Transformer substation automation equipment fault diagnosis method and system based on event knowledge graph

Abstract

The invention discloses a transformer substation automation equipment fault diagnosis method and system based on an event knowledge graph, and the method comprises the steps: building a final event knowledge graph of an automation equipment fault, and taking a signal set related to each type of fault of each piece of equipment as a fault signal sub-graph, fault signal data involved in the fault signal sub-atlas are collected, and a fault signal graph data set is obtained; dividing the fault signal diagram data set into a training set, a verification set and a test set, inputting the training set and the verification set into the space-time diagram convolutional neural network basic model, and performing training and verification on the automatic equipment fault diagnosis model to obtain the automatic equipment fault diagnosis model. And the automation equipment fault diagnosis model classifies the fault signal sub-atlas in the test set, and outputs fault equipment and fault types. According to the invention, the accuracy of fault diagnosis is improved, fault equipment can be accurately positioned, and a fault disposal suggestion is provided.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a method and system for diagnosing faults of substation automation equipment based on an event knowledge graph. Background Art

[0002] With the development of fault diagnosis technology for substation automation equipment, some methods use fault signature tables for fault diagnosis, while others use knowledge graphs for fault reasoning. Existing methods can be summarized as follows: They acquire equipment operating data, utilize pre-set state diagnosis models and power knowledge graphs for diagnosis, and output the results to the power dispatch system. Real-time monitoring of fault alarm data builds a feature set, which is then used to locate the fault using pre-trained models and provide feedback on the fault location. Fault information databases and system data are used to identify and analyze the fault type and location. Operating data is acquired, classified, and integrated, and a fault detection model is constructed to quickly and accurately locate the fault node.

[0003] The existing methods have the following problems:

[0004] (1) The existing automated equipment fault diagnosis method that is not based on knowledge graphs is ineffective. It only uses the status signal data at a certain moment, organizes the fault template first, and then combines it with the threshold judgment method to determine whether the equipment signal at a certain moment has a certain fault. This judgment method is fixed and single and cannot accurately judge the fault situation.

[0005] (2) Existing methods of utilizing knowledge graphs use multiple data sources, such as equipment ledgers, SCD (Substation Configuration Description), and PMS (Plant Management System) defect records, to construct conventional knowledge graphs for automation equipment. However, there is a lack of effective entity alignment methods. Alignment is performed by manually defining similarity sets or by simple text similarity, resulting in poor alignment effects and failing to effectively resolve the problem of one entity corresponding to multiple names.

[0006] (3) Conventional knowledge graphs are currently used for automated equipment fault diagnosis. However, the information constructed in conventional knowledge graphs is not comprehensive enough, the fault information is poorly correlated, and no correlation graph is constructed around specific fault information. As a result, the fault result information is inaccurately inferred, and accurate fault and fault handling suggestions cannot be given, which cannot effectively assist staff in troubleshooting. Summary of the Invention

[0007] In view of this, the present invention provides a substation automation equipment fault diagnosis method and system based on event knowledge graph.

[0008] The present invention discloses a method for fault diagnosis of substation automation equipment based on event knowledge graph, which includes:

[0009] Establish the final event knowledge graph of automation equipment failure;

[0010] Take the signal set related to each type of fault of each device in the final event knowledge graph of automation equipment fault as the fault signal sub-graph, and collect the fault signal data involved in the fault signal sub-graph to obtain the fault signal graph dataset;

[0011] The fault signal graph dataset is divided into training set, validation set and test set. The training set and validation set are input into the spatiotemporal graph convolutional neural network basic model to train and verify the automated equipment fault diagnosis model. The automated equipment fault diagnosis model is obtained. The automated equipment fault diagnosis model classifies the fault signal sub-graphs in the test set and outputs the faulty device and fault type.

[0012] Furthermore, the establishment of the final event knowledge graph of the automation equipment failure includes:

[0013] By labeling the automation equipment field corpus data, mixing it with general corpus data in a preset ratio, and training the ALBERT basic model multiple times, we obtain a feature extraction model that can extract high-dimensional features of the automation equipment field corpus data.

