uses this pre-defined trie to rapidly traverse each original heterogeneous label data, quickly capturing keyword matches, searching for all possible matching fragments in the pre-defined trie, calculating the corresponding confidence scores, and finally selecting the highest confidence score as the candidate label. For example, it might quickly match "zero-sequence differential protection" (because "zero" and "differential" are both keywords), outputting the matching result (i.e., the candidate label) and its corresponding confidence score.

[0028] Specifically, the aforementioned confidence level can be calculated based on factors such as the favorability ratio of the matched characters, which is existing technology and will not be elaborated here.

[0029] Step S103: Input each candidate label into the preset power professional word vector model so that the preset power professional word vector model can calculate the similarity between each candidate label and each preset standard label. The standard label with the highest similarity is selected as the selected label, and the corresponding original heterogeneous label data is replaced according to the selected label to obtain the final power system mapping model. Specifically, the aforementioned preset power industry-specific word vector model is a neural network model. After candidate labels are input into the model, the model retrieves a preset standard label database and calculates the semantic similarity between the input candidate label and the standard labels in the database to determine the selected label. This semantic similarity can be represented by calculating the cosine similarity between the two. Cosine similarity measures the directional proximity of two vectors, with values ​​between 0 and 1; the closer to 1, the more semantically similar.

[0030] Preferably, a threshold (such as 0.85) can be set. If the maximum similarity is higher than this threshold, it is determined to be semantically equivalent, and the selected label is determined; otherwise, the candidate label is marked as abnormal, and it is necessary to check whether the matching result of the corresponding original heterogeneous label data is incorrect.

[0031] Preferably, before inputting candidate labels into the model, a confidence threshold (e.g., 0.9) can be set. Candidate labels with a confidence level not less than the confidence threshold are considered to have basically reliable results after matching by the AC automaton, and can therefore be directly input into the preset power professional word vector model for final confirmation. For candidate labels with a confidence level less than the confidence threshold, it is considered that the results after matching by the AC automaton are relatively ambiguous. Therefore, string similarity algorithms can be used to fine-tune these candidate labels at the character level in advance. For example, the character "zero difference" can be corrected to the closest reasonable string: "zero-order differential", and then input into the model to match the selected label.

[0032] Preferably, by standardizing the original heterogeneous label data, regardless of how the source system names it, it is uniformly mapped to standardized power terms, which are then injected into the mapping model to ensure semantic elimination and provide support for subsequent in-depth analysis.

[0033] Preferably, by constructing a unified data model—the power system mapping model—professional barriers are completely broken down, achieving true unity of equipment data in semantics, spatiotemporal context, and logic, thus providing a solid data foundation for advanced applications.

[0034] Preferably, the AC automaton can achieve second-level data standardization, upgrading the anomaly handling from the traditional model that relies on manual labor and takes several hours to an automated intelligent process at the minute or even second level, improving handling efficiency by more than 40 times.

[0035] In a preferred embodiment, the training of the aforementioned preset power industry term vector model includes: Obtain several heterogeneous data label sample groups with real labels; wherein, each group of the above heterogeneous data label sample includes a heterogeneous label sample, a positive example heterogeneous label sample whose real semantic similarity with the above heterogeneous label sample is not less than a preset similarity threshold, and a negative example heterogeneous label sample whose real semantic similarity with the above heterogeneous label sample is less than the above preset similarity threshold; the above real label is used to represent the real similarity between the heterogeneous label sample and the positive example heterogeneous label sample and the negative example heterogeneous label sample, respectively; Specifically, the aforementioned heterogeneous data label sample groups can be obtained from power literature, regulations, and configuration files. Therefore, the model trained based on these data can learn the deeper meanings of power-related terms.

[0036] The above heterogeneous data label sample group is input into the power professional word vector model to be trained for iterative training until the loss function converges, and the trained preset power professional word vector model is obtained. In each iteration of training, the first predicted semantic similarity between the current heterogeneous label sample and the corresponding positive heterogeneous label sample, and the second predicted semantic similarity between the current heterogeneous label sample and the corresponding negative heterogeneous label sample are calculated based on the current heterogeneous data label sample group. The current loss function is calculated based on the first predicted semantic similarity, the second predicted semantic similarity, and the corresponding real labels. If the current loss function converges, the current power industry word vector model is used as the aforementioned preset power industry word vector model; otherwise, the model parameters in the current power industry word vector model are adjusted, and training continues.

