Power industry form filling standard auditing method, system and device and medium

Abstract

The embodiment of the invention provides a power industry form filling specification auditing method, system and device and a medium, and belongs to the field of power industry work order auditing. The method comprises the following steps: constructing a standardized framework of work order auditing cue words, and designing initial cue words based on the standardized framework; performing benchmark testing on the initial cue word by using the test set to obtain an initial evaluation index; performing error analysis on the benchmark test result, identifying an error type and a root cause, and optimizing the initial cue word according to an error analysis result to obtain an optimized cue word; performing iterative verification on the optimized cue word, and if the comparison between the evaluation index and the initial evaluation index reaches a preset target, generating a target cue word meeting an auditing requirement; and using the target prompt word to carry out normative auditing on the to-be-audited power industry form through the large language model, and outputting an auditing result. By establishing a set of quantifiable and iterable cue word optimization system, the accuracy and auditing efficiency of work order filling and auditing are improved.

Description

Technical Field

[0001] This invention relates to the field of work order review technology in the power industry, specifically to a method, system, equipment, and medium for reviewing the form filling standards in the power industry. Background Technology

[0002] Power industry work orders (such as equipment maintenance work orders, fault reporting work orders, and safe operation work orders) are the core document carriers for the safe operation of the power grid, and their completion standards directly affect work safety, work efficiency, and operation and maintenance quality.

[0003] Currently, the industry generally uses manual review, which has three significant problems: First, manual review is inefficient, with an average review time of 3-5 minutes per work order, making it difficult to meet the work order processing needs of the power grid's 24 / 7 operation. Second, review standards are inconsistent; different reviewers have differing understandings of "standardization," leading to varying compliance judgments for the same work order from different reviewers. Third, the application of large-scale models lacks systematicity; existing review schemes based on large-scale models mostly rely on random prompt word design, without establishing a standardized prompt word framework and scientific optimization process, resulting in large fluctuations in review accuracy. While existing technologies, such as the CN114581234A patent, propose a work order review model, they only focus on data preprocessing and do not address the key bottleneck of prompt word design. Summary of the Invention

[0004] The purpose of this invention is to provide a method, system, device, and medium for reviewing the form filling specifications in the power industry. By establishing a quantifiable and iterative prompt word optimization system, the accuracy and efficiency of work order filling review are improved, providing the power industry with a safe, reliable, efficient, and intelligent work order review solution.

[0005] To achieve the above objectives, embodiments of the present invention provide a method for reviewing form filling specifications in the power industry, including: Construct a standardized framework for work order review prompts, and design initial prompts based on the standardized framework; The initial prompt words are benchmarked using a test set containing multiple labeled samples to obtain initial evaluation metrics; Error analysis is performed on the benchmark test results to identify error types and root causes, and the initial prompt words are optimized based on the error analysis results to obtain optimized prompt words; The optimized prompt words are iteratively verified. If the evaluation indicators reach the preset target compared with the initial evaluation indicators, target prompt words that meet the review requirements are generated. Using the target prompt words, the power industry forms to be reviewed are subjected to a standardization review through a large language model, and the review results are output.

[0006] Optionally, the standardization framework includes role definition, review specifications, output instructions, and boundary constraint elements; wherein, the role definition is used to limit the large language model to power work order review experts with industry experience; The review guidelines are based on the power industry work order filling standards and are further refined into language standards, content elements, and compliance requirements. The output instruction requires the large language model to output the audit results in a given structured data format; The boundary constraints are used to clarify the scope of the audit, define exceptions, and provide boundary cases.

[0007] Optionally, the evaluation metrics include accuracy, precision, recall, and F1 score.

[0008] Optionally, the accuracy calculation formula is as follows: ; The formula for calculating accuracy is as follows: ; The recall rate is calculated using the following formula: ; The formula for calculating the F1 score is as follows: ; In the formula, TP represents a correctly identified violation work order, TN represents a correctly identified compliant work order, FP represents a misjudged violation work order, and FN represents a missed violation work order.

