Engineering feasibility file security compliance examination system and method

Abstract

The invention discloses an engineering feasibility file security compliance examination system and method. Through the pre-stored first instruction template and the pre-stored second instruction template customized for different first auditing types, pertinence, accuracy and standardization of auditing different to-be-audited project feasibility files are ensured, and standard errors existing in manual auditing of different to-be-audited project feasibility files are avoided. Through the pre-trained data extraction model and auditing model, the required information can be automatically identified and extracted from the file, and corresponding auditing processing is performed, so that the auditing efficiency is improved, relatively high accuracy can be kept, manual auditing is not needed, the difficulty of auditing the feasibility file of the project to be audited is reduced, and the auditing efficiency of the project to be audited is improved. And errors caused by manual auditing are avoided. According to the technical scheme, audit of multiple target first audit types can be supported, comprehensive audit of project feasibility files is achieved, various problems possibly existing in the files can be found easily, and the reliability of audit results is improved.

Description

Technical Field

[0001] The present application relates to the field of building information management, and in particular to a system and method for reviewing the safety and compliance of engineering feasibility documents. Background Art

[0002] In the field of engineering project management, engineering feasibility study reports are an important basis for evaluating project feasibility, technical feasibility and economic rationality. However, with the increasing scale and complexity of engineering projects, the traditional review of engineering feasibility documents faces many challenges. These challenges mainly include low review efficiency, uneven review quality, and difficulty in fully covering all compliance and safety requirements.

[0003] Traditional review methods mainly rely on manual reading and comparison, which is not only time-consuming and labor-intensive, but also inefficient. When faced with a large number of complex engineering feasibility study reports, manual review is often difficult to complete in a short period of time, resulting in a longer review cycle and affecting project progress. Moreover, manual review depends on the professional level and experience of the reviewers, but there are differences between different reviewers in terms of professional ability, work attitude and sense of responsibility, which may lead to subjectivity and uncertainty in the review results. Some minor errors or omissions may be ignored due to human factors, which in turn affects the accuracy and reliability of the review results.

[0004] In addition, due to differences in audit standards between different projects and institutions, the same document may receive different audit conclusions in different environments, which increases the uncertainty and controversy of the audit results. Therefore, there is an urgent need for an efficient, accurate and comprehensive review method to replace the traditional manual review method.

[0005] Based on this, there is an urgent need for an automated and standardized engineering feasibility document safety compliance review system. Summary of the invention

[0006] The present application provides a system and method for the security and compliance review of engineering feasibility documents, which is used to solve the problems of low efficiency, great difficulty in review, and inconsistent standards in the existing traditional review method that relies on manual review for efficiency and quality.

[0007] In a first aspect, the present application provides a system for reviewing the safety and compliance of engineering feasibility documents, the system comprising:

[0008] A file acquisition module, used to acquire the feasibility file of the project to be reviewed and its corresponding target review information; wherein the target review information includes the first review type of each target;

[0009] An instruction template storage module is used to store first instruction templates and second instruction templates corresponding to each first audit type; wherein the first instruction template of any first audit type is used to instruct to identify and extract information required for the first audit type from the engineering feasibility file, and the second instruction template of any first audit type is used to instruct to perform the first audit type audit on the audit data of the first audit type;

[0010] A data extraction module is used to obtain the first instruction templates corresponding to the first audit types of the respective targets from the instruction template storage module; obtain the audit data corresponding to the first audit types of the respective targets through a pre-trained data extraction model based on the data extraction prompts corresponding to the first audit types of the respective targets; wherein the data extraction prompts include the first instruction templates corresponding to the first audit types of the respective targets and the feasibility files of the projects to be audited;

[0011] A data review module is used to obtain, for each target first review type, a second instruction template of the target first review type from the instruction template storage module; determine the review result of the feasibility document of the project to be reviewed on the target first review type based on the review prompt corresponding to the target first review type through a pre-trained review model supporting the target first review type; wherein the review prompt corresponding to the target first review type includes the second instruction template of the target first review type and the review data of the target first review type;

[0012] The audit result output module is used to output the audit results corresponding to the first audit type of each target.

[0013] In a second aspect, the present application also provides a method for safety compliance review of engineering feasibility documents, the method comprising:

[0014] Obtaining the feasibility documents of the project to be reviewed and its corresponding target review information; wherein the target review information includes the first review type of each target;

[0015] Acquire the first instruction templates corresponding to the first audit types of the targets from the pre-saved first instruction templates corresponding to the first audit types;

[0016] Acquire the audit data corresponding to the first audit type of each target through a pre-trained data extraction model based on the data extraction prompts corresponding to the first audit type of each target; wherein the data extraction prompts include the first instruction templates corresponding to the first audit type of each target and the feasibility file of the project to be audited;

[0017] For each target first audit type, obtain the second instruction template of the target first audit type from the pre-saved second instruction templates corresponding to each first audit type; determine the audit result of the feasibility document of the project to be audited on the target first audit type based on the audit prompt corresponding to the target first audit type through the pre-trained audit model supporting the target first audit type; wherein the audit prompt corresponding to the target first audit type includes the second instruction template of the target first audit type and the audit data of the target first audit type;

[0018] Output the audit results corresponding to the first audit type of each target.

[0019] The beneficial effects of this application are as follows:

[0020] 1. Through the pre-stored first instruction template and second instruction template customized for different first audit types, the pertinence, accuracy and standardization of the audit of different feasibility documents of the projects to be audited are ensured, avoiding the standard error existing in the manual audit of different feasibility documents of the projects to be audited.

[0021] 2. Through pre-trained data extraction models and audit models, the system can automatically identify and extract required information from documents and perform corresponding audit processing, which not only improves the audit efficiency but also maintains a high degree of accuracy. It does not require manual audit, reduces the difficulty of auditing the feasibility documents of the project to be audited, and avoids errors in manual audit.

