An intelligent retrieval method and system for power grid digitization construction achievement resources

Abstract

The application discloses an intelligent retrieval method and system for power grid digital construction achievement resources, and comprises the following steps: obtaining a retrieval word; inputting the retrieval word into a trained deep network joint model based on a word segmentation model and a power grid digital construction achievement knowledge graph; performing word segmentation on the retrieval word based on the word segmentation model to obtain semantic features in the deep network joint model; obtaining a matching result with different weight orders by performing deep network reasoning on the semantic features in the power grid digital construction achievement knowledge graph; and outputting the matching result as a search result. Advantages: the unstructured information of the power grid digital construction achievement is associated and modeled by using the knowledge graph, a structured power grid digital construction achievement knowledge graph is constructed, more accurate and efficient search experience can be provided for business personnel, business data related to the retrieval keyword can be displayed, and the opening cost of the power grid digital construction achievement resources is effectively reduced.

Description

TECHNICAL FIELD

[0001] The application relates to an intelligent retrieval method and system for power grid digitization construction achievement resources, and belongs to the technical field of natural language processing. BACKGROUND

[0002] In order to show the overall picture of the digitization construction achievements, the power grid company gathers all the digitization achievements such as infrastructure, data resources, shared services, data models, algorithm models, digitization applications and knowledge cases scattered in the cloud platform, data center, business center and technical center in a full amount, designs a scientific hierarchical classification system, forms a unified service window, and has accessed TB level achievements. However, these digitization achievements are unstructured and exist in the form of documents, videos and various forms, so it is difficult for grassroots business personnel to accurately search and utilize the achievements. The traditional retrieval method adopts a word similarity matching degree, and the business personnel using the traditional retrieval method to find the required achievements has the problems of slow query and difficult acquisition, which leads to a high learning achievement of the digitization construction achievements by the business personnel of various professions and departments.

[0003] With the development of the field of artificial intelligence, deep networks play a greater role, and knowledge graph is a representation form of deep networks. The knowledge graph stores the attributes and relationships of entities in the form of triples, and can be understood as a semantic deep network connected by knowledge points. It can be used to better query unstructured information. For example, the Chinese patent with the application number CN202111540151.2 discloses an intelligent search system based on a knowledge graph. The patent uses a twice-training triple extraction model method to make the enterprise-level search engine more intelligent. For another example, the Chinese patent with the application number CN202111194340.9 discloses a task-based intelligent dialogue construction method based on a knowledge graph, which reduces the configuration of intent and word slots and facilitates the association between different tasks. It can be seen that the knowledge graph can effectively support the retrieval of a large amount of unstructured data by establishing the association between businesses, between businesses and data, and between data. The unstructured data of the power grid digitization construction achievements urgently need the help of tools such as the knowledge graph to realize intelligent, efficient, fast and flexible information retrieval.

[0004] When the traditional retrieval method is used to retrieve the power grid digitization construction achievements, there is a problem of long retrieval time and inaccuracy caused by unstructured achievement data, which does not meet the needs of users to conveniently find various achievements. SUMMARY

[0005] The technical problem to be solved by the application is to overcome the defects of the prior art and provide an intelligent retrieval method and system for power grid digitization construction achievement resources.

[0006] To solve the above technical problems, the present application provides an intelligent retrieval method for power grid digital construction achievement resources, comprising:

[0007] Obtaining a retrieval term;

[0008] Inputting the retrieval term into a trained deep network joint model based on a word segmentation model and a power grid digital construction achievement knowledge graph, performing word segmentation on the retrieval term based on the word segmentation model in the deep network joint model to obtain semantic features, and obtaining a matching result with different weight orders by the power grid digital construction achievement knowledge graph through deep network reasoning on the semantic features;

[0009] Outputting the matching result as a search result.

[0010] Further, the training of the deep network joint model based on the word segmentation model and the power grid digital construction achievement knowledge graph comprises:

[0011] The word segmentation model adopts a BERT Chinese natural language processing model;

[0012] Obtaining a GCN pre-training model;

[0013] Performing joint network training on the BERT Chinese natural language processing pre-training model and the GCN pre-training model using the constructed power grid digital construction achievement knowledge graph dataset until convergence is achieved, thereby obtaining the trained deep network joint model based on the word segmentation model and the power grid digital construction achievement knowledge graph.

[0014] Further, the construction of the power grid digital construction achievement dataset comprises:

[0015] According to business requirements, combing retrieval scenarios and collecting retrieval cases;

[0016] Taking the corresponding achievement ontology object as the center, determining the topological structure of the center and other associated attributes, and associating with other related data to form a power grid digital construction achievement knowledge graph dataset with the achievement ontology as the core and other related attributes as branches and leaves.