[0037] Specifically, the loss function described above can be calculated using the following formula: In the formula, L represents the loss function, A represents heterogeneous labeled samples, N represents negative heterogeneous labeled samples, and P represents positive heterogeneous labeled samples. Indicates the second predicted semantic similarity. Indicates the first predicted semantic similarity. This represents the margin threshold, which can be set to 0.5. This is used to force a sufficient semantic similarity gap between heterogeneous labeled samples and positive and negative heterogeneous labeled samples to avoid confusion. This represents the weighting coefficient, which can be set to 0.5. This represents the true label corresponding to the first predicted semantic similarity. This represents the true label corresponding to the second predicted semantic similarity.

[0038] Preferably, the semantic similarity mentioned above can be represented by cosine similarity.

[0039] In this preferred embodiment, the power industry word vector model was trained using several heterogeneous data label sample groups with real labels.

[0040] Step S104: Based on the final power system mapping model, identify anomalies in each power device.

[0041] In a preferred embodiment, the above-mentioned anomaly identification of each power device based on the final power system mapping model includes: Using each power device in the final power system mapping model as a node, the topological connection relationship between each power device as an edge, and the selected labels and the original heterogeneous data labels as node information, a device association knowledge graph is constructed. Indicatively, each electrical device, such as "line protection device A", "circuit breaker QF1", and "bay I", becomes a node. Subsequently, the relationships between devices are defined. For example, "protection device A" can control "circuit breaker QF1". "Circuit breaker QF1" belongs to "bay I" and becomes an edge connecting the nodes. These edges define the static logical relationships between entities.

[0042] Specifically, when replacing the corresponding original heterogeneous label data based on the selected label, the original heterogeneous label data may itself be standard label data, so it does not need to be replaced or the replacement results in the original heterogeneous label data. Therefore, when constructing a device association knowledge graph, some nodes correspond to selected labels, while others correspond to original heterogeneous data labels.

[0043] Specifically, entities (such as protection devices, circuit breakers, bays, etc.) and the relationships between entities (such as topology, function, and logical relationships between equipment) are extracted from the final power system mapping model. Entities are used as nodes and relationships are used as edges. At the same time, corresponding node information is added to each node to obtain the equipment relationship knowledge graph.

[0044] As an illustration, the data "Zero-sequence overcurrent protection stage 1 setting: 1000A" is an attribute of the entity "Line Protection Device A" (i.e., the node information mentioned above). However, these attribute values ​​change over time, such as setting modifications and alarm signal resets. Therefore, the constructed device association knowledge graph becomes a dynamic knowledge graph that can reflect the real-time status of the system.

[0045] Obtain the power time-series data of each power device within a preset time period; A three-dimensional feature matrix is ​​constructed based on the preset time period, power time series data, and the topological location of the corresponding power equipment in the above-mentioned equipment association knowledge graph. Specifically, a three-dimensional feature matrix refers to a feature matrix based on the time dimension, spatial dimension, and frequency dimension.

[0046] The aforementioned three-dimensional feature matrix is ​​input into a preset anomaly identification model so that the preset anomaly identification model can obtain the anomaly score of each power device based on the aforementioned three-dimensional feature matrix; Specifically, the aforementioned pre-defined anomaly detection model is an isolated forest model, and its training process follows machine learning principles. During training, a three-dimensional feature matrix is ​​first acquired for several historical time periods. Assuming each historical time period generates a vector containing 100 feature values ​​(e.g., 20 temporal features + 30 spatial correlation features + 50 spectral features), three months of data constitute a massive dataset containing millions of feature vectors. This dataset is then input into the isolated forest model. The model randomly selects a feature from the input three-dimensional feature matrix and a random splitting value, dividing the dataset into two parts. This process is then recursively repeated along this spatial dimension, randomly splitting a feature matrix. A normal, dense data point requires many "cuts" to be isolated into a subspace, while an anomaly, being far from the normal data group, may only require a few cuts to isolate. Finally, after training, the algorithm generates a "forest" composed of many "decision trees." This forest remembers the pattern of cutting the normal data space—that is, how many random cuts are needed to isolate the baseline distribution of a data point—thus obtaining the trained model.