[0009] Optionally, the initial prompt words can be optimized based on the error analysis results, including: strengthening the mandatory nature of the element description, adding boundary case examples, introducing thought chain guidance, clearly defining boundary constraints, strengthening output format requirements, and requiring suggestions to include one or more specific modification examples.

[0010] Optionally, the optimized prompts are iteratively validated. If the evaluation metrics meet the preset targets compared to the initial evaluation metrics, target prompts that meet the review requirements are generated, including: The evaluation metrics obtained by running the optimized prompts on the test set are quantitatively compared with the initial evaluation metrics to calculate the improvement. The improvement is compared with a preset target threshold. If the evaluation index reaches or exceeds the preset target threshold, the current optimized prompt word is determined to be a target prompt word that meets the review requirements, and the version of the target prompt word is locked for use in the production environment. Record the optimized prompt words, corresponding evaluation metrics, and improvement rates in the optimization knowledge base during this iteration.

[0011] Optionally, using the target prompt words, a large language model is used to perform a standardization review on the power industry forms to be reviewed, and the review results are output, including: The target prompt words are combined with the content of the work order to be reviewed to form an initial query context. Key entities and operation verbs are extracted from the content of the work order to be reviewed to generate a cognitive focus list. This list is then appended to the context in the form of instructions, requiring the large language model to verify each item in the list one by one during the reasoning process. The system calls upon a large language model to process the initial query context and instructs the large language model to output the final JSON audit results and the structured thought chain of the reasoning process. It also maps and associates the various check results in the structured thought chain with the items in the cognitive focus list. The thought chain must follow a preset logical framework of language standardization check, core element completeness check, sensitive word and compliance scan, and boundary condition compliance judgment. Based on the key violations or low-confidence items identified in the main review process, adversarial prompts are generated. The role of these adversarial prompts is set as experts in defending the compliance of work orders. Their task is to find compliant explanations or exceptions for the key violations or low-confidence items, and then submit the adversarial prompts and work order content to the large language model again to obtain adversarial verification defense arguments. The final JSON audit result and the structured thought chain and adversarial verification defense arguments in the reasoning process are weighted and logically conflict-detected using predefined rules. If the adversarial verification defense arguments cannot provide support from the boundary case library or industry standard provisions, the final JSON audit result is adopted; if the adversarial verification defense arguments successfully provide support from the boundary case library or industry standard provisions, the final JSON audit result is corrected and the violation is marked as compliant.

[0012] Secondly, the present invention also provides an auditing system for the standardization of form filling in the power industry, comprising: A construction unit is used to build a standardized framework for work order review prompts and to design initial prompts based on the standardized framework. The testing unit is used to perform benchmark testing on the initial prompt words using a test set containing multiple labeled samples, and to obtain initial evaluation metrics. The optimization unit is used to perform error analysis on the benchmark test results, identify error types and root causes, and optimize the initial prompt words based on the error analysis results to obtain optimized prompt words; The verification unit is used to iteratively verify the optimized prompt words. If the evaluation indicators reach the preset target compared with the initial evaluation indicators, the target prompt words that meet the review requirements are generated. The review unit is used to perform a standardization review of the power industry forms to be reviewed using the target prompt words and a large language model, and output the review results.

[0013] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the above-described power industry form filling specification review method.

[0014] Fourthly, the present invention also provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described auditing method for the power industry form filling specifications.

[0015] By implementing the above technical solutions and establishing a quantifiable and iterative prompt word optimization system, the accuracy and efficiency of work order filling and review have been improved.

[0016] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a method for reviewing form filling specifications in the power industry, provided by an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of an auditing system for form filling specifications in the power industry, provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0018] Various embodiments of this disclosure will be described more fully in the following detailed description. This disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of this disclosure to the specific embodiments disclosed herein, but rather this disclosure should be understood to cover all adjustments, equivalents, and / or alternatives falling within the spirit and scope of the various embodiments of this disclosure.

[0019] In the following, the terms “comprising” or “may include”, which may be used in various embodiments of this disclosure, indicate the presence of the disclosed functions or operations and do not limit the addition of one or more functions or operations. Furthermore, as used in various embodiments of this disclosure, the terms “comprising,” “having,” and their cognates are intended only to indicate a specific feature, number, step, operation, or combination of the foregoing and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, or combinations of the foregoing, or the possibility of adding one or more features, numbers, steps, operations, or combinations of the foregoing.

