Intelligent search method based on electric power data knowledge graph
By building a power data knowledge graph, the complexity and accuracy of power data search are solved, intelligent search is realized, the efficiency and accuracy of power data acquisition are improved, and the construction of high-elastic power grids and power data automation are supported.
Patent Information
- Application Number
- CN202411084792.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-08
- Publication Date
- 2024-12-06
AI Technical Summary
Traditional power data search methods cannot effectively process power data with huge quantities, scattered sources and diverse formats, resulting in complex searches, low efficiency, low accuracy, and difficult to obtain power data business knowledge.
By building a power data knowledge graph, grab the entities in structured data, extract the entity relationships in the data source, establish a knowledge graph index structure, match user search requests, return accurate and personalized query results, and use intelligent search technology to reduce data acquisition Difficulty.
It improves the accuracy and efficiency of power data search, reduces the labor intensity of power data workers, supports the construction of diversified and integrated high-elastic power grids, improves the degree of power data automation, and enhances the company's competitiveness and economic benefits.
Smart Images

Figure CN119088823A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of information search technology, and in particular to an intelligent search method based on power data knowledge graph. Background Art
[0002] With the rise of new technologies such as mobile Internet and Internet of Things, the construction of smart grids by power grid companies is continuously advancing, and the power data assets owned by power grid companies are growing at an unprecedented rate. Faced with huge amounts of power data with scattered sources and various formats, how to discover new knowledge, create new value, and enhance new capabilities is an urgent problem that power grid companies need to solve. With the continuous improvement of the function construction of power grid enterprise data centers, a large amount of business system power data is connected to the data center, including structured data, unstructured data, and semi-structured data in the enterprise.
[0003] These power data are huge in quantity, scattered in source, and lack correlation between data. However, the traditional data search method in the store can only search for unstructured data such as documents, and cannot fully meet the needs of power data knowledge search when facing complex power systems, resulting in a complex power data search process, low search efficiency and inaccurate search, and difficulty in obtaining power data business knowledge. Faced with huge data resources, power grid companies are unable to quickly and accurately search for data knowledge, identify data relationships, and mine data value. Summary of the invention
[0004] The purpose of the present invention is to overcome the problems in the prior art that the search process for power data business knowledge is complex and the search is inaccurate, and that it is difficult to acquire power data business knowledge. An intelligent search method based on a power data knowledge graph is provided. By searching for related knowledge cards through the constructed knowledge graph, the user's search needs can be understood more deeply, and more accurate and personalized query results can be returned to the user according to the needs. The search process is simpler, the search results are more accurate, and the power data business knowledge can be acquired quickly and accurately.
[0005] In order to achieve the above object, the present invention adopts the following technical solutions: An intelligent search method based on power data knowledge graph, characterized by comprising the following steps: S1: capturing a first entity from the power data structured data, and processing the captured first entity; S2: extract the second entity from the data source, build the relationship between the first entity and the second entity, obtain the entity with a unique identifier, and build a power data knowledge graph; S3: Build a knowledge graph index structure and synchronize the changes in the power data knowledge graph; S4: Obtain user search requests, match user search requests, map matching results to entities and relationships in the knowledge graph, and return search results based on the structure in the knowledge graph.
[0006] The present invention constructs a knowledge graph of electric power data, and conducts intelligent search on this basis. Through intelligent search, the difficulty for users to obtain the required electric power data knowledge is reduced, the feedback accuracy of search knowledge is improved, the efficiency and accuracy of electric power data work is improved, the labor intensity of electric power data workers is reduced, and better support is provided for the construction and implementation of a multi-integrated and highly elastic power grid. At the same time, the degree of automation of electric power data is improved to ensure that the company's various electric power data work is accurate, efficient, and intelligent. Solve the current problems of electric power data from the root, provide the company with advanced, efficient, and intelligent electric power data capabilities, enhance the company's competitiveness in power supply services after the power sales side is liberalized, and bring more economic benefits to power grid companies.
[0007] Preferably, step S4 includes: using a bidirectional maximum matching algorithm to segment the user's search request; classifying the user's search request according to the segmentation result; determining the entity targeted by the user's search request to obtain a triple to be retrieved; forming a search statement to establish a match for the triple to be retrieved, and performing an intelligent search.