[0022] 3. It can support multiple objective first audit types, so as to realize the comprehensive review of engineering feasibility documents. This multi-dimensional audit method helps to discover various problems that may exist in the documents and improve the reliability of the audit results. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0024] Figure 1 A schematic diagram of the structure of a system for reviewing the safety and compliance of engineering feasibility documents provided in an embodiment of the present application;

[0025] Figure 2 A schematic diagram of the workflow of a safety compliance review system for engineering feasibility documents provided for this application;

[0026] Figure 3A process diagram of a method for security compliance review of engineering feasibility documents provided in an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0028] In order to improve the efficiency and quality of the safety and compliance review of engineering feasibility documents and achieve standardization and automation, the present application provides a system and method for the safety and compliance review of engineering feasibility documents.

[0029] Embodiment 1:

[0030] This application provides a system for reviewing the safety and compliance of engineering feasibility documents. Figure 1 A schematic diagram of the structure of a system for reviewing the safety and compliance of engineering feasibility documents provided in an embodiment of the present application, the system comprising:

[0031] The file acquisition module 11 is used to acquire the feasibility file of the project to be reviewed and its corresponding target review information; wherein the target review information includes the first review type of each target;

[0032] The instruction template storage module 12 is used to store the first instruction template and the second instruction template corresponding to each first audit type respectively; wherein the first instruction template of any first audit type is used to instruct to identify and extract the information required for the first audit type from the engineering feasibility file, and the second instruction template of any first audit type is used to instruct to perform the first audit type audit on the audit data of the first audit type;

[0033] The data extraction module 13 is used to obtain the first instruction templates corresponding to the first audit types of the respective targets from the instruction template storage module 12; obtain the audit data corresponding to the first audit types of the respective targets based on the data extraction prompts corresponding to the first audit types of the respective targets through the pre-trained data extraction model; wherein the data extraction prompts include the first instruction templates corresponding to the first audit types of the respective targets and the feasibility documents of the project to be audited;

[0034] The data review module 14 is used to obtain the second instruction template of the target first review type from the instruction template storage module 12 for each target first review type; determine the review result of the feasibility document of the project to be reviewed on the target first review type based on the review prompt corresponding to the target first review type through a pre-trained review model supporting the target first review type; wherein the review prompt corresponding to the target first review type includes the second instruction template of the target first review type and the review data of the target first review type;

[0035] The audit result output module 15 is used to output the audit results corresponding to the first audit types of each target.

[0036] In today's engineering project management field, the safety compliance review of engineering feasibility documents is a key link to ensure the smooth implementation of projects and avoid potential risks. However, with the expansion of project scale and the increase in complexity, traditional manual review methods can no longer meet the needs of efficient, accurate and comprehensive review. In order to solve this problem, the present application provides an engineering feasibility document safety compliance review system (hereinafter referred to as the system), which integrates advanced information extraction, intelligent review and automated report generation technology to achieve efficient and automated review of engineering feasibility documents, thereby improving the accuracy and efficiency of the review, reducing human errors, and providing a reliable basis for project decision-making. The following is a detailed introduction to each module included in the system:

[0037] 1. Instruction template storage module 12

[0038] The instruction template storage module 12 stores the first instruction template and the second instruction template corresponding to each first audit type. The first instruction template of any audit type defines the information required for extracting the first audit type from the engineering feasibility file, and the second instruction template of the audit type defines the audit of the first audit type for the information required for the first audit type.

[0039] In a possible implementation, each first audit type includes one or more of the following: entity name consistency review, numerical consistency review, standard and specification document reference review, and technical terms compliance review. The first instruction template and the second instruction target of each first audit type are exemplarily described below:

[0040] For entity name consistency review, the first instruction template of the entity name consistency review contains regular expressions or specific text matching rules for identifying all important entity names in the file (such as company name, project name, product model, etc.). Based on these rules, the system can search and extract all relevant entity names in the file to prepare for subsequent consistency verification. The second instruction template of the entity name consistency review can define an algorithm for comparing whether different entity names are consistent.

[0041] For numerical consistency review, the first instruction template for numerical consistency review defines the extraction rules of numerical values ​​(such as identifying numerical values ​​by specific formats or units), as well as the numerical fields that need to be compared or verified (such as cost budget, return on investment, etc.). The system follows these rules to extract all relevant numerical data from the file. The second instruction template for numerical consistency review can include logical judgments to verify whether the numerical value is within a reasonable range.

[0042] For the review of references to standards and regulations, the first instruction template for the review of references to standards and regulations contains the reference format, numbering rules or keyword list of standards and regulations. The system uses these templates to find and extract all referenced standards and regulations in the document so that the accuracy and compliance of these references can be verified later. The second instruction template for the review of references to standards and regulations can specify how to verify the validity and applicability of the referenced documents.

[0043] For the technical terms compliance review, the first instruction template for the technical terms compliance review may include a list of technical terms that need to be reviewed, technical terms keywords or phrases. Based on these templates, the system searches and extracts relevant technical terms content in the file, laying the foundation for subsequent technical compliance assessment. The second instruction template for the technical terms compliance review may include specific steps and scoring criteria for evaluating whether the technical terms comply with specific standards or specifications.

[0044] In a possible implementation, during the safety and compliance review of the engineering feasibility document, the integrity review of the document is an important step to ensure that the document content is complete and without omissions, which helps to identify and correct incomplete information caused by negligence or errors. Based on this, in the present application, the instruction template storage module 12 also stores the third instruction templates corresponding to each second review type, so as to guide the system to conduct an integrity review of the document through the third instruction template.

[0045] The second review types include one or more of the following: chapter content integrity review, reference basis integrity review, and argument logic and content integrity review. The third instruction template corresponding to each second review type is described below:

[0046] A. Completeness review of chapter content.