[0014] For automation equipment failure events, we extract entity, relationship, and attribute triplets related to the failure event from the failure cases, ledgers, SCD, and PMS defect record data of the automation equipment to build a primary event knowledge graph. Entity types include faulty equipment, fault type, fault level, fault signal, and fault handling suggestions.

[0015] Using the trained feature extraction model, the entities in the primary event knowledge graph are aligned, and entities with different text descriptions but the same semantics are inferred and merged. These entities are modified into the same text description, and the entities are merged to complete the entity alignment and obtain the final event knowledge graph of automation equipment failure.

[0016] Furthermore, by labeling the automation equipment professional corpus data, mixing the general corpus data in a preset ratio, and performing multiple rounds of training on the ALBERT basic model, a feature extraction model capable of extracting high-dimensional features of the automation equipment field corpus data is obtained, including:

[0017] By crawling web pages, we collect and annotate corpus data in the field of automation equipment, and then mix the annotated corpus data in the field of automation equipment with corpus data in the general field according to a preset ratio to obtain mixed corpus data of automation equipment. This data is input into the ALBERT basic model for multiple rounds of training. The ALBERT basic model converts the mixed corpus data into word embedding vectors, and then uses the self-attention mechanism to extract and represent features of the word embedding vectors. After training, a feature extraction model that can extract high-dimensional features of the mixed corpus data is obtained; each word embedding vector consists of word embedding, position embedding and segment embedding.

[0018] Furthermore, the trained feature extraction model is used to align entities in the primary event knowledge graph, infer and merge entities with different text descriptions but representing the same semantics, modify these entities to the same text description, merge the entities, complete entity alignment, and obtain the final event knowledge graph of automation equipment failure, including:

[0019] Using the trained feature extraction model, the entities in the primary event knowledge graph are aligned. By converting all entities of the same type into word embedding vectors, high-dimensional features are extracted. The cosine similarity method is used to calculate the similarity of high-dimensional features, and a similarity threshold is set. When the calculated similarity exceeds the threshold, it is determined that the entities have different text descriptions but represent the same semantics. These entities are modified to the same text description, and the entities are merged to complete the entity alignment, thus obtaining the final simplified event knowledge graph of automation equipment failure.

[0020] Furthermore, the signal set related to each type of fault of each device in the final event knowledge graph of automation equipment fault is taken as a fault signal sub-graph, and the fault signal data involved in the fault signal sub-graph is collected to obtain a fault signal graph dataset, including:

[0021] The signal set related to each type of fault of each device in the final event knowledge graph is taken as the fault signal sub-graph, and the fault signal data involved in the fault signal sub-graph is collected. It is necessary to collect the signal timing data generated within a period of time under the normal state and fault state of the automation equipment, and annotate the fault signal sub-graph data according to the fault device and fault type to which the fault signal sub-graph belongs as a label to obtain the fault signal graph dataset.

[0022] Furthermore, the training set and the validation set are input into the spatiotemporal graph convolutional neural network basic model to train and validate the automated equipment fault diagnosis model, thereby obtaining the automated equipment fault diagnosis model, including:

[0023] The training set and validation set are input into the basic model of the spatiotemporal graph convolutional neural network to train and verify the automated equipment fault diagnosis model. The spatiotemporal graph convolutional neural network is divided into a spatial feature extraction module and a temporal feature extraction module. The features extracted by the graph convolutional neural network and the gated convolutional neural network are fused, and the fault signal sub-graphs are classified according to the fused features to obtain the automated equipment fault diagnosis model.

[0024] Furthermore, the spatial feature extraction module uses a graph convolutional neural network to extract the full-graph features of the fault signal sub-graph data and learn the spatial feature distribution between fault signals of different categories; the temporal feature extraction module uses a gated convolutional neural network to extract the temporal feature distribution of the fault signal.