[0008] Preferably, the index structure organizes and indexes the collected data using a grid file, and calculates the aggregation value required by the user when creating the index.
[0009] Preferably, the word segmentation of the user's search request includes: dividing the user's search request according to the maximum length of the word, and then matching it from left to right in the preset dictionary; dividing the user's search request according to the maximum length of the word, and then matching it from right to left in the preset dictionary; if the word segmentation results of the two matches are the same, the word segmentation result is returned; if the word segmentation results of the two matches are different, the word segmentation result with a smaller number of word segmentations is returned; if the number of word segmentations is the same, the word segmentation result with a larger word length is returned.
[0010] Preferably, entities are extracted from user search requests based on a fuzzy recognition algorithm, and the type of user search request is used as a relationship to obtain triples to be retrieved; triples matching the triples to be retrieved are searched in the knowledge graph, and the information similarity is calculated; the triples obtained in the knowledge graph and the corresponding similarities are returned to the user as intelligent search results.
[0011] Preferably, the classifying of the user search request according to the word segmentation result includes: performing consciousness recognition on the word segmentation result to obtain the code and purpose of the user search request as the type of the user search request.
[0012] Preferably, the step S1 includes: analyzing the power data, using a shallow semantic model to calculate a word set with a hierarchical relationship for the power data entity, the hierarchical relationship is represented by similarity, and a similar word set of a certain power data entity word vector is obtained; the similar word set is classified, and a semantic real-time extension association related to the word vector is established.
[0013] Preferably, when matching the word segmentation, if the match is successful, the word segmentation is extracted from the search request; if the match fails, the length of the matching word is minus 1 and the word segmentation is performed again until all words in the search request are split.
[0014] Preferably, the index structure is stored in a key-value database and read in a key-based manner.
[0015] Preferably, the similar word sets are classified to include: keywords, primary word associations calculated using a preview model based on the keywords, secondary word associations obtained based on the primary word association set, and tertiary word associations obtained based on the secondary word association set.
[0016] Therefore, the present invention has the following beneficial effects: 1. By constructing a knowledge graph for power data planning business, we can manage business knowledge in a structured manner, combine business needs with data resources, and conduct intelligent search based on the knowledge graph. Through intelligent search, we can reduce the difficulty for users to obtain the required power data knowledge, improve the feedback accuracy of search knowledge, improve the efficiency and accuracy of power data work, reduce the labor intensity of power data workers, and better support the construction and implementation of a diversified, integrated, and highly elastic power grid.
[0017] 2. Through knowledge graphs and intelligent search technology, help users better understand the relationship between various types of knowledge in power data and help users better discover the value of data. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] Figure 1 It is a flowchart of the overall steps of the intelligent search method based on the power data knowledge graph in the present invention.
[0019] Figure 2 Schematic diagram of the index structure in the first embodiment.
[0020] Figure 3 This is a schematic diagram of a search engine model based on the power data knowledge graph in Example 2. DETAILED DESCRIPTION
[0021] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments: Embodiment 1: like Figure 1In the illustrated embodiment, an intelligent search method based on the electric power data knowledge graph can be seen, and its operation process is as follows: Step 1, capture the first entity from the electric power data structured data, and use the ontology library to process and store the captured first entity; Step 2, extract the second entity from the data source, build the relationship between the first entity and the second entity, obtain the entity with a unique identifier, and build the electric power data knowledge graph; Step 3, build the knowledge graph index structure, and synchronously update the changes in the electric power data knowledge graph; Step 4, obtain the user search request, match the user search request, map the matching results to the entities and relationships of the knowledge graph, and return the search results based on the structure in the knowledge graph.
[0022] The knowledge graph is designed to describe various entities or concepts that exist in the real world. Each entity or concept is identified by a globally unique ID (IDentity). The knowledge graph can be viewed as a huge network graph, where nodes represent entities or concepts, and edges are composed of attributes or relationships. The knowledge graph is mainly used to build and maintain the above entities and relationships, providing support for recommendation systems, semantic understanding, question answering, etc. in subsequent intelligent search.
