Method and system for checking consistency of cross-modal information of overhaul plan based on large model

Abstract

The application provides a method and system for checking the cross-modal information consistency of a maintenance plan based on a large model, and relates to the technical field of big data analysis. The method comprises the following steps: identifying a target maintenance plan through a power large model, determining key maintenance content, an unstructured proportion, and an attachment type complexity; obtaining a text data complexity and a business risk index based on key maintenance content analysis, as a first checking difficulty coefficient and a second checking difficulty coefficient; determining a third checking difficulty coefficient according to the unstructured proportion and the attachment type complexity evaluation; taking the expected review time limit of the target maintenance plan as a constraint, formulating an adaptive checking depth and an adaptive checking breadth based on the first checking difficulty coefficient, the second checking difficulty coefficient, and the third checking difficulty coefficient; and performing cross-modal information consistency checking on the target maintenance plan according to the adaptive checking depth and the adaptive checking breadth. The dynamic allocation of resources for checking the cross-modal information consistency of a maintenance plan is achieved.

Description

Technical Field

[0001] This invention relates to the field of big data analytics, and in particular to a method and system for cross-modal information consistency verification of maintenance plans based on large models. Background Technology

[0002] Power system maintenance plans are core technical documents that ensure the safe and stable operation of power equipment and prevent large-scale power outages. They typically contain multiple modalities of information, including text descriptions, tabular data, drawings, and voice recordings. The consistency of this cross-modal information directly determines the accuracy of the maintenance plan's execution. Information conflicts, such as discrepancies between the maintenance times described in the text and the outage periods in the table, or mismatches between the associated equipment marked on the drawings and the equipment models described in the text, can lead to maintenance errors, equipment damage, and even personnel safety risks.

[0003] However, the existing maintenance plan review mechanism lacks flexibility and adopts a one-size-fits-all fixed verification model. This leads to inefficiency in simple maintenance plans due to excessive verification, and makes it difficult to fully investigate information conflicts in high-risk and complex maintenance plans due to insufficient verification resources. Ultimately, it is difficult to achieve an effective balance between maintenance safety assurance and improved review efficiency.

[0004] Therefore, there is an urgent need for a method and system for cross-modal information consistency verification of maintenance plans based on large models to address the shortcomings of existing technologies. Summary of the Invention

[0005] This invention addresses the technical problem of the lack of flexibility in the existing maintenance plan review mechanism, which makes it difficult to balance maintenance safety and review efficiency. It provides a method and system for cross-modal information consistency verification of maintenance plans based on a large model.

[0006] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: In a first aspect, the present invention provides a method for cross-modal information consistency verification of maintenance plans based on a large model, including: The target maintenance plan is identified through a large power model, and the key maintenance contents, unstructured proportion, and mixed nature of attachment types are determined. Based on the analysis of the key maintenance content, the complexity of the text data and the business risk index are obtained as the first and second verification difficulty coefficients. The difficulty coefficient of the third verification is determined based on the unstructured ratio and the mixedness of attachment types. Based on the expected review time limit of the target maintenance plan, the appropriate verification depth and breadth are determined according to the first verification difficulty coefficient, the second verification difficulty coefficient, and the third verification difficulty coefficient. The target maintenance plan is subjected to cross-modal information consistency verification according to the aforementioned adaptation verification depth and adaptation verification breadth.

[0007] Secondly, this invention provides a maintenance plan cross-modal information consistency verification system based on a large model, comprising: The maintenance plan feature recognition module is used to identify target maintenance plans through a large power model, and to determine key maintenance content, unstructured proportion, and the degree of mixing of attachment types. The first and second verification difficulty coefficient acquisition modules are used to analyze and obtain the text data complexity and business risk index based on the key maintenance content, which serve as the first and second verification difficulty coefficients. The third verification difficulty coefficient acquisition module is used to determine the third verification difficulty coefficient based on the unstructured ratio and the mixedness of attachment types. The verification depth and breadth acquisition module is used to determine the appropriate verification depth and breadth based on the first verification difficulty coefficient, the second verification difficulty coefficient, and the third verification difficulty coefficient, constrained by the expected review time limit of the target maintenance plan. The consistency verification module is used to perform cross-modal information consistency verification on the target maintenance plan according to the adaptation verification depth and adaptation verification breadth.

[0008] The beneficial effects of this invention are: Compared to existing technologies, this application first identifies the target maintenance plan through a large-scale power model, determining key maintenance content, the proportion of unstructured data, and the degree of mixing of attachment types. This provides comprehensive and reliable data support for the subsequent construction of verification difficulty coefficients and the formulation of adaptive verification strategies. Secondly, based on the analysis of key maintenance content, it obtains text data complexity and business risk indices, which serve as the first and second verification difficulty coefficients. This effectively avoids insufficient verification or resource waste caused by uniform verification standards, improving the targeting and rationality of cross-modal information consistency verification. Thirdly, a third verification difficulty coefficient is determined based on the assessment of the proportion of unstructured data and the degree of mixing of attachment types, supplementing the dimensions of verification difficulty assessment and making the overall difficulty assessment more comprehensive. This provides crucial support for the subsequent formulation of a fully adapted verification strategy. Furthermore, constrained by the expected review timeframe of the target maintenance plan, the application formulates the adapted verification depth and breadth based on the first, second, and third verification difficulty coefficients. This achieves a reasonable allocation of verification resources, effectively avoiding problems such as over-verification, insufficient verification, or verification that fails to complete within the time limit. Finally, cross-modal information consistency verification was performed on the target maintenance plan according to the depth and breadth of the adaptation verification, which effectively avoided information conflict omissions and over-verification problems, and improved the accuracy and efficiency of cross-modal information consistency verification.

[0009] Through the aforementioned technical solution, this application utilizes power big data as its core data support. By employing a large power model, it accurately identifies key maintenance content, the proportion of unstructured data, and the complexity of attachment types in the target maintenance plan. This allows for the construction of first, second, and third verification difficulty coefficients. Furthermore, by constraining the expected review timeframe of the target maintenance plan, it determines the appropriate verification depth and breadth, thereby enabling cross-modal information consistency verification. This effectively overcomes the shortcomings of existing review mechanisms, such as lack of flexibility and a one-size-fits-all approach. It achieves dynamic allocation of verification resources based on the inherent risks, complexity, and urgency of the maintenance task. This approach avoids the inefficiency caused by excessive verification of simple plans while ensuring sufficient verification of high-risk, complex plans. It improves review efficiency while guaranteeing cross-modal information consistency and the safety of maintenance operations, achieving a dynamic balance between safety and efficiency. Attached Figure Description

[0010] Figure 1 A flowchart illustrating the cross-modal information consistency verification method for maintenance plans based on a large model provided by this invention; Figure 2 This is a schematic diagram of the structure of the maintenance plan cross-modal information consistency verification system based on a large model provided by the present invention.

[0011] In the attached diagram, the components represented by each number are as follows: Maintenance plan feature recognition module 11, first and second verification difficulty coefficient acquisition module 12, third verification difficulty coefficient acquisition module 13, verification depth and breadth acquisition module 14, consistency verification module 15. Detailed Implementation

[0012] 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.

[0013] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0014] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0015] Example 1, as Figure 1 As shown, this embodiment of the invention provides a method for cross-modal information consistency verification of maintenance plans based on a large model, including: S10: Identify target maintenance plans through a large power model, and determine key maintenance contents, unstructured proportions, and the degree of mixing of accessory types.

[0016] In the power system maintenance plan management scenario, maintenance plans typically include multimodal information such as text descriptions, tabular data, drawings, and voice recordings. They also suffer from problems such as scattered key information, large differences in the proportion of unstructured data, and diverse attachment types. Traditional manual identification methods are inefficient, prone to missing key information, and difficult to accurately extract core maintenance elements and quantify the degree of data structuring and the diverse characteristics of attachments. This results in a lack of reliable data foundation for subsequent cross-modal information consistency verification.

[0017] To address the aforementioned issues, this application uses a large power model to identify target maintenance plans, determine key maintenance content, unstructured proportions, and the degree of mixing of attachment types.

[0018] Specifically, step S10 in the method includes: The preset key maintenance indicators and the target maintenance plan are input into the power big data model for information identification, and the key maintenance content is output. The preset key maintenance indicators include at least the maintenance object, maintenance time, scope of impact, operation type, work content, related equipment, safety measures and power outage requirements. The target maintenance plan is input into the power big data model for unstructured data identification, and the proportion of unstructured data and the proportion of attachment types are output. The proportion of unstructured data is used as the unstructured ratio, and the attachment type mixing degree is determined based on the proportion of attachment types.

