A Method and System for Calculating and Verifying Protection Setting Values ​​Based on Large Model Parameter Identification

Abstract

This invention discloses a method for verifying protection setting calculations based on large model parameter identification. Specifically, the method comprises the following steps: Step 1, obtaining the file to be verified and parsing it into a structured document; Step 2, segmenting the structured document into parts for original equipment parameters, setting calculations, and setting values; Step 3, constructing prompt text and training it using a large language model to obtain structured power equipment parameter data; Step 4, constructing a power grid topology model; Step 5, calculating the setting calculation results and setting values; Step 6, extracting the setting calculation and setting value parts from Step 2 and comparing them with the corresponding setting calculation and setting value results from Step 5, and outputting a verification result report. This invention also discloses a protection setting calculation verification system based on large model parameter identification, which solves the problems of high workload, low verification efficiency, and high error rate associated with manual verification of protection setting calculation sheets in existing technologies.

Description

Technical Field

[0001] This invention belongs to the field of power system automation technology, specifically relating to a method for calculating and verifying protection setting values ​​based on large model parameter identification, and also relating to a system for calculating and verifying protection setting values ​​based on large model parameter identification. Background Technology

[0002] With the rapid construction and large-scale operation of new energy power stations (including wind farms, photovoltaic power stations, and hybrid energy power stations), the accuracy of the relay protection setting calculation sheets directly affects the safe and stable operation of the power grid. The protection setting calculation sheets are the core basis for protection device setting work, covering per-unit calculations of equipment parameters (such as per-unit values ​​of positive-sequence impedance and zero-sequence impedance), short-circuit current calculations (including symmetrical and asymmetrical calculations for fault types such as phase-to-phase short circuits and single-phase grounding), and protection setting calculations (including setting values ​​for various setting principles and recommended setting values). To ensure the correctness of the protection setting calculation sheets and setting lists, they usually need to be verified.

[0003] Currently, the data collection and parameter identification for protection setting calculation sheets and setting sheets rely entirely on manual methods: Verification personnel must review scanned PDF files page by page, verifying the setting calculation process based on their professional knowledge. The verification process includes checking the electrical parameters of the protected equipment, its positive sequence impedance per unit value, zero sequence impedance per unit value, short-circuit point positive sequence impedance, zero sequence impedance, and phase-to-phase fault and ground fault short-circuit currents, comparing the calculation results with the setting sheet. Verification personnel must calculate the various settings of the protection device to be verified, and for each protection setting, calculate its upper and lower limits according to the protection setting procedure, then compare them with the actual set values ​​in the setting sheet to verify whether they are within a reasonable range.

[0004] A single calibration calculation sheet typically contains 50 or even hundreds of independent calculation steps. Manual verification by checkers takes 3-8 hours per sheet, resulting in a heavy workload and low efficiency. Furthermore, manual verification is prone to errors due to clerical mistakes, data entry errors, or lack of experience, leading to a high error rate. The calculation steps, data input basis, and decision-making logic in the verification process are untraceable, making error localization difficult, and the same problem may recur, increasing the difficulty and cost of verification. In summary, existing manual verification methods are insufficient to meet the calculation verification requirements for rapid and reliable grid connection of new energy power plants. Summary of the Invention

[0005] One objective of this invention is to provide a method for calculating and verifying protection setting values ​​based on large model parameter identification. This method utilizes large model intelligent parameter identification and automatic calculation technology for protection settings to solve the problems of large workload, low verification efficiency, and high error rate caused by manual verification in the prior art.

[0006] Another objective of this invention is to provide a protection setting calculation and verification system based on large model parameter identification.

