Intelligent generation method and system for stable boundary of power flow section of power grid

Abstract

The invention discloses an intelligent generation method and system for a stable boundary of a power grid flow section. A power grid data processing module preprocesses power grid voice scheduling data and original power grid text scheduling data; the model adjustment module transmits the preprocessed original power grid text scheduling data and the preprocessed power grid text scheduling data to the power flow section stability boundary generation large model, and adjusts the pre-training weight of the power flow section stability boundary generation large model; and the section stability boundary generation module transmits the power grid dispatching data to the power flow section stability boundary intelligent generation model of the multi-role strategy for structured analysis to obtain a structured text of the power grid power flow section stability boundary. According to the method, an end-to-end power flow section stable boundary intelligent generation structured analysis task can be realized based on a multi-role strategy, so that powerful support and guarantee are provided for power grid multi-tense section boundary reasoning work.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a method and system for intelligently generating a stable boundary of a power grid power flow section. Background Art

[0002] With the transformation of global energy structure and the increase of electricity demand, power grid systems are facing increasingly complex operating environments and challenges. The generation of stable boundaries of power flow sections is the core task of safe and stable operation of power systems. Especially in the context of multiple business systems, the intelligent generation of stable boundaries of power flow sections is becoming more and more important. The generation of stable boundaries is the key to ensuring the safe operation of power grids. However, existing methods often rely on static data when identifying and predicting stable boundaries, and fail to fully consider the interrelationships of key information in the text, resulting in inflexibility when dealing with complex working conditions and only being able to handle simple information extraction tasks. In addition, traditional power flow analysis and stability boundary generation often require a lot of manual intervention, consume human resources, and human factors may lead to misjudgment. With the acceleration of power system construction, power grid personnel need to make decisions quickly and accurately to deal with emergencies and load fluctuations. The traditional intelligent generation method of stable boundaries of power flow sections cannot meet the requirements of real-time decision-making, which is prone to safety hazards and economic losses. Therefore, improving the degree of automation and reducing human intervention have become urgent tasks to be solved.

[0003] In order to solve the above problems, this paper proposes a method for intelligent generation of power grid flow section stability boundaries based on a generative pre-training model. First, the power grid dispatching data is obtained, including power grid dispatching voice instructions, power grid dispatching accident plans, UHV interconnected power grid stability and reactive voltage dispatching operation regulations, and safety automatic device dispatching operation management regulations. Then, the power grid dispatching data is input into the large model for generating the flow section stability boundary, and the instructions of the large model are adjusted. Finally, according to the original text of the power grid, the model can complete the end-to-end intelligent generation of the flow section stability boundary structured analysis based on multiple role strategies, providing support for the multi-temporal section boundary reasoning of the power grid. Summary of the invention

[0004] The purpose of the present invention is to provide a method for intelligently generating stable boundaries of power grid flow sections on the one hand, and to provide a system for intelligently generating stable boundaries of power grid flow sections on the other hand. The system and method can realize end-to-end intelligent generation of stable boundaries of power grid flow sections based on a multi-role strategy, thereby providing strong support and guarantee for the reasoning of multi-temporal section boundaries of power grids.

[0005] To achieve this purpose, the present invention designs a power grid power flow section stability boundary intelligent generation system, which includes: a power grid data processing module, a model adjustment module and a section stability boundary generation module;

[0006] The power grid data processing module is used to pre-process the power grid voice dispatch data, convert the power grid voice dispatch data into pre-processed power grid text dispatch data; pre-process the original power grid text dispatch data to obtain the pre-processed original power grid text dispatch data;

[0007] The model adjustment module is used to transmit the preprocessed original power grid text dispatch data and the preprocessed power grid text dispatch data to the large model for generating the stable boundary of the power flow section, and adjust the pre-trained weights of the large model for generating the stable boundary of the power flow section to obtain the intelligent generation model of the stable boundary of the power flow section with a multi-role strategy;

[0008] The section stability boundary generation module is used to transmit the preprocessed original power grid text dispatching data and the preprocessed power grid text dispatching data to the multi-role strategy power flow section stability boundary intelligent generation model for structured analysis to obtain the structured text of the power grid power flow section stability boundary.

[0009] Furthermore, the method for preprocessing the power grid voice dispatching data includes: extracting features from the power grid voice dispatching data using MFCC, normalizing the features when using the MFCC features, and the MFCC formula of the j-th dimension of the i-th frame is as follows:

[0010]

[0011] Where i = 1, 2, ..., I is the speech parameter, j = 1, 2 ..., J m is the MFCC dimension, M is the number of filters, m is the filter, M MFCC F×J m dimensional MFCC feature matrix, S i (m) is the i-th frame speech signal.

[0012] Furthermore, after feature extraction of the power grid voice dispatch data, the extracted features are downsampled, and after linear transformation, they are encoded using multiple feature fusion modules. A CTC layer is added after the feature fusion layer to map the output of the encoder to a character sequence to obtain the final power grid dispatch speech recognition result.