[0025] Furthermore, after outputting the faulty device and the fault type, the method further includes:

[0026] Based on the final event knowledge graph, the fault results of the automation equipment fault diagnosis model are completed.

[0027] Furthermore, the fault diagnosis model of the automated equipment is supplemented with fault results based on the final event knowledge graph, including:

[0028] The fault diagnosis model of automated equipment is used to infer fault results. The input of the fault diagnosis model of automated equipment is the fault signal timing data generated by the automated equipment in the final event knowledge graph, and the output is a complete fault diagnosis result to achieve fault diagnosis. The complete fault diagnosis result is obtained by combining the final event knowledge graph and reasoning based on the faulty equipment and fault type. Among them, the complete fault diagnosis result includes the faulty equipment, fault type, fault level, fault-related equipment, fault judgment signal, and fault handling suggestions.

[0029] The present invention also discloses a substation automation equipment fault diagnosis system based on an event knowledge graph, which implements any of the above-mentioned substation automation equipment fault diagnosis methods based on an event knowledge graph, and includes:

[0030] Graph construction module, used to build the final event knowledge graph of automation equipment failure;

[0031] The acquisition module is used to take the signal set related to each type of fault of each device in the final event knowledge graph of automation equipment faults as a fault signal sub-graph, and collect the fault signal data involved in the fault signal sub-graph to obtain a fault signal graph dataset;

[0032] The fault diagnosis module is used to divide the fault signal graph dataset into a training set, a validation set, and a test set, and input the training set and validation set into the spatiotemporal graph convolutional neural network basic model to train and verify the automation equipment fault diagnosis model to obtain the automation equipment fault diagnosis model. The automation equipment fault diagnosis model classifies the fault signal sub-graphs in the test set and outputs the faulty device and fault type.

[0033] Due to the adoption of the above technical solution, the present invention has the following advantages:

[0034] 1. The final event knowledge graph constructed by this invention is more accurate and clearer, effectively solving the problem of different descriptions for the same entity during the graph construction process. This effect comes from the entity alignment of the automation equipment event knowledge graph based on the Albert-T text understanding model.

[0035] 2. This invention utilizes device-generated signal time series data and fault signal sub-graphs for model training, improving fault diagnosis accuracy, accurately locating faulty equipment, and providing fault resolution recommendations. During the experiment, 2,000 sets of data were collected from 30 automation devices in three substation categories (measurement and control devices, time synchronization devices, and telecontrol devices), covering 10 types of faults. Training and testing were performed (training set / validation set / test set: 1,400 / 400 / 200 sets). The test results showed a 97.5% accuracy rate in diagnosing faulty equipment and fault type. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments described in the embodiments of the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0037] Figure 1 The present invention is a flowchart of a method for diagnosing substation automation equipment faults based on an event knowledge graph according to an embodiment of the present invention. DETAILED DESCRIPTION

[0038] The present invention will be further described with reference to the accompanying drawings and embodiments. The embodiments described are only a part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by those skilled in the art should fall within the scope of protection of the embodiments of the present invention.

[0039] See also Figure 1 The present invention provides an embodiment of a method for diagnosing faults of substation automation equipment based on an event knowledge graph, which comprises the following steps:

[0040] S1: Establish the final event knowledge graph of automation equipment failure;

[0041] In one embodiment, corpus data in the field of automation equipment is collected and annotated by crawling web pages, and keywords for automation equipment are annotated. It can be mixed with corpus data in the general field at a ratio of 7:3 and input into the ALBERT (pre-trained model) basic model for multiple rounds of training. ALBERT converts the input corpus data into word embedding vectors, and each word embedding vector consists of three parts: word embedding, position embedding, and segment embedding. The model converts corpus data in the field of automation equipment into word embedding vectors, and then uses the self-attention mechanism feature to extract and represent the word embedding vectors. After training, a feature extraction model that can extract high-dimensional features of corpus data in the field of automation equipment is obtained.