[0047] Specifically, the anomaly detection process of the pre-defined anomaly detection model is a multi-dimensional joint outlier detection process based on unsupervised machine learning. It does not rely on a fixed threshold, but instead establishes a "normal baseline" by learning historical, normal multi-dimensional data patterns. This allows it to keenly detect latent anomalies that are not obvious in a single dimension, but reveal "unique" characteristics under the joint analysis of multi-dimensional features.

[0048] Specifically, after the pre-trained anomaly detection model is completed, it first constructs a three-dimensional feature matrix and converts it into a feature vector before inputting it into the model. Each tree in the forest attempts to isolate a new data point and records the path length (number of cuts) required to isolate it. The model calculates an average path length. If the average path length of the input data is significantly shorter than the average path length of the normal data points, it indicates that it is easily isolated and is therefore considered an anomaly. Finally, the model outputs an anomaly score, and based on the anomaly score corresponding to the data point, the abnormal power equipment can be identified.

[0049] For each power device, if the above-mentioned abnormal score is greater than the preset score threshold, it is preliminarily determined that the above-mentioned power device has an abnormality; otherwise, it is preliminarily determined that the above-mentioned power device does not have an abnormality.

[0050] Specifically, the anomaly score is typically between 0 and 1. The closer the score is to 1, the greater the likelihood that the data point is an anomaly. Therefore, in anomaly identification, a threshold is set (e.g., 0.65). When the anomaly score is greater than 0.65, it is preliminarily determined that the corresponding power equipment is abnormal.

[0051] In this preferred embodiment, anomalies were identified for each power device based on the final power system mapping model.

[0052] In another preferred embodiment, the construction of a three-dimensional feature matrix based on a preset time period, power time-series data, and the topological location of the corresponding power equipment in the final power system mapping model includes: Perform a Fourier transform on the above power time-series data to obtain the power data spectrum; Specifically, power time-series data is a sequence of raw analog sampled values. For example, within one power frequency cycle (20 milliseconds), 80 consecutive instantaneous current values ​​collected at a rate of 4000 points per second constitute a sampling sequence S. The sampled values ​​mainly include power data of various analog quantities, such as current, voltage, power, and phase angle.

[0053] The amplitude of the preset harmonic order in the above power data spectrum is extracted to obtain the frequency feature vector; To illustrate, the amplitudes of preset harmonic orders (e.g., 2nd to 13th harmonics) are extracted from the power data spectrum obtained after FFT transformation, and finally a feature vector in the frequency dimension is formed.

[0054] Based on the final power system mapping model, the associated power equipment is identified, and the phase difference of the power time series data of the aforementioned associated power equipment is calculated. Specifically, phase difference is the phase angle difference between the power time-series data of two related power devices, which can be calculated based on the signals corresponding to the power time-series data. Phase difference is a crucial parameter for determining power direction and equipment synchronization. For example, in differential protection, it is necessary to compare the phases of the currents at both ends of the line. Under normal circumstances, their phase difference is close to 180 degrees (current directions are opposite). If a fault occurs inside the line, the phases of the currents at both ends may become the same, causing a significant change in phase difference, which is a key criterion for protection operation.

[0055] Based on the preset time period, frequency feature vector, and phase difference of all power equipment, the above three-dimensional feature matrix is ​​constructed.

[0056] Specifically, based on a preset time period, data change characteristics (such as data abrupt change gradients) of power time-series data within that preset time period can be extracted as time dimension features, frequency feature vectors as frequency dimension features, and phase difference as spatial dimension features, thus constructing the aforementioned three-dimensional feature matrix. Illustratively, the three-dimensional feature matrix can be represented by the following formula: In the formula, Let m represent the three-dimensional feature matrix, n represent the time dimension, n represent the spatial dimension, and p represent the frequency dimension. This indicates the sampling interval for power time-series data, i.e., the aforementioned preset time period. Indicates phase difference, This represents the power data spectrum after performing a Fourier transform on the power time-series data. This represents the power time series data corresponding to the kth harmonic.