[0023] The acquisition of power data is driven by business needs. It is a highly parallel automatic collection of big data about the power industry on the company's information system and the Internet. The data acquisition process quickly collects the data into the system, using the ontology construction, knowledge reasoning and other related technologies in the knowledge graph, and relying on the mature knowledge graph construction technology in the Internet field, combined with the traditional data storage model of the power industry, intelligently and quickly identify and acquire power industry knowledge. When the user puts forward a search demand, information discovery interacts with the intelligent search technology knowledge deduction system, and gives a deep search task based on knowledge deduction, so as to obtain more task-oriented data, and integrate the acquired data to finally meet the user's search needs.
[0024] The intelligent search method based on the power data knowledge graph provided in this embodiment, by constructing the power data knowledge graph and conducting intelligent search on this basis, reduces the difficulty for users to obtain the required power data knowledge through intelligent search, improves the feedback accuracy of search knowledge, improves the efficiency and accuracy of power data work, reduces the labor intensity of power data workers, and better supports the construction and implementation of a multi-integrated and highly elastic power grid. It can further improve the refinement and intelligence of power data, reduce the workload of front-line designers, and at the same time improve the automation of power data, ensuring that the company's various power data work is accurate, efficient, and intelligent. Solve the current problems of power data from the root, provide the company with advanced, efficient, and intelligent power data capabilities, enhance the company's competitiveness in power supply services after the power sales side is liberalized, and bring more economic benefits to power grid companies.
[0025] The technical scheme and technical effects of the present invention will be further described below through specific examples. The following examples are explanations of the present invention and the present invention is not limited to the following examples.
[0026] Step 1: Capture the first entity from the power data structured data, and process the captured first entity.
[0027] Classify the data collected by the data source, including structured data (relational database), semi-structured data (HTML, XML, etc.) and unstructured data (text, images, etc.).
[0028] Analyze the power data, use the shallow semantic model to calculate the word set with hierarchical relationship of the power data entity, where the hierarchical relationship is expressed by similarity, and obtain the similar word set of a certain power data entity word vector; classify the similar word set, and establish the semantic real-time extended association related to the word vector.
[0029] When grading, the classification is carried out according to: keywords < first-level word association < second-level word association < third-level word association, where the first-level word association is a word set calculated based on the keywords using the preview model, the second-level word association is a word set obtained based on the first-level word association, and the third-level word association is a word set obtained based on the second-level word association.
[0030] The data sources of this embodiment include open link data, online encyclopedias, industry websites, etc.
[0031] Step 2: Extract the second entity from the data source, build the relationship between the first entity and the second entity, obtain the entity with a unique identifier, and build a power data knowledge graph.
[0032] Electric power data is multi-source and heterogeneous. Studying the association calculation of electric power data knowledge is a very important technology for relationship extraction in constructing the electric power data knowledge graph. This embodiment adopts an extraction model based on the attention graph convolutional network to realize the electric power data knowledge relationship extraction based on semantic annotation.
[0033] This step includes entity alignment. Entity alignment, as a key technology in the knowledge fusion process, is also called entity matching. It is the process of inferring whether different entities from different data sets are mapped to the same object in the physical world. It can eliminate cross-system data inconsistencies such as entity conflicts and unclear pointing in heterogeneous data.
[0034] Step 3: Build a knowledge graph index structure and synchronously update changes in the power data knowledge graph.
[0035] The modeling of the knowledge graph involves the addition, modification, and deletion of nodes, and the addition, modification, and deletion of relationships. When users perform a large number of operations on the front-end page, all changes need to be updated and synchronized to the knowledge base, which requires a large number of queries and comparisons, greatly affecting performance. The operation of more than a dozen nodes and edges greatly reduces the experience, and batch operations cannot be completed.
[0036] Therefore, the indexing technology of this embodiment is used to solve the problem of optimizing the batch operation performance of the graph modeling engine, and an efficient knowledge graph indexing structure is designed to improve the matching efficiency and accuracy when the knowledge graph is updated, while also improving the matching efficiency and accuracy of subsequent knowledge graph user search requests.
[0037] Use grid files to organize and index the collected data, that is, reorganize and store the data according to the index dimension. By dividing the data space into grid units, the index table only needs to store the information of the grid units, which greatly reduces the size of the index table.