[0047] For each chapter in the document structure, the third instruction template lists the key elements or sub-chapters that each chapter should contain. Based on these templates, the system checks whether the document covers all necessary chapters and sub-chapters, and whether the content of each chapter is complete and without omissions.

[0048] B. Completeness review of citation basis.

[0049] The third instruction template for the completeness review of the citation basis includes verification rules for the citation format, necessity assessment criteria for the citation content, etc. The system compares the rules in the template with the actual citations in the document to identify possible missing citations or citation formats that do not conform to the specifications, thereby ensuring the integrity and compliance of the document citations.

[0050] C. Review of argument logic and content integrity.

[0051] The review of argument logic and content integrity requires the system to not only check the physical integrity of the document content (i.e., whether there are any missing parts), but also evaluate the logical coherence of the document content and the comprehensiveness of the argument. Based on this, the third instruction template for the review of argument logic and content integrity can include the verification logic of the argument framework, the necessity check of key arguments, etc. The system simulates the reading process of human readers to analyze whether the document's argument logic is clear, whether the arguments are sufficient, and whether the conclusions are reasonable, thereby ensuring that the document also meets the integrity requirements at the content level.

[0052] 2. File acquisition module 11

[0053] The file acquisition module 11 is the entrance of the system and is responsible for receiving or retrieving the engineering feasibility documents to be reviewed and their corresponding target review information.

[0054] In the present application, the file acquisition module 11 supports multiple file acquisition methods to flexibly respond to the needs of different scenarios. Exemplarily, it allows users to directly upload one or more engineering feasibility files. Users can select locally stored documents and upload them to the system through the user interface (UI) provided by the system to accommodate files that require immediate review or are not within the scope of the system's preset data source. In another exemplary embodiment, the file acquisition module 11 can also establish a stable data interface (such as an API) with a database or file storage system for storing engineering feasibility files, so that the system can automatically obtain engineering feasibility files to be reviewed regularly or according to trigger conditions to meet large-scale and continuous project review needs.

[0055] In addition to the feasibility documents of the project to be reviewed, the file acquisition module 11 is also responsible for receiving or determining the target review information corresponding to these documents. These target review information are important bases for guiding subsequent review work, including but not limited to the target first review type. Exemplarily, the user can input or select specific target review information through the display interface provided by the system. In another exemplary embodiment, the target review information can also be pre-configured in the file acquisition module 11 to reduce the workload of the user manually entering the target review information.

[0056] For example, a user uploads a project feasibility document in PDF format through the system's web interface and fills out a form to indicate the type of review to be conducted, such as "financial feasibility", "technical feasibility", "legal compliance", etc. The system's file acquisition module 11 acquires the project feasibility document to be reviewed and its corresponding target review information and saves it.

[0057] In a possible implementation, when there is a need to conduct an integrity review of the feasibility documents of the project to be reviewed, the target review information obtained by the document acquisition module 11 may carry a target second review type in addition to various first godhead types.

[0058] In a possible implementation, after obtaining the feasibility file of the project to be reviewed (hereinafter referred to as the file), the file acquisition module 11 can perform file verification on the feasibility file of the project to be reviewed, for example, file format, file size, file integrity, etc. Only when it is determined that the feasibility file of the project to be reviewed passes the file verification, the feasibility file of the project to be reviewed and its target review information are stored in the database. When it is determined that the feasibility file of the project to be reviewed does not pass the file verification, a notification message indicating that an error has occurred in the feasibility file of the project to be reviewed can be displayed through the display interface corresponding to the system.

[0059] For example, the file acquisition module 11 can check whether the format of the file meets the system's preset support range, such as PDF, Word (.docx), Excel (.xlsx), etc. If the file format is not supported, the system will immediately notify the user through the display interface, prompting the file format error and suggesting that the user upload a file that meets the requirements.

[0060] For example, the file acquisition module 11 can verify whether the size of the file is within a reasonable range set by the system. A file that is too large may consume too many system resources and affect the efficiency of the review; while a file that is too small may not be effectively reviewed due to incomplete content. If the file size does not meet the requirements, the system will issue a warning to the user through the display interface and may require the user to adjust the file size or re-upload.

[0061] For another example, the file acquisition module 11 can check the integrity of the file. This includes verifying whether the file is downloaded or uploaded completely without missing any part; and checking whether the file has not been tampered with or damaged. For example, the hash value or checksum of the file is calculated and compared with the expected value. If the file integrity verification fails, the system will notify the user that the file may be damaged or there is a security risk, and suggest that the user re-acquire the file.

[0062] In a possible implementation, the file acquisition module 11 can also perform structural processing on the obtained engineering feasibility documents to be reviewed, aiming to convert the contents of the engineering feasibility documents (such as PDF, Word documents, scanned copies, etc.) that may originally exist in unstructured or semi-structured forms into a structured data format that can be processed more directly and efficiently by the computer, and then transmit the structured engineering feasibility documents to be reviewed to the modules for subsequent processing in the system (including the data extraction module 13 and the data review module 14). Specifically, when the file acquisition module 11 receives the engineering feasibility documents to be reviewed, it will first identify the format and type of the engineering feasibility documents to be reviewed. For files in non-text format (such as PDF scanned copies), the file acquisition module 11 can use OCR technology to convert the text in the image into an editable text format. Then, the file acquisition module 11 will use natural language processing (NLP) technology or a predefined set of rules to parse the text content, identify key information units in the file, such as chapter titles, paragraphs, tables, lists, etc., and try to organize these information units according to a certain logical structure (such as a tree structure, a key-value pair form, etc.).