[0020] In various embodiments of this disclosure, the expression "or" or "at least one of A and / or B" includes any combination or all combinations of the words listed simultaneously. For example, the expression "A or B" or "at least one of A and / or B" may include A, may include B, or may include both A and B.

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] See Figure 1 The diagram shows a flowchart of the review method for the power industry form filling specifications in a specific embodiment, including the following execution steps: Step 100: Construct a standardized framework for work order review prompts, and design initial prompts based on the standardized framework.

[0023] Specifically, the standardized framework includes role definitions, review specifications, output instructions, and boundary constraints. The role definitions limit the large language model to power work order review experts with industry experience. The review specifications, based on power industry work order completion standards, are refined into language specifications, content elements, and compliance requirements. The output instructions require the large language model to output review results in a given structured data format. The boundary constraints clarify the review scope, define exceptions, and provide boundary cases.

[0024] In one specific implementation, the audit specifications adopt a three-tiered construction method of "classification-quantification-scenario-based approach." First, the audit specifications are divided into three categories: language specifications, content elements, and compliance requirements, with clear quantitative indicators set for each category. The language specifications require "the use of standard Chinese, prohibition of English abbreviations and special symbols, and a word count of 150-300 characters"; the content elements require "must include four elements: equipment name, maintenance steps, safety measures, and acceptance standards; the absence of any one element constitutes non-compliance"; and the compliance requirements are refined based on the "Power Industry Work Order Filling Specification" (DL / T1055-2022), specifying that "the use of vague expressions such as 'guarantee safety' and 'ensure safety' is strictly prohibited, and the sensitive word library contains 127 power industry-specific sensitive words."

[0025] For example, the role is clearly defined as "Power Industry Work Order Review Expert," with a professional background; the review standards are based on the "Power Industry Work Order Filling Standard" (DL / T1055-2022) and are refined into three categories of standards: language standards, content elements, and compliance requirements; the output instructions are required to be returned in JSON format, including compliance judgment, list of violations, confidence score, and specific suggestions; the boundary constraints clearly define the "sensitive word library" (containing 127 power industry-specific sensitive words) and "element completeness" (must include four elements: equipment name, operation steps, safety measures, and acceptance standards).

[0026] Example prompts for equipment maintenance work orders: "You are an expert with senior work order review qualifications in the power industry, responsible for reviewing the standardization of equipment maintenance work orders. The review standards include: 1) The language must be standard Chinese, and English abbreviations and special symbols are prohibited; 2) It must include four elements: equipment name, maintenance steps, safety measures, and acceptance standards; 3) Vague expressions such as 'guarantee safety' are strictly prohibited. The sensitive word list is attached. The output must be in strict JSON format with four required fields: is_compliant (boolean value, indicating whether it is compliant), violations (string array, listing all violations), confidence (0.0-1.0 floating-point number, indicating the confidence level), and suggestions (string, providing specific modification suggestions): {'is_compliant': bool, 'violations': [str], 'confidence': float, 'suggestions': str}." In one specific implementation, to improve output quality, explicit constraints are set for each field: `is_compliant` must be true or false, the `violations` array cannot be empty, the `confidence` value must be in the range of 0.0-1.0, and `suggestions` must provide actionable modification suggestions. For example, when a "lack of safety measures" violation is detected, the `suggestions` field must include specific suggestions such as "Please supplement specific safety measures, such as 'test for power after power failure'," rather than simply writing "Please supplement safety measures." This structured output design not only facilitates automatic system parsing but also ensures the practical value of the audit results.

[0027] In one specific implementation, the boundary constraints are implemented using a three-step approach: "defining the scope, defining exceptions, and providing examples." First, the scope of review is clearly defined, limited to the review of work order completion standards within the power industry, excluding assessments of the technical feasibility of the work order content. Second, boundary exceptions are defined; for example, vague expressions such as "pay attention to safety" or "ensure safety" are prohibited in safety measure descriptions, while specific measures such as power-off testing and wearing insulating gloves are permitted. Finally, boundary case examples are used to reinforce understanding, such as: "Pay attention to safety during equipment maintenance" in a work order → violation; "Power-off testing and wearing insulating gloves during equipment maintenance" in a work order → compliance.