[0007] The technical solution adopted in this invention is a protection setting calculation and verification method based on large model parameter identification, which is implemented according to the following steps: Step 1: Obtain the file to be verified, parse the file, and convert it into a structured document; Step 2: Segment the structured document into three parts: original equipment parameters, tuning calculation, and setpoints. Step 3: Construct prompt text, train it using a large language model, and obtain structured power equipment parameter data; Step 4: Construct a power grid topology model; Step 5: Using a symbolic computation engine, the tuning calculation results and setpoint results are obtained; Step 6: Extract the tuning calculation part and the setting part from Step 2, compare them with the corresponding tuning calculation results and setting results from Step 5, and output the verification result report.

[0008] In step 1, the protection setting calculation sheet to be verified is a PDF file. An OCR tool is used to convert the PDF file into a structured document. The structured document is Markdown text data including titles, paragraphs, tables, and formulas.

[0009] The specific process of step 2 is as follows: Step 2.1: Filter the Markdown text data in Step 1 to remove text content that is irrelevant to the proofreading. Text content that is irrelevant to the proofreading includes soft padding and control words. Step 2.2: Reduce text length by splitting the Markdown text data into paragraphs and by length; Step 2.3: Divide the Markdown text data into the original device parameters, the tuning calculation, and the setpoints.

[0010] In step 3, the large language model uses any one of qwen3, deepseek, glm, or gpt. Through prompt words, the large language model is guided to output structured power equipment parameter data in a predetermined format.

[0011] Step 4 is as follows: Step 4.1: By combining structured power equipment parameter data with power grid expert knowledge, the connection relationships of the power equipment are obtained; Step 4.2: Treat the power equipment as nodes in the power grid topology graph, the power equipment parameter data as node attributes, and the electrical connections between the equipment as edges in the power grid topology graph.

[0012] The setting calculation results in step 5 include the positive sequence and zero sequence impedance of each power equipment, the positive sequence and zero sequence impedance of each short circuit point, the short circuit current, and the short circuit current of the transformer branch; the setting results include the protection setting range of each power equipment, and the protection setting range includes the lower limit, upper limit, and recommended value.

[0013] Step 6 specifically involves: Step 6.1: Calculate the percentage deviation between the tuning calculation part and the setting part in Step 2 and the corresponding tuning calculation result and setting result; Step 6.2: Determine whether there are any abnormalities in the setpoints based on the calculated thresholds. Setpoints with deviations exceeding the thresholds are marked as abnormal. Analyze the causes of abnormal setpoints to determine if they are due to incorrect parameter input, differences in calculation methods, or equipment configuration issues. Finally, generate a verification result report, which includes a summary table of parameter extraction results, a summary table of standard calculation results, a deviation comparison analysis table, an abnormal setpoint warning list, and setpoint adjustment suggestions.

[0014] Another technical solution adopted in this invention is a relay protection setting calculation and verification system, comprising: a file receiving module, a file conversion module, a segmentation and display module, a verification management module, and an output module. The file receiving module is used to receive the protection setting calculation sheet to be verified; the file conversion module is used to convert the received protection setting calculation sheet to be verified into a structured document; the segmentation and display module is used to segment the structured document and arrange and display the segmentation results; the verification management module is used to perform verification calculations; and the output module is used to output the verification results.

[0015] Another feature of the present invention is that the verification management module includes a setting calculation module and a setting value verification module. The setting calculation module is used to calculate the setting value and compare the setting calculation part of the protection setting calculation book with the setting value setting calculation result. The setting value verification module is used to verify the setting value sheet result and compare the setting value item in the setting value sheet with the calculated setting value result.

[0016] The beneficial effects of this invention are: (1) This invention supports both offline and online document parsing modes through OCR document parsing, which can adapt to application scenarios with different security level requirements and has good flexibility and applicability; (2) The present invention uses a general large language model for parameter extraction, which can accurately understand the professional terms and calculation formulas in the field of protection tuning, overcome the limitations of general document processing solutions that lack professional knowledge, and improve the accuracy of parameter extraction; (3) The present invention constructs a power grid topology model, obtains the setting calculation results and setting results, realizes the standardized generation of the verification basis, and ensures the reliability and consistency of the verification results; (4) The present invention outputs a verification result report, which makes it easier for verification personnel to quickly locate problems and make decisions, thereby improving the quality and traceability of verification work.