[0038] The index structure pre-calculates the aggregate values required by the user when creating the index, greatly improving the performance of aggregate value queries. In order to improve the reading efficiency of the index table, the index structure uses key-value database storage and reads in a key-based manner without the need for full table scanning, which improves the reading performance of the index.
[0039] For example, this embodiment establishes a 2-dimensional index structure on index dimensions X and Y, such as Figure 2As shown. The index structure uses grid files to divide the data space into many small units, which are called GFUs (Grid File Units). In this way, the records in the table will fall into the corresponding GFU according to the value of the index dimension, and all records falling in the same GFU will be stored continuously in the data file, called a slice. Each GFU is stored in the index table in the form of Key / Value. The index structure is the lower left corner coordinate of the GFU in the data space. The GFU Value consists of two parts: Header and Location. Header is a user-defined pre-calculated UDF. For example, the Sum of the numerical column in each GFU can be pre-calculated, which can greatly speed up the query of aggregated values. Location records the file name, start offset, and end offset of the data slice corresponding to the GFU.
[0040] For example, Figure 2 The GFU Key corresponding to the GFU in the upper right corner is 10_17. If the sum of dimension Z is pre-calculated when creating the index, the header is the sum(Z) value of all records falling in the GFU, and Location is the location information of the data slices that store all records falling in the GFU. In this embodiment, the file name is f, the start offset is 4, and the end offset is 20. The granularity of the data slice is determined by the segmentation strategy given by the user when creating the index. The segmentation strategy includes the minimum value and segmentation interval of each index dimension. As shown in the figure above, when creating the index, the minimum value of the specified index dimension X is 1, and the segmentation interval size is 3; and the minimum value of the index dimension Y is 11, and the segmentation interval size is 2. In addition, in order to speed up the reading of the index table, the index table of the index structure is saved in HBase in the form of Key / Value, so that the index information can be quickly read in a Key-based manner.
[0041] The creation process of the index structure is a Map Reduce task. In the Map phase of index creation, for each record, first read the value of its index dimension, then calculate the GFU to which it belongs based on the value of the index dimension and the segmentation strategy, so as to obtain the GFU Key of the record, and finally send the GFU Key and the record as a key-value pair to the Reduce function. In the Reduce phase of index creation, the Reducer receives a list of records with the same GFU Key, that is, all records belonging to the same data slice. The ultimate goal is to output each data slice in sequence and record the correspondence between the GFU Key and the position of the data slice in the file. At the same time, it is also necessary to pre-calculate the corresponding value of each record in the GFU according to the UDF specified by the user. Specifically, first initialize the data slice start offset, end offset, data slice size, output file name and header to save the pre-calculated value. In the second step, for each record with the same GFU Key, first pre-calculate its UDF value and accumulate it in the header, then accumulate the size of the current record in the slice Size, and calculate its end offset after the entire record list is processed. Then, GFUValue is calculated and the GFU Key / GFU Value pair is written to HBase. This completes the index creation.
[0042] Step 4: Obtain user search requests, match user search requests, map matching results to entities and relationships in the knowledge graph, and return search results based on the structure within the knowledge graph.
[0043] Specifically include: 1. Use the bidirectional maximum matching algorithm to segment the user's search request.
[0044] The user search request is split according to the maximum length of the word, and then matched from left to right in the preset dictionary. If the match is successful, the word segmentation is extracted from the search request; if the match fails, the length of the matching word is minus 1 and the word segmentation is performed again until all the words in the search request are split.
[0045] The user search request is split according to the maximum length of the word, and then matched from right to left in the preset dictionary; if the match is successful, the word segmentation is extracted from the search request; if the match fails, the length of the matching word is minus 1 and the word segmentation is performed again until all the words in the search request are split.
[0046] If the segmentation results of the two matches are the same, the segmentation results are returned. If the segmentation results of the two matches are different, the segmentation result with fewer segmentations is returned. If the number of segmentations is the same, the segmentation result with a larger word length is returned.
[0047] 2. Classify user search requests based on word segmentation results.