[0028] To ensure the accuracy of boundary constraints, a dynamic maintenance mechanism was established during implementation. Based on historical review data, a power industry work order boundary case library was built, containing over 500 typical boundary cases, and updated quarterly. Simultaneously, a confidence assessment mechanism was introduced; when the model's confidence level in judging boundary situations falls below 0.7, a manual review process is automatically triggered. This refined implementation of boundary constraints effectively reduced the false positive and false negative rates, achieving a boundary issue handling accuracy rate of 95.6%.

[0029] Step 101: Benchmark the initial prompt words using a test set containing multiple labeled samples to obtain initial evaluation metrics.

[0030] Specifically, the evaluation metrics include accuracy, precision, recall, and F1 score.

[0031] More specifically, the accuracy calculation formula is as follows: ; The formula for calculating accuracy is as follows: ; The recall rate is calculated using the following formula: ; The formula for calculating the F1 score is as follows: ; In the formula, TP represents a correctly identified violation work order, TN represents a correctly identified compliant work order, FP represents a misjudged violation work order, and FN represents a missed violation work order.

[0032] Step 102: Perform error analysis on the benchmark test results, identify the error types and root causes, and optimize the initial prompt words based on the error analysis results to obtain optimized prompt words.

[0033] Specifically, the initial prompts are optimized based on the error analysis results, including: strengthening the mandatory nature of element descriptions, adding boundary case examples, introducing thought chain guidance, clearly defining boundary constraints, strengthening output format requirements, and requiring suggestions to include one or more specific modification examples.

[0034] In one specific implementation, the proportion of typical error types is shown in Table 1: Table 1. Distribution and Cause Analysis of Work Order Review Error Types

[0035] In one specific implementation, targeted optimization is carried out, and precise adjustments are made based on the error analysis results. For example, optimization strategies include: strengthening the mandatory description of elements by changing "must include" to "must include"; adding boundary case examples, such as "Work order 'Safety precautions must be taken during equipment inspection' → Reason for violation: No safety measures are clearly defined, suggestion: Please supplement specific safety measures, such as power off and power testing"; and introducing a chain-of-thought approach, requiring the model to analyze according to a four-step process: "language check → element verification → sensitive word scanning → comprehensive judgment".

[0036] To address the issue of false alarms, optimization strategies include: clearly defining boundary constraints and adding provisions prohibiting vague expressions such as "caution" and "ensure safety" in safety measure descriptions, while allowing specific measures such as "power off and test for voltage" and "wearing insulated gloves"; adding boundary examples to the prompts to clarify that "work order 'caution during equipment maintenance' → violation; work order 'power off and test for voltage and wear insulated gloves during equipment maintenance' → compliance."

[0037] To address the formatting error issue, optimization strategies include: strengthening format constraints by changing "output must be in JSON format" to "output must be in strict JSON format, and no additional explanatory text is allowed; formatting errors will invalidate the review results"; and adding format examples to the prompts to demonstrate the correct JSON output structure.

[0038] To address the issue of vague suggestions, optimization strategies include: requiring suggestions to include specific modification examples, changing "Please add safety measures" to "Please add specific safety measures, such as 'power off and test for voltage, wear insulated gloves'"; and adding suggestion templates to the prompts, requiring suggestions to include "specific modification content".

[0039] Each optimization underwent A / B testing, applying the pre- and post-optimization prompts to 100 test samples respectively, calculating the changes in metrics, and verifying the optimization effect. For example, the optimization for the missed detection problem reduced the missed detection rate from 25.4% to 11.3%, an improvement of 14.1 percentage points.

[0040] Step 103: Iteratively verify the optimized prompt words. If the evaluation indicators reach the preset target compared with the initial evaluation indicators, then generate target prompt words that meet the review requirements.