[0017] This invention achieves fully automated relay protection setting calculation and verification by deeply integrating large language model parameter extraction technology with power grid topology analysis and protection setting calculation technology. It realizes full-process automation from structured parameter extraction of PDF format protection setting calculation document, setting calculation and verification, and output of calculation and verification results, which greatly improves the efficiency of verification work, realizes automated and highly reliable verification, and improves efficiency, accuracy and traceability. Detailed Implementation

[0018] The present invention will now be described in detail with reference to specific embodiments. The described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0019] Example 1 This invention relates to a protection setting calculation and verification method based on large-model electrical parameter identification, which is implemented according to the following steps: Step 1: Obtain the file to be verified, parse the file, and convert it into a structured document; Step 2: Segment the structured document into three parts: original equipment parameters, tuning calculation, and setpoints. Step 3: Extract the original parameters of the equipment from Step 2, construct the prompt text, and train it using a large language model to obtain structured power equipment parameter data; Step 4: Construct a power grid topology model using structured power equipment parameter data; Step 5: Based on the power grid topology model, a symbolic computing engine is used to obtain the setting calculation results and setting results; Step 6: Extract the tuning calculation part and the setting part from Step 2, compare them with the corresponding tuning calculation results and setting results from Step 5, and output the verification result report.

[0020] Example 2 This invention relates to a protection setting calculation and verification method based on large model parameter identification, which is implemented according to the following steps: Step 1: Obtain the protection setting calculation sheet to be verified, parse the protection setting calculation sheet, and convert it into a structured document; The documents to be verified include two types: protection setting calculation sheet and setting value sheet, both of which are PDF files. The PDF files are converted into structured documents using an OCR tool. The structured documents are Markdown text data including titles, paragraphs, tables, and formulas. Step 2: Preprocess the structured document by dividing it into the original equipment parameters section, the tuning calculation section, and the setting value section; Step 2.1: Filter the Markdown text data in Step 1 to remove text content that is irrelevant to the proofreading. Text content that is irrelevant to the proofreading includes soft padding and control words. Step 2.2: Reduce text length by splitting the Markdown text data into paragraphs and by length; Step 2.3: For the protection setting calculation sheet, divide the Markdown text data into the original equipment parameter section and the setting calculation section; for the setting sheet, identify the setting section.

[0021] Step 3: Extract the original parameters of the equipment from Step 2, construct the prompt text, and train it using a large language model to obtain structured power equipment parameter data; The large language model can be any one of qwen3, deepseek, glm, or gpt. By calling the large language model interface through prompt text, the model is trained by inputting prompt text and is guided to output structured power equipment parameter data in a predetermined JSON format. The JSON output is then validated for format and semantics, checking whether required fields are complete, numerical formats are correct, and parameter ranges are reasonable. If the validation fails, correct prompts are automatically constructed, and the large language model is called again for correction until valid structured parameter data is obtained.

[0022] Step 4: Construct a power grid topology model; Step 4.1: By combining structured power equipment parameter data with power grid expert knowledge, the connection relationships of the power equipment are obtained; Step 4.2: Treat the power equipment as nodes in the power grid topology graph, the power equipment parameter data as node attributes, and the electrical connections between the equipment as edges in the power grid topology graph.

[0023] Step 5: Based on the power grid topology model, a symbolic computing engine is used to obtain the setting calculation results and setting results; The setting calculation results include the positive sequence and zero sequence impedances of each power device, the positive sequence and zero sequence impedances of each short-circuit point, the short-circuit current, and the short-circuit current of the transformer branch. The setting results include the protection setting ranges for each power device, which include the lower limit, upper limit, and recommended value.

[0024] Step 6: Extract the tuning calculation part and the setting part from Step 2, compare them with the corresponding tuning calculation results and setting results from Step 5, and output the verification result report.

