Information retrieval methods, systems, and media based on deep learning in the field of power grid control.

Abstract

This invention discloses a method, system, and medium for information retrieval in the field of power grid control based on deep learning. The method includes a training data construction module for constructing training data for training a deep learning translation model; a model training module for training the deep learning translation model using the training data obtained from the training data construction module; a question parsing module for parsing user-inputted source language into source language text recognizable by the deep learning translation model; an SQL post-processing module for converting the target language text output by the deep learning translation model into database-executable statements; and an SQL execution module for executing the database-executable statements obtained from the SQL post-processing module to retrieve the desired information from the database. Utilizing the method and system described in this invention lowers the barrier to entry for operating power grid databases, enabling staff to quickly retrieve the information they need and improving work efficiency. It also increases the depth of data mining and enhances the value-added monetization capabilities of data.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of deep learning and power grid control information retrieval, specifically involving a deep learning-based method, system, and medium for information retrieval in the field of power grid control. Background Technology

[0002] After years of operation, the power grid dispatching system has stored massive amounts of power grid operation data in databases or file systems, exhibiting characteristics such as large scale, diverse types, and wide scope. Currently, querying power grid dispatching information typically requires interaction with the database (or file system) using a procedural query language like Structured Query Language (SQL). This necessitates skilled SQL professionals to perform this operation. SQL has a high learning curve; even for computer professionals, writing numerous accurate SQL statements for different databases and application scenarios is a complex task. In recent years, with the continuous development of power grid operations, the number of database tables and the volume of data in the power grid dispatching information field have surged, posing challenges to dispatchers' query work. Often, only experienced dispatchers can quickly write the correct SQL statements to find relevant answers in the database based on the given questions. This is extremely unfriendly to less experienced dispatchers. Based on this situation, a rapid information retrieval method is needed to allow dispatchers to quickly obtain the search results based on their query questions. Summary of the Invention

[0003] This invention addresses the problem of cumbersome and inefficient analysis of massive amounts of data in power grid control systems. It utilizes semantic parsing technology to train an AI agent that can convert natural language into executable SQL statements, thereby lowering the threshold for data querying and providing users with more user-friendly information retrieval services.

[0004] A deep learning-based information retrieval method for power grid control, which achieves one of the objectives of this invention, includes the following steps:

[0005] S1. Collect statement pairs of natural language questions and SQL statements from multiple power grid control fields;

[0006] S2. Convert the collected natural language questions into source statements that can be used to train deep learning translation models; convert the collected SQL statements into target statements expressed in natural language.

[0007] The conversion process in step S2 can be implemented using software coding based on existing technology, which can convert natural language questions into source text that can be input into a deep learning translation model.

[0008] Furthermore, step S2 also includes preprocessing the source statement, which includes word segmentation, special character replacement, and adding special characters.

[0009] Furthermore, the process of converting the collected SQL statements into target statements expressed in natural language includes the following steps:

[0010] S201. Construct a mapping relationship between database table names and column names and natural language representations;

[0011] S202. Based on the above-described mapping relationship, convert the SQL statement in step S1 into a target statement expressed in natural language.

[0012] The mapping relationship refers to the one-to-one mapping between information expressed in natural language and the table or column name in the database. For example, if "PLANT_BASIC" is a table in the database used to store "power plant information", and the column named "PLANT_TYPE" in the table represents the power plant type, then two mapping relationships can be obtained: "PLANT_BASIC power plant basic information" and "PLANT_BASIC.PLANT_TYPE power plant basic information.power plant type".

[0013] Furthermore, step S202 also includes preprocessing the target statement, which includes word segmentation, special character replacement, and adding special characters.

[0014] S3. Build a deep learning translation model. Use the source and target sentences obtained in step S2 to train the deep learning translation model to obtain a Text-to-SQL agent.

[0015] Through the training of the above model, the Text-to-SQL agent can translate source statements into target statements. The Text-to-SQL agent can adopt a deep learning translation model based on Transformer, which can improve the accuracy of SQL generation compared with other models.

[0016] S4. Use the Text-to-SQL agent to convert questions expressed in natural language into text in the target language;

[0017] Further, step S4 includes the following steps:

[0018] S401. Convert the natural language question input by the user to obtain the source language text that can be input into the Text-to-SQL agent;

[0019] S402. Use the Text-to-SQL agent to predict the source language text and obtain the target language text.

[0020] The conversion process in step S401 can be implemented using software coding based on existing technology, which can convert natural language questions into source text that can be input into a deep learning translation model.