[0013] Furthermore, the method for preprocessing the original power grid text dispatch data includes: firstly splitting the original power grid text dispatch data into paragraphs, uniformly representing the digital data and digital serial numbers in the split text paragraphs, and deleting useless characters in the unstructured text after the uniform representation.

[0014] Furthermore, the pre-trained weights of the large model for generating the stable boundary of the tidal section are adjusted to obtain the method for the intelligent generation model of the stable boundary of the tidal section with a multi-role strategy, including: using the LoRA adjustment technology to adjust the pre-trained weights of the large model for generating the stable boundary of the tidal section.

[0015] Furthermore, the LoRA adjustment technology is used to adjust the pre-trained weights of the large model for generating the stable boundary of the tidal section, specifically including: for a pre-trained weight W of the large model for generating the stable boundary of the tidal section, it is expressed as:

[0016] W′=W+△W

[0017] Among them, W represents the pre-trained weight of the large model for generating the stable boundary of the tidal section, W′ represents the application weight of the large model for generating the stable boundary of the tidal section after adjustment, and △W is the LoRA weight adjusted by the LoRA adjustment technology on the pre-trained weight of the large model for generating the stable boundary of the tidal section;

[0018] △W is represented by the product of low-rank matrix A and low-rank matrix B:

[0019] △W=A·B

[0020] The low-rank matrix A is a parameter for the LoRA adjustment technology to adjust the pre-trained weights of the large model for generating the stable boundary of the tidal section. The low-rank matrix B is a parameter for the LoRA adjustment technology to adjust the pre-trained weights of the large model for generating the stable boundary of the tidal section. The dimension of A is (d×r), and the dimension of B is (r×d). r is a hyperparameter smaller than the dimension d of the pre-trained weight matrix, which represents the rank.

[0021] Furthermore, the preprocessed original power grid text dispatching data and the preprocessed power grid text dispatching data are transmitted to the multi-role strategy power flow section stable boundary intelligent generation model for analysis, and the method for obtaining the structured text of the power grid power flow section stable boundary includes: the multi-role strategy power flow section stable boundary intelligent generation model analyzes the preprocessed original power grid text dispatching data and the preprocessed power grid text dispatching data, and constructs different instruction descriptions according to the power grid dispatching requirements in the preprocessed original power grid text dispatching data and the preprocessed power grid text dispatching data; the multi-role strategy power flow section stable boundary intelligent generation model plays different roles based on different instruction descriptions; different roles cooperate with each other to parse the preprocessed original power grid text dispatching data and the preprocessed power grid text dispatching data, thereby obtaining the structured text of the power grid power flow section stable boundary.

[0022] Further, different instruction descriptions are constructed according to the preprocessed original power grid text dispatch data and the power grid dispatch requirements in the preprocessed power grid text dispatch data, the power flow section stability boundary intelligent generation model of the multi-role strategy plays different roles based on different instruction descriptions, and different roles cooperate with each other to parse the preprocessed original power grid text dispatch data and the preprocessed power grid text dispatch data. The method includes: constructing three instruction descriptions, and the three instruction descriptions are: "Please find the power facilities in the article", "Please find the section boundary requirements in each power facility", and "Please proofread these power facilities and section boundary requirements and fill them into the template";

[0023] The intelligent generation model of power flow section stability boundary of multiple role strategy plays three roles based on three instruction descriptions, namely, the power facility identification role, the power flow section stability boundary requirement identification role and the structured proofreading role. The power facility identification role completes the identification of power facilities in the power grid dispatching text data according to the instruction description of "please find the power facilities in the article". The power flow section stability boundary requirement identification role uses the preprocessed original power grid text dispatching data and the preprocessed power grid text dispatching data to identify the power facilities determined by the power facility identification role, and obtains the power flow section stability boundary requirements of the power facilities. The structured proofreading role accumulates the errors in the power facilities determined by the power facility identification role and the power flow section stability boundary requirements of the power facilities according to the instruction description of "please proofread these power facilities and section boundary requirements and fill them into the template". The power facilities determined by the power facility identification role, the power flow section stability boundary requirements of the power facilities, and the accumulated errors in the power facilities determined by the power facility identification role and the power flow section stability boundary requirements of the power facilities constitute the structured text of the power grid power flow section stability boundary.

[0024] A method for intelligently generating a stable boundary of a power grid flow section based on the power grid flow section stable boundary intelligent generation system comprises:

[0025] Preprocessing the power grid voice dispatching data to convert the power grid voice dispatching data into preprocessed power grid text dispatching data; preprocessing the original power grid text dispatching data to obtain the preprocessed original power grid text dispatching data;

[0026] The preprocessed original power grid text dispatch data and the preprocessed power grid text dispatch data are transmitted to the large model for generating the stable boundary of the power flow section, and the pre-trained weights of the large model for generating the stable boundary of the power flow section are adjusted to obtain the intelligent generation model of the stable boundary of the power flow section with a multi-role strategy;

[0027] The preprocessed original power grid text dispatch data and the preprocessed power grid text dispatch data are transmitted to the intelligent generation model of the power flow section stability boundary of the multi-role strategy for structured analysis to obtain the structured text of the power grid power flow section stability boundary.