[0025] Step 6.1: Calculate the percentage deviation between the tuning calculation part and the setting part in Step 2 and the corresponding tuning calculation result and setting result; Step 6.2: Determine whether there are any abnormalities in the setpoints based on the calculated thresholds. Setpoints with deviations exceeding the thresholds are marked as abnormal. Analyze the causes of abnormal setpoints to determine if they are due to incorrect parameter input, differences in calculation methods, or equipment configuration issues. Finally, generate a verification result report, which includes a summary table of parameter extraction results, a summary table of standard calculation results, a deviation comparison analysis table, an abnormal setpoint warning list, and setpoint adjustment suggestions.

[0026] Example 3 This invention relates to a protection setting calculation and verification system based on large model parameter identification, comprising: a file receiving module, a file conversion module, a segmentation and display module, a verification management module, and an output module. The file receiving module receives the protection setting calculation sheet to be verified; the file conversion module converts the received protection setting calculation sheet to be verified into a structured document; the segmentation and display module segments the structured document and displays the segmentation results; the verification management module performs the verification calculation; and the output module outputs the verification results.

[0027] The verification management module includes a setting calculation module and a setting value verification module. The setting calculation module is used to calculate the setting value and compares the setting calculation part of the protection setting calculation book with the setting value setting calculation result. The setting value verification module is used to verify the setting value sheet result and compares the setting value item in the setting value sheet with the calculated setting value result.

[0028] Example 4 This invention relates to a protection setting calculation and verification method based on large model parameter identification, which is implemented according to the following steps: Step 1: Obtain the protection setting calculation sheet to be verified for a photovoltaic power station. The protection setting calculation sheet is a scanned PDF file. Use an OCR tool to convert the PDF file into a structured document. The structured document is Markdown text data including titles, paragraphs, tables, and formulas. Step 2: Preprocess the structured document by dividing it into the original equipment parameters section, the tuning calculation section, and the setting value section; Step 2.1: Filter the Markdown text data in Step 1 to remove text content that is irrelevant to the proofreading. Text content that is irrelevant to the proofreading includes soft padding and control words. Step 2.2: Reduce text length by splitting the Markdown text data into paragraphs and by length; Step 2.3: Divide the original parameter chapter into the equipment original parameter section, divide the main transformer protection setting calculation chapter, the 35kV bus protection setting calculation chapter, the collector line protection setting calculation chapter, and the SVG protection setting calculation chapter into the setting calculation section, and divide the setting value sheet chapter into the setting value section.

[0029] Step 3: Extract the original parameters of the equipment from Step 2, construct the prompt text, and train it using a large language model to obtain structured power equipment parameter data; The large language model uses glm. With prompt text, the large language model is guided to extract parameters according to a predetermined format. A total of 42 main transformer protection parameters, 12 bus protection parameters, 65 collector line protection parameters, and 15 SVG protection parameters were extracted. The extraction took 5 minutes and output structured power equipment parameter data.

[0030] Step 4: Construct a power grid topology model; Step 4.1: By combining structured power equipment parameter data with power grid expert knowledge, the connection relationships of the power equipment are obtained; Step 4.2: Treat the power equipment as nodes in the power grid topology graph, the power equipment parameter data as node attributes, and the electrical connections between the equipment as edges in the power grid topology graph. Construct a power grid topology model based on the power plant equipment parameters.

[0031] Step 5: Based on the power grid topology model, a symbolic computing engine is used to obtain the setting calculation results and setting results; short-circuit calculation and protection setting calculation are performed to generate standard protection setting data. The calculation takes 2 minutes.

[0032] Step 6: Extract the tuning calculation part and the set value part from Step 2, and compare and verify the parameters with the corresponding tuning calculation results and set value results from Step 5. If the deviation of 3 set values ​​exceeds 10%, output the verification result report, which includes deviation analysis and adjustment suggestions. The total verification time is 15 minutes.