[0041] Specifically, when executing step 103, the following steps can be performed: S1030: The evaluation metrics obtained by running the optimized prompts on the test set are quantitatively compared with the initial evaluation metrics to calculate the improvement.

[0042] For example, the test was retested using the optimized prompts, and the changes in metrics before and after optimization were compared. Through multiple iterations, the F1 score improved from an initial 0.68 to over 0.92, the accuracy improved from 72.3% to 92.1%, and the recall improved from 65.8% to 88.7%.

[0043] In one specific implementation, to verify the effectiveness of the prompt word standardization framework, this embodiment conducted a three-month pilot application in a provincial power grid company. A comparison of key indicators before and after implementation is shown in Table 2.

[0044] Table 2 Comparison of Key Indicators Before and After the Implementation of the Prompt Word Standardization Framework

[0045] Comparative data shows that the implementation of the prompt word standardization framework improved the accuracy rate by 19.8 percentage points, reduced the false positive rate and the missed detection rate by 10.8 and 14.1 percentage points respectively, improved the review efficiency by 68.7%, and improved the consistency of results by 16.8%. These significant improvements validate the effectiveness of the prompt word standardization framework and provide a solid foundation for subsequent prompt word optimization.

[0046] S1031: Compare the improvement rate with the preset target threshold. If the evaluation index reaches or exceeds the preset target threshold, the currently optimized prompt word is determined to be a target prompt word that meets the review requirements, and the version of the target prompt word is locked for use in the production environment.

[0047] S1032: Record the optimized prompt words, corresponding evaluation indicators, and improvement rates in the optimization knowledge base during this iteration.

[0048] In one specific implementation, the iterative verification adopts a closed-loop mechanism of "optimization-testing-evaluation-decision" to ensure continuous improvement in optimization results. After each optimization, benchmark testing is immediately performed to calculate the changes in various indicators. The optimized indicators are compared with the baseline indicators to calculate the improvement. If the improvement reaches the preset target (e.g., an F1 score increase of ≥0.05), the optimized prompt words are included in the next iteration; otherwise, the cause of the error is analyzed, and a new round of optimization is carried out. During the iterative verification process, an indicator tracking mechanism is implemented to record the changes in key indicators in each round of optimization. Implementation data shows that after three rounds of iterative optimization, the F1 score improved from the initial 0.68 to 0.92, the accuracy improved from 72.3% to 92.1%, the recall improved from 65.8% to 88.7%, the false positive rate decreased from 18.7% to 7.9%, and the false negative rate decreased from 25.4% to 11.3%. To ensure the sustainability of optimization, an optimization record library is established during implementation to record the detailed content, reasons for optimization, optimization strategies, and optimization effects of each round of optimization. The optimization record library contains over 200 optimization records, forming a complete optimization knowledge base to provide a reference for subsequent similar tasks. Simultaneously, an optimization effectiveness evaluation mechanism is implemented, conducting effectiveness evaluations after each round of optimization to ensure the correctness of the optimization direction.

[0049] Step 104: Using the target prompt words, perform a standardization review on the power industry forms to be reviewed through a large language model, and output the review results.

[0050] Specifically, when performing step 104, the following steps can be performed: S1040: Combine the target prompt words with the work order content to be reviewed to form an initial query context, extract key entities and operation verbs from the work order content to be reviewed, generate a cognitive focus list, and append the list to the context in the form of instructions, requiring the large language model to verify each item in the list one by one during the reasoning process.

[0051] S1041: Call the large language model to process the initial query context, and instruct the large language model to output the final JSON audit result and the structured thinking chain of the reasoning process, and map and associate the various check results in the structured thinking chain with the items in the cognitive focus list; wherein, the thinking chain must follow the preset logical framework of language standardization check, core element completeness check, sensitive word and compliance scan, and boundary condition compliance judgment.

[0052] S1042: Generate adversarial warning words based on key violations or low-confidence items identified in the main review process.

[0053] In this context, the role of the adversarial prompt is defined as an expert in work order compliance defense. Its task is to find compliant explanations or exceptions for key violations or low-confidence items, and then submit the adversarial prompt and work order content to the large language model again to obtain adversarial verification defense arguments.