[0019] In this embodiment, preset key maintenance indicators and target maintenance plans are first input into a power big data model for information identification, and key maintenance content is output. The preset key maintenance indicators include at least the maintenance object, maintenance time, scope of impact, work type, work content, related equipment, safety measures, and power outage requirements. Key maintenance content refers to the specific information extracted from the target maintenance plan corresponding to the preset key maintenance indicators, and serves as the basis for subsequent verification difficulty assessment and consistency verification. Specifically, the power big data model first performs multimodal data parsing on the target maintenance plan, such as text segmentation, table structure extraction, drawing element recognition, and speech-to-text conversion. Then, it filters and extracts information corresponding to each preset key maintenance indicator from the parsed multimodal data, integrating them to form the key maintenance content.

[0020] For example, if the text of the target maintenance plan states: the maintenance target is the circuit breaker of pole #15 of the 10kV Chengxi Line, the maintenance time is November 2, 2024, the operation type is emergency repair, the safety measures include power outage, voltage testing, and grounding, the affected area is the power supply area of ​​poles #10-#20 of the Chengxi Line, and the power outage requirement is from 8:00 to 12:00 on November 2, 2024. The preset key maintenance indicators, including the maintenance object, maintenance time, affected area, operation type, work content, related equipment, safety measures, and power outage requirements, are input into the power big data model along with the aforementioned target maintenance plan for information identification. Based on the semantic matching and information extraction logic of the preset key maintenance indicators, the power big data model outputs the following key maintenance content: the maintenance object is the 10kV Chengxi Line #15 pole circuit breaker, the maintenance time is November 2, 2024, the affected area is the power supply area of ​​Chengxi Line #10-#20 poles, the operation type is emergency repair, the work content is not clearly recorded, the related equipment is not clearly recorded, the safety measures are power outage, voltage testing, and grounding wire installation, and the power outage requirement is from 8:00 to 12:00 on November 2, 2024.

[0021] It should be noted that large-scale power models are mature existing technologies, and those skilled in the art can directly obtain and apply publicly available pre-trained large-scale power models such as the State Grid's Guangming Power Large-Scale Model and the China Southern Power Grid's Big Watt Power Model. Moreover, based on the specific needs of the maintenance plan verification scenario involved in this application, such existing large-scale power models can be fine-tuned without having to start model training from scratch.

[0022] Secondly, the target maintenance plan is input into the power system big data model for unstructured data identification, outputting the proportion of unstructured data and the proportion of attachment types. Unstructured data refers to data that cannot be directly stored and represented using a fixed format or database table structure, such as scanned copies of handwritten maintenance records, on-site equipment photos, voice briefing documents, and non-standardized text descriptions. Structured data refers to data with a fixed format that can be directly parsed into tables or key-value pairs, such as standardized maintenance tables and equipment parameter tables exported from databases. The proportion of unstructured data refers to the percentage of unstructured data in the target maintenance plan relative to the total data volume, such as the percentage of bytes. The proportion of attachment types refers to the percentage of each type of attachment in the target maintenance plan relative to the total number of attachments, such as the percentage of drawings, photos, voice files, scanned copies, etc., relative to the total number of attachments.

[0023] Specifically, the power big data model first classifies and identifies all data in the target maintenance plan, distinguishing between structured and unstructured data, calculating the capacity of both and the proportion of unstructured data; at the same time, it classifies all attachments in the target maintenance plan by type, counts the number of attachments of each type, and calculates the proportion of attachment types.

[0024] For example, if the total data size of the target maintenance plan is 100MB, of which 20MB is structured data such as standardized maintenance forms and 80MB is unstructured data such as on-site photos and voice files, then the proportion of unstructured data is 80%. If the target maintenance plan includes 10 attachments, of which 3 are drawings, 4 are photos, 2 are voice files and 1 is a scan, then the proportion of drawings is 30%, photos are 40%, voice files are 20% and scans are 10%.

[0025] Finally, the proportion of unstructured data is used as the unstructured ratio, and the attachment type hybridity is determined based on the attachment type proportion. Attachment type hybridity refers to the diversity of attachment types in the target maintenance plan, used to quantify the dispersion of attachment types. The higher the attachment type hybridity, the more diverse the attachment types, and the higher the complexity of cross-modal verification.

[0026] For example, the heterogeneity of attachment types can be calculated using the entropy method. A larger entropy value indicates a more uniform distribution of attachment types and a higher degree of heterogeneity. For instance, suppose there are n types of attachments in total, and the proportion of the i-th attachment type is p. i The formula for calculating the heterogeneity H of attachment types is: H = -Σ(p i ×lnp i ), i=1,2,...,n.

[0027] For example, if the target maintenance plan has two types of attachments, accounting for 90% and 10% respectively, then the attachment type heterogeneity H = -(0.9×ln0.9+0.1×ln0.1)≈0.325; if the target maintenance plan has four types of attachments, each accounting for 25%, then the attachment type heterogeneity H = -(4×0.25×ln0.25)≈1.386. Obviously, the more diverse the attachment types, the higher the attachment type heterogeneity.

[0028] In summary, compared to existing technologies, this application identifies target maintenance plans using a large-scale power model, determining key maintenance content, the proportion of unstructured data, and the degree of mixing of attachment types. This provides comprehensive and reliable data support for subsequent verification difficulty coefficient construction and adaptive verification strategy formulation, replacing traditional manual identification methods and significantly improving information extraction efficiency and accuracy.

[0029] S20: Based on the analysis of the key maintenance content, obtain the text data complexity and business risk index, which serve as the first verification difficulty coefficient and the second verification difficulty coefficient.

[0030] In the process of cross-modal information consistency verification of power maintenance plans, traditional verification methods fail to quantitatively assess the text quality differences and potential risks of key maintenance content, relying solely on a uniform verification standard. This leads to difficulties in accurately matching verification resources with actual verification needs. For example, maintenance plans with ambiguous text and incomplete information are prone to omissions due to insufficient verification, high-risk work plans are prone to safety hazards due to inadequate verification standards, while low-risk plans with clear text suffer from resource waste due to over-verification. Therefore, it is necessary to quantify the complexity of text data and the business risk index based on the extracted key maintenance content through professional analysis, and convert these into a first verification difficulty coefficient and a second verification difficulty coefficient, providing a basis for difficulty assessment in the subsequent development of adaptive verification strategies.

[0031] To address the aforementioned issues, this application analyzes and obtains the text data complexity and business risk index based on the key maintenance content, using these as the first and second verification difficulty coefficients.

[0032] Specifically, step S20 in the method includes: Based on the maintenance objects and work types of the key maintenance content, the appropriate text ambiguity recognizer, text completeness recognizer, and business risk assessor are matched and invoked. The key maintenance content is input into the adaptive text ambiguity recognizer and the adaptive text completeness recognizer respectively, and the predicted text ambiguity and predicted text completeness are output. The text data complexity is determined based on the predicted text ambiguity and predicted text completeness, and is used as the first verification difficulty coefficient. The first verification difficulty coefficient is positively correlated with the predicted text ambiguity and negatively correlated with the predicted text completeness. The key maintenance content is input into the adaptive business risk assessor, which outputs a predicted business risk index as the second verification difficulty coefficient.

[0033] In this embodiment, the adaptive text ambiguity recognizer, the adaptive text completeness recognizer, and the adaptive business risk assessor are first invoked based on the maintenance object and job type of the key maintenance content. The adaptive text ambiguity recognizer is a model optimized and trained for a specific maintenance object and job type, used to assess the degree of ambiguity of textual information in the key maintenance content; the adaptive text completeness recognizer is a model optimized and trained for a specific maintenance object and job type, used to assess the degree of completeness of textual information in the key maintenance content; and the adaptive business risk assessor is a model optimized and trained for a specific maintenance object and job type, used to assess the potential business risks of the maintenance operation.

[0034] Specifically, different maintenance objects and work types have different textual expression habits, information integrity requirements and risk points, so it is necessary to match appropriate recognizers and evaluators to improve the accuracy of evaluation.

[0035] For example, if the maintenance object is a main transformer and the operation type is routine maintenance, the system calls up the adapted text ambiguity recognizer, adapted text completeness recognizer, and adapted business risk assessor, which are specifically trained for routine maintenance scenarios of main transformers.

[0036] Secondly, the key maintenance content is input into the adaptive text ambiguity recognizer and the adaptive text completeness recognizer, respectively, and the predicted text ambiguity and predicted text completeness are output. The predicted text ambiguity is an index output by the adaptive text ambiguity recognizer, used to quantify the degree of ambiguity of text information in the key maintenance content, with a value range of [0,1]. A higher predicted text ambiguity indicates more ambiguous text. The predicted text completeness is an index output by the adaptive text completeness recognizer, used to quantify the degree of completeness of text information in the key maintenance content, with a value range of [0,1]. A higher predicted text completeness indicates more complete text.

