Power equipment text data processing method and system

Abstract

The invention discloses a power equipment text data processing method, which comprises the following steps of: 100, performing entity identification on non-structural text data in power equipment text data to extract entity information in the non-structural text data; 200, performing synonym elimination and entity disambiguation processing on the entity information to obtain cleaned entity information; 300, extracting the relation between the cleaned entity information pairs to combine the entity information pairs into a triple form of'first entity-relation-second entity '; and 400, storing the entity information in the triple form. Correspondingly, the invention further discloses a power equipment text data processing system. According to the method, the knowledge in the unstructured text of the power equipment can be efficiently processed and stored, so that the operation and maintenance decision of operation and maintenance personnel of the power equipment is assisted.

Description

Technical Field

[0001] The present invention relates to a method and system for processing equipment text, and in particular to a method and system for processing and storing knowledge in the field of electric power equipment. Background Art

[0002] A large amount of equipment-related data is generated during the operation and maintenance of power equipment. In addition to recording and saving images, graphs, and numerical data related to equipment reliability, inspectors also form text information to record the status of the equipment. Since these accumulated insulation status text data do not have a fixed structure format, it is difficult to process and store them in a unified manner. In the existing production management system, usually only the images, graphs and other data results obtained by the inspection are stored, which also results in low data utilization and the loss of a large amount of original records and information, bringing potential risks to the safe and stable operation of the power grid.

[0003] In the processing of text data, traditional relational databases use association tables to store the relationships between entities. For data with more complex relationships, data storage can only be achieved by adding a large number of relationship connection tables. In recent years, due to the defects of relational databases in processing complex relationships, such as slow speed and high resource consumption, graph database technology has gradually received attention.

[0004] At present, graph databases are mostly related to power equipment in applications of high-security equipment reliability, mainly focusing on power grid topology management, and there are also a small number of studies on equipment asset management. At present, when graph databases are applied to power grid topology storage, they mainly rely on their own graph attributes to manage the reliability of power equipment more clearly. When managing the assets of power equipment, the advantages of its visual display and efficient processing of asset relationships are mainly used to achieve the aggregation of equipment-related information. As for the storage and application of power equipment reliability text data, there is currently no precedent for graph databases, and there is also a lack of methods and platforms to provide system processing and storage for text data, and there is no unified format specification requirement for text records. Summary of the invention

[0005] One of the purposes of the present invention is to provide a method for processing text data of electric power equipment, which can efficiently process and store the knowledge of electric power equipment, and thereby assist the operation and maintenance decision-making of electric power equipment operation and maintenance personnel.

[0006] According to the above invention purpose, the present invention proposes a method for processing text data of electric power equipment, which comprises the steps of:

[0007] 100: performing entity recognition on unstructured text data in the power equipment text data to extract entity information therein;

[0008] 200: performing synonym elimination and entity disambiguation processing on the entity information to obtain cleansed entity information;

[0009] 300: extracting the relationship between the cleaned entity information pairs to combine the entity information pairs into a triple form of “first entity-relationship-second entity”;

[0010] 400: Store the entity information in triple form.

[0011] Furthermore, in step 100 of the power equipment text data processing method described in the present invention, entity recognition is performed on the unstructured text data in the power equipment text data to extract the entity information therein, including: text segmentation is performed on the unstructured text data based on a hidden Markov model to extract the entity information therein.

[0012] Furthermore, in step 200 of the method for processing electric power equipment text data of the present invention, removing synonyms from the entity information includes the following steps:

[0013] Use the word embedding model to convert each word in the entity information into a corresponding word vector;

[0014] Calculate the cosine similarity of the word vectors between each entity information, and identify synonyms based on the cosine similarity;

[0015] Eliminate synonyms.

[0016] Furthermore, in the electric power equipment text data processing method described in the present invention, step 200 also includes: forming a synonym database based on the identified synonyms.