[0054] S1043: Using predefined rules, the final JSON audit result and the structured thought chain and adversarial verification defense arguments of the reasoning process are weighted and logically conflicted. If the adversarial verification defense arguments cannot provide support from the boundary case library or industry standard provisions, the final JSON audit result is adopted; if the adversarial verification defense arguments successfully provide support from the boundary case library or industry standard provisions, the final JSON audit result is corrected and the violation is marked as compliant.

[0055] In this embodiment, the prompt word standardization framework not only solves the standardization problem of work order review in the power industry, but also provides a reusable methodology for the application of large-scale models in specific industry scenarios. By transforming industry expertise into structured prompt word elements, the review process becomes more professional, standardized, and quantifiable. This framework can be extended to other sub-sectors of the power industry, such as substations, transmission lines, and distribution centers, forming industry-specific prompt word libraries and further improving the accuracy and professionalism of the review process. Simultaneously, the implementation of the framework provides a standardized path for the application of large-scale models in the industry, promoting the deep application of large-scale model technology in vertical fields, and possessing significant industry promotion value and application prospects.

[0056] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0057] like Figure 2 As shown, the following is an embodiment of the power industry form filling specification review system provided in this disclosure. It belongs to the same inventive concept as the power industry form filling specification review method in the above embodiments. For details not described in detail in the embodiments of the power industry form filling specification review system, please refer to the embodiments of the power industry form filling specification review method described above.

[0058] The power industry form completion standard review system includes: A construction unit is used to build a standardized framework for work order review prompts and to design initial prompts based on the standardized framework. The testing unit is used to perform benchmark testing on the initial prompt words using a test set containing multiple labeled samples, and to obtain initial evaluation metrics. The optimization unit is used to perform error analysis on the benchmark test results, identify error types and root causes, and optimize the initial prompt words based on the error analysis results to obtain optimized prompt words; The verification unit is used to iteratively verify the optimized prompt words. If the evaluation indicators reach the preset target compared with the initial evaluation indicators, the target prompt words that meet the review requirements are generated. The review unit is used to perform a standardization review of the power industry forms to be reviewed using the target prompt words and a large language model, and output the review results.

[0059] By optimizing the initial prompt words, the accuracy rate of prompt word review was increased from 72.3% to 92.1%, achieving the core objective of this application. Moreover, the implementation of this solution not only solved the accuracy problem of work order review in the power industry, but also provided a reusable optimization methodology for the application of large models in specific industry scenarios, which has significant industry promotion value.

[0060] Figure 3 This is a schematic diagram of the hardware structure of an electronic device that implements various embodiments of the present invention.

[0061] The method for reviewing the power industry form filling specifications provided in this application embodiment can be applied to electronic devices. Those skilled in the art will understand that the electronic device structure involved in the embodiments of this invention does not constitute a limitation on the electronic device. An electronic device may include more or fewer components than illustrated, or combine certain components, or have different component arrangements. In the embodiments of this invention, electronic devices include, but are not limited to, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of this application described and / or claimed herein.

[0062] Electronic devices may include processors, external memory interfaces, internal memory, universal serial bus (USB) interfaces, charging management modules, power management modules, batteries, wireless communication modules, audio modules, speakers, microphones, sensor modules, buttons, cameras, displays, and SIM card interfaces, etc.

[0063] It is understood that the structures illustrated in the embodiments of this application do not constitute a specific limitation on the electronic device. In other embodiments of this application, the electronic device may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0064] A processor may include one or more processing units, such as: a central processing unit (CPU), an application processor (AP), a modem processor, a graphics processing unit (GPU), an image signal processor (ISP), a controller, memory, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural network processing unit (NPU). Different processing units may be independent devices or integrated into one or more processors.

[0065] The processor can serve as the nerve center and command center of an electronic device. The controller can generate operation control signals based on the instruction opcode and timing signals to control the fetching and execution of instructions.