[0037] For example, if the key maintenance content clearly records all preset key maintenance indicators and is clearly and unambiguously described, after inputting into the adaptive text ambiguity recognizer and the adaptive text completeness recognizer respectively, the output results in a predicted text ambiguity of 0.1 and a predicted text completeness of 0.95; if the key maintenance content is missing some preset key maintenance indicators and some descriptions are ambiguous, after inputting into the adaptive text ambiguity recognizer and the adaptive text completeness recognizer respectively, the output results in a predicted text ambiguity of 0.7 and a predicted text completeness of 0.3.

[0038] Secondly, the text data complexity is determined based on the predicted text ambiguity and predicted text completeness, serving as the first verification difficulty coefficient. This first verification difficulty coefficient is positively correlated with predicted text ambiguity and negatively correlated with predicted text completeness. Text data complexity refers to the complexity of textual information within key maintenance content, directly affecting the difficulty of cross-modal information consistency verification. The more ambiguous and incomplete the text, the more time is required for information verification and completion during verification, resulting in higher verification difficulty.

[0039] For example, the first verification difficulty coefficient can be calculated using a weighted summation formula: first verification difficulty coefficient C1=α×F+β×(1-I), where F is the predicted text ambiguity, I is the predicted text completeness, α and β are weight coefficients and α+β=1, and α and β can be determined through statistical analysis of historical verification data, such as setting α=0.6 and β=0.4.

[0040] For example, if the predicted text ambiguity F=0.5, predicted text completeness I=0.6, α=0.6, and β=0.4, substituting these values ​​into the formula for calculating the first verification difficulty coefficient, we get C1=0.6×0.5+0.4×(1-0.6)=0.46. The first verification difficulty coefficient reflects the impact of the complexity of the text data of the key maintenance content in the target maintenance plan on cross-modal information consistency verification. The higher the first verification difficulty coefficient, the higher the text ambiguity and the lower the text completeness of the key maintenance content. This means that the information verification and completion costs required during the cross-modal information consistency verification process are higher, and the verification difficulty is greater.

[0041] Finally, the key maintenance content is input into the adaptive business risk assessor, which outputs a predicted business risk index as the second verification difficulty coefficient. The predicted business risk index is an indicator output by the adaptive business risk assessor used to quantify the potential business risks of maintenance operations. Its value ranges from [0,1]. A higher predicted business risk index indicates a higher business risk, requiring more rigorous verification of cross-modal information consistency during validation to avoid risks.

[0042] For example, for maintenance work on a 110kV main transformer, if the affected area is the core urban area, the predicted business risk index output by the adaptive business risk assessor may be 0.9; for maintenance work on a 10kV branch line, if the affected area is a single residential community, the predicted business risk index output by the adaptive business risk assessor may be 0.3.

[0043] Furthermore, the construction method of the "adaptive text ambiguity recognizer, adaptive text completeness recognizer, and adaptive business risk assessor" includes: Based on the maintenance objects and operation types of the key maintenance contents, information retrieval is performed using power big data to obtain a sample set of key maintenance contents, a sample set of text ambiguity, a sample set of text completeness, and a sample set of business risk indices. Using the sample key maintenance content set as input data and the sample text ambiguity set as supervision data, the power large model is adjusted and trained to generate an adapted text ambiguity recognizer. Using the sample key maintenance content set as input data and the sample text completeness set as supervision data, the power large model is adjusted and trained to generate an adapted text completeness recognizer. Using the sample set of key maintenance contents as input data and the sample set of business risk indices as monitoring data, the power large model is adjusted and trained to generate an adapted business risk assessor.

[0044] In this embodiment, firstly, constrained by the maintenance objects and operation types of key maintenance content, information retrieval is performed based on power big data to obtain a sample key maintenance content set, a sample text fuzziness set, a sample text completeness set, and a sample business risk index set. Power big data refers to the massive amounts of power-related data accumulated by power companies, including historical maintenance plans, equipment operation records, fault handling reports, safety accident statistics, industry standards and specifications, etc.; the sample key maintenance content set refers to the collection of historical key maintenance content retrieved from power big data that matches the maintenance objects and operation types of the current key maintenance content; the sample text fuzziness set refers to the set of text fuzziness annotation values ​​for each sample in the corresponding sample key maintenance content set, which can be annotated by power industry technicians according to the text fuzziness assessment dimension; the sample text completeness set refers to the set of text completeness annotation values ​​for each sample in the corresponding sample key maintenance content set, which can be annotated by power industry technicians according to the text completeness assessment dimension; the sample business risk index set refers to the set of business risk index annotation values ​​for each sample in the corresponding sample key maintenance content set, which can be annotated by power industry technicians according to the business risk assessment dimension.

[0045] Specifically, the search constraints are that the maintenance object and the operation type are consistent. For example, if the current key maintenance content is a 110kV main transformer and the operation type is routine maintenance, then the search will retrieve all historical maintenance plans in the power big data where the maintenance object is a 110kV main transformer and the operation type is routine maintenance. The key maintenance content will be extracted to form a sample key maintenance content set. Then, power industry technicians will label each sample in the sample key maintenance content set with text fuzziness, text completeness, and business risk index to form the corresponding sample text fuzziness set, sample text completeness set, and sample business risk index set.

[0046] Secondly, using a set of key maintenance content samples as input data and a set of text ambiguity samples as supervisory data, a targeted adjustment and training method is used to train the large-scale power model to generate an adapted text ambiguity recognizer. Specifically, the targeted adjustment and training refers to fine-tuning the pre-trained large-scale power model using a small number of samples, enabling the generated adapted text ambiguity recognizer to accurately output the predicted text ambiguity of key maintenance content under specific maintenance objects and work types.

[0047] For example, the underlying network parameters of the large-scale power model can be frozen, such as the first 10 Transformer layers. Only the sample key maintenance content set is used as input data, and the sample text ambiguity set is used as supervision data to train the top-level network parameters, such as the last 3 Transformer layers and the output layer, to avoid the model forgetting general power domain knowledge. For instance, the sample key maintenance content set and the sample text ambiguity set can be divided into training set, validation set, and test set in a 7:2:1 ratio. The sample key maintenance content in the training set is used as input data, and the corresponding sample text ambiguity is used as supervision data. The training process can use mean squared error as the loss function, AdamW as the optimizer, a learning rate of 1e-5, and 20 training rounds. After each training round, the validation set is used for verification. Training stops when the loss function value of the validation set no longer decreases for 3 consecutive rounds, resulting in an adapted text ambiguity recognizer.

[0048] Furthermore, using a sample set of key maintenance content as input data and a sample set of text completeness as supervisory data, the power system's large-scale model is specifically adjusted and trained to generate an adapted text completeness recognizer. For example, the training method for the adapted text completeness recognizer is consistent with that of the adapted text ambiguity recognizer; that is, by fine-tuning the power system's large-scale model, the generated adapted text completeness recognizer can accurately output the predicted text completeness of key maintenance content under specific maintenance objects and work types.

[0049] Finally, using a sample set of key maintenance items as input data and a sample set of business risk indices as supervisory data, the power system's large-scale model is specifically adjusted and trained to generate an adapted business risk assessor. For example, the training method for the adapted business risk assessor is consistent with that of the adapted text ambiguity recognizer; that is, by fine-tuning the power system's large-scale model, the generated adapted business risk assessor can accurately output the predicted business risk index of key maintenance items under specific maintenance objects and work types.

[0050] In summary, compared to existing technologies, this application analyzes and obtains text data complexity and business risk index based on the key maintenance content, using these as the first and second verification difficulty coefficients. Thus, through in-depth analysis of the key maintenance content, the text data complexity and business risk index are accurately quantified and converted into the first and second verification difficulty coefficients. This provides a difficulty assessment basis for the subsequent formulation of adaptive verification strategies, effectively avoiding insufficient verification or resource waste caused by unified verification standards, and improving the targeting and rationality of cross-modal information consistency verification.

[0051] S30: Determine the third verification difficulty coefficient based on the unstructured ratio and the mixedness of attachment types.

[0052] In cross-modal information consistency verification of power maintenance plans, the processing difficulty of unstructured data and the diversity of attachment types directly affect the verification efficiency and accuracy. However, traditional verification methods do not quantify these two key influencing factors. If only a single standard is used for verification, maintenance plans with a high proportion of unstructured data and mixed attachment types may experience missed detections due to insufficient verification difficulty adaptation, while plans with a high degree of structure and a single attachment type may suffer from over-verification and wasted resources. Therefore, it is necessary to conduct a comprehensive evaluation based on the obtained unstructured data proportion and attachment type mixing to determine a third verification difficulty coefficient that can quantify the impact of these factors on verification difficulty. This will supplement the key difficulty assessment dimension for the subsequent development of a fully adaptable verification strategy.

[0053] To address the aforementioned issues, this application determines the third verification difficulty coefficient based on the unstructured proportion and the mixed nature of attachment types.