[0017] Furthermore, in step 300 of the method for processing text data of electric power equipment according to the present invention, extracting the relationship between the cleaned entity information pairs includes the steps of:

[0018] The word embedding model is used to convert the entity information pairs whose relationships need to be extracted into the first entity vector and the second entity vector respectively;

[0019] Convert the text corresponding to the entity information pair into a text vector sequence;

[0020] Using a deep self-attention neural network to perform information analysis on the text vector sequence, outputting a text vector sequence with high information density, and using maximum pooling to convert the text vector sequence with high information density into a one-dimensional text vector;

[0021] Concatenate the first entity vector, the second entity vector and the one-dimensional text vector to obtain a feature vector;

[0022] A fully connected neural network is used to classify the feature vectors and identify the relationship between the entity information pairs.

[0023] Another object of the present invention is to provide a power equipment text data processing system that can efficiently process and store power equipment knowledge, thereby assisting power equipment operation and maintenance personnel in making operation and maintenance decisions.

[0024] According to the above invention purpose, the present invention proposes a power equipment text data processing system, which includes:

[0025] An entity extraction module performs entity recognition on unstructured text data in the power equipment text data to extract entity information therein;

[0026] A preprocessing module, which performs synonym elimination and entity disambiguation processing on the entity information to obtain cleaned entity information;

[0027] A relation extraction module, which extracts the relations between the cleaned entity information pairs to combine the entity information pairs into a triple form of "first entity-relationship-second entity";

[0028] A storage module stores the entity information in the triple form.

[0029] Furthermore, in the electric power equipment text data processing system of the present invention, the entity extraction module performs text segmentation on the unstructured text data based on a hidden Markov model to extract entity information therein.

[0030] Furthermore, in the electric power equipment text data processing system of the present invention, the preprocessing module removes synonyms from the entity information, including the following steps:

[0031] Use the word embedding model to convert each word in the entity information into a corresponding word vector;

[0032] Calculate the cosine similarity of the word vectors between each entity information, and identify synonyms based on the cosine similarity;

[0033] Eliminate synonyms.

[0034] Furthermore, in the electric power equipment text data processing system described in the present invention, the preprocessing module also forms a synonym database based on the identified synonyms.

[0035] Furthermore, in the electric power equipment text data processing system of the present invention, the relationship extraction module extracts the relationship between the cleaned entity information pairs, including the steps of:

[0036] The word embedding model is used to convert the entity information pairs whose relationships need to be extracted into the first entity vector and the second entity vector respectively;

[0037] Convert the text corresponding to the entity information pair into a text vector sequence;

[0038] Using a deep self-attention neural network to perform information analysis on the text vector sequence, outputting a text vector sequence with high information density, and using maximum pooling to convert the text vector sequence with high information density into a one-dimensional text vector;

[0039] Concatenate the first entity vector, the second entity vector and the one-dimensional text vector to obtain a feature vector;

[0040] A fully connected neural network is used to classify the feature vectors and identify the relationship between the entity information pairs.

[0041] The power equipment text data processing method and system described in the present invention have the following advantages and beneficial effects:

[0042] The electric power equipment text data processing method and system described in the present invention can efficiently process and store electric power equipment knowledge, especially knowledge in unstructured text, and thereby assist electric power equipment operation and maintenance personnel in making operation and maintenance decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A flowchart of the steps of the method for processing text data of electric power equipment described in the present invention in one implementation manner is shown.

[0044] Figure 2 A schematic diagram showing the relationship between entity information pairs extracted by the power equipment text data processing method described in the present invention in one implementation manner.

[0045] Figure 3 A schematic diagram of the architecture of the electric power equipment text data processing system according to one embodiment of the present invention is shown. DETAILED DESCRIPTION

[0046] The power equipment text data processing method and system described in the present invention will be further explained and illustrated below in conjunction with the accompanying drawings and specific embodiments of the specification. However, such explanation and illustration do not constitute an improper limitation on the technical solution of the present invention.