[0063] 3. Data extraction module 13

[0064] The data extraction module 13 is used to extract the information required for each target first audit type from the feasibility file of the project to be audited. This information is then passed to the data audit module 14 for further processing. Exemplarily, the data extraction module 13 can obtain the feasibility file of the project to be audited and the first audit type of each target from the file acquisition module 11. Then, the first instruction template corresponding to each target first audit type is obtained from the instruction template storage module 12. The data extraction module 13 obtains the data extraction prompts corresponding to each target first audit type according to the feasibility file of the project to be audited and the first instruction templates corresponding to each target first audit type. The data extraction module 13 calls the pre-trained data extraction model and inputs the acquired data extraction prompts corresponding to each target first audit type into the data extraction model. The data extraction model processes the input data extraction prompts to obtain the audit data corresponding to each target first audit type.

[0065] In a possible implementation, the pre-trained data extraction model can be a large model that supports processing data extraction of each first audit type. Exemplarily, the data extraction module 13 can input the data extraction prompts corresponding to each target first audit type separately into the data extraction model for processing, or can input the data extraction prompts corresponding to each target first audit type together into the data extraction model for processing.

[0066] In another possible implementation, the pre-trained data extraction model may also be a plurality of models that support data extraction of different first audit types, that is, each first audit type has a corresponding model that supports the first audit type. Exemplarily, for each target first audit type, the data extraction prompt corresponding to the target first audit type may be input into the data extraction model that supports the target first audit type for processing to obtain the audit data of the target first audit type.

[0067] In a possible implementation, the data extraction model is trained in the following manner:

[0068] Obtaining first sample data extraction prompts and first standard audit data corresponding to each of the first audit types for any first sample project feasibility file in the data extraction sample set;

[0069] Acquire the first expected audit data corresponding to each first audit type respectively based on the first sample data extraction prompts respectively corresponding to each first audit type through the basic data extraction model;

[0070] Based on the first standard audit data and the first expected audit data, the basic data extraction model is trained to obtain a trained data extraction model.

[0071] In order to accurately extract the audit data corresponding to each target first audit type, in this application, it is necessary to construct a comprehensive data extraction sample set. The data extraction sample set contains multiple first sample engineering feasibility files, and each first sample engineering feasibility file determines the corresponding first sample data extraction prompts and first standard audit data for each first audit type. Among them, the first standard audit data is manually extracted and verified from the first sample engineering feasibility file based on audit standards and professional knowledge (such as manual audit files, expert library experience, etc.), ensuring its accuracy and authority.

[0072] Extract the first sample data extraction prompts corresponding to each first audit type of any sample project feasibility file from the data extraction sample set, and then input the first sample data extraction prompts corresponding to each first audit type into the basic data extraction model. The input first sample data extraction prompts are processed by the basic data extraction model to obtain the first expected audit data corresponding to each first audit type.

[0073] In a possible implementation, if the pre-trained data extraction model is desired to be a large model that supports processing data extraction of each first audit type, the loss values ​​corresponding to each first audit type are determined according to the first expected audit data and the first standard audit data corresponding to each first audit type. According to the loss values ​​corresponding to each first audit type, the parameters in the basic data extraction model are adjusted to obtain a trained data extraction model.

[0074] In another possible implementation, if the pre-trained data extraction model is a plurality of models supporting data extraction of different first audit types, then for each first audit type, the loss value of the first audit type is determined according to the first expected audit data and the first standard audit data of the first audit type. According to the loss value of the first audit type, the parameters in the basic data extraction model corresponding to the first audit type are adjusted to obtain a trained data extraction model supporting data extraction of the first audit type.

[0075] It should be noted that the basic data extraction model can be an untrained model or a trained model with natural language understanding capabilities.

[0076] The data extraction sample set includes a plurality of first sample engineering feasibility files, and the above steps can be performed for each first sample engineering feasibility file. When a preset convergence condition is met, the data extraction model training is completed.

[0077] Among them, satisfying the preset convergence condition may be that the number of iterations for training the basic data extraction model reaches the set maximum number of iterations, or the change in the determined loss value remains within a very small range in multiple consecutive iterations, etc. In specific implementations, the setting can be flexibly performed and is not specifically limited here.

[0078] As a possible implementation method, when conducting model training, the first sample engineering feasibility file in the data extraction sample set can be divided into training samples and test samples. The basic data extraction model is first trained based on the training samples, and then the reliability of the trained data extraction model is verified based on the test samples.

[0079] 4. Data review module 14

[0080] The data audit module 14 uses a pre-trained audit model to determine the audit results of the feasibility documents of the project to be audited on each target first audit type based on the second instruction template and the extracted audit data.

[0081] Exemplarily, the data audit module 14 first receives the audit data corresponding to each target first audit type extracted by the data extraction module 13, and these data have been accurately extracted according to the first instruction template of each target first audit type. Subsequently, the data extraction module 13 will access the instruction template storage module 12 to obtain the second instruction template corresponding to each target first audit type. For each target first audit type, the second instruction template of the target first audit type and the audit data of the target audit type are combined to obtain the audit prompt corresponding to the target first audit type. The audit model that supports the target first audit type is called, and the audit prompt corresponding to the target first audit type is input into the audit model. The audit prompt corresponding to the input target first audit type is processed by the audit model to obtain the audit result of the feasibility document of the project to be audited on the target first audit type.

[0082] Among them, the audit models corresponding to each first audit type can be deployed on different models or on the same large model.

[0083] In a possible implementation, the basic audit model can be trained based on the sample sets of each first audit type collected in advance. For any first audit type, based on the sample audit prompts in the sample set of the first audit type and the corresponding standard audit results, to obtain a trained audit model that supports the first audit type. The sample audit prompts include sample audit data and a second instruction template.

[0084] It should be noted that the training process of the audit model that supports any first audit type may be similar to the above-mentioned process of training the data extraction model, and will not be described in detail here.