[0054] Specifically, step S30 in the method includes: The third verification difficulty coefficient is determined by weighting the unstructured proportion and the attachment type mixing degree, wherein the third verification difficulty coefficient is positively correlated with the unstructured proportion and the attachment type mixing degree.

[0055] In this embodiment of the application, the third verification difficulty coefficient is an indicator used to quantify the impact of the unstructured ratio and the mixedness of accessory types in the target maintenance plan on the verification difficulty. It is positively correlated with the unstructured ratio and the mixedness of accessory types. That is, the higher the unstructured ratio and the higher the mixedness of accessory types, the larger the third verification difficulty coefficient and the higher the verification difficulty.

[0056] For example, the calculation of the third verification difficulty coefficient can be performed using a weighted summation formula: Third verification difficulty coefficient C3 = γ × S + δ × H, where S is the unstructured proportion, H is the attachment type mix degree, γ and δ are weighting coefficients and γ + δ = 1. γ and δ can be dynamically determined based on the importance of the unstructured proportion and attachment type mix degree to the verification difficulty, for example, γ = 0.7 and δ = 0.3.

[0057] For example, if the unstructured proportion S=0.8, the attachment type mixing degree H=1.2, γ=0.7, δ=0.3, then the difficulty coefficient of the third verification C3=0.7×0.8+0.3×1.2=0.92.

[0058] In summary, compared to existing technologies, this application determines the third verification difficulty coefficient based on the aforementioned unstructured content ratio and attachment type mix. Thus, through a comprehensive evaluation of the unstructured content ratio and attachment type mix, the impact of these two factors on cross-modal information consistency verification is accurately quantified, and the third verification difficulty coefficient is determined. This supplements the dimensions of verification difficulty assessment, making the overall difficulty assessment more comprehensive and providing crucial support for the subsequent development of a fully adaptable verification strategy.

[0059] S40: Based on the expected review time limit of the target maintenance plan, formulate the appropriate verification depth and breadth according to the first verification difficulty coefficient, the second verification difficulty coefficient and the third verification difficulty coefficient.

[0060] In the cross-modal information consistency verification of power maintenance plans, traditional verification methods do not take into account the constraints of multi-dimensional verification difficulty coefficients and expected review time limits. They only use fixed verification depth and breadth to carry out the work, which leads to inaccurate verification problems for high-difficulty maintenance plans due to insufficient verification resources and insufficient depth / breadth, while low-difficulty plans suffer from resource waste due to excessive verification. In addition, some plans have not completed the verification due to the lack of consideration of time limit constraints.

[0061] To address the aforementioned issues, this application uses the expected review timeframe of the target maintenance plan as a constraint, and formulates an adaptive verification depth and an adaptive verification breadth based on the first verification difficulty coefficient, the second verification difficulty coefficient, and the third verification difficulty coefficient.

[0062] Specifically, step S40 in the method includes: Obtain the verification depth weight distribution and the verification breadth weight distribution; The first verification difficulty coefficient, the second verification difficulty coefficient, and the third verification difficulty coefficient are weighted and fused according to the verification depth weight distribution to obtain the verification depth difficulty coefficient; The first verification difficulty coefficient, the second verification difficulty coefficient, and the third verification difficulty coefficient are weighted and fused according to the verification breadth weight distribution to obtain the verification breadth difficulty coefficient. Based on the expected review time limit, an appropriate verification depth and an appropriate verification breadth are determined according to the verification depth difficulty coefficient and the verification breadth difficulty coefficient.

[0063] In this embodiment, the verification depth weight distribution and verification breadth weight distribution are first obtained. The verification depth weight distribution refers to the weight allocation ratio of the first verification difficulty coefficient, the second verification difficulty coefficient, and the third verification difficulty coefficient when calculating the verification depth difficulty coefficient; the verification breadth weight distribution refers to the weight allocation ratio of the above three coefficients when calculating the verification breadth difficulty coefficient. The purpose of the weight distribution is to distinguish the degree of influence of different verification difficulty coefficients on the verification depth and verification breadth.

[0064] Secondly, the first, second, and third verification difficulty coefficients are weighted and fused according to the verification depth weight distribution to obtain the verification depth difficulty coefficient. The verification depth difficulty coefficient is an indicator used to quantify the verification depth requirements of the target maintenance plan; a larger coefficient indicates a deeper required verification depth. Specifically, the weighted fusion uses a weighted summation formula: Verification depth difficulty coefficient = w1 × First verification difficulty coefficient + w2 × Second verification difficulty coefficient + w3 × Third verification difficulty coefficient, where w1, w2, and w3 are the weights of the first, second, and third verification difficulty coefficients in the verification depth weight distribution, respectively, and w1 + w2 + w3 = 1.

[0065] For example, if the verification depth weight distribution is w1=0.32, w2=0.42, w3=0.26, the first verification difficulty coefficient is 0.6, the second verification difficulty coefficient is 0.8, and the third verification difficulty coefficient is 0.7, then the verification depth difficulty coefficient is 0.32×0.6+0.42×0.8+0.26×0.7=0.71.

[0066] Next, the first, second, and third verification difficulty coefficients are weighted and fused according to the verification breadth weight distribution to obtain the verification breadth difficulty coefficient. The verification breadth difficulty coefficient is an indicator used to quantify the verification breadth requirements of the target maintenance plan; a larger coefficient indicates a wider required verification breadth. Specifically, the weighted fusion uses a weighted summation formula: Verification Breadth Difficulty Coefficient = v1 × First Verification Difficulty Coefficient + v2 × Second Verification Difficulty Coefficient + v3 × Third Verification Difficulty Coefficient, where v1, v2, and v3 are the weights of the first, second, and third verification difficulty coefficients in the verification breadth weight distribution, respectively, and v1 + v2 + v3 = 1.

[0067] For example, if the breadth weight distribution of the verification is v1=0.35, v2=0.3, v3=0.35, the difficulty coefficient of the first verification is 0.6, the difficulty coefficient of the second verification is 0.8, and the difficulty coefficient of the third verification is 0.7, then the breadth difficulty coefficient of the verification is 0.35×0.6+0.3×0.8+0.35×0.7=0.695.

[0068] Finally, constrained by the expected review timeframe, the appropriate verification depth and breadth are determined based on the difficulty coefficients of the verification depth and breadth. This is because the expected review timeframe determines the maximum available time for verification work. If the expected review timeframe is short, even with a high difficulty coefficient, the verification depth and breadth should be appropriately reduced to ensure timely completion; conversely, if the expected review timeframe is long, the verification depth and breadth can be increased based on the difficulty coefficients to ensure thorough verification.

[0069] Furthermore, the "obtaining the verification depth weight distribution and the verification breadth weight distribution" includes: Based on the aforementioned key maintenance content, and by retrieving historical qualified power maintenance records using power big data, the correlation analysis of cross-modal information consistency verification depth with the first verification difficulty coefficient, the second verification difficulty coefficient, and the third verification difficulty coefficient is performed, and the first depth correlation, the second depth correlation, and the third depth correlation are output. The verification depth weight distribution is determined based on the first depth correlation degree, the second depth correlation degree, and the third depth correlation degree, wherein the verification depth weight and the depth correlation degree are positively correlated. Based on the aforementioned key maintenance content, and by retrieving historical qualified power maintenance records using power big data, the correlation analysis of the cross-modal information consistency verification breadth with the first verification difficulty coefficient, the second verification difficulty coefficient, and the third verification difficulty coefficient is performed, and the first breadth correlation degree, the second breadth correlation degree, and the third breadth correlation degree are output. The verification breadth weight distribution is determined based on the first breadth correlation degree, the second breadth correlation degree, and the third breadth correlation degree, wherein the verification breadth weight and the breadth correlation degree are positively correlated.

[0070] In this embodiment, firstly, based on key maintenance content as constraints, historical qualified power maintenance records are retrieved using power big data. Then, the correlation analysis of the cross-modal information consistency verification depth with the first, second, and third verification difficulty coefficients is performed, outputting the first, second, and third depth correlations. Here, historical qualified power maintenance records refer to maintenance plans that have passed cross-modal information consistency verification in the past and have not encountered any information inconsistencies during subsequent execution. The cross-modal information consistency verification depth refers to the actual verification depth used in the historical qualified power maintenance records, such as model level and semantic reasoning complexity. The correlation analysis uses the Pearson correlation coefficient to calculate the linear correlation between two indicators, with a value range of [-1, 1]. A value closer to 1 indicates a stronger positive correlation, closer to -1 indicates a stronger negative correlation, and closer to 0 indicates no correlation.

