Photovoltaic project report-oriented style specialization instruction generation method, system and device, and storage medium

Abstract

The invention provides a photovoltaic project report-oriented style specialization instruction generation method, system and device and a storage medium. The method comprises the following steps of S1, collecting and labeling photovoltaic field text style data; s2, constructing a COT seed instruction template in the photovoltaic field; s3, diffusing and generating a style migration COT instruction by using an MCTS algorithm; and S4, evaluating and applying the instruction set. According to the method, the instruction set is automatically generated, so that the manual labeling and editing cost is reduced, and the labor cost is reduced; an automatic instruction generation process accelerates the implementation speed of a text style migration task and remarkably improves the overall efficiency; the method can be popularized and applied to text style migration tasks in other fields and is not limited to photovoltaic reports; through a high-quality instruction set, the pre-training large model can be guided to generate a specialized text which is more in line with a photovoltaic field specification.

Description

Technical Field

[0001] The present invention relates to the fields of natural language processing and intelligent management of photovoltaic projects, and particularly relates to a method, system, device and storage medium for generating style-specialized instructions for photovoltaic project reports. Background Art

[0002] In the management and operation of photovoltaic projects, the compilation of project reports is crucial for project evaluation and decision-making. Project reports usually gather various data sources, such as publicly available Internet data, enterprise internal data, third-party professional statistical data, etc., and cover various data types, such as geographical locations, climatic conditions, policies and regulations, and photovoltaic module parameters. However, due to their diverse presentation forms, different compilation formats, wide practical uses and many other characteristics, the overall style of these data materials has great heterogeneity and is difficult to be directly used in photovoltaic project reports. A large amount of manpower is required for re-integration and screening, and their expression forms need to be specialized and standardized before they can be used for subsequent unified and comprehensive evaluation and decision-making of projects.

[0003] In order to achieve the unification of text styles, previous researchers have proposed a large number of style transfer models. However, the training cost of these models is relatively high and requires high-quality manual style annotation, which is not only time-consuming and laborious, but also extremely inefficient when dealing with long text content. In recent years, pre-trained large models (such as GPT, etc.) have attracted wide attention in various industries, and their application performance in text style transfer is also very prominent. However, their transfer ability highly depends on the quality of the input instructions and requires a large amount of manual instruction construction work. Summary of the Invention

[0004] The first object of the present invention is to provide a method for generating style-specialized instructions for photovoltaic project reports in view of the above-mentioned problems.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for generating style-specialized instructions for photovoltaic project reports includes the following steps:

[0007] S1: Collect and annotate text style data in the photovoltaic field;

[0008] S2: Construct a COT seed instruction template in the photovoltaic field;

[0009] S3: Use the MCTS algorithm to diffuse and generate COT instructions for style transfer;

[0010] S4: Instruction set evaluation and application.

[0011] While adopting the above technical solutions, the present invention can also adopt or combine the following technical solutions:

[0012] As a preferred technical solution of the present invention: step S1 also includes the following sub-steps:

[0013] S11: Data source collection and classification;

[0014] S12: Labeling and classification of style inconsistencies.

[0015] As a preferred technical solution of the present invention: in step S11, the data source can be classified into the following types: photovoltaic project proposals, technical information, policy interpretation, market analysis reports, academic papers and research reports, news reports and comments.

[0016] As a preferred technical solution of the present invention: in step S12, the style inconsistency includes the following aspects: formality, technicality, regional differences, and language differences.

[0017] As a preferred technical solution of the present invention: step S2 also includes the following sub-steps:

[0018] S21: Content understanding, extracting key information from the source text and analyzing its logical structure and semantics;

[0019] S22: semantic analysis, confirming the meaning of professional terms and technical concepts;

[0020] S23: Style feature comparison, comparing the differences between the source style and the target style, and determining the adjustment plan;

[0021] S24: Logical reorganization, reorganizing sentence and paragraph structures without changing the information;

[0022] S25: Language adjustment, replacing vocabulary according to the target style, adjusting the tone, and ensuring that the text is easy to understand;

[0023] S26: Proofreading review to check whether the converted text is accurate and conforms to the target style.

[0024] As a preferred technical solution of the present invention: the MCTS algorithm in step S3 further includes the following sub-steps:

[0025] S31: Initialize the search tree;

[0026] S32: Select a path;

[0027] S33: Extend and generate new instructions;

[0028] S34: simulated style transfer;

[0029] S35: Backtracking optimization instructions.