[0021] Further, step S401 includes the following steps:

[0022] Step 1: Based on the mapping relationship between database table names and column names and natural language expressions, convert the natural language text input by the user into source language text that can be input into the Text-to-SQL agent;

[0023] Step 2: Preprocess the source language text that can be input into the Text-to-SQL agent.

[0024] The preprocessing includes word segmentation, special character replacement, and adding special characters.

[0025] S5. Convert the target language text obtained in step S4 into executable SQL statements.

[0026] Further, step S5 includes the following steps:

[0027] S501. Post-process the target language text;

[0028] The post-processing includes replacing or deleting special characters.

[0029] S502. Construct a mapping relationship between database table names and column names and natural language representations;

[0030] S503. Based on the above-described mapping relationship, convert the target language text obtained in step S5 into an executable SQL statement.

[0031] S6. Run the SQL statement obtained in step S5 to retrieve query information from the database.

[0032] The second objective of this invention, a deep learning-based information retrieval system for power grid control, includes:

[0033] Training data construction module: Used to construct training data for training deep learning translation models;

[0034] Model training module: Used to train the deep learning translation model based on the training data obtained from the training data construction module, to obtain a Text-to-SQL agent;

[0035] Question parsing module: used to parse the user's input source language into Text-to-SQL intelligent agent-recognized source language text;

[0036] SQL post-processing module: used to convert the target language text output by the Text-to-SQL agent into database executable statements;

[0037] SQL Execution Module: Used to run executable statements obtained from the SQL Post-Processing Module and retrieve the required information from the database.

[0038] Furthermore, it also includes a data preprocessing module for preprocessing the statements input to the deep learning translation model and / or Text-to-SQL agent. The preprocessing includes word segmentation, special character replacement, and adding special characters.

[0039] Furthermore, it also includes a mapping relationship building module for building mapping relationships between database table names and / or column names and natural language representations.

[0040] A non-transitory computer-readable storage medium for achieving the third objective of the present invention stores a computer program thereon, characterized in that, when the computer program is executed by a processor, it implements any one of the steps of the deep learning-based information retrieval method for power grid control.

[0041] Beneficial effects:

[0042] The method and system described in this invention lower the barrier to operating the power grid database, realize rapid and intelligent retrieval of structured data in the field of power grid control information, save a lot of time writing SQL manually, improve work efficiency, and at the same time increase the depth of data mining and improve the ability to monetize data value. Attached Figure Description

[0043] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0044] The following detailed embodiments are provided to explain the technical solutions of the claims of this invention, so that those skilled in the art can understand the claims. The scope of protection of this invention is not limited to the following specific embodiments. Any modifications made by those skilled in the art that incorporate the technical solutions of the claims but differ from the following detailed embodiments are also within the scope of protection of this invention.

[0045] Furthermore, in the description of this invention, the terms "first," "second," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance.

[0046] The following is combined Figure 1 An embodiment of the system described in this invention is described, including:

[0047] Training data building module: Used to build training data for training deep learning translation models;

[0048] Model training module: Used to train the deep learning translation model based on the training data obtained from the training data construction module, to obtain a Text-to-SQL agent;

[0049] Question parsing module: used to parse the user's input source language into Text-to-SQL intelligent agent-recognized source language text;

[0050] SQL post-processing module: used to convert the target language text output by the Text-to-SQL agent into database executable statements;

[0051] SQL Execution Module: Used to run executable statements obtained from the SQL Post-Processing Module and retrieve the required information from the database.

[0052] Furthermore, it also includes a data preprocessing module for preprocessing the statements input to the deep learning translation model and / or Text-to-SQL agent. The preprocessing includes word segmentation, special character replacement, and adding special characters.

[0053] Furthermore, it also includes a mapping relationship building module for building mapping relationships between database table names and / or column names and natural language representations.

[0054] The modules mentioned above are applied to different stages. This invention includes two stages: the model training stage and the model application stage, such as... Figure 1 As shown, the model training phase includes: training data construction module, data preprocessing module, and model training module; the model application phase includes: question parsing module, SQL post-processing module, and SQL execution module.

[0055] During the model training phase, the training data construction module organizes natural language questions, database table structures, and SQL statements to construct multiple source and target statements for training the deep learning translation model; the data preprocessing module preprocesses the statements constructed in the training data construction module; and the model training module trains the deep learning translation model based on the data obtained from the data preprocessing module to obtain a Text-to-SQL agent, thereby automatically converting natural language questions into SQL statements.