[0033] Example 5 This invention relates to a protection setting calculation and verification method based on large model parameter identification, which is implemented according to the following steps: Step 1: Obtain the protection setting calculation sheet to be verified for a wind farm. The setting calculation sheet is in text PDF format, with a total of 12 pages. After parsing the protection setting calculation sheet for 30 seconds, convert it into a Markdown format document. Step 2: Filter the Markdown text data from Step 1, removing text content irrelevant to the verification, including soft pressure plates and control words; reduce the text length and split the Markdown text data by paragraph and length; divide the Markdown text data into the equipment original parameter part, the setting calculation part, and the 35kV collector line protection setting part.

[0034] Step 3: Extract the original parameters of the equipment from Step 2, construct the prompt text, and train it using a large language model to obtain structured power equipment parameter data; The large language model uses GPT. With prompt text, the large language model is guided to extract 28 power line protection parameters according to a predetermined format. The extraction takes 2 minutes and outputs structured power equipment parameter data.

[0035] Step 4: Construct a power grid topology model; Step 4.1: By combining structured power equipment parameter data with power grid expert knowledge, the connection relationships of the power equipment are obtained; Step 4.2: Treat the power equipment as nodes in the power grid topology graph, the power equipment parameter data as node attributes, and the electrical connections between the equipment as edges in the power grid topology graph.

[0036] Step 5: Based on the power grid topology model, a symbolic computing engine is used to obtain the setting calculation results and setting results, which takes 1 minute. Step 6: Extract the tuning calculation part and the set value part from Step 2, and compare them with the corresponding tuning calculation results and set value results from Step 5. The comparison and verification found that one set value was abnormal, and a verification report was output. The total verification time was 5 minutes.

[0037] Example 6 This invention relates to a protection setting calculation and verification method based on large model parameter identification, which is implemented according to the following steps: Step 1: Obtain the protection setting calculation sheet to be verified for a certain energy storage power station. The protection setting calculation sheet to be verified is a scanned PDF format, with a total of 8 pages. Parse the protection setting calculation sheet, which takes 1 minute, and convert it into a structured document. Step 2: Preprocess the structured document by dividing it into the original equipment parameters section, the tuning calculation section, and the setting value section; Step 3: Extract the original parameters of the equipment from Step 2, construct the prompt text, and train it using a large language model to obtain structured power equipment parameter data; The large language model uses Deepseek, which guides the large language model to extract 35 parameters, including main transformer differential protection, gas protection, and overcurrent protection, according to a predetermined format, and outputs structured power equipment parameter data.

[0038] Step 4: Construct a power grid topology model; Step 4.1: By combining structured power equipment parameter data with power grid expert knowledge, the connection relationships of the power equipment are obtained; Step 4.2: Treat the power equipment as nodes in the power grid topology graph, the power equipment parameter data as node attributes, and the electrical connections between the equipment as edges in the power grid topology graph.

[0039] Step 5: Based on the power grid topology model, a symbolic computing engine is used to obtain the setting calculation results and setting results. The calculation takes 1 minute. Step 6: Extract the tuning calculation part and the setting part from Step 2, and compare them with the corresponding tuning calculation results and setting results from Step 5. The verification found that the two setting deviations were large, and the verification report was output. The total verification time was 5 minutes.

[0040] This invention was used to verify the protection setting calculation sheet of a photovoltaic power station. The calculation sheet includes protection settings for 8 types of equipment, with more than 200 parameters. Traditional manual verification takes about 4 hours to complete the entire verification process, while the method of this invention can complete it in only about 15 minutes, improving efficiency by about 16 times and significantly reducing the verification error rate. Furthermore, this invention can automatically generate a detailed verification report, including deviation analysis and adjustment suggestions for each parameter, and automatically generate a report containing calculation steps, data input basis, and decision logic, achieving full traceability. This effectively solves the core problems of low efficiency, error susceptibility, and lack of systematic recording in manual verification. The parameter extraction accuracy of this invention reaches over 95%, the verification results are reliable, and it can effectively handle protection setting calculation sheets of different formats and sizes, achieving automated parameter extraction and verification, significantly improving verification efficiency and reducing human error.