[0030] The second objective of the present invention is to provide a style specialization instruction generation system for photovoltaic project reports.

[0031] To achieve this, the above objectives of the present invention are realized through the following technical solutions:

[0032] A collection and classification module, which is used to collect and classify data sources;

[0033] A COT seed instruction template construction module, which is used to construct a COT seed instruction template and convert the source text into text in the target style;

[0034] An MCTS algorithm diffusion module, which is used to diffuse and generate COT instructions for style transfer;

[0035] An instruction set evaluation and application module, which is used to accurately evaluate the performance of the instruction set in the style specialization task, so that the large-scale pre-trained model can be successfully applied to the style specialization task of photovoltaic project reports.

[0036] The third objective of the present invention is to provide an electronic device.

[0037] To achieve this, the above objectives of the present invention are realized through the following technical solutions:

[0038] An electronic device, including a memory and a processor, wherein an executable program is stored in the memory, and the processor is configured to run the executable program to execute the steps of the style specialization instruction generation method for photovoltaic project reports as described above.

[0039] The fourth objective of the present invention is to provide a non-volatile storage medium.

[0040] To achieve this, the above objectives of the present invention are realized through the following technical solutions:

[0041] A non-volatile storage medium, wherein an executable program is stored in the non-volatile storage medium, and when the executable program is executed by a processor, it is used to realize the steps of the style specialization instruction generation method for photovoltaic project reports as described above.

[0042] The present invention provides a method, system, device, and storage medium for generating style-specialized instructions for photovoltaic project reports, which have the following beneficial effects: The instruction set automatically generated by the present invention reduces the costs of manual annotation and editing, and reduces the labor costs; The automated instruction generation process speeds up the implementation of the text style migration task and significantly improves the overall efficiency; The method of the present invention can be extended and applied to text style migration tasks in other fields, not limited to photovoltaic reports; Through high-quality instruction sets, it is possible to guide pre-trained large models to generate more professional texts that conform to the specifications of the photovoltaic field. Description of the Drawings

[0043] Figure 1 It is a flowchart of the method for generating style-specialized instructions for photovoltaic project reports provided by the present invention. Detailed Embodiments

[0044] The present invention will be further described in detail with reference to the accompanying drawings and specific embodiments.

[0045] As Figure 1 shown, a method for generating style-specialized instructions for photovoltaic project reports includes the following steps:

[0046] S1: Collect and annotate text style data in the photovoltaic field;

[0047] S2: Construct a COT seed instruction template for the photovoltaic field;

[0048] S3: Use the MCTS algorithm to diffuse and generate COT instructions for style migration;

[0049] S4: Instruction set evaluation and application.

[0050] Step S1 further includes the following sub-steps:

[0051] S11: Data source collection and classification;

[0052] S12: Annotation and classification of style inconsistencies.

[0053] In step S11, in order to ensure that the collected text data can comprehensively cover different scenarios in the photovoltaic field, it is first necessary to identify and classify different types of data sources. The data sources can be classified according to the following types: photovoltaic project proposals, technical materials, policy interpretations, market analysis reports, academic papers and research reports, news reports and comments.

[0054] Photovoltaic project proposals: Include the overall project plan, technical selection, economic evaluation, etc., and the style is usually more formal and professional.

[0055] Technical materials: Such as the specification manuals and installation manuals of photovoltaic modules. These types of documents are highly technical and contain a large number of professional terms and technical parameters.

[0056] Policy interpretation: including policy documents issued by national or local governments, interpretation of industry standards, etc. The document style is relatively rigorous and may also contain some legal language.

[0057] Market analysis reports: For example, reports on market trends, investment analysis, and economic returns for the photovoltaic industry. These texts often combine large amounts of data and analysis, and tend to be concise or contain analytical reasoning.

[0058] Academic Papers and Research Reports: Research literature and academic papers on the cutting edge of photovoltaic technology, which are highly technical in style and contain a large number of formulas, charts and in-depth analysis.

[0059] News reports and comments: News reports on the photovoltaic industry tend to be popular and easy to understand, suitable for non-professional readers.

[0060] In step S12, in order to efficiently transfer the style, different style features need to be labeled and classified. The style inconsistency includes the following aspects: formality, technicality, regional differences, and language differences.

[0061] Formalism: Some passages in the text may use extremely formal, technical language, while other passages may be colloquial or concise.

[0062] Technicality: Some paragraphs contain a lot of technical terms, suitable for professional readers, while other parts tend to be more non-professional.