[0056] In the model application phase, the question parsing module preprocesses the natural language questions input by the user to generate source language text; the Text-to-SQL agent of the model training module automatically converts the source language text into target language text; the SQL post-processing module post-processes the target language text output by the Text-to-SQL agent to convert it into SQL statements that can be executed by the database; and the SQL execution module runs the SQL statements obtained by the SQL post-processing module, performs queries within the database engine, and obtains the search results needed by the user.

[0057] The following steps detail one embodiment of the method described in this invention.

[0058] S1. Collect statement pairs of natural language questions and SQL statements from multiple power grid control fields;

[0059] Based on the needs of power grid control operations, we collected pairs of natural language questions and SQL statements. The following is an example of a natural language question and SQL statement pair:

[0060] Statement pair 1:

[0061] Question: What type of power plant is ABC?

[0062] The corresponding SQL query is: SELECT PLANT_TYPE FROM PLANT_BASIC WHERE NAME = 'ABC Power Plant';

[0063] The natural language question and the SQL statement have a one-to-one correspondence, meaning that the information requested by the natural language question can be retrieved by running the SQL statement in the database.

[0064] S2. Convert the collected natural language questions into source statements that can be used to train deep learning translation models; convert the collected SQL statements into target statements expressed in natural language.

[0065] Each natural language question is modeled along with its corresponding candidate database table structure to construct text input feature1. This type of text can be input into a deep learning translation model as its source sentence, with the following format: <q> and< / q> The text between these lines represents the original question. <t> and< / t> The space between them represents the database table structure. <c>Previously, it was the database name. <c>The column names follow, separated by the @ character, in the following format:

[0066] <q> Original question text< / q> | <t>Table 1 Name <c> @field1 @field2< / c> < / t> <t>Table 2 Name <c> @field1 @field2< / c> < / t> ...

[0067] The above statement can be converted into the following text:

[0068] <q> What is the type of ABC power plant?< / q> | <t>Basic information about power plants <c> @Power Plant ID@Power Plant Name@Power Plant Type...< / c> < / t> <t>Basic information about substations <c> @SubstationID@SubstationName...< / c> < / t> ...

[0069] To improve the prediction accuracy of the deep learning model, the source text needs to be preprocessed. This preprocessing includes word segmentation, specifically segmenting the text input feature 1 to obtain the text input feature 2. During word segmentation, special symbols such as... <q>、 <t> 、 <c>Do not perform word segmentation and retain the entire character. After segmenting the above text, the following text is obtained as the source sentence for the training sentences of the deep learning translation model:

[0070] <q> What is the type of ABC power plant?< / q> | <t>Basic information about power plants <c> Power Plant ID@Power Plant Name@Power Plant Type...< / c> < / t> <t>Basic information about substations <c> @SubstationID@SubstationName...< / c> < / t> ……;

[0071] Convert the SQL statement into a target statement expressed in natural language, which specifically includes the following steps:

[0072] S201. Construct the mapping relationship between the database table name and column name and the natural language expression;

[0073] For example, map the table name and column name of "Power Plant Basic Information" to the natural language expression. The front is the table name, or table name.column name, and the back is the meaning of the natural language it represents, as shown below:

[0074] PLANT_BASIC Power Plant Basic Information

[0075] PLANT_BASIC.NAME Power Plant Basic Information.Power Plant Name

[0076] PLANT_BASIC.PLANT_TYPE Power Plant Basic Information.Power Plant Type

[0077] In this step, preprocessing needs to be performed. The preprocessing includes special character replacement. In order to avoid the conflict between the dot in the above mapping relationship and the decimal point in the WHERE query condition of the SQL statement, which affects the training of the deep learning model, it is necessary to replace the dot in the above expression mapping relationship with a character that does not appear in the SQL statement. In this embodiment, it is replaced with @.

[0078] S202. Convert the SQL statement in step S1 into a target statement expressed in natural language according to the above expression mapping relationship;

[0079] After converting the SQL in the above statement pair 1: SELECT PLANT_TYPE FROM PLANT_BASIC WHERE NAME = 'ABC Power Plant', it is as follows:

[0080] Processed SQL:

[0081] SELECT Power Plant Basic Information@Power Plant Type FROM Power Plant Basic Information WHERE Power Plant Basic Information@Power Plant Name = 'ABC Power Plant'

[0082] The SQL processed above is preprocessed to obtain the target statement for inputting into the deep learning translation model. This processing includes word segmentation; during segmentation, keywords in the SQL are not segmented but retained as whole characters, such as SELECT, WHERE, FROM, JOIN, etc.; table names and column names are segmented; the preprocessing also includes adding special characters, adding separators between different components in the SQL. The segmented SQL statement is shown below and serves as the target statement for inputting into the deep learning translation model:

[0083] SELECT Power Plant Basic Information @ Power Plant Type FROM Power Plant Basic Information WHERE Power Plant Basic Information @ Power Plant Name = 'ABC Power Plant'

[0084] S3. Build a deep learning translation model. Use the source and target sentences obtained in the above steps to train the deep learning translation model and obtain a Text-to-SQL agent.