[0071] Specifically, the retrieval constraints are the maintenance objects and operation types of the key maintenance content. For example, if the current key maintenance object is a 10kV circuit breaker and the operation type is emergency repair, then all historical qualified power maintenance records in the power big data with 10kV circuit breakers as the maintenance object and emergency repair as the operation type are retrieved. The cross-modal information consistency verification depth, first verification difficulty coefficient, second verification difficulty coefficient, and third verification difficulty coefficient of each historical qualified power maintenance record are extracted. Then, the Pearson correlation coefficients between the cross-modal information consistency verification depth and the three difficulty coefficients are calculated to obtain the first depth correlation, second depth correlation, and third depth correlation.

[0072] For example, if 1,000 historical qualified power maintenance records are retrieved, the correlation coefficient between the cross-modal information consistency verification depth and the first verification difficulty coefficient is 0.6, the correlation coefficient with the second verification difficulty coefficient is 0.8, and the correlation coefficient with the third verification difficulty coefficient is 0.5, that is, the first depth correlation is 0.6, the second depth correlation is 0.8, and the third depth correlation is 0.5.

[0073] Secondly, the verification depth weight distribution is determined based on the first, second, and third depth correlation degrees, where the verification depth weight is positively correlated with the depth correlation degree. Specifically, the verification depth weight distribution can be obtained by normalization, which involves dividing the first, second, and third depth correlation degrees by the sum of their respective depth correlation degrees to obtain the corresponding weights.

[0074] For example, if the first depth correlation is 0.6, the second depth correlation is 0.8, and the third depth correlation is 0.5, then the sum of the three correlations is 0.6 + 0.8 + 0.5 = 1.9. Therefore, the weight of the first depth correlation is 0.6 / 1.9 ≈ 0.32, the weight of the second depth correlation is 0.8 / 1.9 ≈ 0.42, and the weight of the third depth correlation is 0.5 / 1.9 ≈ 0.26. The verification depth weight distribution is: 0.32, 0.42, 0.26.

[0075] Secondly, constrained by key maintenance content, historical qualified power maintenance records are retrieved based on power big data. Correlation analysis is then performed on the breadth of cross-modal information consistency verification with the first, second, and third verification difficulty coefficients, outputting the first, second, and third breadth correlation scores. Specifically, the calculation logic for the first, second, and third breadth correlation scores is consistent with the aforementioned first, second, and third depth correlation scores. Therefore, the calculation method is also consistent with the first, second, and third depth correlation scores, i.e., calculating the Pearson correlation coefficient between the breadth of cross-modal information consistency verification and the first, second, and third verification difficulty coefficients in historical qualified power maintenance records, which serves as the first, second, and third breadth correlation scores.

[0076] For example, if 1000 historical qualified power maintenance records are retrieved, and the correlation coefficients between the cross-modal information consistency verification breadth and the first verification difficulty coefficient are calculated to be 0.7, the correlation coefficients with the second verification difficulty coefficient are 0.6, and the correlation coefficients with the third verification difficulty coefficient are 0.7, then the first breadth correlation degree, the second breadth correlation degree, and the third breadth correlation degree are 0.7, 0.6, and 0.7, respectively.

[0077] Finally, the validation breadth weight distribution is determined based on the first, second, and third breadth relevance, where the validation breadth weight is positively correlated with the breadth relevance. Specifically, the validation breadth weight distribution can also be obtained through normalization by dividing the first, second, and third breadth relevance by the sum of their respective values ​​to obtain the corresponding weights.

[0078] For example, if the first breadth relevance, the second breadth relevance, and the third breadth relevance are 0.7, 0.6, and 0.7 respectively, then the sum of the three relevances is 0.7 + 0.6 + 0.7 = 2.0. Therefore, the weight of the first breadth relevance is 0.7 / 2.0 = 0.35, the weight of the second breadth relevance is 0.6 / 2.0 = 0.3, and the weight of the third breadth relevance is 0.7 / 2.0 = 0.35. That is, the breadth weight distribution is 0.35, 0.3, and 0.35.

[0079] Furthermore, the phrase "using the expected review time limit as a constraint, and formulating an appropriate verification depth and breadth based on the verification depth difficulty coefficient and verification breadth difficulty coefficient" includes: Using the expected review time limit as a constraint, and employing a large power model, the maximum verification depth and maximum verification breadth are obtained based on the key maintenance content, the proportion of unstructured components, and the mixed nature of attachment types. The ratio of the verification depth difficulty coefficient to the preset standard verification depth difficulty coefficient is used as the depth compensation coefficient, and the product of the depth compensation coefficient and the initial verification depth is used as the compensation verification depth, wherein the initial verification depth is set based on the maintenance object and the operation type. The ratio of the verification breadth difficulty coefficient to the preset standard verification breadth difficulty coefficient is used as the breadth compensation coefficient, and the product of the breadth compensation coefficient and the initial verification breadth is used as the compensated verification breadth, wherein the compensated verification breadth is set based on the maintenance object and the operation type. If the compensated verification depth is less than or equal to the maximum verification depth, the compensated verification depth is used as the adaptation verification depth; if the compensated verification depth is greater than the maximum verification depth, the maximum verification depth is used as the adaptation verification depth. If the compensated verification breadth is less than or equal to the maximum verification breadth, the compensated verification breadth is used as the adapted verification breadth; if the compensated verification breadth is greater than the maximum verification breadth, the maximum verification breadth is used as the adapted verification breadth.

[0080] In this embodiment, firstly, constrained by the expected review timeframe, a large power model is used to analyze the maximum verification depth and maximum verification breadth based on key maintenance content, unstructured proportion, and attachment type mix. The maximum verification depth refers to the highest verification depth that can be completed within the expected review timeframe; exceeding this depth will prevent timely verification. The maximum verification breadth refers to the broadest verification scope that can be completed within the expected review timeframe; exceeding this breadth will prevent timely verification.

[0081] Specifically, the power big data model analyzes historical verification data to establish a mapping relationship model between expected review time limits, key maintenance content, unstructured data ratio, attachment type mix, maximum verification depth, and maximum verification breadth. For example, if the expected review time limit is 2 hours, the key maintenance content is routine maintenance of a 110kV main transformer, the unstructured data ratio is 0.8, and the attachment type mix is ​​1.2, then the power big data model, based on historical data mapping, obtains a maximum verification depth of 5 levels and a maximum verification breadth of 80%.

[0082] Secondly, the ratio of the verification depth difficulty coefficient to the preset standard verification depth difficulty coefficient is used as the depth compensation coefficient. The product of the depth compensation coefficient and the initial verification depth is used as the compensated verification depth. The initial verification depth is set based on the maintenance object and the type of work. The preset standard verification depth difficulty coefficient refers to the standard difficulty coefficient set by the power company based on the industry average and its own management requirements, such as 0.5, which serves as the calculation benchmark for the depth compensation coefficient. The initial verification depth refers to the basic verification depth set for a specific maintenance object and type of work, such as level 3 for conventional transformer maintenance. The depth compensation coefficient is used to dynamically adjust the initial verification depth according to the actual difficulty coefficient, ensuring that the verification depth matches the actual difficulty.

[0083] Specifically, the formula for calculating the depth compensation coefficient is: Depth compensation coefficient = Verification depth difficulty coefficient / Preset standard verification depth difficulty coefficient, and the formula for calculating the compensated verification depth is: Compensated verification depth = Depth compensation coefficient × Initial verification depth.

[0084] For example, if the verification depth difficulty coefficient is 0.71, the preset standard verification depth difficulty coefficient is 0.5, and the initial verification depth is level 3, then the depth compensation coefficient is 0.71 / 0.5=1.42, and the compensated verification depth is 1.42×3=4.26 levels.

[0085] Secondly, the ratio of the verification breadth difficulty coefficient to the preset standard verification breadth difficulty coefficient is used as the breadth compensation coefficient. The product of the breadth compensation coefficient and the initial verification breadth is used as the compensated verification breadth. The compensated verification breadth is set based on the maintenance object and the type of work. The preset standard verification breadth difficulty coefficient refers to the standard breadth difficulty coefficient set by the power company, such as 0.5. The initial verification breadth refers to the basic verification breadth set for a specific maintenance object and type of work, such as 60% for conventional transformer maintenance.

[0086] Specifically, the formula for calculating the breadth compensation coefficient is: Breadth compensation coefficient = Verification breadth difficulty coefficient / Preset standard verification breadth difficulty coefficient, and the formula for calculating the compensated verification breadth is: Compensated verification breadth = Breadth compensation coefficient × Initial verification breadth.

[0087] For example, if the verification breadth difficulty coefficient is 0.695, the preset standard verification breadth difficulty coefficient is 0.5, and the initial verification breadth is 60%, then the breadth compensation coefficient = 0.695 / 0.5 = 1.39, and the compensated verification breadth = 1.39 × 60% = 83.4%.