[0028] Beneficial effects of the present invention: Under the situation of global energy structure transformation and continuous growth in electricity demand, the operating environment of the power grid system is becoming more and more complex, and the challenges it faces are becoming increasingly severe. The generation of the stable boundary of the power flow section occupies a core position in the safe and stable operation of the power system. Especially in the context of multiple interconnected business systems, the intelligent generation of the stable boundary of the power flow section is extremely critical. After all, the generation of the stable boundary is the top priority to ensure the safe operation of the power grid. Most of the current methods rely only on static data when identifying and predicting the stable boundary, and cannot fully consider the inherent connection between the key information in the text, which makes it lack of sufficient flexibility in the face of complex working conditions and can only handle relatively simple information extraction tasks. The traditional power flow analysis and stability boundary generation process often require a lot of manual intervention, which not only consumes human resources, but also may cause judgment bias due to the existence of human factors.

[0029] In response to the above difficulties, this paper innovatively proposes a large-scale power grid power flow section stability boundary intelligent generation scheme based on a generative pre-training model. The first step is to collect power grid dispatching data, which covers power grid dispatching voice instructions, power grid dispatching accident plans, ultra-high voltage interconnected power grid stability and reactive voltage dispatching operation regulations, and safety automatic device dispatching operation management regulations. The second step is to input the acquired power grid dispatching data into the large model for generating the power flow section stability boundary, and perform instruction adjustment operations on the large model. The third step is to use the original text of the power grid to enable the model to realize the end-to-end power flow section stability boundary intelligent generation structured analysis task based on multiple role strategies, thereby providing strong support and guarantee for the multi-temporal section boundary reasoning work of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a structural schematic diagram of the present invention;

[0031] Figure 2 A step diagram of the model building method of the present invention;

[0032] Figure 3 A diagram showing the steps of a method for power grid dispatching speech recognition according to the present invention;

[0033] Figure 4 It is the overall framework diagram of the intelligent generation method of the stable boundary of the tidal section based on the multi-role strategy of the present invention;

[0034] Figure 5 This is a diagram of the model adjustment method architecture of the present invention. DETAILED DESCRIPTION

[0035] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0036] Example 1

[0037] like Figure 1 As shown, a power grid power flow section stability boundary intelligent generation system includes: a power grid data processing module, a model adjustment module and a section stability boundary generation module;

[0038] The power grid data processing module is used to pre-process the power grid voice dispatch data, convert the power grid voice dispatch data into pre-processed power grid text dispatch data; pre-process the original power grid text dispatch data to obtain the pre-processed original power grid text dispatch data;

[0039] The model adjustment module is used to transmit the preprocessed original power grid text dispatch data and the preprocessed power grid text dispatch data to the large model for generating the stable boundary of the power flow section, and adjust the pre-trained weights of the large model for generating the stable boundary of the power flow section to obtain the intelligent generation model of the stable boundary of the power flow section with a multi-role strategy;

[0040] The section stability boundary generation module is used to transmit the preprocessed original power grid text dispatching data and the preprocessed power grid text dispatching data to the multi-role strategy power flow section stability boundary intelligent generation model for structured analysis to obtain the structured text of the power grid power flow section stability boundary.

[0041] The present invention extracts descriptive information such as power grid operation mode, section flow, equipment status, etc. from power grid dispatching information based on a large model. Power grid dispatching data refers to various information collected during the operation and management of the power system. The power grid dispatching data set covers multiple dimensions such as power grid dispatching voice, power grid dispatching accident plan, UHV interconnected power grid stability and reactive voltage dispatching operation regulations, and safety automatic device dispatching operation management regulations. We collected the original data of power grid dispatching and preprocessed the power grid dispatching data to obtain valid data in the power grid.

[0042] Power grid dispatching speech refers to the voice calls or conference recordings made by dispatchers in their daily work. These recordings contain information such as the communication content between dispatchers and other power grid personnel, operation instructions, and operational decisions. During the communication process, dispatchers and staff usually use pre-defined standard terms to improve communication efficiency. Therefore, a corpus containing various types of recording data can be established for these recording contents. The corpus should at least include common terms for power equipment models, professional descriptions of the dispatching process, and some general information. These corpora can be used for subsequent analysis to correspond the feature-extracted recording content to specific information, where these specific information can be regarded as text attributes.

[0043] The power grid dispatching accident plan refers to the emergency response plan formulated to deal with possible accidents or emergencies during the operation of the power grid system. These plans include faulty equipment, emergency decisions, response measures, backup equipment, and accident and handling process records, so the dispatching accident plan contains power grid description information.