[0057] In this preferred embodiment, a three-dimensional feature matrix is ​​constructed based on a preset time period, power time series data, and the topological location of the corresponding power equipment in the final power system mapping model.

[0058] In another preferred embodiment, after initially determining that the aforementioned power equipment is malfunctioning, the method further includes: From the aforementioned knowledge graph of device associations, we extract the associated device information of the associated devices that are associated with the aforementioned power devices; Specifically, taking the abnormal power equipment as the center, we traverse along the edges connected to it in the equipment association knowledge graph to obtain the information of the associated equipment, which serves as evidence of multi-source association.

[0059] Based on a preset consistency rule built on power knowledge, the information of the aforementioned associated equipment is verified for faults, and the final anomaly identification result of the aforementioned power equipment is obtained.

[0060] Specifically, the information from associated devices is compared with consistency rules constructed based on power knowledge. For example, a consistency rule could be: "A genuine CT disconnection must be accompanied by a severe anomaly in the related current measurement." If the isolated forest model has preliminarily determined that the CT device is abnormal, and the associated device information is "synchronous device voltage difference <5%", the consistency rule shows that this contradicts the preliminary conclusion obtained from the isolated forest model. Therefore, the preliminary conclusion is a false alarm, and the final anomaly identification result for the CT device is no anomaly. Conversely, if the associated device information is "MU sampling value distortion" and "related protection activated", the consistency rule determines that the preliminary conclusion is correct, the CT device is indeed abnormal, and its final anomaly identification result is anomaly.

[0061] Preferably, based on the information of associated equipment, cross-validation was performed on the power equipment initially judged to be abnormal, thereby significantly improving the accuracy of the conclusion.

[0062] The preferred dual-channel detection mechanism, which combines a three-dimensional feature matrix of time, space, frequency, and device-related knowledge graph, can not only detect explicit anomalies, but also accurately identify implicit and complex anomalies that are difficult to detect by traditional methods, significantly reducing the false alarm rate.

[0063] In this preferred embodiment, the power equipment initially identified as abnormal was finally verified based on the associated device information and consistency rules.

[0064] In another preferred embodiment, it further includes: Several anomaly handling strategies were developed to obtain the final anomaly identification results for power equipment with anomalies. Specifically, an exception handling strategy library can be configured in advance, which records the corresponding handling strategies when each power device malfunctions.

[0065] A power system simulation model is constructed based on the above power system mapping model; Specifically, after obtaining the anomaly handling strategy, to ensure the safety and effectiveness of the strategy, it is not immediately implemented in the power system, but is first verified through simulation in a power system simulation model built based on the power system mapping model.

[0066] The above-mentioned anomaly handling strategies are executed in the power system simulation model, and several power system operation data are obtained after execution. Based on the aforementioned power system operation data, security verification is performed. The anomaly handling strategy that passes the security verification is taken as the target anomaly handling strategy, and the target anomaly handling strategy is executed in the aforementioned power system.

[0067] Specifically, the safety verification mainly includes: logical correctness rule verification: verifying whether the activation conditions and action logic of the anomaly handling strategy conform to the actual operation mode of the current power system; system safety rule verification: verifying whether the key parameters of the power system such as power flow, voltage, and frequency remain within a safe and stable range after the anomaly handling strategy is executed; equipment safety rule verification: verifying whether related electrical equipment such as transformers, lines, and circuit breakers will be overloaded or subjected to impacts exceeding their tolerance during and after the execution of the anomaly handling strategy; and protection coordination rule verification: verifying whether the protection setting coordination relationship of the power system is still correct after the anomaly handling strategy is executed, and whether it will cause protection maloperation or failure to operate.