[0048] The classifying of the user search request according to the word segmentation result includes: performing consciousness recognition on the word segmentation result to obtain the code and purpose of the user search request as the type of the user search request.
[0049] 3. Determine the entity targeted by the user's search request and obtain the triples to be retrieved.
[0050] Entities are extracted from user search requests based on the fuzzy recognition algorithm, and the type of user search request is used as a relationship to obtain triples to be retrieved. The pattern of triples to be retrieved is "Entity 1, relationship,?".
[0051] 4. Form a search statement that establishes a match for the triples to be retrieved and perform intelligent search.
[0052] According to the index structure, the knowledge graph is searched for triples that match the triples to be retrieved (the pattern is "entity 1, relationship, entity 2"), and the information similarity is calculated; the triples obtained in the knowledge graph and the corresponding similarities are returned to the user as intelligent search results.
[0053] Embodiment 2: This embodiment provides an intelligent search method based on the knowledge graph of power data, and brings in a specific application scenario to execute the method. Power design work is an important part of building a multi-source integrated and highly elastic power grid in the form of the Internet. In the power design and planning business, due to a series of complex processes involving load statistics, main transformer selection, main wiring design, flow calculation, electrical equipment selection, relay protection design, selection of geographical location engineering construction, etc., it often relies on a large amount of manpower, and there are problems such as insufficient understanding of the business by different business personnel, and it is difficult to ensure the optimality of the design solution.
[0054] The above process also involves a large amount of power design business knowledge. The dispersion and heterogeneity of this knowledge make it difficult for users to effectively obtain and use relevant information. At the same time, the field of power design technology also needs to search for structured data, but traditional information retrieval can only search for unstructured data such as documents, which cannot fully meet the power design knowledge search needs, resulting in a complex and inaccurate search process for power design business knowledge, and difficulty in obtaining power design business knowledge.
[0055] An intelligent search method based on a knowledge graph of electric power data has the following operation flow: Step 1, obtaining structured data of electric power data from a data source, capturing entities from the structured data, and processing the captured entities; Step 2, forming relationships between entities based on knowledge links, obtaining entities with unique identifiers, storing entities and relationships in a knowledge base, and constructing a knowledge graph of electric power data; Step 3, constructing a knowledge graph index structure, and synchronously updating changes in the knowledge graph of electric power data; Step 4, obtaining user search requests, matching user search requests, mapping matching results to entities and relationships in the knowledge graph, and returning search results based on the structure in the knowledge graph.
[0056] The intelligent search in this embodiment is based on knowledge extraction and entity association technology, realizing the application of knowledge graph technology in data intelligent search. Through artificial intelligence technologies such as natural language processing and knowledge graph, artificial intelligence is implemented in search engine products.
[0057] The technical scheme and technical effects of the present invention will be further described below through specific examples. The following examples are explanations of the present invention and the present invention is not limited to the following examples.
[0058] In this embodiment, the search engine model based on the power data knowledge graph is as follows: Figure 3 As shown, it includes: data processing, knowledge graph construction and knowledge query.
[0059] Specific: The first step: data processing, mainly involves capturing entities from authoritative structured data, and storing them in a specified collection after being processed by the ontology library.
[0060] Investigate and analyze the sources and storage methods of power data, extract and convert source data through various channels and methods, and label all raw data to facilitate subsequent management and use.
[0061] The current mainstream big data technologies include large-scale horizontally scalable database technology, distributed file system HDFS and MapReduce data processing technology, scalable NoSQL technology, distributed indexing and retrieval technology for specific applications, and parallel data processing technology.
[0062] With the popularity of the concept of big data, unstructured data has gradually become synonymous with big data. Unstructured data includes the following types: text, interactive data, images, audio, video, etc. The technologies for processing unstructured data include: dynamic metadata discovery (for text data, dynamically extract metadata from the result set, maintain and control metadata on a continuous basis); classification setting (annotate and reference input data); entity extraction and semantic analysis (extract entities and relationships expressed in data streams); multimedia recognition technology (various technologies for interpreting images, audio and video).