[0044] The UHV interconnected power grid stability and reactive power voltage dispatch operation regulations include information such as system stability requirements, reactive power dispatch, and voltage control strategies.

[0045] The regulations on dispatching and operating management of safety automatic devices refer to the specifications and standards formulated for the dispatching and operating management of safety automatic devices in power systems, including information such as equipment classification and functions, operating detection and data recording, and evaluation and improvement mechanisms.

[0046] In the above technical solution, the method for preprocessing the power grid voice dispatching data includes: using MFCC to extract features of the power grid voice dispatching data, and normalizing the features when using MFCC features. The MFCC formula of the j-th dimension of the i-th frame is as follows:

[0047]

[0048] Where i = 1, 2, ..., I is the speech parameter, j = 1, 2 ..., J m is the MFCC dimension, M is the number of filters, m is the filter, M MFCC F×J m dimensional MFCC feature matrix, S i (m) is the i-th frame speech signal.

[0049] In order to extract effective information of power grid dispatching speech, the present invention adopts Transformer-CTC model to recognize power grid dispatching speech. The recognition method architecture is as follows: Figure 3 shown.

[0050] First, it is necessary to extract features from the input speech signal, and the present invention uses MFCC for feature extraction. MFCC is a commonly used feature extraction method for speech signal processing. This method first pre-emphasizes the audio signal and enhances the high-frequency part. Then, the signal is framed and a window function is applied to reduce the edge effect. Next, a fast Fourier transform is performed to obtain a frequency domain representation, and the spectrum is converted to a Mel frequency scale to better capture the auditory characteristics of the human ear. Subsequently, the logarithm of each frequency component of the Mel spectrum is taken to emphasize energy changes, and finally a discrete cosine transform is applied to extract MFCC features. Generally, 20-40 filters are used, and when using MFCC features, the features need to be normalized to ensure that the importance of different feature dimensions is the same.

[0051] In the above technical solution, after feature extraction of the power grid voice dispatch data, the extracted features are downsampled, and after linear transformation, they are encoded using multiple feature fusion modules, and a CTC layer is added after the feature fusion layer to map the output of the encoder to a character sequence to obtain the final power grid dispatch speech recognition result.

[0052] Each feature fusion module contains a self-attention layer, a feedforward neural network layer, and a convolutional layer, and adopts the structure of a Transformer encoder. The vector feature sequence is input into the feedforward network, which consists of two linear transformation layers and a switch activation function, introducing nonlinear transformations to better capture the complex relationship of the input features. Placing the feedforward network module before the multi-head attention module enables the model to fully utilize the multi-head attention module's ability to model global context, promote the effective transfer of information, and alleviate the gradient vanishing problem.

[0053] Finally, a CTC layer is added after the feature fusion layer to map the encoder output to a character sequence. The CTC layer uses the CTC loss function for model training, does not require alignment labels, and can handle input and output sequences of variable length. During the training process, the CTC decoder is used to decode the output of the CTC layer to obtain the final power grid dispatch speech recognition result.

[0054] In the above technical scheme, the method for preprocessing the original power grid text dispatching data includes: firstly splitting the original power grid text dispatching data into paragraphs, uniformly representing the digital data and digital serial numbers in the split text paragraphs, and deleting useless characters in the unstructured text after the unified representation.

[0055] Traditional entity relationship extraction tasks are often aimed at simple text extraction tasks, and the extracted relationship entities are usually commonly used words. However, the relationship entity information structure that needs to be extracted in this paper is relatively complex. In order to solve the above problems and complete the power grid dispatching data analysis task, the entities that need to be extracted in this paper include: "power facilities", "control requirements", "control basis", "operation content", and "operating status and conditions".

[0056] For example, the original text of the power grid dispatching text is "Based on the 600 kV bus voltage of Beiyue Station, the sum of the reactive power exchange between the main transformer and the 600 kV system, switch on and off the 120 kV low-voltage reactive compensation equipment in the station, and control the sum of the reactive power exchange between the Beiyue main transformer and the 600 kV system to not exceed 200,000 kVar. When the system encounters the following situations, the 120 kV reactive compensation equipment of Beiyue Station will be switched on and off: (1) The sum of the reactive power exchange between the Beiyue main transformer and the 600 kV system exceeds 200,000 kVar; (2) The 600 kV bus voltage of Beiyue exceeds the limit."

[0057] The extraction results are as follows: Result 1: {“Power Facilities”: “Beiyue Station”, “Control Requirements”: “The sum of the reactive power exchange between Beiyue main transformer and the 600 kV system shall not exceed 200,000 kVar”, “Control Basis”: “The sum of the 600 kV bus voltage, main transformer and 600 kV system reactive power exchange at this station”, “Operation Content”: “Switch on and off the 110 kV low-voltage reactive power compensation equipment at the station”}. Result 2: {“Power Facilities”: “System”, “Operation Status and Conditions”: “[“The sum of the reactive power exchange between Beiyue main transformer and the 600 kV system exceeds 200,000 kVar”, “Beiyue 600 kV bus voltage exceeds the limit”]”, “Operation Content”: “120 kV reactive power compensation equipment at Beiyue Station is switched on and off”}.