[0068] Specifically, the entire verification process is as follows: Step 1: Simulation environment initialization. Based on the power system mapping model, a real-time simulation model (i.e., the aforementioned power system mapping model) is quickly constructed. This model replicates the current operating state, including the real-time status of the working power supply line, backup power supply line, busbars, transformers, loads, and all circuit breakers and disconnectors; current power flow, voltage, and current data; the logic model of the automatic transfer switch itself, and related line protection and transformer protection models. Step 2: Injecting anomaly handling strategies and acquiring corresponding power system operating data. In the simulation environment, anomaly conditions consistent with the real scenario are set, for example, simulating the failure of the measurement function of a protection device reporting "CT disconnection". Then, the anomaly handling strategy is executed (e.g., issuing a series of instructions such as "trip the working power supply line switch" and "close the backup power supply line switch" according to a preset sequence). Step 3: Safety verification based on power system operating data.

[0069] Specifically, after the security verification is completed, there will be two results: pass or fail. If all security verifications pass, the system is deemed to have passed and is allowed to execute the corresponding anomaly handling strategy in the actual power system; conversely, if any security verification fails, the system is deemed to have failed.

[0070] Preferably, the result (success or failure) and the reason for each exception handling can be fed back to the strategy library to continuously optimize the exception handling strategy.

[0071] In this preferred embodiment, the final exception handling strategy is determined by simulating and verifying each exception handling strategy.

[0072] Based on the above method embodiments, the present invention provides corresponding apparatus embodiments.

[0073] like Figure 2 As shown, an embodiment of the present invention provides an abnormal equipment identification device based on a unified data foundation model for power equipment, comprising: The system includes an initial mapping model construction module, a label matching module, a mapping model optimization module, and an anomaly detection module. The aforementioned initial mapping model construction module is used to obtain the original heterogeneous label data of each power device in the power system and the power device, and to construct the initial power system mapping model. The aforementioned tag matching module is used to match the aforementioned original heterogeneous tag data according to a preset dictionary tree constructed based on power industry terminology, and to obtain the candidate tag that is most similar to each of the original heterogeneous tag data in the preset dictionary tree. The aforementioned mapping model optimization module is used to input each candidate label into a preset power professional word vector model, so that the preset power professional word vector model calculates the similarity between each candidate label and each preset standard label, takes the standard label with the highest similarity as the selected label, and replaces the corresponding original heterogeneous label data according to the selected label to obtain the final power system mapping model. The aforementioned anomaly identification module is used to identify anomalies in each power device based on the final power system mapping model.

[0074] In a preferred embodiment, the mapping model optimization module includes: Sample data acquisition unit and model training unit; The aforementioned sample data acquisition unit is used to acquire several heterogeneous data label sample groups with real labels; wherein, each group of the aforementioned heterogeneous data label samples includes a heterogeneous label sample, a positive example heterogeneous label sample whose real semantic similarity with the aforementioned heterogeneous label sample is not less than a preset similarity threshold, and a negative example heterogeneous label sample whose real semantic similarity with the aforementioned heterogeneous label sample is less than the aforementioned preset similarity threshold; the aforementioned real label is used to represent the real similarity between the heterogeneous label sample and the positive example heterogeneous label sample and the negative example heterogeneous label sample, respectively; The aforementioned model training unit is used to input the aforementioned heterogeneous data label sample group into the power industry word vector model to be trained for iterative training until the loss function converges, thereby obtaining the trained preset power industry word vector model. In each iteration of training, the first predicted semantic similarity between the current heterogeneous label sample and the corresponding positive heterogeneous label sample, and the second predicted semantic similarity between the current heterogeneous label sample and the corresponding negative heterogeneous label sample are calculated based on the current heterogeneous data label sample group. The current loss function is calculated based on the first predicted semantic similarity, the second predicted semantic similarity, and the corresponding real labels. If the current loss function converges, the current power industry word vector model is used as the aforementioned preset power industry word vector model; otherwise, the model parameters in the current power industry word vector model are adjusted, and training continues.

[0075] It should be noted that the device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without creative effort. The above schematic diagram is merely an example of an abnormal device identification device based on a unified data base modeling of power equipment, and does not constitute a limitation on an abnormal device identification device based on a unified data base modeling of power equipment. It may include more or fewer components than illustrated, or combine certain components, or use different components.

[0076] Based on the above method embodiments, the present invention provides corresponding terminal device embodiments.

[0077] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the abnormal device identification method based on the unified data base modeling of power equipment described in any embodiment of the present invention.