[0063] Data quality usually refers to the quality of data values, including accuracy, completeness and consistency. Data accuracy means that the data does not contain errors or outliers, completeness means that the data does not contain missing values, and consistency means that the data is the same in each data source. Broadly speaking, data quality also includes the validity of the data as a whole, for example, whether the data as a whole is credible, whether the data sampling is reasonable, etc. The main task of data quality analysis is to detect whether there is dirty data in the original data. Dirty data generally refers to data that does not meet the requirements and cannot be directly analyzed. Usually, there are problems such as incompleteness (missing values), inconsistency, and data anomalies in the original data. These dirty data will reduce the quality of the data and affect the results of data analysis. Therefore, before conducting data analysis, the data needs to be cleaned, integrated, converted, and other processes to improve the quality of the data.
[0064] 1) Missing value processing.
[0065] Missing data generally refers to the missing of data collection and the missing of variable values during the storage process. Both will cause inaccurate analysis results and lead to deviations in the nature and characteristics of the overall data inferred from the sample data.
[0066] In R language, missing values are usually represented by NA. The is.na() function can be used to determine whether missing values exist. In addition, the complete.cases() function can be used to determine missing values by identifying whether the sample data is complete. complete.cases() is a function in the stats package that returns a logical vector. Each element of the vector indicates whether each observation is complete. Complete means that all variables in the observation do not contain missing values.
[0067] Common methods for dealing with missing values are to delete the records where the missing values are located, replace the mean, and interpolate. When the proportion of missing values is relatively small, the deletion method can be used to reduce the amount of sample data in exchange for data integrity. Using the mean to replace missing values is a common practice in dealing with missing values. In addition, interpolation methods such as mean / mode interpolation, model interpolation, and multiple interpolation can also be used to achieve missing value filling.
[0068] 2) Outlier processing.
[0069] The full name of outliers in statistics is suspected outliers, also known as outliers, and the analysis of outliers is also called outlier analysis. Outliers refer to "extreme values" that appear in samples. The data values appear abnormally large or small, and their distribution deviates significantly from the rest of the observed values. Outlier analysis is to test whether there are unreasonable data in the data. In data analysis, we can neither ignore the existence of outliers nor simply exclude them from data analysis.
[0070] When processing data, the method of handling outliers depends on the specific situation. Sometimes, outliers may be normal values, but they are abnormally large or small. Therefore, in many cases, it is necessary to first analyze the possible causes of outliers and then determine how to handle them.
[0071] Common methods for handling outliers include: deleting records containing outliers; interpolation, treating outliers as missing values and handling them using the missing value processing method, with the advantage of using existing data to replace or interpolate outliers; no processing, directly performing data analysis on the data set containing outliers.
[0072] 3) Consistency analysis.
[0073] Data inconsistency refers to the contradiction of data, which is mainly caused by different data sources, resulting in inconsistent data encoding. In addition, asynchronous data updates can also easily lead to data inconsistency. For example, the address information of the same line may be different in different data tables. This is when the data is updated, the asynchronous update leads to data inconsistency.
[0074] When data inconsistencies occur, the data needs to be cleaned and integrated to remove redundant data, unify variable names, unify data units, etc., and use the latest data to eliminate inconsistencies.
[0075] The second step: knowledge graph construction, mainly extracting entities from the data source, forming the relationship between entities through the knowledge linking stage, forming entities with unique identifiers, and finally storing these entities and relationships in the knowledge base.
[0076] Electric power data is multi-source and heterogeneous. Studying the association calculation of power data knowledge is a very important technology for relationship extraction in constructing the power data knowledge graph. The association calculation model based on POS_CBOW is used to realize the power data knowledge relationship extraction based on semantic annotation.
[0077] To ensure the availability of the knowledge graph, one is accuracy, and the other is efficiency. These two points depend on the correctness of semantic reasoning and the efficiency of the algorithm. In order to ensure the efficient availability of the knowledge graph, this embodiment uses an efficient computing model, Word2vec, to expand the relationship extraction to assist semantic calculation. Word2vec is an unsupervised machine learning algorithm characterized by high accuracy and fast calculation. After training based on large-scale corpus, a word vector model of semantic relations and linguistic laws is generated. The characteristics of high efficiency and scalability make it the key to knowledge base generation and semantic analysis, which is of great significance for building a knowledge graph of power data.