[0042] By crawling web pages, we collect and annotate corpus data in the field of automation equipment, and mix the annotated corpus data in the field of automation equipment with the corpus data in the general field according to a preset ratio (for example, 7:3) to obtain the mixed corpus data of automation equipment and input it into the ALBERT (pre-trained model) basic model for multiple rounds of training. The ALBERT basic model converts the mixed corpus data into word embedding vectors, and then uses the self-attention mechanism to extract and represent features of the word embedding vectors. After training, a feature extraction model that can extract high-dimensional features of the mixed corpus data is obtained; each word embedding vector consists of word embedding, position embedding and segment embedding.

[0043] Corpus data in the field of automation equipment includes text such as words and sentences, which can be annotated using keyword annotation.

[0044] For automation equipment failure events, we extract entity, relationship, and attribute triplets related to the failure events from automation equipment failure cases, ledgers, SCD, and PMS defect record data to construct a primary event knowledge graph (including entities such as fault equipment, fault type, fault level, fault signal, fault-related equipment, and fault handling suggestions).

[0045] Using the trained automation equipment feature extraction model, the entities in the primary event knowledge graph are aligned. By converting all entities of the same type into word embedding vectors, high-dimensional features are extracted. The cosine similarity method is used to calculate the similarity of high-dimensional features, and a similarity threshold is set. When the calculated similarity exceeds the threshold, it is judged that the entities have different text descriptions but represent the same semantics. These entities are modified to the same text description, and the entities are merged to complete the entity alignment, thus obtaining the final simplified event knowledge graph of automation equipment failure.

[0046] S2: Take the signal set related to each type of fault of each device in the final event knowledge graph of automation equipment faults as the fault signal sub-graph, and collect the fault signal data involved in the fault signal sub-graph to obtain the fault signal graph dataset;

[0047] In one embodiment, the signal set related to each type of fault of each device in the final event knowledge graph is taken as the fault signal sub-graph, and the fault signal data involved in the fault signal sub-graph is collected. It is necessary to collect the signal timing data generated within a period of time under the normal state and fault state of the automation equipment, and annotate the fault signal sub-graph data according to the fault device and fault type to which the fault signal sub-graph belongs as a label to obtain a fault signal graph dataset.

[0048] S3: Divide the fault signal graph dataset into a training set, a validation set, and a test set. Input the training set and validation set into the spatiotemporal graph convolutional neural network basic model to train and verify the automated equipment fault diagnosis model. The automated equipment fault diagnosis model is obtained. The automated equipment fault diagnosis model classifies the fault signal sub-graphs in the test set and outputs the faulty device and fault type.

[0049] In one embodiment, based on a labeled fault signal graph dataset, it can be divided into a training set, a validation set, and a test set in a ratio of 7:2:1. The training set and validation set are then input into a spatiotemporal graph convolutional neural network basic model to train an automated equipment fault diagnosis model. The spatiotemporal graph convolutional neural network is divided into a spatial feature extraction module and a temporal feature extraction module. The spatial feature extraction module uses a graph convolutional neural network to extract the full-graph features of the fault signal sub-graph data and learn the spatial feature distribution between different types of fault signals. The temporal feature extraction module uses a gated convolutional neural network (Gated CNNs) to extract the temporal feature distribution of the fault signal. The features extracted by the graph convolutional neural network and the gated convolutional neural network are then added and fused. The fault signal sub-graphs are classified based on the fused features to obtain an automated equipment fault diagnosis model. This model can classify the fault signal sub-graphs in the test set and output the faulty device and fault type.

[0050] In one embodiment, the fault diagnosis model for automated equipment is supplemented with fault results based on the final event knowledge graph. The model input is the time series data of fault signals generated by automated equipment, and the model output is the faulty device and fault type. The final event knowledge graph is then used to query the fault-related graph subgraphs to complete the fault diagnosis results (including the faulty device, fault type, fault level, fault-related devices, fault determination signal, and fault handling recommendations), thus achieving fault diagnosis.