[0017] Further, the joint network training on the BERT Chinese natural language processing pre-training model and the GCN pre-training model using the constructed power grid digital construction achievement knowledge graph dataset until convergence is achieved comprises:

[0018] The GCN pre-training model is used for relation extraction of the power grid digitization construction achievement knowledge graph dataset, the BERT Chinese natural language processing model is used for word segmentation processing of achievement entities of the power grid digitization construction achievement knowledge graph dataset, the GCN pre-training model is used for encoding the semantic relationship of the achievement into a vector, then the semi-supervised learning mode is used for the encoded dataset, the nodes in the entity pair are trained, negative sampling is performed when the loss function is calculated, and the network model is continuously optimized until the network model converges.

[0019] An intelligent retrieval system for power grid digitization construction achievement resources comprises:

[0020] An acquisition module is configured to acquire a search term;

[0021] A processing module is configured to input the search term into a deep network joint model based on a word segmentation model and a power grid digitization construction achievement knowledge graph, perform word segmentation on the search term based on the word segmentation model to obtain semantic features in the deep network joint model, and obtain a matching result with different weights after ordering by performing deep network reasoning on the semantic features in the power grid digitization construction achievement knowledge graph.

[0022] An output module is configured to output the matching result as a search result.

[0023] A computer-readable storage medium storing one or more programs, the one or more programs comprising instructions that, when executed by a computing device, cause the computing device to perform any of the methods.

[0024] A computing device comprises,

[0025] One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs comprise instructions for performing any of the methods.

[0026] The present application has the following beneficial effects:

[0027] The unstructured information of the power grid digitization construction achievement is associated and modeled using a knowledge graph, a structured power grid digitization construction achievement knowledge graph is constructed, more accurate and efficient search experience can be provided for business personnel, business data related to the search keyword can be displayed, and the present application can effectively reduce the opening cost of the power grid digitization construction achievement resources, can fully release the resources and capabilities of the existing construction achievements, and can support rapid search. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 The present application is a flowchart;

[0029] Figure 2 The figure is a grid digital construction achievement knowledge graph modeling graph in the application. DETAILED DESCRIPTION

[0030] The application will be further described below with reference to the drawings. The following examples are only used to more clearly illustrate the technical solutions of the application, and cannot be used to limit the protection scope of the application.

[0031] The application discloses an intelligent retrieval method for grid digital construction achievement resources, and the use mode is shown in the accompanying Figure 1 , and the steps are as follows:

[0032] Step 1: input search terms

[0033] Step 2: the retrieval module searches the achievements according to the search terms

[0034] Step 3: output the search results

[0035] Further, the retrieval module in step 2 is composed of a BERT-based word segmentation model and a GCN-based grid digital construction achievement knowledge graph, and the operation steps of the retrieval module are as follows:

[0036] Step (1): the BERT-based word segmentation model performs word segmentation processing on the search terms to obtain semantic features

[0037] Step (2): the GCN-based grid digital construction achievement knowledge graph performs deep network reasoning on the semantic features to obtain matching results with different weights

[0038] Further, the BERT-based word segmentation model in step (2) is characterized in that BERT (Bidirectional Encoder Representation from Transformers) is one of the best pre-training models in natural language processing learning representation, has strong language representation ability and feature extraction ability, the application uses the Google open source BERT-based Chinese pre-training model, TensorFlow version BERT-base-Chinese, relies on Huggingface-Transformers 3.1.0 and PaddleHub to call the model, the BERT Chinese processing pre-training model can effectively preprocess the search word, and the GCN pre-training model is jointly trained in the network until the model converges, and a deep network joint model based on the word segmentation model and the power grid digital construction achievement knowledge graph is obtained; the GCN pre-training model refers to a simple framework model with pre-set parameters, which becomes a trained model after being optimized by different scenarios and different data sets, and the application has a TensorFlow version of the GCN pre-training model.

[0039] Further, the GCN-based power grid digital construction achievement knowledge graph in step (2) is characterized in that the application uses a semi-supervised graph convolutional network (GCN, Graph Convolutional Network) to construct the knowledge graph, the input is a graph of any size and shape, the specific input is a feature vector of nodes and structures, and the goal is to learn a nonlinear formula to represent the input features and generate node-level output. GCN can encode the information of adjacent relationship nodes, and combine the structure and input features together by using a graph, which facilitates the classification regression task in the back end and can reduce the operation cost of the knowledge graph. The power grid digital construction achievement knowledge graph modeling method is shown in the accompanying Figure 2 , and the steps are as follows:

[0040] Step S1: search case collection

[0041] Step S2: define achievement entities and label attribute association information to obtain a power grid digital construction achievement knowledge graph data set;

[0042] Step S3: use a deep network to process and train the data set to obtain a network model that is continuously optimized until convergence;

[0043] Step S4: search service online test

[0044] In step S1, according to the business demand, the search scene is sorted out, and a large number of search cases are collected.