[0078] The third step: knowledge query. Users interact with the search engine through the retrieval platform. The search engine uses the knowledge graph to deeply understand the user's request, and then returns the search results to the user after a series of operations such as request parsing, retrieval, and recommendation.
[0079] Search engines use the knowledge graph they build to search for relevant knowledge cards. Through knowledge graph search, we can better understand the user's search needs and return more accurate and personalized query results to users based on their needs. Knowledge graph search returns the answer directly, rather than a recommended link or similar content.
[0080] Through intelligent search of power design knowledge, predictions and assessments can be made on the cost, duration, material supply, etc. of power design planning business projects, thereby improving the accuracy of business personnel's decision-making.
[0081] The above-described embodiment is only a preferred solution of the present invention and does not limit the present invention in any form. There are other variations and modifications without exceeding the technical solution described in the claims.
Claims
1. An intelligent search method based on power data knowledge graph, characterized in that: include: S1: Capture a first entity from the power structured data and process the captured first entity; S2: Extract the second entity from the data source, build the relationship between the first entity and the second entity, obtain the entity with a unique identifier, and build a power data knowledge graph; S3: Build a knowledge graph index structure and synchronize updates to the power data knowledge graph; S4: Obtain user search requests, match user search requests, map matching results to entities and relationships in the knowledge graph, and return search results based on the structure within the knowledge graph.
2. The intelligent search method based on power data knowledge graph according to claim 1 is characterized in that: The step S4 includes: using a bidirectional maximum matching algorithm to segment the user's search request; classifying the user's search request according to the segmentation result; determining the entity targeted by the user's search request to obtain a triple to be retrieved; forming a search statement to establish a match for the triple to be retrieved, and performing an intelligent search.
3. The intelligent search method based on power data knowledge graph according to claim 1 is characterized in that: The index structure organizes and indexes the collected data using a grid file, and calculates the aggregation value required by the user when creating the index.
4. The intelligent search method based on power data knowledge graph according to claim 2 is characterized in that: The word segmentation of the user's search request includes: dividing the user's search request according to the maximum length of the word, and then matching it from left to right in the preset dictionary; dividing the user's search request according to the maximum length of the word, and then matching it from right to left in the preset dictionary; if the word segmentation results of the two matches are the same, the word segmentation result is returned; if the word segmentation results of the two matches are different, the word segmentation result with fewer word segments is returned; if the number of word segments is the same, the word segmentation result with the larger word length is returned.
5. The intelligent search method based on power data knowledge graph according to claim 2 or 4, characterized in that: Based on the fuzzy recognition algorithm, entities are extracted from the user search request, and the type of the user search request is used as a relationship to obtain the triples to be retrieved; the knowledge graph is searched for triples that match the triples to be retrieved, and the information similarity is calculated; the triples obtained in the knowledge graph and the corresponding similarities are returned to the user as intelligent search results.
6. The intelligent search method based on power data knowledge graph according to claim 2 or 4, characterized in that: The classifying of the user search request according to the word segmentation result includes: performing consciousness recognition on the word segmentation result to obtain the code and purpose of the user search request as the type of the user search request.
7. An intelligent search method based on power data knowledge graph according to claim 1, 2, 3 or 4, characterized in that: The step S1 includes: analyzing the power data, using a shallow semantic model to calculate a word set with a hierarchical relationship of the power data entity, the hierarchical relationship is represented by similarity, and a similar word set of a certain power data entity word vector is obtained; the similar word set is graded, and a real-time semantic extension association related to the word vector is established.
8. The intelligent search method based on power data knowledge graph according to claim 4 is characterized in that: When matching a word, if the match is successful, the word is extracted from the search request; if the match fails, the length of the matching word is minus 1 and the word is segmented again until all words in the search request are split.
9. The intelligent search method based on power data knowledge graph according to claim 1, 2 or 3, characterized in that: The index structure is stored in a key-value database and read based on the key.
10. The intelligent search method based on power data knowledge graph according to claim 7, characterized in that: The similar word set is classified into: keywords, first-level word associations calculated based on the keywords using a preview model, second-level word associations obtained based on the first-level word association set, and third-level word associations obtained based on the second-level word association set.