[0051] The present invention also provides an embodiment of a substation automation equipment fault diagnosis system based on an event knowledge graph, which implements the substation automation equipment fault diagnosis method based on an event knowledge graph described in the above embodiment, and includes:

[0052] Graph construction module, used to build the final event knowledge graph of automation equipment failure;

[0053] The acquisition module is used to take the signal set related to each type of fault of each device in the final event knowledge graph of automation equipment faults as a fault signal sub-graph, and collect the fault signal data involved in the fault signal sub-graph to obtain a fault signal graph dataset;

[0054] The fault diagnosis module is used to divide the fault signal graph dataset into a training set, a validation set, and a test set, and input the training set and validation set into the spatiotemporal graph convolutional neural network basic model to train and verify the automation equipment fault diagnosis model to obtain the automation equipment fault diagnosis model. The automation equipment fault diagnosis model classifies the fault signal sub-graphs in the test set and outputs the faulty device and fault type.

[0055] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A method for fault diagnosis of substation automation equipment based on event knowledge graph, characterized in that: include: Establish the final event knowledge graph of automation equipment failure; Take the signal set related to each type of fault of each device in the final event knowledge graph of automation equipment fault as the fault signal sub-graph, and collect the fault signal data involved in the fault signal sub-graph to obtain the fault signal graph data set; The fault signal graph dataset is divided into training set, validation set and test set. The training set and validation set are input into the spatiotemporal graph convolutional neural network basic model to train and verify the automation equipment fault diagnosis model. The automation equipment fault diagnosis model is obtained. The automation equipment fault diagnosis model classifies the fault signal sub-graphs in the test set and outputs the faulty equipment and fault type.

2. The method for fault diagnosis of substation automation equipment based on event knowledge graph according to claim 1 is characterized in that: The final event knowledge graph of the automation equipment failure is established, including: By annotating the automation equipment field corpus data, mixing the general corpus data in a preset ratio, and training the ALBERT basic model for multiple rounds, a feature extraction model that can extract high-dimensional features of the automation equipment field corpus data is obtained; For automation equipment failure events, extract entities, relationships, and attribute triplets related to the failure events from the failure cases, ledgers, SCD, and PMS defect record data of the automation equipment to build a primary event knowledge graph; the types of entities include faulty equipment, fault type, fault level, fault signal, and fault handling suggestions; Using the trained feature extraction model, the entities in the primary event knowledge graph are aligned, and entities with different text descriptions but representing the same semantics are inferred and merged. These entities are modified into the same text description, and the entities are merged to complete the entity alignment and obtain the final event knowledge graph of automation equipment failure.

3. The method for fault diagnosis of substation automation equipment based on event knowledge graph according to claim 2 is characterized in that: The method obtains a feature extraction model that can extract high-dimensional features of corpus data in the field of automation equipment by annotating professional corpus data of automation equipment, mixing general corpus data according to a preset ratio, and training the ALBERT basic model for multiple rounds, including: The corpus data in the field of automation equipment is collected and annotated by crawling web pages, and the annotated corpus data in the field of automation equipment is mixed with the corpus data in the general field according to a preset ratio to obtain the mixed corpus data of automation equipment and input it into the ALBERT basic model for multiple rounds of training. The ALBERT basic model converts the mixed corpus data into word embedding vectors, and then uses the self-attention mechanism to extract and represent features of the word embedding vectors. After training, a feature extraction model that can extract high-dimensional features of mixed corpus data is obtained; each word embedding vector is composed of word embedding, position embedding and segment embedding.

4. The method for fault diagnosis of substation automation equipment based on event knowledge graph according to claim 2 is characterized in that: The trained feature extraction model is used to align entities in the primary event knowledge graph, infer and merge entities with different text descriptions but representing the same semantics, modify these entities to the same text description, merge the entities, complete entity alignment, and obtain the final event knowledge graph of automation equipment failure, including: Using the trained feature extraction model, the entities in the primary event knowledge graph are aligned. By converting all entities of the same type into word embedding vectors, high-dimensional features are extracted. The cosine similarity method is used to calculate the similarity of high-dimensional features, and a similarity threshold is set. When the calculated similarity exceeds the threshold, it is judged that the entities have different text descriptions but represent the same semantics. These entities are modified to the same text description, and the entities are merged to complete the entity alignment, thus obtaining a simplified final event knowledge graph of automation equipment failure.