[0058] Preprocessing the text data of power grid dispatch is a key step to improve the effect of subsequent analysis and model training. First, the dispatch data text is split into paragraphs, which can structure the information for subsequent processing. Each paragraph usually contains relevant dispatch instructions, operating status or operation decisions, which helps to retain contextual information.

[0059] Next, the numbers and serial numbers in the intercepted text paragraphs are uniformly represented. For example, serial numbers such as "1." and "2." can be uniformly converted to "first", "second", etc., and numbers can be standardized, which helps to eliminate ambiguity caused by different expressions.

[0060] Finally, data cleaning is a necessary step. The cleaning process includes deleting useless characters in unstructured text, such as line breaks, spaces, special symbols, etc. These useless characters often interfere with subsequent text analysis and models. By removing these noises, the quality of the power grid dispatch text paragraphs to be identified can be effectively improved, thereby enhancing the accuracy of the intelligent generation of the stability boundary of the large power grid power flow section.

[0061] In the above technical scheme, the pre-trained weights of the large model for generating the stable boundary of the tidal section are adjusted to obtain the method for the intelligent generation model of the stable boundary of the tidal section with a multi-role strategy, including: using the LoRA adjustment technology to adjust the pre-trained weights of the large model for generating the stable boundary of the tidal section.

[0062] LoRA (Low-Rank Adaptation) fine-tuning technology is an efficient model fine-tuning method that aims to reduce training costs by reducing the number of training parameters. The core idea is to decompose the weight matrix into the sum of low-rank matrices, so as to fine-tune only the low-rank part. The main architecture is as follows Figure 5 shown.

[0063] This paper uses ChatGLM3 as a pre-trained model. The pre-trained model has already learned and has common features. For specific tasks, it only needs to be fine-tuned on the pre-trained model without training from scratch. LoRA trains two low-rank matrices.

[0064] In the above technical solution, the LoRA adjustment technology is used to adjust the pre-trained weights of the large model for generating the stable boundary of the tidal section, specifically including: for a pre-trained weight W of the large model for generating the stable boundary of the tidal section, it is expressed as:

[0065] W′=W+△W

[0066] Among them, W represents the pre-trained weight of the large model for generating the stable boundary of the tidal section, W′ represents the application weight of the large model for generating the stable boundary of the tidal section after adjustment, and △W is the LoRA weight adjusted by the LoRA adjustment technology on the pre-trained weight of the large model for generating the stable boundary of the tidal section;

[0067] △W is represented by the product of low-rank matrix A and low-rank matrix B:

[0068] △W=A·B

[0069] The low-rank matrix A is a parameter for the LoRA adjustment technology to adjust the pre-trained weights of the large model for generating the stable boundary of the tidal section. The low-rank matrix B is a parameter for the LoRA adjustment technology to adjust the pre-trained weights of the large model for generating the stable boundary of the tidal section. The dimension of A is (d×r), and the dimension of B is (r×d). r is a hyperparameter smaller than the dimension d of the pre-trained weight matrix, which represents the rank.

[0070] For text data input, the unified definition is JSON format. The content of JSON data is different for different roles. For example, in the power facility identification role, the JSON content is, {"instruction":"You are now a power facility identification model. Please help me find the power facilities in the article. If there are multiple power facilities, use "_" to connect.","input":"According to the 600 kV bus voltage of Beiyue Station, the sum of the reactive power exchange between the main transformer and the 600 kV system, the 120 kV low-voltage reactive compensation equipment on the station side is switched on and off to control the sum of the reactive power exchange between the Beiyue main transformer and the 600 kV system to not exceed 200,000 kvar. When the system encounters the following situations, the 120 kV reactive compensation equipment of Beiyue Station is switched on and off: (1) The sum of the reactive power exchange between the Beiyue main transformer and the 600 kV system exceeds 200,000 kvar; (2) The 600 kV bus voltage of Beiyue exceeds the limit.","output":"Beiyue Station_System"}.

[0071] The JSON content in the role of identifying the boundary requirements of the flow section stability is, {"instruction":"You are now a flow section stability boundary requirement identification model. Please help me find the flow section stability boundary requirements in the article.","input":"According to the 600 kV bus voltage of Beiyue Station, the sum of the reactive exchange between the main transformer and the 600 kV system, the 120 kV low-voltage reactive compensation equipment on the station side is switched on and off to control the sum of the reactive exchange between the Beiyue main transformer and the 600 kV system to be no more than 200,000 kvar. When the system encounters the following situations, the 120 kV reactive compensation equipment of Beiyue Station is switched on and off: (1) The sum of the reactive exchange between the Beiyue main transformer and the 600 kV system exceeds 200,000 kvar; (2) Beiyue 600 kV bus voltage exceeds the limit. ","output":"{"Beiyue Station":{"Control requirements":"The sum of the reactive power exchange between Beiyue main transformer and 600 kV system shall not exceed 200,000 kVar","Control basis":"The sum of the 600 kV bus voltage, main transformer and 600 kV system reactive power exchange at this station","Operation content":"Switch on and off the 110 kV low-voltage reactive power compensation equipment at the station"},"System":{"Operation status and conditions":"["The sum of the reactive power exchange between Beiyue main transformer and 600 kV system exceeds 200,000 kVar","Beiyue 600 kV bus voltage exceeds the limit"]","Operation content":"Switch on and off the 120 kV reactive power compensation equipment at Beiyue Station"}}.