[0088] Furthermore, if the compensated verification depth is less than or equal to the maximum verification depth, the compensated verification depth is used as the adapted verification depth; if the compensated verification depth is greater than the maximum verification depth, the maximum verification depth is used as the adapted verification depth. In this way, by constraining the verification depth through the expected review timeframe, it is ensured that the adapted verification depth meets both the actual difficulty requirements and can be completed within the expected review timeframe.

[0089] For example, if the compensation verification depth is 4.26 levels and the maximum verification depth is 5 levels, the output adaptation verification depth is 4.2 levels; conversely, if the compensation verification depth is 5.5 levels and the maximum verification depth is 5 levels, the output adaptation verification depth is 5 levels.

[0090] Finally, if the compensated verification breadth is less than or equal to the maximum verification breadth, the compensated verification breadth is used as the adapted verification breadth; if the compensated verification breadth is greater than the maximum verification breadth, the maximum verification breadth is used as the adapted verification breadth. In this way, by constraining the verification breadth through the expected review time limit, it is ensured that the adapted verification breadth both meets the actual difficulty requirements and can be completed within the expected review time limit.

[0091] For example, if the compensation check breadth is 83.4% and the maximum check breadth is 80%, then the output adaptation check breadth is 80%.

[0092] Furthermore, the breadth of verification refers to the coverage ratio and scope of the core fields to be verified, while the depth of verification refers to the model hierarchy and semantic reasoning complexity of the information consistency review model invoked.

[0093] In this embodiment, the verification breadth refers to the coverage ratio and coverage scope of the core fields to be verified. The core fields to be verified refer to the specific data fields corresponding to preset key maintenance indicators, such as the name of the maintenance object, the start and end points of the maintenance time, and the area number of the affected area. The coverage ratio refers to the proportion of the number of core fields actually verified to the total number of core fields to be verified. The coverage scope refers to the business scenario scope corresponding to the core fields actually verified, such as whether the verification of the affected area field covers all involved power supply areas, or whether the verification of the associated equipment field covers all related equipment.

[0094] For example, there are 10 core fields to be verified, and 8 core fields are actually verified, so the coverage ratio is 80%; the associated device field involves 5 devices, and the cross-modal information consistency of all 5 devices is actually verified, so the coverage is 100%.

[0095] In this embodiment, the verification depth refers to the model level and semantic reasoning complexity of the invoked information consistency review model. The information consistency review model refers to a model used to perform cross-modal information consistency verification, such as a text-table consistency verification model or a text-drawing consistency verification model. The model level reflects the network structure complexity of the information consistency review model. Semantic reasoning complexity refers to the reasoning depth of the information consistency review model when making information consistency judgments, such as simple keyword matching or phrase semantic matching.

[0096] For example, if the intermediate layer of the information consistency review model is called and sentence logic matching is used for reasoning, the verification depth is higher; if the basic layer is called and only simple keyword matching is used for reasoning, the verification depth is lower.

[0097] In summary, compared to existing technologies, this application uses the expected review timeframe of the target maintenance plan as a constraint, and formulates an appropriate verification depth and breadth based on the first, second, and third verification difficulty coefficients. In this way, an appropriate verification depth and breadth are formulated that precisely matches the verification difficulty and meets the expected review timeframe requirements, achieving a reasonable allocation of verification resources and effectively avoiding problems such as over-verification, under-verification, or verification exceeding the timeframe, while balancing the efficiency and accuracy of cross-modal information consistency verification.

[0098] S50: Perform cross-modal information consistency verification on the target maintenance plan according to the adaptation verification depth and adaptation verification breadth.

[0099] The aforementioned steps, through thorough quantitative analysis and precise calculation, yielded the depth and breadth of the adaptation verification. In the final execution stage of the cross-modal information consistency verification of the power maintenance plan, the target maintenance plan can be verified for cross-modal information consistency according to the depth and breadth of the adaptation verification, ensuring that the verification work fully covers the core elements and accurately matches the actual needs.

[0100] In this embodiment, the target maintenance plan is subjected to cross-modal information consistency verification according to the depth and breadth of the adaptation verification. Specifically, cross-modal information consistency verification refers to verifying whether the descriptions of the same maintenance indicator are consistent among different modal information (such as text, tables, drawings, voice, etc.) in the target maintenance plan. The maintenance indicator includes at least the maintenance object, maintenance time, scope of impact, work type, work content, associated equipment, safety measures, and power outage requirements.

[0101] For example, if the target maintenance plan is: the routine maintenance plan for the #2 main transformer of the 110kV Chengdong Substation, which includes a text-based work instruction, a table-based power outage schedule, CAD-formatted equipment wiring diagrams, and a voice-based safety briefing record, then the following verifications are performed according to the depth and breadth of the adaptation verification: whether the maintenance time described in the text as 9:00-17:00 on October 15, 2024 is consistent with the power outage period recorded in the table as 8:30-17:30 on October 15, 2024, and whether the associated equipment marked in the drawing as the low-voltage side switch of the #2 main transformer is consistent with the associated equipment described in the text as the low-voltage side switch of the #2 main transformer, etc.

[0102] In summary, compared to existing technologies, this application performs cross-modal information consistency verification on the target maintenance plan according to the aforementioned adaptation verification depth and breadth. This achieves comprehensive coverage and accurate verification of the multimodal core information of the target maintenance plan, effectively avoiding information conflict omissions and over-verification issues, improving the accuracy and efficiency of cross-modal information consistency verification, and providing a reliable guarantee for the safe and compliant execution of the maintenance plan.

[0103] In summary, the embodiments of this application have at least the following technical effects: Compared to existing technologies, this application first identifies target maintenance plans through a large-scale power model, determining key maintenance content, the proportion of unstructured data, and the degree of mixing of attachment types. This provides comprehensive and reliable data support for subsequent verification difficulty coefficient construction and adaptive verification strategy formulation, replacing traditional manual identification methods and significantly improving information extraction efficiency and accuracy.

[0104] Secondly, this application analyzes and obtains the text data complexity and business risk index based on the key maintenance content, using these as the first and second verification difficulty coefficients. Thus, through in-depth analysis of the key maintenance content, the text data complexity and business risk index are accurately quantified and converted into the first and second verification difficulty coefficients. This provides a difficulty assessment basis for the subsequent formulation of adaptation verification strategies, effectively avoiding insufficient verification or resource waste caused by unified verification standards, and improving the targeting and rationality of cross-modal information consistency verification.

[0105] Furthermore, this application determines the third verification difficulty coefficient based on the aforementioned unstructured proportion and attachment type mix. Thus, through a comprehensive evaluation of the unstructured proportion and attachment type mix, the impact of these two factors on cross-modal information consistency verification is accurately quantified, and the third verification difficulty coefficient is determined. This supplements the dimensions of verification difficulty assessment, making the overall difficulty assessment more comprehensive and providing crucial support for the subsequent development of a fully adapted verification strategy.

[0106] Furthermore, this application uses the expected review timeframe of the target maintenance plan as a constraint, and formulates an appropriate verification depth and breadth based on the first verification difficulty coefficient, the second verification difficulty coefficient, and the third verification difficulty coefficient. In this way, an appropriate verification depth and breadth are formulated that precisely matches the verification difficulty and meets the expected review timeframe requirements, achieving a reasonable allocation of verification resources and effectively avoiding problems such as over-verification, under-verification, or verification failing to complete within the time limit, while balancing the efficiency and accuracy of cross-modal information consistency verification.

[0107] Finally, this application performs cross-modal information consistency verification on the target maintenance plan according to the aforementioned adaptation verification depth and adaptation verification breadth. This achieves comprehensive coverage and accurate verification of the multimodal core information of the target maintenance plan, effectively avoiding information conflict omissions and over-verification issues, improving the accuracy and efficiency of cross-modal information consistency verification, and providing a reliable guarantee for the safe and compliant execution of the maintenance plan.

[0108] Through the aforementioned technical solution, this application utilizes power big data as its core data support. By employing a large power model, it accurately identifies key maintenance content, the proportion of unstructured data, and the complexity of attachment types in the target maintenance plan. This allows for the construction of first, second, and third verification difficulty coefficients. Furthermore, by constraining the expected review timeframe of the target maintenance plan, it determines the appropriate verification depth and breadth, thereby enabling cross-modal information consistency verification. This effectively overcomes the shortcomings of existing review mechanisms, such as lack of flexibility and a one-size-fits-all approach. It achieves dynamic allocation of verification resources based on the inherent risks, complexity, and urgency of the maintenance task. This approach avoids the inefficiency caused by excessive verification of simple plans while ensuring sufficient verification of high-risk, complex plans. It improves review efficiency while guaranteeing cross-modal information consistency and the safety of maintenance operations, achieving a dynamic balance between safety and efficiency.