[0063] Regional differences: The expression styles of policies, regulations and industry standards in different regions may vary, and the corresponding regional characteristics need to be marked.

[0064] Differences in register: Some texts may prefer first-person or third-person narration, and these differences in register also need to be noted.

[0065] Style feature extraction: Extract obvious style features from different types of text and build a style library as a reference for subsequent style transfer.

[0066] Step S2 converts the source text into the target style text through COT (chain of thought) template design. Template design is not just a superficial language conversion, but also combines logical reasoning and semantic analysis to ensure that the converted text information remains intact. It also includes the following sub-steps:

[0067] S21: Content understanding, extracting key information from the source text and analyzing its logical structure and semantics;

[0068] S22: semantic analysis, confirming the meaning of professional terms and technical concepts;

[0069] S23: Style feature comparison. Compare the differences between the source style and the target style to determine the adjustment plan;

[0070] S24: Logical restructuring. Without changing the information, reorganize the sentence patterns and paragraph structures;

[0071] S25: Language adjustment. Replace words according to the target style and adjust the tone to ensure that the text is easy to understand;

[0072] S26: Proofreading and review. Check whether the converted text is accurate and conforms to the target style.

[0073] Task description: Please convert the text of the following photovoltaic project report from the "source style" to the "target style". During the conversion process, the integrity and accuracy of the information need to be maintained.

[0074] Source style features: Describe in detail the style features of the source text, such as dense professional terms, formal tone, complex sentence patterns, etc.

[0075] Target style features: Describe in detail the style features of the target text, such as easy to understand, concise sentence patterns, friendly tone, etc.

[0076] Text to be converted: Provide the source text that needs to be style-transferred.

[0077] Instruction template conversion example:

[0078] Source text: "The project plans to build a 100-megawatt solar power station, adopting advanced solar panels and high-efficiency inverter technologies."

[0079] Target text: "This project aims to build a photovoltaic power station with an installed capacity of 100 MW, adopting advanced monocrystalline silicon photovoltaic modules and high-efficiency inverter technologies."

[0080] In step S3, the MCTS algorithm also includes the following sub-steps:

[0081] S31: Initialize the search tree: Use the COT seed instruction as the root node to establish the initial state of the search tree. For example, the initial instruction is "Please rewrite the following text into a concise and formal style.";

[0082] S32: Select a path: Starting from the root node, select a path according to a certain strategy, and select child nodes until reaching the leaf node;

[0083] S33: Expand to generate new instructions: Generate new child nodes at the leaf node to generate different instruction variants, for example:

[0084] "Please adjust the following text to a formal style that conforms to the photovoltaic industry standards."

[0085] "Rewrite the following text in a more accessible tone suitable for non - professionals.";

[0086] S34: Simulate style transfer: Use these newly generated instructions to rewrite the text and simulate the style transfer effect. Evaluate whether the output text conforms to the target style;

[0087] S35: Backtrack and optimize instructions: Update the simulation results to each node on the path, optimize the instruction selection strategy, and select the instructions with the best effect.

[0088] Example of extended instructions

[0089] "Please rewrite the data section in the following photovoltaic report in an accessible and professional way."

[0090] "Please adjust the following content to a formal and objective style that meets the report standards."

[0091] Finally, by continuously expanding and evaluating the instruction set, ensure that the generated instructions perform excellently in the style transfer task.

[0092] The instruction set generated through the above process enables the large - scale pre - trained model to be successfully applied to the style specialization task of photovoltaic project reports.

[0093] Definition of evaluation metrics: To accurately evaluate the performance of the instruction set in the style specialization task, define a set of evaluation metrics, including:

[0094] Style consistency: Evaluate whether the text after transfer is consistent with the target style in terms of overall style, including the unity of tone, diction, and sentence pattern.

[0095] Information integrity: Ensure that no key information is lost or misinterpreted during the style transfer process, and all technical details and data remain accurate.

[0096] Readability: Evaluate whether the text after transfer is more accessible, especially its readability for non - technical readers.

[0097] Transfer efficiency: Evaluate the speed and efficiency in large - scale document processing to ensure that the instruction set can respond quickly and process multiple documents.

[0098] Introduction of automatic evaluation tools: During the application of the instruction set, introduce automatic evaluation tools to quantitatively analyze the transfer effect. Through natural language processing techniques, use tools based on automatic evaluation metrics such as BLEU, ROUGE, GLEU, etc. to measure the similarity of the text after transfer to the target style and information fidelity.