[0041] This invention presents a protection setting calculation and verification method based on large-scale model parameter identification. The large-scale language model possesses powerful natural language understanding capabilities, enabling it to comprehend professional terminology and semantic relationships in the field of power protection, accurately identifying and extracting the required parameter information from unstructured text descriptions. Through prompts, the large-scale language model is guided to output structured data according to a predetermined format, achieving automatic conversion from natural language to structured data. This invention does not require pre-defined complex rule expressions, can adapt to calculation documents from different manufacturers and in different formats, and exhibits strong generalization ability and robustness. A symbolic calculation engine implements the calculation process, thereby achieving traceability, completing the calculation process, and generating a setting calculation report output.

Claims

1. A method for calculating and verifying protection setting values ​​based on large model parameter identification, characterized in that, The specific steps are as follows: Step 1: Obtain the file to be verified, parse the file, and convert it into a structured document; Step 2: Segment the structured document into three parts: original equipment parameters, tuning calculation, and setpoints. Step 3: Construct prompt text, train it using a large language model, and obtain structured power equipment parameter data; Step 4: Construct a power grid topology model; Step 5: Using a symbolic computation engine, the tuning calculation results and setpoint results are obtained; Step 6: Extract the tuning calculation part and the setting part from Step 2, compare them with the corresponding tuning calculation results and setting results from Step 5, and output the verification result report; In step 1, the protection setting calculation sheet to be verified is a PDF file. The PDF file is converted into a structured document using an OCR tool. The structured document is Markdown text data including titles, paragraphs, tables, and formulas. The specific process of step 2 is as follows: Step 2.1: Filter the Markdown text data in Step 1 to remove text content that is irrelevant to the proofreading. Text content that is irrelevant to the proofreading includes soft padding and control words. Step 2.2: Reduce text length by splitting the Markdown text data into paragraphs and by length; Step 2.3: Divide the Markdown text data into the original device parameters, the tuning calculation, and the setpoints. The large language model can use any one of qwen3, deepseek, glm, or gpt. With prompt words, the large language model is guided to output structured power equipment parameter data in a predetermined format. Step 4 specifically involves: Step 4.1: By combining structured power equipment parameter data with power grid expert knowledge, the connection relationships of the power equipment are obtained; Step 4.2: Treat the power equipment as nodes in the power grid topology graph, the power equipment parameter data as node attributes, and the electrical connections between the equipment as edges in the power grid topology graph. In step 5, the setting calculation results include the positive sequence and zero sequence impedances of each power device, the positive sequence and zero sequence impedances of each short-circuit point, the short-circuit current, and the short-circuit current of the transformer branch. The setting results include the protection setting ranges for each power device, which include the lower limit, upper limit, and recommended value.

2. A protection setting calculation and verification system based on large model parameter identification, characterized in that, include: The module includes a file receiving module, a file conversion module, a segmentation and display module, a verification and management module, and an output module. The file receiving module is used to receive the protection setting calculation sheet to be verified. The file conversion module is used to convert the received protection setting calculation sheet to be verified into a structured document; The segmentation and display module is used to segment structured documents and arrange and display the segmentation results; the verification and management module is used to perform verification calculations; and the output module is used to output the verification results.

3. The protection setting calculation and verification system based on large model parameter identification according to claim 2, characterized in that, The verification management module includes a setting calculation module and a setting value verification module. The setting calculation module is used to calculate the setting value and match and compare the setting calculation part of the protection setting calculation book with the setting value setting calculation result. The setting verification module is used to verify the setting sheet results by matching and comparing the setting items in the setting sheet with the calculated setting results.