[0109] Example 2, as Figure 2 As shown, based on the same inventive concept as the cross-modal information consistency verification method for maintenance plans based on a large model provided in Embodiment 1, this embodiment of the invention also provides a cross-modal information consistency verification system for maintenance plans based on a large model, including: The maintenance plan feature identification module 11 is used to identify target maintenance plans through the power big data model, and determine key maintenance contents, unstructured proportions, and the degree of mixing of attachment types. The first and second verification difficulty coefficient acquisition module 12 is used to analyze and obtain the text data complexity and business risk index based on the key maintenance content, and use them as the first verification difficulty coefficient and the second verification difficulty coefficient. The third verification difficulty coefficient acquisition module 13 is used to determine the third verification difficulty coefficient based on the unstructured ratio and the mixedness of attachment types. The verification depth and breadth acquisition module 14 is used to formulate an appropriate verification depth and an appropriate verification breadth based on the first verification difficulty coefficient, the second verification difficulty coefficient, and the third verification difficulty coefficient, with the expected review time limit of the target maintenance plan as a constraint. The consistency verification module 15 is used to perform cross-modal information consistency verification on the target maintenance plan according to the adaptation verification depth and adaptation verification breadth.

[0110] The maintenance plan feature recognition module 11 is specifically used for: The preset key maintenance indicators and the target maintenance plan are input into the power big data model for information identification, and the key maintenance content is output. The preset key maintenance indicators include at least the maintenance object, maintenance time, scope of impact, operation type, work content, related equipment, safety measures and power outage requirements. The target maintenance plan is input into the power big data model for unstructured data identification, and the proportion of unstructured data and the proportion of attachment types are output. The proportion of unstructured data is used as the unstructured ratio, and the attachment type mixing degree is determined based on the proportion of attachment types.

[0111] Specifically, the first and second verification difficulty coefficient acquisition modules 12 are used for: Based on the maintenance objects and work types of the key maintenance content, the appropriate text ambiguity recognizer, text completeness recognizer, and business risk assessor are matched and invoked. The key maintenance content is input into the adaptive text ambiguity recognizer and the adaptive text completeness recognizer respectively, and the predicted text ambiguity and predicted text completeness are output. The text data complexity is determined based on the predicted text ambiguity and predicted text completeness, and is used as the first verification difficulty coefficient. The first verification difficulty coefficient is positively correlated with the predicted text ambiguity and negatively correlated with the predicted text completeness. The key maintenance content is input into the adaptive business risk assessor, which outputs a predicted business risk index as the second verification difficulty coefficient.

[0112] Furthermore, the construction method of the "adaptive text ambiguity recognizer, adaptive text completeness recognizer, and adaptive business risk assessor" includes: Based on the maintenance objects and operation types of the key maintenance contents, information retrieval is performed using power big data to obtain a sample set of key maintenance contents, a sample set of text ambiguity, a sample set of text completeness, and a sample set of business risk indices. Using the sample key maintenance content set as input data and the sample text ambiguity set as supervision data, the power large model is adjusted and trained to generate an adapted text ambiguity recognizer. Using the sample key maintenance content set as input data and the sample text completeness set as supervision data, the power large model is adjusted and trained to generate an adapted text completeness recognizer. Using the sample set of key maintenance contents as input data and the sample set of business risk indices as monitoring data, the power large model is adjusted and trained to generate an adapted business risk assessor.

[0113] The third verification difficulty coefficient acquisition module 13 is specifically used for: The third verification difficulty coefficient is determined by weighting the unstructured proportion and the attachment type mixing degree, wherein the third verification difficulty coefficient is positively correlated with the unstructured proportion and the attachment type mixing degree.

[0114] Specifically, the verification depth and breadth acquisition module 14 is used for: Obtain the verification depth weight distribution and the verification breadth weight distribution; The first verification difficulty coefficient, the second verification difficulty coefficient, and the third verification difficulty coefficient are weighted and fused according to the verification depth weight distribution to obtain the verification depth difficulty coefficient; The first verification difficulty coefficient, the second verification difficulty coefficient, and the third verification difficulty coefficient are weighted and fused according to the verification breadth weight distribution to obtain the verification breadth difficulty coefficient. Based on the expected review time limit, an appropriate verification depth and an appropriate verification breadth are determined according to the verification depth difficulty coefficient and the verification breadth difficulty coefficient.

[0115] Furthermore, the "obtaining the verification depth weight distribution and the verification breadth weight distribution" includes: Based on the aforementioned key maintenance content, and by retrieving historical qualified power maintenance records using power big data, the correlation analysis of cross-modal information consistency verification depth with the first verification difficulty coefficient, the second verification difficulty coefficient, and the third verification difficulty coefficient is performed, and the first depth correlation, the second depth correlation, and the third depth correlation are output. The verification depth weight distribution is determined based on the first depth correlation degree, the second depth correlation degree, and the third depth correlation degree, wherein the verification depth weight and the depth correlation degree are positively correlated. Based on the aforementioned key maintenance content, and by retrieving historical qualified power maintenance records using power big data, the correlation analysis of the cross-modal information consistency verification breadth with the first verification difficulty coefficient, the second verification difficulty coefficient, and the third verification difficulty coefficient is performed, and the first breadth correlation degree, the second breadth correlation degree, and the third breadth correlation degree are output. The verification breadth weight distribution is determined based on the first breadth correlation degree, the second breadth correlation degree, and the third breadth correlation degree, wherein the verification breadth weight and the breadth correlation degree are positively correlated.

[0116] Furthermore, the phrase "using the expected review time limit as a constraint, and formulating an appropriate verification depth and breadth based on the verification depth difficulty coefficient and verification breadth difficulty coefficient" includes: Using the expected review time limit as a constraint, and employing a large power model, the maximum verification depth and maximum verification breadth are obtained based on the key maintenance content, the proportion of unstructured components, and the mixed nature of attachment types. The ratio of the verification depth difficulty coefficient to the preset standard verification depth difficulty coefficient is used as the depth compensation coefficient, and the product of the depth compensation coefficient and the initial verification depth is used as the compensation verification depth, wherein the initial verification depth is set based on the maintenance object and the operation type. The ratio of the verification breadth difficulty coefficient to the preset standard verification breadth difficulty coefficient is used as the breadth compensation coefficient, and the product of the breadth compensation coefficient and the initial verification breadth is used as the compensated verification breadth, wherein the compensated verification breadth is set based on the maintenance object and the operation type. If the compensated verification depth is less than or equal to the maximum verification depth, the compensated verification depth is used as the adaptation verification depth; if the compensated verification depth is greater than the maximum verification depth, the maximum verification depth is used as the adaptation verification depth. If the compensated verification breadth is less than or equal to the maximum verification breadth, the compensated verification breadth is used as the adapted verification breadth; if the compensated verification breadth is greater than the maximum verification breadth, the maximum verification breadth is used as the adapted verification breadth.

[0117] Furthermore, the breadth of verification refers to the coverage ratio and scope of the core fields to be verified, while the depth of verification refers to the model hierarchy and semantic reasoning complexity of the information consistency review model invoked.

[0118] The consistency verification module 15 is specifically used for: The target maintenance plan is subjected to cross-modal information consistency verification according to the aforementioned adaptation verification depth and adaptation verification breadth.

[0119] In summary, the embodiments of this application have at least the following technical effects: Compared to existing technologies, this application firstly uses a maintenance plan feature identification module 11 to identify target maintenance plans through a large power model, determining key maintenance content, unstructured data ratio, and accessory type mixing degree, providing comprehensive and reliable data support for subsequent verification difficulty coefficient construction and adaptive verification strategy formulation. Secondly, the first and second verification difficulty coefficient acquisition modules 12 analyze key maintenance content to obtain text data complexity and business risk index, which serve as the first and second verification difficulty coefficients, effectively avoiding insufficient verification or resource waste caused by unified verification standards, and improving the pertinence and rationality of cross-modal information consistency verification. Thirdly, the third verification difficulty coefficient acquisition module 13 determines the third verification difficulty coefficient based on the unstructured data ratio and accessory type mixing degree assessment, supplementing the verification difficulty assessment dimension, making the overall difficulty assessment more comprehensive, and providing key support for subsequent formulation of a fully adapted verification strategy. Furthermore, the verification depth and breadth acquisition module 14, constrained by the expected review timeframe of the target maintenance plan, determines the appropriate verification depth and breadth based on the first, second, and third verification difficulty coefficients. This achieves a reasonable allocation of verification resources and effectively avoids issues such as over-verification, insufficient verification, or verification failing to complete within timeout. Finally, the consistency verification module 15 performs cross-modal information consistency verification on the target maintenance plan according to the appropriate verification depth and breadth, effectively avoiding information conflict omissions and over-verification issues, and improving the accuracy and efficiency of cross-modal information consistency verification.

[0120] In this way, by using power big data as the core data support, the inefficiency caused by excessive verification of simple plans is avoided, and high-risk and complex plans are fully verified. While ensuring the consistency of cross-modal information in maintenance plans and the safety of maintenance operations, the efficiency of review is improved, and a dynamic balance between safety and efficiency is achieved.