[0072] In the above technical scheme, the preprocessed original power grid text dispatching data and the preprocessed power grid text dispatching data are transmitted to the multi-role strategy power flow section stability boundary intelligent generation model for analysis, and the method for obtaining the structured text of the power grid power flow section stability boundary includes: the multi-role strategy power flow section stability boundary intelligent generation model analyzes the preprocessed original power grid text dispatching data and the preprocessed power grid text dispatching data, and constructs different instruction descriptions according to the power grid dispatching requirements in the preprocessed original power grid text dispatching data and the preprocessed power grid text dispatching data. The multi-role strategy power flow section stability boundary intelligent generation model plays different roles based on different instruction descriptions, and different roles cooperate with each other to parse the preprocessed original power grid text dispatching data and the preprocessed power grid text dispatching data, thereby obtaining the structured text of the power grid power flow section stability boundary.

[0073] In order to quickly extract key information from power grid dispatch text, this paper proposes an intelligent generation method for stable boundary of power flow section based on multi-role strategy based on ChatGLM-6B large language model and LoRA fine-tuning. This method breaks down complex tasks into several simple tasks, and allows the model to play different roles through different task descriptions. These roles cooperate with each other to effectively parse power grid dispatch data. The extracted key information will be converted into structured data for subsequent visualization and further analysis. The overall framework is as follows Figure 4 shown.

[0074] In the above technical scheme, different instruction descriptions are constructed according to the preprocessed original power grid text dispatch data and the power grid dispatch requirements in the preprocessed power grid text dispatch data, the power flow section stable boundary intelligent generation model of the multi-role strategy plays different roles based on different instruction descriptions, and different roles cooperate with each other to parse the preprocessed original power grid text dispatch data and the preprocessed power grid text dispatch data. The method includes: constructing three instruction descriptions, and the three instruction descriptions are: "Please find the power facilities in the article", "Please find the section boundary requirements in each power facility", and "Please proofread these power facilities and section boundary requirements and fill them into the template";

[0075] The intelligent generation model of power flow section stability boundary of multiple role strategy plays three roles based on three instruction descriptions, namely, the power facility identification role, the power flow section stability boundary requirement identification role and the structured proofreading role. The power facility identification role completes the identification of power facilities in the power grid dispatching text data according to the instruction description of "please find the power facilities in the article". The power flow section stability boundary requirement identification role uses the preprocessed original power grid text dispatching data and the preprocessed power grid text dispatching data to identify the power facilities determined by the power facility identification role, and obtains the power flow section stability boundary requirements of the power facilities. The structured proofreading role accumulates the errors in the power facilities determined by the power facility identification role and the power flow section stability boundary requirements of the power facilities according to the instruction description of "please proofread these power facilities and section boundary requirements and fill them into the template". The power facilities determined by the power facility identification role, the power flow section stability boundary requirements of the power facilities, and the accumulated errors in the power facilities determined by the power facility identification role and the power flow section stability boundary requirements of the power facilities constitute the structured text of the power grid power flow section stability boundary.

[0076] like Figure 4 As shown in the figure, this paper constructs different instruction descriptions based on the requirements of the three tasks, so that the model plays different roles based on different instruction descriptions, and different roles write to each other to jointly complete the ultimate goal of analyzing the stability boundary of the power grid flow section. The division of labor of the three roles is as follows:

[0077] ①Identification role of power facilities

[0078] The power facility identification role needs to complete the power facilities in the power grid dispatching text data, including power facilities such as "plant station", "busbar", "line", "transformer", "switch", and "unit", for example, "Three Gorges Left Bank Power Plant 500 kV busbar" and "Beiyue Station".

[0079] ② The stability boundary of the tidal section requires identification of roles

[0080] The role of identifying the stable boundary requirements of a power section needs to take the power dispatching text data and the power facilities output by the power facility identification model as input, and identify the stable boundary requirements of the power section of the power facilities according to the content of the power dispatching text, including the stable boundary requirements of the power section such as "control requirements", "control basis", "operation content", "operating status and conditions", etc. The stable boundary requirements of the power section in each aspect can be empty or there can be multiple stable boundary requirements of the power section, which are stored in the form of a list.