[0085] The model adopts the existing Transformer-based deep learning translation model, which can improve the accuracy of SQL generation compared with other models. The encoder is responsible for encoding the input sequence (i.e., the source statement), and the decoder is responsible for generating the text (target statement). The encoder and decoder are connected through a self-attention mechanism.

[0086] The Text-to-SQL intelligent agent can learn autonomously and translate source statements into target statements.

[0087] S4. Use a Text-to-SQL agent to convert the question expressed in natural language into text in the target language, including the following steps:

[0088] S401. Convert the natural language question input by the user into source language text that can be input into the Text-to-SQL agent;

[0089] For example, a user might enter a natural language question: What is the email address for the EFG power plant?

[0090] Step 1: Based on the mapping relationship between database table names and column names and natural language expressions, convert the natural language text input by the user into source language text that can be input into the Text-to-SQL agent;

[0091] Step 2: Preprocess the source language text that can be input into the Text-to-SQL agent, including word segmentation.

[0092] The converted source language text is shown below:

[0093] <q> What is the EFG power plant email address?< / q> | <t>Basic information about power plants <c> @Power Plant ID@Power Plant Name@Power Plant Type...< / c> < / t> <t>Basic information about substations <c> @SubstationID@SubstationName...< / c> < / t> ...

[0094] S402. Use a Text-to-SQL agent to predict the source language text to obtain the target language text. The target language text adopts the SQL language framework, but the table names and column names are pseudo-SQL statements expressed in natural language. It cannot be executed and needs further processing to convert the natural language into the corresponding tables or columns in the actual database, and to replace special characters with original characters before it can be executed in the database.

[0095] The pseudo-SQL statement is as follows:

[0096] SELECT Power Plant Basic Information@Email Address FROM Power Plant Basic Information WHERE Power Plant Basic Information@Power Plant Name = 'EFG Power Plant'

[0097] S5. Convert the target language text obtained in the above steps into executable SQL statements, including the following steps:

[0098] Step 1: Post-process the target language text, including replacing or deleting special characters.

[0099] The symbols in the target language text above are processed using SQL block segmentation. The word separators between different SQL blocks are removed, resulting in the following statement:

[0100] SELECT Power Plant Basic Information @ Email Address FROM Power Plant Basic Information WHERE Power Plant Basic Information @ Power Plant Name = 'EFG Power Plant'

[0101] Step 2: Based on the mapping relationship between the constructed database table names and column names and the natural language expressions, replace the table names and column names in the result statement with the original English expressions in the database.

[0102] The mapping relationship between the database table names and column names and the natural language representation has been established in the preceding steps.

[0103] The converted result is shown below:

[0104] SELECT PLANT_BASIC.EMAIL FROM PLANT_BASIC WHERE PLANT_BASIC.NAME='EFG Power Plant'

[0105] S6. Run the SQL statement obtained from the above steps to retrieve query information from the database.

[0106] The above executable SQL statement is then executed within the database engine to retrieve the user's desired search results. Example:

[0107] EFG_plant@163.com

[0108] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0109] This application also provides a computer-readable storage medium storing a computer program, which includes program instructions. When executed by a processor, the program instructions implement the various steps of the deep learning-based information retrieval method for power grid control, which will not be described in detail here.

[0110] The computer-readable storage medium can be the data transmission apparatus or the internal storage unit of a computer device provided in any of the foregoing embodiments, such as the hard disk or memory of the computer device. The computer-readable storage medium can also be an external storage device of the computer device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., provided on the computer device.

[0111] Furthermore, the computer-readable storage medium may include both internal storage units and external storage devices of the computer device. The computer-readable storage medium is used to store the computer program and other programs and data required by the computer device. The computer-readable storage medium may also be used to temporarily store data that is to be output or has already been output.

[0112] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied 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.