[0121] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0122] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0123] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0124] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0125] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0126] Although preferred embodiments of the invention have been described, those skilled in the art, once they have learned the basic inventive concept, can make other changes and modifications to these embodiments.

[0127] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for cross-modal information consistency verification of maintenance plans based on a large model, characterized in that, The methods include: The target maintenance plan is identified through a large power model, and the key maintenance contents, unstructured proportion, and mixed nature of attachment types are determined. Based on the analysis of the key maintenance content, the complexity of the text data and the business risk index are obtained as the first and second verification difficulty coefficients. The difficulty coefficient of the third verification is determined based on the unstructured ratio and the mixedness of attachment types. Based on the expected review time limit of the target maintenance plan, the appropriate verification depth and breadth are determined according to the first verification difficulty coefficient, the second verification difficulty coefficient, and the third verification difficulty coefficient. The target maintenance plan is subjected to cross-modal information consistency verification according to the aforementioned adaptation verification depth and adaptation verification breadth.

2. The method for cross-modal information consistency verification of maintenance plans based on a large model according to claim 1, characterized in that, The target maintenance plan is identified through a large power model, determining key maintenance content, the proportion of unstructured components, and the degree of mixture of attachment types, including: The preset key maintenance indicators and the target maintenance plan are input into the power big data model for information identification, and the key maintenance content is output. The preset key maintenance indicators include at least the maintenance object, maintenance time, scope of impact, operation type, work content, related equipment, safety measures and power outage requirements. The target maintenance plan is input into the power big data model for unstructured data identification, and the proportion of unstructured data and the proportion of attachment types are output. The proportion of unstructured data is used as the unstructured ratio, and the attachment type mixing degree is determined based on the proportion of attachment types.

3. The method for cross-modal information consistency verification of maintenance plans based on a large model according to claim 2, characterized in that, Based on the analysis of the key maintenance content, the complexity of the text data and the business risk index are obtained, including: Based on the maintenance objects and work types of the key maintenance content, the appropriate text ambiguity recognizer, text completeness recognizer, and business risk assessor are matched and invoked. The key maintenance content is input into the adaptive text ambiguity recognizer and the adaptive text completeness recognizer respectively, and the predicted text ambiguity and predicted text completeness are output. The text data complexity is determined based on the predicted text ambiguity and predicted text completeness, and is used as the first verification difficulty coefficient. The first verification difficulty coefficient is positively correlated with the predicted text ambiguity and negatively correlated with the predicted text completeness. The key maintenance content is input into the adaptive business risk assessor, which outputs a predicted business risk index as the second verification difficulty coefficient.

4. The method for cross-modal information consistency verification of maintenance plans based on a large model according to claim 3, characterized in that, The construction methods for the adaptive text ambiguity recognizer, the adaptive text completeness recognizer, and the adaptive business risk assessor include: Based on the maintenance objects and operation types of the key maintenance contents, information retrieval is performed using power big data to obtain a sample set of key maintenance contents, a sample set of text ambiguity, a sample set of text completeness, and a sample set of business risk indices. Using the sample key maintenance content set as input data and the sample text ambiguity set as supervision data, the power large model is adjusted and trained in a targeted manner to generate an adapted text ambiguity recognizer. Using the sample key maintenance content set as input data and the sample text completeness set as supervision data, the power large model is adjusted and trained to generate an adapted text completeness recognizer. Using the sample set of key maintenance contents as input data and the sample set of business risk indices as monitoring data, the power large model is adjusted and trained to generate an adapted business risk assessor.

5. The method for cross-modal information consistency verification of maintenance plans based on a large model according to claim 1, characterized in that, The third verification difficulty coefficient is determined by weighting the unstructured proportion and the attachment type mixing degree, wherein the third verification difficulty coefficient is positively correlated with the unstructured proportion and the attachment type mixing degree.

6. The method for cross-modal information consistency verification of maintenance plans based on a large model according to claim 1, characterized in that, Constrained by the expected review timeframe of the target maintenance plan, and based on the first verification difficulty coefficient, the second verification difficulty coefficient, and the third verification difficulty coefficient, an adaptive verification depth and an adaptive verification breadth are determined, including: Obtain the verification depth weight distribution and the verification breadth weight distribution; The first verification difficulty coefficient, the second verification difficulty coefficient, and the third verification difficulty coefficient are weighted and fused according to the verification depth weight distribution to obtain the verification depth difficulty coefficient; The first verification difficulty coefficient, the second verification difficulty coefficient, and the third verification difficulty coefficient are weighted and fused according to the verification breadth weight distribution to obtain the verification breadth difficulty coefficient. Based on the expected review time limit, an appropriate verification depth and an appropriate verification breadth are determined according to the verification depth difficulty coefficient and the verification breadth difficulty coefficient.

7. The method for cross-modal information consistency verification of maintenance plans based on a large model according to claim 6, characterized in that, Obtain the verification depth weight distribution and verification breadth weight distribution, including: Based on the aforementioned key maintenance content, and by retrieving historical qualified power maintenance records using power big data, the correlation analysis of cross-modal information consistency verification depth with the first verification difficulty coefficient, the second verification difficulty coefficient, and the third verification difficulty coefficient is performed, and the first depth correlation, the second depth correlation, and the third depth correlation are output. The verification depth weight distribution is determined based on the first depth correlation degree, the second depth correlation degree, and the third depth correlation degree, wherein the verification depth weight and the depth correlation degree are positively correlated. Based on the aforementioned key maintenance content, and by retrieving historical qualified power maintenance records using power big data, the correlation analysis of the cross-modal information consistency verification breadth with the first verification difficulty coefficient, the second verification difficulty coefficient, and the third verification difficulty coefficient is performed, and the first breadth correlation degree, the second breadth correlation degree, and the third breadth correlation degree are output. The verification breadth weight distribution is determined based on the first breadth correlation degree, the second breadth correlation degree, and the third breadth correlation degree, wherein the verification breadth weight and the breadth correlation degree are positively correlated.

8. The method for cross-modal information consistency verification of maintenance plans based on a large model according to claim 6, characterized in that, Constrained by the expected review timeframe, an appropriate verification depth and breadth are determined based on the verification depth difficulty coefficient and verification breadth difficulty coefficient, including: Using the expected review time limit as a constraint, and employing a large power model, the maximum verification depth and maximum verification breadth are obtained based on the key maintenance content, the proportion of unstructured components, and the mixed nature of attachment types. The ratio of the verification depth difficulty coefficient to the preset standard verification depth difficulty coefficient is used as the depth compensation coefficient, and the product of the depth compensation coefficient and the initial verification depth is used as the compensation verification depth, wherein the initial verification depth is set based on the maintenance object and the operation type. The ratio of the verification breadth difficulty coefficient to the preset standard verification breadth difficulty coefficient is used as the breadth compensation coefficient, and the product of the breadth compensation coefficient and the initial verification breadth is used as the compensated verification breadth, wherein the compensated verification breadth is set based on the maintenance object and the operation type. If the compensated verification depth is less than or equal to the maximum verification depth, the compensated verification depth is used as the adaptation verification depth; if the compensated verification depth is greater than the maximum verification depth, the maximum verification depth is used as the adaptation verification depth. If the compensated verification breadth is less than or equal to the maximum verification breadth, the compensated verification breadth is used as the adapted verification breadth; if the compensated verification breadth is greater than the maximum verification breadth, the maximum verification breadth is used as the adapted verification breadth.

9. The method for cross-modal information consistency verification of maintenance plans based on a large model according to claim 8, characterized in that, The breadth of verification refers to the coverage ratio and scope of the core fields to be verified, while the depth of verification refers to the model hierarchy and semantic reasoning complexity of the information consistency review model being invoked.

10. A maintenance plan cross-modal information consistency verification system based on a large model, characterized in that, The method for performing cross-modal information consistency verification of maintenance plans based on a large model as described in any one of claims 1-9 includes: The maintenance plan feature recognition module is used to identify target maintenance plans through a large power model, and to determine key maintenance content, unstructured proportion, and the degree of mixing of attachment types. The first and second verification difficulty coefficient acquisition modules are used to analyze and obtain the text data complexity and business risk index based on the key maintenance content, which serve as the first and second verification difficulty coefficients. The third verification difficulty coefficient acquisition module is used to determine the third verification difficulty coefficient based on the unstructured ratio and the mixedness of attachment types. The verification depth and breadth acquisition module is used to determine the appropriate verification depth and breadth based on the first verification difficulty coefficient, the second verification difficulty coefficient, and the third verification difficulty coefficient, constrained by the expected review time limit of the target maintenance plan. The consistency verification module is used to perform cross-modal information consistency verification on the target maintenance plan according to the adaptation verification depth and adaptation verification breadth.

Images