[0081] ③Structured proofreading role

[0082] Considering the possible error accumulation in the power facility identification and flow section stability boundary identification tasks, for example, the identification results of the power facility generated by the power facility identification model may be inconsistent with the power facility identification model of the flow section stability boundary requirement model. Therefore, the structured proofreading model is used to integrate and proofread the results of the first two models. This model ensures that each power facility meets the classification results of power facility identification. On the other hand, plain text information is not conducive to further analysis and visualization of power grid dispatching data. Therefore, the structured proofreading model fills the proofreading results into the preset JSON structure template, thereby outputting structured information that can be parsed by JSON, as shown in the figure above.

[0083] Example 2

[0084] like Figure 2 As shown, the steps of the intelligent generation method of the stable boundary of the power grid flow section are as follows:

[0085] 1. Construct the power grid dispatching data and pre-process the power grid text dispatching data and power grid voice dispatching data;

[0086] 2. Input the grid dispatching data into the stable boundary of the power flow section to generate the Chat-CLM3 large model, and use the LoRA method to fine-tune the instructions of the Chat-CLM3 large model;

[0087] 3. Construct a multi-role strategy power flow section stability boundary intelligent generation model, generate structured text, and complete the intelligent generation of power flow section stability boundary of large power grid.

[0088] The method for intelligently generating a stable boundary of a power grid flow section based on the intelligent generation system for the stable boundary of a power grid flow section comprises:

[0089] Preprocessing the power grid voice dispatching data to convert the power grid voice dispatching data into preprocessed power grid text dispatching data; preprocessing the original power grid text dispatching data to obtain the preprocessed original power grid text dispatching data;

[0090] The preprocessed original power grid text dispatch data and the preprocessed power grid text dispatch data are transmitted to the large model for generating the stable boundary of the power flow section, and the pre-trained weights of the large model for generating the stable boundary of the power flow section are adjusted to obtain the intelligent generation model of the stable boundary of the power flow section with a multi-role strategy;

[0091] The preprocessed original power grid text dispatch data and the preprocessed power grid text dispatch data are transmitted to the intelligent generation model of the power flow section stability boundary of the multi-role strategy for structured analysis to obtain the structured text of the power grid power flow section stability boundary.

[0092] Example 3

[0093] A computer program product comprises a computer program, wherein when the computer program is executed by a processor, the steps of the method described in Embodiment 2 are implemented.

[0094] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0095] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take 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.

[0096] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A system that specifies the functions of a box or multiple boxes.

[0097] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction system, which is implemented in the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0098] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0099] The contents not described in detail in this specification belong to the prior art known to professional and technical personnel in this field.

Claims

1. A power grid power flow section stability boundary intelligent generation system, characterized in that: It includes: Power grid data processing module, model adjustment module and section stability boundary generation module; The power grid data processing module is used to pre-process the power grid voice dispatch data and convert the power grid voice dispatch data into pre-processed power grid text dispatch data; Preprocessing the original power grid text dispatching data to obtain preprocessed original power grid text dispatching data; The model adjustment module is used to transmit the preprocessed original power grid text dispatch data and the preprocessed power grid text dispatch data to the large model for generating the stable boundary of the power flow section, and adjust the pre-trained weights of the large model for generating the stable boundary of the power flow section to obtain the intelligent generation model of the stable boundary of the power flow section with a multi-role strategy; The section stability boundary generation module is used to transmit the preprocessed original power grid text dispatching data and the preprocessed power grid text dispatching data to the multi-role strategy power flow section stability boundary intelligent generation model for structured analysis to obtain the structured text of the power grid power flow section stability boundary.

2. According to claim 1, a power grid power flow section stability boundary intelligent generation system is characterized by: The method for preprocessing the power grid voice dispatching data includes: extracting features of the power grid voice dispatching data using MFCC, and normalizing the features when using the MFCC features. The MFCC formula of the j-th dimension of the i-th frame is as follows: Where i = 1, 2, ..., I is the speech parameter, j = 1, 2 ..., J m is the MFCC dimension, M is the number of filters, m is the filter, M MFCC F×J m dimensional MFCC feature matrix, S i (m) is the i-th frame speech signal.

3. The intelligent generation system of power grid power flow section stability boundary according to claim 2 is characterized by: After feature extraction of the power grid voice dispatch data, the extracted features are downsampled and encoded using multiple feature fusion modules after linear transformation. A CTC layer is added after the feature fusion layer to map the output of the encoder to a character sequence to obtain the final power grid dispatch speech recognition result.

4. The intelligent generation system of power grid power flow section stability boundary according to claim 1 is characterized by: The method for preprocessing original power grid text dispatch data includes: firstly splitting the original power grid text dispatch data into paragraphs, uniformly representing the digital data and digital serial numbers in the split text paragraphs, and deleting useless characters in the unstructured text after the uniform representation.

5. The intelligent generation system of power grid power flow section stability boundary according to claim 1 is characterized by: The method for adjusting the pre-trained weights of the large model for generating the stable boundary of the tidal section to obtain the intelligent generation model of the stable boundary of the tidal section with a multi-role strategy includes: adjusting the pre-trained weights of the large model for generating the stable boundary of the tidal section by using the LoRA adjustment technology.