[0085] In a possible implementation, if the target audit information includes the second audit type of each target, the file acquisition module 11 can transfer the feasibility file of the project to be audited and the second audit type of each target to the data audit model. The data audit module 14 can obtain the third instruction template of the second audit type of the target from the instruction module storage module 12 for each second audit type of the target. For each second audit type of the target, the third instruction template of the second audit type of the target and the feasibility file of the project to be audited are combined to obtain the audit prompt corresponding to the second audit type of the target. The audit model that supports the second audit type of the target is called, and the audit prompt corresponding to the second audit type of the target is input into the audit model. The audit prompt corresponding to the input second audit type of the target is processed by the audit model to obtain the audit result of the feasibility file of the project to be audited on the second audit type of the target.

[0086] It should be noted that, regardless of the target second audit type or the target first audit type, their audit results may include but are not limited to audit conclusions, audit problem points, modification opinions, etc. Moreover, when the feasibility documents of the project to be audited are audited through the corresponding audit model, the audit model may be audited in combination with the knowledge in the pre-built knowledge base, thereby improving the knowledge in the knowledge base, which may include but is not limited to the national standard document library, local policy documents, local standard document library, and approval and adjustment documents, so as to further guide the audit model to audit the feasibility documents of the project to be audited through the knowledge in the knowledge base.

[0087] The audit models corresponding to each second audit type can be deployed on different models or on the same large model. The audit models corresponding to each second audit type can share part or all of the models with the audit models corresponding to each first audit type, that is, a certain model supports both the first audit type and the second audit type, or they can be independent of each other.

[0088] 5. Audit result output module 15

[0089] The audit result output module 15 is responsible for outputting the audit results output by the data audit module 14. Exemplarily, the audit result output module 15 organizes the audit results into a format that is easy to understand and use, and outputs them to users or other systems through a system interface or an API interface. Among them, each audit result includes the audit results corresponding to the first audit type of each target. Optionally, each audit result also includes the audit results corresponding to the second audit type of each target.

[0090] In order to improve user experience and interoperability between systems, the audit result output module 15 will format the collected audit results. This includes but is not limited to arranging the results in a specific logical order, applying a unified naming convention, and converting them into easy-to-read tables, charts or reports. At the same time, in response to the needs of the API interface, the audit result output module 15 will also convert the audit results into standard JSON, XML and other data structures so that other systems can seamlessly connect and parse them.

[0091] In a possible implementation, the audit result output module 15 supports automatic generation of audit reports based on pre-configured report generation templates. These templates can be customized according to different user preferences, organizational specifications, or legal and regulatory requirements, including but not limited to report structure, content layout, style design, etc. When generating an audit report, the audit result output module 15 will intelligently insert the audit results corresponding to the first audit type and the second audit type of each target into the report template to ensure the comprehensiveness and accuracy of the report content.

[0092] In order to meet the access habits and needs of different users, the audit result output module 15 supports outputting audit results through multiple channels. On the one hand, it can be directly displayed to users in the form of a graphical interface through a system interface (such as a web page, mobile application, etc.), providing an intuitive data overview and detail query function. On the other hand, the audit result output module 15 also provides an API interface, allowing other systems to obtain audit results by calling these interfaces, thereby realizing automatic processing and sharing of data.

[0093] The beneficial effects of this application are as follows:

[0094] 1. Through the pre-stored first instruction template and second instruction template customized for different first audit types, the pertinence, accuracy and standardization of the audit of different feasibility documents of the projects to be audited are ensured, avoiding the standard error existing in the manual audit of different feasibility documents of the projects to be audited.

[0095] 2. Through pre-trained data extraction models and audit models, the system can automatically identify and extract required information from documents and perform corresponding audit processing, which not only improves the audit efficiency but also maintains a high degree of accuracy. It does not require manual audit, reduces the difficulty of auditing the feasibility documents of the project to be audited, and avoids errors in manual audit.

[0096] 3. It can support multiple objective first audit types, so as to realize the comprehensive review of engineering feasibility documents. This multi-dimensional audit method helps to discover various problems that may exist in the documents and improve the reliability of the audit results.

[0097] Embodiment 2:

[0098] In order to accurately conduct the first audit type audit on the feasibility documents of the audited project, based on the above, in this application, the audit model supporting any first audit type is trained in the following manner:

[0099] Obtain a second sample data extraction prompt corresponding to any second sample engineering feasibility file in the first audit type sample set and its corresponding first standard audit result; wherein the second sample data extraction prompt includes the second sample engineering feasibility file and the first instruction template of the first audit type;

[0100] Acquire the second expected audit data of the first audit type from the second sample engineering feasibility file through the basic data extraction model based on the second sample data extraction prompt;

[0101] Determine, by means of a basic audit model, a first expected audit result of the second sample engineering feasibility file for the first audit type based on the audit prompt corresponding to the second sample engineering feasibility file; wherein the audit prompt corresponding to the second sample engineering feasibility file includes the second expected audit data and a second instruction template corresponding to the first audit type;

[0102] Based on the first expected audit result and the first standard audit result, the parameters in the basic data extraction model and the basic audit model are jointly adjusted to obtain a trained data extraction model and audit model that support the first audit type.

[0103] In the present application, for each first audit type, the data extraction model and the audit model supporting the first audit type can be jointly trained, and in order to obtain the data extraction model and the audit model supporting the first audit type, a sample set of the first audit type (referred to as the first audit type sample set) is collected in advance, wherein the first audit type sample set includes multiple second sample engineering feasibility files. For each second sample engineering feasibility file, its corresponding second sample data extraction prompt and first standard audit result are prepared, the second sample data extraction prompt includes the second sample engineering feasibility file itself and the first instruction template of the first audit type, and the first standard audit result is manually verified according to the audit standards and professional knowledge (such as manual audit files, expert library experience, etc.), ensuring its accuracy and authority.