[0066] The processor may also include memory for storing instructions and data. In some embodiments, the memory in the processor is a cache memory. This memory can store instructions or data that the processor has just used or that are used repeatedly. If the processor needs to use the instruction or data again, it can retrieve it directly from this memory. This avoids repeated accesses, reduces processor latency, and thus improves system efficiency.

[0067] An external storage interface (ESI) can be used to connect external memory cards, such as microSD cards, to expand the storage capacity of electronic devices. The external memory card communicates with the processor through the ESI to perform data storage functions, such as saving music and video files on the external memory card.

[0068] Internal memory can be used to store computer executable program code, which includes instructions. The processor executes various functional applications and data processing of electronic devices by running the instructions stored in internal memory. Internal memory can include a program storage area and a data storage area. Internal memory can include high-speed random access memory, and can also include non-volatile memory, such as at least one disk storage device, flash memory device, universal flash storage (UFS), etc.

[0069] Wireless communication functionality in electronic devices can be achieved through antennas, wireless communication modules, modem processors, and baseband processors.

[0070] Wireless communication modules can provide solutions for wireless communication applications in electronic devices, including wireless local area networks (WLANs) (such as wireless fidelity (Wi-Fi) networks), Bluetooth (BT), global navigation satellite system (GNSS), frequency modulation (FM), near field communication (NFC), and infrared (IR) technologies.

[0071] Electronic devices can implement audio functions through audio modules, speakers, receivers, microphones, headphone jacks, and application processors.

[0072] Electronic devices can achieve shooting functions through ISPs, cameras, video codecs, GPUs, displays, and application processors.

[0073] Electronic devices can achieve display functions through GPUs, displays, and application processors.

[0074] A GPU is a microprocessor for image processing, connected to the display screen and application processor. GPUs perform mathematical and geometric calculations for graphics rendering. A processor may include one or more GPUs, which execute program instructions to generate or modify display information.

[0075] A display screen is used to display images, videos, etc. A display screen includes a display panel.

[0076] The storage medium provided in this application stores a program product capable of implementing the auditing method for the form filling standards in the power industry.

[0077] The review method for standardized form filling in the power industry includes: constructing a standardized framework for work order review prompts, and designing initial prompts based on the standardized framework; benchmarking the initial prompts using a test set containing multiple labeled samples to obtain initial evaluation indicators; performing error analysis on the benchmark test results to identify error types and root causes, and optimizing the initial prompts based on the error analysis results to obtain optimized prompts; iteratively verifying the optimized prompts, and if the evaluation indicators meet the preset target compared with the initial evaluation indicators, generating target prompts that meet the review requirements; using the target prompts, performing a standardization review of the power industry forms to be reviewed through a large language model, and outputting the review results.

[0078] In some possible implementations, the subject matter of this disclosure, namely, the method and system for reviewing the form filling specifications in the power industry, can be implemented as a program product, which includes program code. When the program product is run on a terminal device, the program code is used to cause the terminal device to perform the steps described in the "Exemplary Methods" section above according to various exemplary embodiments of this disclosure.

[0079] The storage medium disclosed herein may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0080] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for reviewing the standardized form filling practices in the power industry, characterized in that, include: Construct a standardized framework for work order review prompts, and design initial prompts based on the standardized framework; The initial prompt words are benchmarked using a test set containing multiple labeled samples to obtain initial evaluation metrics; Error analysis is performed on the benchmark test results to identify error types and root causes, and the initial prompt words are optimized based on the error analysis results to obtain optimized prompt words; The optimized prompt words are iteratively verified. If the evaluation indicators reach the preset target compared with the initial evaluation indicators, target prompt words that meet the review requirements are generated. Using the target prompt words, the power industry forms to be reviewed are subjected to a standardization review through a large language model, and the review results are output.

2. The method for reviewing the power industry form filling specifications according to claim 1, characterized in that, The standardized framework includes role definitions, review specifications, output instructions, and boundary constraints; wherein, the role definitions are used to limit the large language model to power work order review experts with industry experience. The review guidelines are based on the power industry work order filling standards and are further refined into language standards, content elements, and compliance requirements. The output instruction requires the large language model to output the audit results in a given structured data format; The boundary constraints are used to clarify the scope of the audit, define exceptions, and provide boundary cases.