6. The intelligent generation system of power grid power flow section stability boundary according to claim 1 is characterized by: The LoRA adjustment technology is used to adjust the pre-trained weights of the large model for generating the stable boundary of the tidal section, specifically including: for a pre-trained weight W of the large model for generating the stable boundary of the tidal section, it is expressed as: W′=W+△W Among them, W represents the pre-trained weight of the large model for generating the stable boundary of the tidal section, W′ represents the application weight of the large model for generating the stable boundary of the tidal section after adjustment, and △W is the LoRA weight adjusted by the LoRA adjustment technology on the pre-trained weight of the large model for generating the stable boundary of the tidal section; △W is represented by the product of low-rank matrix A and low-rank matrix B: △W=A·B The low-rank matrix A is a parameter for the LoRA adjustment technology to adjust the pre-trained weights of the large model for generating the stable boundary of the tidal section. The low-rank matrix B is a parameter for the LoRA adjustment technology to adjust the pre-trained weights of the large model for generating the stable boundary of the tidal section. The dimension of A is (d×r), and the dimension of B is (r×d). r is a hyperparameter smaller than the dimension d of the pre-trained weight matrix, which represents the rank.

7. The intelligent generation system of power grid power flow section stability boundary according to claim 1 is characterized by: The preprocessed original power grid text dispatching data and the preprocessed power grid text dispatching data are transmitted to the multi-role strategy power flow section stable boundary intelligent generation model for analysis, and the method for obtaining the structured text of the power grid power flow section stable boundary includes: the multi-role strategy power flow section stable boundary intelligent generation model analyzes the preprocessed original power grid text dispatching data and the preprocessed power grid text dispatching data, and constructs different instruction descriptions according to the power grid dispatching requirements in the preprocessed original power grid text dispatching data and the preprocessed power grid text dispatching data; the multi-role strategy power flow section stable boundary intelligent generation model plays different roles based on different instruction descriptions, and different roles cooperate with each other to parse the preprocessed original power grid text dispatching data and the preprocessed power grid text dispatching data, so as to obtain the structured text of the power grid power flow section stable boundary.

8. The intelligent generation system of power grid power flow section stability boundary according to claim 7 is characterized by: According to the preprocessed original power grid text dispatch data and the power grid dispatch requirements in the preprocessed power grid text dispatch data, different instruction descriptions are constructed, and the power flow section stable boundary intelligent generation model of the multi-role strategy plays different roles based on different instruction descriptions, and different roles cooperate with each other to parse the preprocessed original power grid text dispatch data and the preprocessed power grid text dispatch data. The method includes: constructing three instruction descriptions, and the three instruction descriptions are: "please find the power facilities in the article", "please find the section boundary requirements in each power facility", and "please proofread these power facilities and section boundary requirements and fill them into the template"; The intelligent generation model of power flow section stability boundary of multiple role strategy plays three roles based on three instruction descriptions, namely, the power facility identification role, the power flow section stability boundary requirement identification role and the structured proofreading role. The power facility identification role completes the identification of power facilities in the power grid dispatching text data according to the instruction description of "please find the power facilities in the article". The power flow section stability boundary requirement identification role uses the preprocessed original power grid text dispatching data and the preprocessed power grid text dispatching data to identify the power facilities determined by the power facility identification role, and obtains the power flow section stability boundary requirements of the power facilities. The structured proofreading role accumulates the errors existing in the power facilities determined by the power facility identification role and the power flow section stability boundary requirements of the power facilities according to the instruction description of "please proofread these power facilities and section boundary requirements and fill them into the template". The power facilities determined by the power facility identification role, the power flow section stability boundary requirements of the power facilities, and the accumulated errors in the power facilities determined by the power facility identification role and the power flow section stability boundary requirements of the power facilities constitute the structured text of the power grid power flow section stability boundary.

9. The method for intelligently generating a stable boundary of a power grid power flow section according to claim 1, characterized in that: It includes: Preprocessing the power grid voice dispatching data to convert the power grid voice dispatching data into preprocessed power grid text dispatching data; preprocessing the original power grid text dispatching data to obtain the preprocessed original power grid text dispatching data; The preprocessed original power grid text dispatch data and the preprocessed power grid text dispatch data are transmitted to the large model for generating the stable boundary of the power flow section, and the pre-trained weights of the large model for generating the stable boundary of the power flow section are adjusted to obtain the intelligent generation model of the stable boundary of the power flow section with a multi-role strategy; The preprocessed original power grid text dispatch data and the preprocessed power grid text dispatch data are transmitted to the intelligent generation model of the power flow section stability boundary of the multi-role strategy for structured analysis to obtain the structured text of the power grid power flow section stability boundary.

10. A computer program product, comprising a computer program / instruction, which implements the steps of the method according to claim 9 when executed by a processor.

Images