[0104] The following describes the process of jointly training the data extraction model and the audit model that support the first audit type for any first audit type:

[0105] The second sample data extraction prompt and the first standard audit result corresponding to any second sample engineering feasibility file are obtained from the first audit type sample set. The second data extraction prompt is input into the basic data extraction model. Through the basic data extraction model, based on the second sample data extraction prompt, the second expected audit data for the first audit type is extracted from the second sample engineering feasibility file. Then, according to the second expected audit data and the pre-configured second instruction template of the first audit type, the audit prompt corresponding to the second sample engineering feasibility file is determined. The audit prompt corresponding to the second sample engineering feasibility file is input into the basic audit model. The audit prompt corresponding to the second sample engineering feasibility file is sorted out through the basic audit model to determine the first expected audit result of the second sample engineering feasibility file on the first audit type. Then, based on the first expected audit result and the first standard audit result, the loss value is determined. According to the loss value, the parameters in the basic data extraction model and the basic audit model are jointly adjusted to obtain a trained data extraction model and an audit model that supports the first audit type.

[0106] The above operation is performed for each second sample engineering feasibility file in the first audit type sample set, and this process is continuously iterated until the preset convergence conditions are met (such as the number of iterations reaches the upper limit, the loss value changes very little, etc.), thereby obtaining a trained data extraction model and audit model that can support the first audit type.

[0107] As a possible implementation method, when conducting model training, the second sample engineering feasibility file in the first audit type sample set can be divided into training samples and test samples. The basic data extraction model and the basic audit model are first trained based on the training samples, and then the reliability of the above-mentioned trained basic data extraction model and basic audit model is verified based on the test samples.

[0108] Embodiment 3:

[0109] In order to accurately conduct the second audit type audit on the feasibility documents of the audited project, based on the above, in this application, the audit model supporting any second audit type is trained in the following way:

[0110] Obtaining a sample audit prompt corresponding to any third sample engineering feasibility file in the second audit type sample set and its corresponding second standard audit result; wherein the sample audit prompt includes the sample engineering feasibility file and the third instruction template of the second audit type;

[0111] Determine, by means of the basic audit model and based on the sample audit prompt, a second expected audit result of the sample engineering feasibility document in the second audit type;

[0112] Based on the second expected audit result and the second standard audit result, the basic audit model is trained to obtain a trained audit model that supports the second audit type.

[0113] In the present application, for each second audit type, a sample set of the second audit type (referred to as the second audit type sample set) can be collected in advance, wherein the second audit type sample set includes a plurality of third sample engineering feasibility documents. For each third sample engineering feasibility document, a corresponding sample audit prompt and a second standard audit result are prepared, wherein the sample audit prompt includes the third sample engineering feasibility document itself and the third instruction template of the second audit type, and the second standard audit result is manually verified according to the audit standards and professional knowledge (such as manual audit documents, expert library experience, etc.), ensuring its accuracy and authority.

[0114] The following describes the process of joint training of the audit model supporting the second audit type for any second audit type:

[0115] Obtain the sample audit prompt and the second standard audit result corresponding to any third sample engineering feasibility document from the second audit type sample set. Input the sample audit prompt into the basic audit model. Use the basic audit model to sort out the sample audit prompt and determine the second expected audit result of the third sample engineering feasibility document on the second audit type. Then determine the loss value based on the second expected audit result and the second standard audit result. According to the loss value, adjust the parameters in the basic audit model to obtain a trained audit model that supports the second audit type.

[0116] The above operation is performed for each third sample engineering feasibility file in the second audit type sample set, and this process is continuously iterated until the preset convergence conditions are met (such as the number of iterations reaches the upper limit, the loss value changes very little, etc.), thereby obtaining a trained audit model that can support the second audit type.

[0117] As a possible implementation method, when conducting model training, the third sample engineering feasibility document in the second audit type sample set can be divided into training samples and test samples. The basic audit model is first trained based on the training samples, and then the reliability of the above-mentioned trained basic audit model is verified based on the test samples.

[0118] Embodiment 4:

[0119] Figure 2 A workflow diagram of a safety compliance review system for engineering feasibility documents provided for this application, the process includes:

[0120] S201: The engineering feasibility document safety compliance review system obtains a pre-trained data extraction model, an audit model corresponding to each first audit type, and an audit model corresponding to each second audit type.

[0121] In one possible implementation, the data extraction model is trained in the following manner:

[0122] Obtaining first sample data extraction prompts and first standard audit data corresponding to each first audit type for any first sample project feasibility file in the data extraction sample set;

[0123] Through the basic data extraction model, based on the first sample data extraction prompts respectively corresponding to each first audit type, the first expected audit data respectively corresponding to each first audit type is obtained;

[0124] Based on the first standard audit data and the first expected audit data, the basic data extraction model is trained to obtain a trained data extraction model.

[0125] In one possible implementation, the audit model supporting any first audit type is trained in the following manner:

[0126] Obtain a second sample data extraction prompt corresponding to any second sample engineering feasibility file in the first audit type sample set and its corresponding first standard audit result; wherein the second sample data extraction prompt includes the second sample engineering feasibility file and the first instruction template of the first audit type;

[0127] Acquire second expected audit data of the first audit type from the second sample engineering feasibility file through the basic data extraction model based on the second sample data extraction prompt;

[0128] Determine, through the basic audit model, a first expected audit result of the second sample project feasibility file on the first audit type based on the audit prompt corresponding to the second sample project feasibility file; wherein the audit prompt corresponding to the second sample project feasibility file includes the second expected audit data and the second instruction template corresponding to the first audit type;

[0129] Based on the first expected audit result and the first standard audit result, the parameters in the basic data extraction model and the basic audit model are jointly adjusted to obtain a trained data extraction model and audit model that support the first audit type.

[0130] In one possible implementation, the audit model supporting any second audit type is trained in the following manner:

[0131] Obtain a sample audit prompt corresponding to any third sample project feasibility file in the second audit type sample set and its corresponding second standard audit result; wherein the sample audit prompt includes the sample project feasibility file and the third instruction template of the second audit type;

[0132] Determine the second expected audit result of the sample project feasibility document on the second audit type based on the sample audit prompt through the basic audit model;

[0133] Based on the second expected audit result and the second standard audit result, the basic audit model is trained to obtain a trained audit model that supports the second audit type.