[0113] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0114] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0115] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0116] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

[0117] The contents not described in detail in this specification are existing technologies known to those skilled in the art.< / c> < / t> < / q> < / c> < / c>

Claims

1. A deep learning-based information retrieval method for power grid control, characterized in that, include: S1. Collect statement pairs of natural language questions and SQL statements from multiple power grid control fields; S2. Convert the collected natural language questions into source statements that can be used to train deep learning translation models; convert the collected SQL statements into target statements expressed in natural language. S3. Build a deep learning translation model. Use the source and target sentences obtained in step S2 to train the deep learning translation model and obtain a Text-to-SQL agent. S4. Use the Text-to-SQL agent to convert questions expressed in natural language into text in the target language; S5. Convert the target language text obtained in S4 into executable SQL statements; S6. Run the SQL statement obtained from S5 to retrieve query information from the database; The collected natural language questions were converted into source sentences that can be used to train deep learning translation models, including: Each natural language question is modeled with its corresponding candidate database table structure to construct text input feature1. The format includes: the original question text between the first label pairs, the database table structure between the second label pairs, the database name before the third label, and the column name after the third label. Different column names are separated by special characters. Text input feature feature1 is segmented into text input feature feature2. During segmentation, the labels are not segmented. The collected SQL statements are converted into target statements expressed in natural language, including: S201. Construct a mapping relationship between database table names and column names and natural language representations; S202. Based on the above-described mapping relationship, convert the SQL statement in step S1 into a target statement expressed in natural language. S202 performs preprocessing on the target statement, including special character replacement; S5 performs post-processing on the target language text to obtain the result statement. The post-processing includes replacing or deleting special characters; based on the mapping relationship between the constructed database table names and column names and the natural language expressions, the table names and column names in the result statement are replaced with the original English expressions in the database. In S2, the source statement undergoes word segmentation, special character replacement, and special character addition. During word segmentation, special characters and SQL keywords SELECT, WHERE, FROM, and JOIN are not segmented; table names and column names are segmented. Special characters include: first tag pairs, second tag pairs, and third tag pairs. During special character replacement, dots in the mapping relationship are replaced with characters that do not appear in the SQL statement. The mapping relationship includes the mapping between information expressed in natural language and the corresponding table or column names in the database. When adding special characters, separators are added between different components in the SQL. The Text-to-SQL agent uses a Transformer-based deep learning translation model, where the encoder is responsible for encoding the source sentence and the decoder is responsible for generating the target sentence. The encoder and decoder are connected through a self-attention mechanism. In S1, based on the needs of business scenarios in the power grid control field, statements of natural language questions and SQL statements are collected; the natural language questions and SQL statements are in a corresponding relationship, and the information to be retrieved by the natural language questions is obtained by running SQL statements in the database. The target statement uses the SQL language framework, and the table names and column names in the target statement are expressed in natural language as pseudo-SQL statements that cannot be executed. The natural language in the pseudo-SQL statement is converted into the corresponding tables or columns in the actual database, and special characters are replaced with original characters that can be executed in the database. In S5, special characters in the target language text are replaced or deleted; SQL blocks are processed for delimiters in the target language text, and delimiters are removed between different SQL blocks.

2. The information retrieval method for power grid control based on deep learning as described in claim 1, characterized in that, Step S4 includes the following steps: S401. Convert the natural language question input by the user to obtain the source language text that can be input into the Text-to-SQL agent; S402. Use a Text-to-SQL agent to predict the source language text and obtain the target language text.

3. The information retrieval method for power grid control based on deep learning as described in claim 2, characterized in that, Step S401 includes the following steps: Step 1: Based on the mapping relationship between database table names and column names and natural language expressions, convert the natural language text input by the user into source language text that can be input into the Text-to-SQL agent; Step 2: Preprocess the source language text that can be input into the Text-to-SQL agent.

4. A deep learning-based information retrieval system for power grid control using the method described in claim 1, characterized in that, include: Training data building module: Used to build training data for training deep learning translation models; Model training module: Used to train the deep learning translation model based on the training data obtained from the training data construction module, to obtain a Text-to-SQL agent; Question parsing module: used to parse the user's input source language into Text-to-SQL intelligent agent-recognized source language text; SQL post-processing module: used to convert the target language text output by the Text-to-SQL agent into database executable statements; SQL Execution Module: Used to run executable statements obtained from the SQL Post-Processing Module and retrieve the required information from the database.

5. The information retrieval system for power grid control based on deep learning as described in claim 4, characterized in that, It also includes a data preprocessing module for preprocessing statements input to deep learning translation models and / or Text-to-SQL agents.

6. The information retrieval system for power grid control based on deep learning as described in claim 4, characterized in that, It also includes a mapping relationship building module, used to build mapping relationships between database table names and / or column names and natural language representations.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the deep learning-based information retrieval method for power grid control as described in any one of claims 1 to 3.

Images