[0078] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the device. The aforementioned terminal devices may be computing devices such as desktop computers, laptops, handheld computers, and cloud servers. These devices may include, but are not limited to, processors and memory. The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. This processor is the control center of the device, connecting various parts of the device via various interfaces and lines. The aforementioned memory can be used to store the aforementioned computer programs and / or modules. The aforementioned processor implements various functions of the aforementioned device by running or executing the computer programs and / or modules stored in the aforementioned memory, and by calling data stored in the memory. The aforementioned memory may mainly include a program storage area and a data storage area, wherein the program storage area may store the operating system, at least one application program required for a function, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0079] Based on the above method embodiments, the present invention provides corresponding storage medium embodiments.

[0080] Another embodiment of the present invention provides a storage medium including a stored computer program, wherein, when the computer program is running, it controls the device where the storage medium is located to execute the abnormal device identification method based on the unified data base modeling of power equipment described in any embodiment of the present invention.

[0081] In this embodiment, the storage medium is a computer-readable storage medium, and the computer program includes computer program code, which may be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium may include any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0082] The above are preferred embodiments of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for identifying abnormal equipment based on a unified data foundation model for power equipment, characterized in that, include: Obtain the original heterogeneous label data of each power device in the power system, and construct an initial power system mapping model; Based on a pre-defined dictionary tree constructed using electrical engineering terminology, the original heterogeneous label data is matched to obtain the candidate label that is most similar to each of the original heterogeneous label data in the pre-defined dictionary tree. Each candidate label is input into a preset power professional word vector model, so that the preset power professional word vector model calculates the similarity between each candidate label and each preset standard label. The standard label with the highest similarity is selected as the selected label, and the corresponding original heterogeneous label data is replaced according to the selected label to obtain the final power system mapping model. Based on the final power system mapping model, anomalies are identified in each power device.

2. The abnormal equipment identification method based on a unified data base modeling system for power equipment as described in claim 1, characterized in that, The training of the preset power industry-specific word vector model includes: Obtain several heterogeneous data label sample groups with real labels; wherein, each group of heterogeneous data label samples includes a heterogeneous label sample, a positive example heterogeneous label sample whose real semantic similarity with the heterogeneous label sample is not less than a preset similarity threshold, and a negative example heterogeneous label sample whose real semantic similarity with the heterogeneous label sample is less than the preset similarity threshold; the real label is used to represent the real similarity between the heterogeneous label sample and the positive example heterogeneous label sample and the negative example heterogeneous label sample, respectively; The heterogeneous data label sample group is input into the power professional word vector model to be trained for iterative training until the loss function converges, and the trained preset power professional word vector model is obtained. In each iteration of training, the first predicted semantic similarity between the current heterogeneous label sample and the corresponding positive heterogeneous label sample, and the second predicted semantic similarity between the current heterogeneous label sample and the corresponding negative heterogeneous label sample are calculated based on the current heterogeneous data label sample group. The current loss function is calculated based on the first predicted semantic similarity, the second predicted semantic similarity, and the corresponding real labels. If the current loss function converges, the current power industry word vector model is used as the preset power industry word vector model; otherwise, the model parameters in the current power industry word vector model are adjusted, and training continues.

3. The abnormal equipment identification method based on a unified data base modeling system for power equipment according to claim 2, characterized in that, The process of identifying anomalies in each power device based on the final power system mapping model includes: Using each power device in the final power system mapping model as a node, the topological connection relationship between each power device as an edge, and the selected label and the original heterogeneous data label as node information, a device association knowledge graph is constructed. Obtain the power time-series data of each power device within a preset time period; A three-dimensional feature matrix is ​​constructed based on preset time periods, power time-series data, and the topological location of the corresponding power equipment in the equipment association knowledge graph. The three-dimensional feature matrix is ​​input into a preset anomaly identification model so that the preset anomaly identification model can obtain the anomaly score of each power device based on the three-dimensional feature matrix; For each power device, if the abnormal score is greater than a preset score threshold, it is preliminarily determined that the power device has an abnormality; otherwise, it is preliminarily determined that the power device does not have an abnormality.