[0134] S202: The engineering feasibility document safety compliance review system obtains a pre-built knowledge base.

[0135] Among them, the knowledge base includes but is not limited to the national standard document library, local policy documents, local standard document library, and approval and adjustment documents.

[0136] S203: The instruction template storage module of the engineering feasibility document safety and compliance review system obtains the pre-configured first instruction templates and second instruction templates corresponding to each first review type, and the third instruction templates corresponding to each second review type.

[0137] Among them, the first instruction template of any first audit type is used to instruct the identification and extraction of information required for the first audit type from the engineering feasibility document, the second instruction template of any first audit type is used to instruct the first audit type to be audited on the audit data of the first audit type, and the third instruction template of any second audit type is used to instruct the integrity audit of the second audit type to be performed on the engineering feasibility document.

[0138] In a possible implementation, each first audit type includes one or more of the following: entity name consistency review, numerical consistency review, standard and specification document reference review, and technical terms compliance review.

[0139] In a possible implementation, each second review type includes one or more of the following: chapter content integrity review, citation basis integrity review, argument logic and content integrity review.

[0140] S204: The file acquisition module of the engineering feasibility document safety and compliance review system acquires the engineering feasibility document to be reviewed and its corresponding target review information.

[0141] The target audit information includes a first audit type for each target and a second audit type for each target.

[0142] S205: The file acquisition module performs structured processing on the feasibility file of the project to be reviewed.

[0143] S206: The file acquisition module sends the structured engineering feasibility document to be reviewed and its corresponding first audit type for each target to the data extraction module of the engineering feasibility document security and compliance review system, and sends the structured engineering feasibility document to be reviewed and its corresponding second audit type for each target to the data review module of the engineering feasibility document security and compliance review system.

[0144] For each target first review type, the project feasibility document safety compliance review system performs the following steps S207 to S213:

[0145] S207: The data extraction module obtains the first instruction template corresponding to the first audit type of the target from the instruction template storage module.

[0146] S208: The data extraction module assembles the first instruction template corresponding to the first audit type of the target with the feasibility file of the project to be audited to obtain a data extraction prompt.

[0147] S209: The data extraction module obtains the audit data corresponding to the target first audit type based on the data extraction prompt by using a pre-trained data extraction model that supports the first audit type.

[0148] S210: The data extraction module transmits the audit data corresponding to the first audit type of the target to the data audit module.

[0149] S211: The data review module obtains the second instruction template of the target first review type from the instruction template storage module.

[0150] S212: The data audit module assembles the second instruction template with the audit data corresponding to the target first audit type to obtain an audit prompt.

[0151] S213: The data audit module determines the audit result of the feasibility document of the project to be audited for the first audit type of the target through a pre-trained audit model that supports the first audit type of the target, based on the audit prompt and in combination with a pre-built knowledge base.

[0152] For each target second review type, the project feasibility document safety compliance review system performs the following steps S214 to S216:

[0153] S214: The data review module obtains the third instruction template of the second review type of the target from the instruction template storage module.

[0154] S215: The data review module assembles the third instruction template with the feasibility file of the project to be reviewed to obtain a review prompt.

[0155] S216: The data audit module determines the audit result of the feasibility document of the project to be audited for the second audit type of the target through a pre-trained audit model that supports the second audit type of the target, based on the audit prompt and in combination with a pre-built knowledge base.

[0156] S217: The data review module transmits the review results corresponding to the first review type of each target and the review results corresponding to the second review type of each target to the output module of the engineering feasibility document security compliance review system.

[0157] S218: The output module generates and outputs an audit report according to a pre-configured report generation template based on the audit results corresponding to the first audit type of each target and the audit results corresponding to the second audit type of each target.

[0158] Embodiment 5:

[0159] Based on the same inventive concept, this application also provides a method for safety compliance review of engineering feasibility documents. Figure 3 A process diagram of a method for reviewing the safety and compliance of engineering feasibility documents provided in an embodiment of the present application, the process comprising:

[0160] S301: Obtain the feasibility document of the project to be reviewed and its corresponding target review information; wherein the target review information includes the first review type of each target.

[0161] S302: Acquire first instruction templates corresponding to each target first audit type from pre-saved first instruction templates corresponding to each first audit type.

[0162] S303: Through a pre-trained data extraction model, based on the data extraction prompts corresponding to the first audit types of each target, the audit data corresponding to the first audit types of each target are obtained; wherein the data extraction prompts include the first instruction templates corresponding to the first audit types of each target and the feasibility documents of the project to be audited.

[0163] S304: For each target first audit type, obtain the second instruction template of the target first audit type from the pre-saved second instruction templates corresponding to each first audit type; determine the audit result of the feasibility document of the project to be audited on the target first audit type based on the audit prompt corresponding to the target first audit type through a pre-trained audit model supporting the target first audit type; wherein the audit prompt corresponding to the target first audit type includes the second instruction template of the target first audit type and the audit data of the target first audit type.

[0164] S305: Output the audit results corresponding to the first audit types of each target.

[0165] The engineering feasibility document security compliance review method provided in this application is applied to computer equipment, which can be an intelligent terminal, such as a computer, a robot, etc., or a server, such as an application server, a business server, etc.

[0166] In some possible implementations, each first audit type includes one or more of the following: entity name consistency review, numerical consistency review, standard and specification document reference review, and technical terms compliance review.