[0047] In some embodiments of the present invention, the present invention proposes a method for processing text data of electric power equipment, such as Figure 1 As shown, it may include the steps of:

[0048] 100: performing entity recognition on unstructured text data in the power equipment text data to extract entity information therein;

[0049] 200: performing synonym elimination and entity disambiguation processing on the entity information to obtain cleansed entity information;

[0050] 300: extracting the relationship between the cleaned entity information pairs to combine the entity information pairs into a triple form of “first entity-relationship-second entity”;

[0051] 400: Store the entity information in triple form.

[0052] In some more specific embodiments, in step 100, entity recognition is performed on the unstructured text data in the power equipment text data to extract the entity information therein, which may specifically include: performing text segmentation on the unstructured text data based on a hidden Markov model, and importing a common word dictionary, a stop word dictionary, and a power equipment field dictionary to assist in segmentation to extract the entity information therein.

[0053] Since the text entities of power equipment reliability all come from related professional fields, importing domain dictionaries can effectively assist in word segmentation, which is very helpful for improving the accuracy of word segmentation. Text segmentation using hidden Markov models and dictionary assistance can achieve a word segmentation accuracy of more than 86%, greatly reducing the workload of manual text screening.

[0054] In some more specific implementations, the synonym elimination of the entity information in step 200 (i.e. Figure 1 The "coreference resolution" in ) may include the steps of:

[0055] Use the word embedding model to convert each word in the entity information into a corresponding word vector;

[0056] Calculate the cosine similarity of the word vectors between each entity information, and identify synonyms based on the cosine similarity;

[0057] Eliminate synonyms.

[0058] Furthermore, in the method for processing electric power equipment text data of the present invention, step 200 may also include: forming a synonym database based on the identified synonyms.

[0059] Specifically, the word embedding model can be used to set the word vector dimension to 746 dimensions to obtain the word vector corresponding to each word. In text data, entities with the same meaning generally have relatively similar word vector values. Therefore, by calculating the cosine similarity of the word vectors between entities, synonyms in the text data can be obtained. Through further fine-grained screening, synonyms in the text data can be deleted. After deleting synonyms, a synonym database, or a synonym table, can be formed. In subsequent text data processing, the data in the synonym table can be used to directly replace the synonyms in the text, thereby improving the efficiency of synonym deletion.

[0060] After the coreference resolution is completed, the entity information needs to be further disambiguated. Since different substation names may be repeated, the substation detection equipment also needs to be uniquely labeled. In a specific example, the full name identification method can be used to identify and distinguish different equipment of the same type in the same substation when recording the name.

[0061] Co-reference resolution and entity disambiguation avoid the occurrence of redundant entity information and ensure the uniqueness of each entity in the graph, which plays an important role in subsequent entity query and relationship mining.

[0062] In some more specific embodiments, Figure 2 As shown, extracting the relationship between the cleaned entity information pairs in step 300 may include the steps of:

[0063] The word embedding model is used to convert the entity information pairs whose relationships need to be extracted into the first entity vector and the second entity vector respectively;

[0064] Convert the text corresponding to the entity information pair into a text vector sequence;

[0065] Using a deep self-attention neural network to perform information analysis on the text vector sequence, outputting a text vector sequence with high information density, and using maximum pooling to convert the text vector sequence with high information density into a one-dimensional text vector;

[0066] Concatenate the first entity vector, the second entity vector and the one-dimensional text vector to obtain a feature vector;

[0067] A fully connected neural network is used to classify the feature vectors and identify the relationship between the entity information pairs.

[0068] exist Figure 1 In the context of knowledge processing, “knowledge processing” refers to combining entities in the entity library into knowledge triples of “entity-relationship-entity”. Therefore, it is necessary to find the corresponding relationship from the identified entity results.