[0045] In step S2, the corresponding achievement ontology object is retrieved as the center, the topology structure of the center and other associated attributes (such as achievement catalog and achievement type) is determined, and other related data (such as application scenario and reporter) is associated to form a complete power grid digital construction achievement knowledge graph dataset with the achievement ontology as the core and other related attributes as branches and leaves.

[0046] In step S3, the BERT-based Chinese processing pre-training model and the GCN pre-training model are jointly trained for the power grid digital construction achievement knowledge graph dataset. The achievement entity of the dataset is processed by the BERT-based Chinese processing pre-training model, and then encoded into a vector by the GCN pre-training model. The semi-supervised learning mode is used for the encoded dataset, the nodes in the achievement entity pair are trained, negative sampling is performed when calculating the loss function, and the network model is continuously optimized until it converges, thereby obtaining the trained deep network joint model based on the word segmentation model and the power grid digital construction achievement knowledge graph.

[0047] In step S4, the retrieval model is tested in conjunction with the business system, and is continuously improved.

[0048] Application example:

[0049] Referring to the accompanying Figure 2 , the retrieval module is constructed

[0050] In actual application scenarios, the following work is carried out according to the intelligent retrieval business requirements of the power grid digital construction achievement resources:

[0051] (1) Retrieval case collection:

[0052] According to the business requirements, the retrieval scenarios are sorted out, and the following examples are given:

[0053] a. The achievement with the highest access volume in Jiangsu Province;

[0054] b. The component application with the highest score;

[0055] c. The power outage application single repair in the resource business center of Gansu Power Grid;

[0056] d. The infrastructure in Hunan;

[0057] e. ……

[0058] (2) Define achievement entity and annotated attribute association information:

[0059] According to the list of search terms, the matched results are sorted out, the parameter entity is determined, and the relevant attributes of the results are defined and labeled, such as the highest search volume in Jiangsu Province, and the corresponding results of the search data are

Jiangsu UAV

[0060] (3) Construction of knowledge graph and model training and optimization:

[0061] For this power grid digital construction result knowledge graph dataset, the BERT-based Chinese processing pre-training model and the GCN pre-training model are jointly trained, and the iterative operation is performed until the deep network model converges, and the trained deep network joint model based on the word segmentation model and the power grid digital construction result knowledge graph is obtained.

[0062] (4) Online test of search service:

[0063] The service is put online, and the business system is tested and optimized continuously.

[0064] Display search examples:

[0065] Example: Input

Jiangsu UAV

[0066] (1) Match the results whose Jiangsu result name is UAV;

[0067] (2) Match the results whose result name contains 'Jiangsu UAV';

[0068] (3) Match the results whose result description contains 'Jiangsu UAV' and whose result name contains 'UAV';

[0069] (4) Match the results whose organization contains 'Jiangsu' and whose result name contains 'UAV';

[0070] (5) Match the results whose application scenario is 'Jiangsu UAV';

[0071] (6)......

[0072] As can be seen from the above examples, the present application constructs a power grid digital construction result knowledge graph, and integrates a deep network to effectively and accurately search non-structured power grid digital construction result resources, which provides convenience for business personnel to explore company digital results and reduces learning costs.

[0073] The application also provides an intelligent retrieval system for grid digital construction achievement resources, comprising:

[0074] An acquisition module is configured to acquire a search term;

[0075] A processing module is configured to input the search term into a deep network joint model based on a word segmentation model and a grid digital construction achievement knowledge graph, perform word segmentation on the search term based on the word segmentation model to obtain semantic features in the deep network joint model, and obtain a matching result with different weights after ordering by performing deep network reasoning on the semantic features in the grid digital construction achievement knowledge graph.

[0076] An output module is configured to output the matching result as a search result.

[0077] The application also provides a computer readable storage medium storing one or more programs, wherein the one or more programs include instructions, and the instructions, when executed by a computing device, cause the computing device to perform any of the methods.

[0078] The application also provides a computing device, comprising,

[0079] One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for performing any of the methods.