5. The method for fault diagnosis of substation automation equipment based on event knowledge graph according to claim 1 is characterized in that: The method takes a signal set related to each type of fault of each device in the final event knowledge graph of automation equipment fault as a fault signal sub-graph, and collects fault signal data involved in the fault signal sub-graph to obtain a fault signal graph data set, including: Take the signal set related to each type of fault of each device in the final event knowledge graph as the fault signal sub-graph, and collect the fault signal data involved in the fault signal sub-graph. It is necessary to collect the signal timing data generated within a period of time under the normal state and fault state of the automation equipment, and annotate the fault signal sub-graph data according to the fault device and fault type to which the fault signal sub-graph belongs as labels to obtain the fault signal graph data set.

6. The method for fault diagnosis of substation automation equipment based on event knowledge graph according to claim 1 is characterized in that: The training set and the validation set are input into the spatiotemporal graph convolutional neural network basic model to train and validate the automation equipment fault diagnosis model, thereby obtaining the automation equipment fault diagnosis model, including: The training set and validation set are input into the basic model of the spatiotemporal graph convolutional neural network to train and verify the automated equipment fault diagnosis model. The spatiotemporal graph convolutional neural network is divided into a spatial feature extraction module and a temporal feature extraction module. The features extracted by the graph convolutional neural network and the gated convolutional neural network are fused, and the fault signal sub-spectra are classified according to the fused features to obtain the automated equipment fault diagnosis model.

7. The method for fault diagnosis of substation automation equipment based on event knowledge graph according to claim 6 is characterized in that: The spatial feature extraction module uses a graph convolutional neural network to extract the full-graph features of the fault signal sub-graph data and learn the spatial feature distribution between fault signals of different categories. The temporal feature extraction module uses a gated convolutional neural network to extract the temporal feature distribution of the fault signal.

8. The method for fault diagnosis of substation automation equipment based on event knowledge graph according to claim 1 is characterized in that: After outputting the fault device and the fault type, it also includes: Based on the final event knowledge graph, the fault results of the automation equipment fault diagnosis model are completed.

9. The method for fault diagnosis of substation automation equipment based on event knowledge graph according to claim 8 is characterized in that: The method of completing the fault result of the automation equipment fault diagnosis model based on the final event knowledge graph includes: The fault diagnosis model of automated equipment is used to infer fault results. The input of the fault diagnosis model of automated equipment is the fault signal timing data generated by the automated equipment in the final event knowledge graph, and the output is a complete fault diagnosis result to achieve fault diagnosis. The complete fault diagnosis result is obtained by combining the final event knowledge graph and reasoning based on the faulty equipment and fault type. Among them, the complete fault diagnosis result includes the faulty equipment, fault type, fault level, fault-related equipment, fault judgment signal, and fault handling suggestions.

10. A substation automation equipment fault diagnosis system based on event knowledge graph, implementing the substation automation equipment fault diagnosis method based on event knowledge graph according to any one of claims 1 to 9, characterized in that: include: A graph building module, used to build the final event knowledge graph of automation equipment failures; The acquisition module is used to take the signal set related to each type of fault of each device in the final event knowledge graph of automation equipment fault as the fault signal sub-graph, and collect the fault signal data involved in the fault signal sub-graph to obtain the fault signal graph data set; The fault diagnosis module is used to divide the fault signal graph data set into a training set, a validation set and a test set, input the training set and the validation set into the spatiotemporal graph convolutional neural network basic model, train and verify the automation equipment fault diagnosis model, and obtain the automation equipment fault diagnosis model. The automation equipment fault diagnosis model classifies the fault signal sub-graphs in the test set and outputs the faulty equipment and fault type.

Images