[0069] like Figure 2 As shown, the word embedding model can be used to convert the entity pairs whose relationships need to be extracted into two entity vectors with a dimension of 746. On the other hand, the text fragment corresponding to this entity pair is converted into a word vector sequence. Subsequently, the text vector sequence is parsed using a deep self-attention neural network to output a text vector sequence with high information density, and the sequence is converted into a one-dimensional text vector using maximum pooling. The two entity vectors are concatenated with the one-dimensional text vector to obtain a feature vector. Finally, the FC neural network (fully connected neural network) is used to classify the feature vector and identify the association relationship between the entity pairs.

[0070] After collecting a large amount of power equipment reliability text data, the present invention can define the relationship values ​​that need to be determined, and can also perform standardized format requirements on the power equipment text data and then store it.

[0071] Table 1 exemplarily shows a normalized storage format of entity knowledge text set by the present invention, which shows the serial number, name and category of the power equipment reliability entity, as well as the format requirements for filling in when recording.

[0072] Table 1.

[0073]

[0074]

[0075] Table 2 exemplarily shows the first entity and second entity serial numbers and relationship categories of the power equipment reliability entity relationship.

[0076] Table 2.

[0077]

[0078] In this way, in future reliability inspections of power equipment, the values ​​of entities and relationships specified in the present invention can be used to record the equipment inspection status in text form, which will facilitate the import of data into the graph database and the completion of data sharing and mining.

[0079] In another embodiment, the present invention further proposes a power equipment text data processing system, which includes:

[0080] An entity extraction module 800 performs entity recognition on the unstructured text data in the electric power equipment text data to extract entity information therein;

[0081] A pre-processing module 802 performs synonym elimination and entity disambiguation processing on the entity information to obtain cleaned entity information;

[0082] A relationship extraction module 804 extracts the relationship between the cleaned entity information pairs to combine the entity information pairs into a triple form of "first entity-relationship-second entity";

[0083] The storage module 806 stores the entity information in the triple form.

[0084] In some more specific implementations, the entity extraction module 800 performs text segmentation on the unstructured text data based on a hidden Markov model to extract entity information therein.

[0085] In some more specific implementations, the pre-processing module 802 may remove synonyms from the entity information by the following steps:

[0086] Use the word embedding model to convert each word in the entity information into a corresponding word vector;

[0087] Calculate the cosine similarity of the word vectors between each entity information, and identify synonyms based on the cosine similarity;

[0088] Eliminate synonyms.

[0089] In some more specific implementations, the pre-processing module 802 further forms a synonym database based on the identified synonyms.

[0090] In some more specific implementations, the relationship extraction module 804 extracts the relationship between the cleansed entity information pairs, which may include the following steps:

[0091] The word embedding model is used to convert the entity information pairs whose relationships need to be extracted into the first entity vector and the second entity vector respectively;

[0092] Convert the text corresponding to the entity information pair into a text vector sequence;

[0093] Using a deep self-attention neural network to perform information analysis on the text vector sequence, outputting a text vector sequence with high information density, and using maximum pooling to convert the text vector sequence with high information density into a one-dimensional text vector;

[0094] Concatenate the first entity vector, the second entity vector and the one-dimensional text vector to obtain a feature vector;

[0095] A fully connected neural network is used to classify the feature vectors and identify the relationship between the entity information pairs.

[0096] It can be seen that the power equipment text data processing method and system described in the present invention can efficiently process and store power equipment knowledge, especially knowledge in unstructured text, and thereby assist power equipment operation and maintenance personnel in making operation and maintenance decisions.

[0097] It should be noted that the prior art in the protection scope of the present invention is not limited to the embodiments given in the present application documents. All prior art that does not contradict the scheme of the present invention, including but not limited to prior patent documents, prior public publications, prior public uses, etc., can be included in the protection scope of the present invention.

[0098] In addition, the combination of the various technical features in this case is not limited to the combination described in the claims of this case or the combination described in the specific embodiments. All technical features recorded in this case can be freely combined or combined in any way unless there is a contradiction between them.