[0080] Those skilled in the art should understand that the embodiments of the application can be provided as a method, a system, or a computer program product. Therefore, the application can be in the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the application can be in the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0081] The application is described with reference to flowcharts and / or block diagrams according to the methods, devices (systems), and computer program products of the embodiments of the application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices produce a device that implements the functions described in the flowcharts and / or block diagrams. Figure 1apparatuses that implement the functions specified in the flowchart or flowcharts and / or blocks. Figure 1

[0082] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flowchart or flowcharts and / or blocks. Figure 1 apparatuses that implement the functions specified in the flowchart or flowcharts and / or blocks. Figure 1

[0083] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flowchart or flowcharts and / or blocks. Figure 1 apparatuses that implement the functions specified in the flowchart or flowcharts and / or blocks. Figure 1

[0084] The above description is merely that of the preferred embodiments of the application and modifications and alterations are possible without departing from the principles of the application as it is understood that implementations of the application in its broadest form can not be necessarily limited to only the preferred embodiments described and / or illustrated herein.​​​

Claims

1. An intelligent retrieval method for resources related to the achievements of power grid digitalization construction, characterized in that, include: Retrieve search terms; The search terms are input into a trained deep network joint model based on a word segmentation model and a knowledge graph of the achievements of digital power grid construction. In the deep network joint model, the search terms are segmented based on the word segmentation model to obtain semantic features. The knowledge graph of the achievements of digital power grid construction then uses deep network reasoning to obtain matching results with different weights and rankings. Output the matching results as search results; The training of the deep network joint model based on the word segmentation model and the knowledge graph of power grid digital construction achievements includes: The word segmentation model uses the BERT Chinese natural language processing model. Obtain the GCN pre-trained model; Using the constructed knowledge graph dataset of power grid digital construction achievements, the BERT Chinese natural language processing pre-trained model and the GCN pre-trained model were jointly trained until convergence, resulting in a well-trained deep network joint model based on the word segmentation model and the knowledge graph of power grid digital construction achievements. The method of jointly training the BERT Chinese natural language processing pre-trained model and the GCN pre-trained model using the constructed power grid digital construction achievement knowledge graph dataset until convergence includes: using the GCN pre-trained model to extract relations from the power grid digital construction achievement knowledge graph dataset; using the BERT Chinese natural language processing model to perform word segmentation on the achievement entities in the power grid digital construction achievement knowledge graph dataset; then using the GCN pre-trained model to encode the semantic relations of the achievements into vectors; then using a semi-supervised learning mode on the encoded dataset to train the nodes in the entity pairs; performing negative sampling when calculating the loss function; and continuously optimizing until the network model converges.

2. The intelligent retrieval method for resources of power grid digitalization construction achievements according to claim 1, characterized in that, The construction of the knowledge graph dataset of the power grid digitalization construction achievements includes: Based on business needs, identify search scenarios and collect search cases; Centered on the retrieved corresponding result ontology object, the topological structure of the center and other related attributes is determined, and then associated with other relevant data to form a knowledge graph dataset of power grid digital construction results with the result ontology as the core and other related attributes as branches and leaves.

3. An intelligent retrieval system for resources related to the achievements of power grid digitalization construction, characterized in that: include: The acquisition module is used to acquire search terms; The processing module is used to input the search terms into a trained deep network joint model based on a word segmentation model and a knowledge graph of the achievements of digital power grid construction. In the deep network joint model, the search terms are segmented based on the word segmentation model to obtain semantic features. The knowledge graph of the achievements of digital power grid construction then uses deep network reasoning to obtain matching results with different weights and rankings. The output module is used to output the matching results as search results. The training of the deep network joint model based on the word segmentation model and the knowledge graph of power grid digital construction achievements includes: The word segmentation model uses the BERT Chinese natural language processing model. Obtain the GCN pre-trained model; Using the constructed knowledge graph dataset of power grid digital construction achievements, the BERT Chinese natural language processing pre-trained model and the GCN pre-trained model were jointly trained until convergence, resulting in a well-trained deep network joint model based on the word segmentation model and the knowledge graph of power grid digital construction achievements. The method of jointly training the BERT Chinese natural language processing pre-trained model and the GCN pre-trained model using the constructed power grid digital construction achievement knowledge graph dataset until convergence includes: using the GCN pre-trained model to extract relations from the power grid digital construction achievement knowledge graph dataset; using the BERT Chinese natural language processing model to perform word segmentation on the achievement entities in the power grid digital construction achievement knowledge graph dataset; then using the GCN pre-trained model to encode the semantic relations of the achievements into vectors; then using a semi-supervised learning mode on the encoded dataset to train the nodes in the entity pairs; performing negative sampling when calculating the loss function; and continuously optimizing until the network model converges.

4. A computer-readable storage medium for storing one or more programs, characterized in that, The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any of the methods according to claims 1 to 2.

5. A computing device, characterized in that, include, One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the methods according to claims 1 to 2.

Images