[0099] A style specialization instruction generation system for photovoltaic project reports, the system includes the following modules:

[0100] Collection and Classification Module, which is used to collect and classify data sources;

[0101] COT Seed Instruction Template Construction Module, which is used to construct a COT seed instruction template and convert the source text into text in the target style;

[0102] MCTS Algorithm Diffusion Module, which is used to diffusely generate COT instructions for style transfer;

[0103] Instruction Set Evaluation and Application Module, which is used to accurately evaluate the performance of the instruction set in the style specialization task, enabling the large-scale pre-trained model to be successfully applied to the style specialization task of photovoltaic project reports.

[0104] The present invention also provides an electronic device, including a processor and a memory for storing processor-executable instructions. Among them, when the processor is set to execute the executable instructions, the method steps of the style specialization instruction generation method for photovoltaic project reports described above are implemented.

[0105] The present invention also provides a non-volatile storage medium, in which an executable program is stored. When the executable program is executed by a processor, the method steps of the style specialization instruction generation method for photovoltaic project reports described above are implemented.

[0106] The above specific implementation manners are used to explain the present invention. They are only the preferred embodiments of the present invention and do not limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and scope of the claims of the present invention fall within the protection scope of the present invention.

Claims

1. A method for generating style-specialized instructions for photovoltaic project reports, characterized in that: The steps include: S1: Collect and annotate text style data in the photovoltaic field; S2: Building a COT seed instruction template for the photovoltaic sector; S3: Use MCTS algorithm to diffuse and generate COT instructions for style transfer; S4: Instruction set evaluation and application.

2. The method for generating style-specialized instructions for photovoltaic project reports according to claim 1, characterized in that: Step S1 also includes the following sub-steps: S11: Data source collection and classification; S12: Labeling and classification of style inconsistencies.

3. The method for generating style-specialized instructions for photovoltaic project reports according to claim 2, characterized in that: In step S11, the data sources may be classified into the following types: photovoltaic project proposals, technical information, policy interpretations, market analysis reports, academic papers and research reports, news reports and comments.

4. The method for generating style-specialized instructions for photovoltaic project reports according to claim 2, characterized in that: In step S12, the inconsistency of style includes the following aspects: formality, technicality, regional differences, and language differences.

5. The method for generating style-specialized instructions for photovoltaic project reports according to claim 1, characterized in that: Step S2 also includes the following sub-steps: S21: Content understanding, extracting key information from the source text and analyzing its logical structure and semantics; S22: semantic analysis, confirming the meaning of professional terms and technical concepts; S23: Style feature comparison, comparing the differences between the source style and the target style, and determining the adjustment plan; S24: Logical reorganization, reorganizing sentence and paragraph structures without changing the information; S25: Language adjustment, replacing vocabulary according to the target style, adjusting the tone, and ensuring that the text is easy to understand; S26: Proofreading review to check whether the converted text is accurate and conforms to the target style.

6. The method for generating style-specialized instructions for photovoltaic project reports according to claim 1, characterized in that: The MCTS algorithm in step S3 also includes the following sub-steps: S31: Initialize the search tree; S32: Select a path; S33: Extend and generate new instructions; S34: simulated style transfer; S35: Backtracking optimization instructions.

7. A style-specialized instruction generation system for photovoltaic project reports, characterized by: The system includes the following modules: A collection and classification module, wherein the collection and classification module is used to collect and classify data sources; A COT seed instruction template construction module, wherein the COT seed instruction template construction module is used to construct a COT seed instruction template to convert a source text into a text of a target style; An MCTS algorithm diffusion module, wherein the MCTS algorithm diffusion module is used to diffuse COT instructions for generating style transfer; An instruction set evaluation and application module is used to accurately evaluate the performance of instruction sets in style specialization tasks, so that large-scale pre-trained models can be successfully applied to style specialization tasks of photovoltaic project reports.

8. An electronic device, comprising a memory and a processor, characterized in that: An executable program is stored in the memory, and the processor is configured to run the executable program to execute the steps of the method for generating style-specialized instructions for photovoltaic project reports according to any one of claims 1 to 6.

9. A non-volatile storage medium, characterized in that: The non-volatile storage medium stores an executable program, and when the executable program is executed by the processor, the steps of the method for generating style-specialized instructions for photovoltaic project reports according to any one of claims 1 to 6 are implemented.

Images