[0099] It should also be noted that the above-listed embodiments are only specific embodiments of the present invention. Obviously, the present invention is not limited to the above embodiments, and similar changes or modifications made therewith can be directly derived or easily associated with the contents disclosed by those skilled in the art from the present invention, and all should belong to the protection scope of the present invention.

Claims

1. A method for processing text data of electric power equipment, characterized in that: Includes steps: 100: performing entity recognition on unstructured text data in the power equipment text data to extract entity information therein; 200: performing synonym elimination and entity disambiguation processing on the entity information to obtain cleansed entity information; 300: extracting the relationship between the cleaned entity information pairs to combine the entity information pairs into a triple form of "first entity-relationship-second entity"; 400: Store the entity information in triple form.

2. The method for processing text data of electric power equipment according to claim 1, characterized in that: In step 100, entity recognition is performed on the unstructured text data in the electric power equipment text data to extract entity information therein, including: performing text segmentation on the unstructured text data based on a hidden Markov model to extract entity information therein.

3. The method for processing text data of electric power equipment according to claim 1, characterized in that: In step 200, performing synonym elimination on the entity information includes the following steps: Use the word embedding model to convert each word in the entity information into a corresponding word vector; Calculate the cosine similarity of the word vectors between each entity information, and identify synonyms based on the cosine similarity; Eliminate synonyms.

4. The method for processing text data of electric power equipment according to claim 1, characterized in that: Step 200 also includes: forming a synonym database based on the identified synonyms.

5. The method for processing text data of electric power equipment according to claim 1, characterized in that: In step 300, extracting the relationship between the cleaned entity information pairs includes the steps of: using a word embedding model to convert the entity information pairs whose relationship needs to be extracted into a first entity vector and a second entity vector respectively; Convert the text corresponding to the entity information pair into a text vector sequence; Using a deep self-attention neural network to perform information analysis on the text vector sequence, outputting a text vector sequence with high information density, and using maximum pooling to convert the text vector sequence with high information density into a one-dimensional text vector; Concatenate the first entity vector, the second entity vector and the one-dimensional text vector to obtain a feature vector; A fully connected neural network is used to classify the feature vectors and identify the relationship between the entity information pairs.

6. A power equipment text data processing system, characterized in that: It includes: An entity extraction module performs entity recognition on unstructured text data in the power equipment text data to extract entity information therein; A preprocessing module, which performs synonym elimination and entity disambiguation processing on the entity information to obtain cleaned entity information; A relation extraction module, which extracts the relations between the cleaned entity information pairs to combine the entity information pairs into a triple form of "first entity-relationship-second entity"; A storage module stores the entity information in the triple form.

7. The electric power equipment text data processing system according to claim 6, characterized in that: The entity extraction module performs text segmentation on the unstructured text data based on a hidden Markov model to extract entity information therein.

8. The electric power equipment text data processing system according to claim 6, characterized in that: The preprocessing module removes synonyms from the entity information, including the following steps: Use the word embedding model to convert each word in the entity information into a corresponding word vector; Calculate the cosine similarity of the word vectors between each entity information, and identify synonyms based on the cosine similarity; Eliminate synonyms.

9. The electric power equipment text data processing system according to claim 6, characterized in that: The pre-processing module also forms a synonym database based on the identified synonyms.

10. The electric power equipment text data processing system according to claim 6, characterized in that: The relationship extraction module extracts the relationship between the cleaned entity information pairs, including the following steps: The word embedding model is used to convert the entity information pairs whose relationships need to be extracted into the first entity vector and the second entity vector respectively; Convert the text corresponding to the entity information pair into a text vector sequence; Using a deep self-attention neural network to perform information analysis on the text vector sequence, outputting a text vector sequence with high information density, and using maximum pooling to convert the text vector sequence with high information density into a one-dimensional text vector; Concatenate the first entity vector, the second entity vector and the one-dimensional text vector to obtain a feature vector; A fully connected neural network is used to classify the feature vectors and identify the relationship between the entity information pairs.

Images