3. The method for reviewing the power industry form filling specifications according to claim 1, characterized in that, The evaluation metrics include accuracy, precision, recall, and F1 score.

4. The method for reviewing the power industry form filling specifications according to claim 3, characterized in that, The accuracy calculation formula is as follows: ； The formula for calculating accuracy is as follows: ； The recall rate is calculated using the following formula: ； The formula for calculating the F1 score is as follows: ； In the formula, TP represents a correctly identified violation work order, TN represents a correctly identified compliant work order, FP represents a misjudged violation work order, and FN represents a missed violation work order.

5. The method for reviewing the power industry form filling specifications according to claim 1, characterized in that, Based on the error analysis results, the initial prompt words are optimized, including: strengthening the mandatory nature of element descriptions, adding boundary case examples, introducing thought chain guidance, clearly defining boundary constraints, strengthening output format requirements, and requiring suggestions to include one or more specific modification examples.

6. The method for reviewing the power industry form filling specifications according to claim 1, characterized in that, The optimized prompts are iteratively validated. If the evaluation metrics meet the preset targets compared to the initial evaluation metrics, target prompts that meet the review requirements are generated, including: The evaluation metrics obtained by running the optimized prompts on the test set are quantitatively compared with the initial evaluation metrics to calculate the improvement. The improvement is compared with a preset target threshold. If the evaluation index reaches or exceeds the preset target threshold, the current optimized prompt word is determined to be a target prompt word that meets the review requirements, and the version of the target prompt word is locked for use in the production environment. Record the optimized prompt words, corresponding evaluation metrics, and improvement rates in the optimization knowledge base during this iteration.

7. The method for reviewing the power industry form filling specifications according to claim 1, characterized in that, Using the target prompt words, a large language model is used to perform a standardization review of the power industry forms to be reviewed, and the review results are output, including: The target prompt words are combined with the content of the work order to be reviewed to form an initial query context. Key entities and operation verbs are extracted from the content of the work order to be reviewed to generate a cognitive focus list. This list is then appended to the context in the form of instructions, requiring the large language model to verify each item in the list one by one during the reasoning process. The system calls upon a large language model to process the initial query context and instructs the large language model to output the final JSON audit results and the structured thought chain of the reasoning process. It also maps and associates the various check results in the structured thought chain with the items in the cognitive focus list. The thought chain must follow a preset logical framework of language standardization check, core element completeness check, sensitive word and compliance scan, and boundary condition compliance judgment. Based on the key violations or low-confidence items identified in the main review process, adversarial prompts are generated. The role of these adversarial prompts is set as experts in defending the compliance of work orders. Their task is to find compliant explanations or exceptions for the key violations or low-confidence items, and then submit the adversarial prompts and work order content to the large language model again to obtain adversarial verification defense arguments. The final JSON audit result and the structured thought chain and adversarial verification defense arguments in the reasoning process are weighted and logically conflict-detected using predefined rules. If the adversarial verification defense arguments cannot provide support from the boundary case library or industry standard provisions, the final JSON audit result is adopted; if the adversarial verification defense arguments successfully provide support from the boundary case library or industry standard provisions, the final JSON audit result is corrected and the violation is marked as compliant.

8. A verification system for standardized form filling in the power industry, characterized in that, include: A construction unit is used to build a standardized framework for work order review prompts and to design initial prompts based on the standardized framework. The testing unit is used to perform benchmark testing on the initial prompt words using a test set containing multiple labeled samples, and to obtain initial evaluation metrics. The optimization unit is used to perform error analysis on the benchmark test results, identify error types and root causes, and optimize the initial prompt words based on the error analysis results to obtain optimized prompt words; The verification unit is used to iteratively verify the optimized prompt words. If the evaluation indicators reach the preset target compared with the initial evaluation indicators, the target prompt words that meet the review requirements are generated. The review unit is used to perform a standardization review of the power industry forms to be reviewed using the target prompt words and a large language model, and output the review results.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the review method for the power industry form filling specifications as described in any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the review method for the power industry form filling specifications as described in any one of claims 1 to 7.

Images