4. The abnormal equipment identification method based on a unified data base modeling of power equipment according to claim 3, characterized in that, The step of constructing a three-dimensional feature matrix based on a preset time period, power time-series data, and the topological location of the corresponding power equipment in the final power system mapping model includes: Perform a Fourier transform on the power time-series data to obtain the power data spectrum; The amplitude of the preset harmonic order in the power data spectrum is extracted to obtain a frequency feature vector; Based on the final power system mapping model, the associated power devices are identified, and the phase difference of the power time series data of the associated power devices is calculated. The three-dimensional feature matrix is ​​constructed based on the preset time period, frequency feature vector, and phase difference of all power equipment.

5. The abnormal equipment identification method based on a unified data base modeling system for power equipment according to claim 4, characterized in that, After initially determining that the power equipment is malfunctioning, the process also includes: From the device association knowledge graph, the associated device information of the associated devices associated with the power equipment is extracted; Based on a preset consistency rule constructed using power knowledge, the associated equipment information is verified for faults to obtain the final anomaly identification result of the power equipment.

6. The abnormal equipment identification method based on a unified data base modeling system for power equipment according to claim 5, characterized in that, Also includes: Several anomaly handling strategies were developed to obtain the final anomaly identification results for power equipment with anomalies. Construct a power system simulation model based on the power system mapping model; The anomaly handling strategy is executed in the power system simulation model, and several power system operation data are obtained after execution. Security verification is performed based on the power system operation data. The anomaly handling strategy that passes the security verification is taken as the target anomaly handling strategy and is executed in the power system.

7. An abnormal equipment identification device based on a unified data foundation model for power equipment, characterized in that, include: The system includes an initial mapping model construction module, a label matching module, a mapping model optimization module, and an anomaly detection module. The initial mapping model construction module is used to acquire each power device in the power system, as well as the original heterogeneous label data of the power device, and construct an initial power system mapping model. The tag matching module is used to match the original heterogeneous tag data according to a preset dictionary tree constructed based on power industry terminology, and obtain the candidate tag that is most similar to each original heterogeneous tag data in the preset dictionary tree. The mapping model optimization module is used to input each candidate label into a preset power professional word vector model, so that the preset power professional word vector model calculates the similarity between each candidate label and each preset standard label, takes the standard label with the highest similarity as the selected label, and replaces the corresponding original heterogeneous label data according to the selected label to obtain the final power system mapping model. The anomaly identification module is used to identify anomalies in each power device based on the final power system mapping model.

8. The abnormal equipment identification device based on a unified data base modeling system for power equipment according to claim 7, characterized in that, The mapping model optimization module includes: Sample data acquisition unit and model training unit; The sample data acquisition unit is used to acquire several heterogeneous data label sample groups with real labels; wherein, each group of heterogeneous data label samples includes heterogeneous label samples, positive heterogeneous label samples whose real semantic similarity with the heterogeneous label samples is not less than a preset similarity threshold, and negative heterogeneous label samples whose real semantic similarity with the heterogeneous label samples is less than the preset similarity threshold; the real label is used to represent the real similarity between the heterogeneous label sample and the positive heterogeneous label samples and the negative heterogeneous label samples, respectively; The model training unit is used to input the heterogeneous data label sample group into the power professional word vector model to be trained for iterative training until the loss function converges, and to obtain the trained preset power professional word vector model. In each iteration of training, the first predicted semantic similarity between the current heterogeneous label sample and the corresponding positive heterogeneous label sample, and the second predicted semantic similarity between the current heterogeneous label sample and the corresponding negative heterogeneous label sample are calculated based on the current heterogeneous data label sample group. The current loss function is calculated based on the first predicted semantic similarity, the second predicted semantic similarity, and the corresponding real labels. If the current loss function converges, the current power industry word vector model is used as the preset power industry word vector model; otherwise, the model parameters in the current power industry word vector model are adjusted, and training continues.

9. A terminal device, characterized in that, The device includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement an abnormal device identification method based on a unified data base model for power equipment as described in any one of claims 1 to 6.

10. A storage medium, characterized in that, The storage medium includes a stored computer program, wherein, when the computer program is running, it controls the device where the storage medium is located to execute an abnormal device identification method based on a unified data base model for power equipment as described in any one of claims 1 to 6.

Images