[0167] In some possible implementations, the method further comprises:

[0168] If the target audit information includes the second audit type of each target, then for each target second audit type, the third instruction template of the target second audit type is obtained from the third instruction templates corresponding to each pre-saved second audit type; through the pre-trained audit model supporting the target second audit type, based on the audit prompt corresponding to the target second audit type, the audit result of the feasibility document of the project to be audited on the target second audit type is determined; wherein the audit prompt corresponding to the target second audit type includes the third instruction template of the target second audit type and the feasibility document of the project to be audited;

[0169] Output the audit results corresponding to the second audit types of each target.

[0170] In some possible implementations, each second review type includes one or more of the following: chapter content integrity review, citation basis integrity review, and argument logic and content integrity review.

[0171] In some possible implementations, after obtaining the feasibility document of the project to be reviewed and its corresponding target review information, and before obtaining the first instruction templates corresponding to the first review types respectively from the pre-saved first instruction templates respectively corresponding to the first review types, the method further includes:

[0172] The feasibility documents of the project to be reviewed are structured.

[0173] In some possible implementations, outputting the audit results corresponding to the first audit types of the respective targets includes:

[0174] An audit report is generated according to a pre-configured report generation template based on the audit results corresponding to the first audit type of each target and the audit results corresponding to the second audit type of each target.

[0175] In some possible implementations, the data extraction model is trained in the following manner:

[0176] Obtaining first sample data extraction prompts and first standard audit data corresponding to each of the first audit types for any first sample project feasibility file in the data extraction sample set;

[0177] Acquire the first expected audit data corresponding to each first audit type respectively based on the first sample data extraction prompts respectively corresponding to each first audit type through the basic data extraction model;

[0178] Based on the first standard audit data and the first expected audit data, the basic data extraction model is trained to obtain a trained data extraction model.

[0179] In some possible implementations, the audit model supporting any first audit type is trained as follows:

[0180] Obtain a second sample data extraction prompt corresponding to any second sample engineering feasibility file in the first audit type sample set and its corresponding first standard audit result; wherein the second sample data extraction prompt includes the second sample engineering feasibility file and the first instruction template of the first audit type;

[0181] Acquire the second expected audit data of the first audit type from the second sample engineering feasibility file through the basic data extraction model based on the second sample data extraction prompt;

[0182] Determine, by means of a basic audit model, a first expected audit result of the second sample engineering feasibility file for the first audit type based on the audit prompt corresponding to the second sample engineering feasibility file; wherein the audit prompt corresponding to the second sample engineering feasibility file includes the second expected audit data and a second instruction template corresponding to the first audit type;

[0183] Based on the first expected audit result and the first standard audit result, the parameters in the basic data extraction model and the basic audit model are jointly adjusted to obtain a trained data extraction model and audit model that support the first audit type.

[0184] In some possible implementations, the audit model supporting any second audit type is trained as follows:

[0185] Obtaining a sample audit prompt corresponding to any third sample engineering feasibility file in the second audit type sample set and its corresponding second standard audit result; wherein the sample audit prompt includes the sample engineering feasibility file and the third instruction template of the second audit type;

[0186] Determine, by means of the basic audit model and based on the sample audit prompt, a second expected audit result of the sample engineering feasibility document in the second audit type;

[0187] Based on the second expected audit result and the second standard audit result, the basic audit model is trained to obtain a trained audit model that supports the second audit type.

[0188] Since the principle of solving the problem by the above method is similar to that of the engineering feasibility document safety compliance review system, the implementation of the above method can refer to the implementation example of the system, and the repeated parts will not be repeated.

[0189] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A system for reviewing the safety and compliance of engineering feasibility documents, characterized in that: The system comprises: A file acquisition module, used to acquire the feasibility file of the project to be reviewed and its corresponding target review information; wherein the target review information includes the first review type of each target; An instruction template storage module is used to store first instruction templates and second instruction templates corresponding to each first audit type; wherein the first instruction template of any first audit type is used to instruct to identify and extract information required for the first audit type from the engineering feasibility file, and the second instruction template of any first audit type is used to instruct to perform the first audit type audit on the audit data of the first audit type; A data extraction module is used to obtain the first instruction templates corresponding to the first audit types of the respective targets from the instruction template storage module; obtain the audit data corresponding to the first audit types of the respective targets through a pre-trained data extraction model based on the data extraction prompts corresponding to the first audit types of the respective targets; wherein the data extraction prompts include the first instruction templates corresponding to the first audit types of the respective targets and the feasibility files of the projects to be audited; A data review module is used to obtain, for each target first review type, a second instruction template of the target first review type from the instruction template storage module; determine the review result of the feasibility document of the project to be reviewed on the target first review type based on the review prompt corresponding to the target first review type through a pre-trained review model supporting the target first review type; wherein the review prompt corresponding to the target first review type includes the second instruction template of the target first review type and the review data of the target first review type; The audit result output module is used to output the audit results corresponding to the first audit type of each target.

2. The system according to claim 1, characterized in that The first audit types include one or more of the following: entity name consistency review, numerical consistency review, standard and specification document reference review, and technical terms compliance review.

3. The system according to claim 1, characterized in that The instruction template storage module is further used to store third instruction templates corresponding to each second review type; wherein the third instruction template of any second review type is used to instruct to perform an integrity review of the second review type on the engineering feasibility document; The data audit module is further configured to, if the target audit information includes the second audit type of each target, obtain, for each target second audit type, a third instruction template of the target second audit type from the instruction template storage module; determine the audit result of the feasibility document of the project to be audited on the second audit type of the target based on the audit prompt corresponding to the second audit type of the target by using a pre-trained audit model supporting the second audit type of the target; wherein the audit prompt corresponding to the second audit type of the target includes the third instruction template of the second audit type of the target and the feasibility document of the project to be audited; The audit result output module is further used to output the audit results corresponding to the second audit types of each target.

4. The system according to claim 3, characterized in that The second review types include one or more of the following: chapter content integrity review, citation basis integrity review, and argument logic and content integrity review.

5. The system according to claim 1 or 3, characterized in that: The file acquisition module is also used to perform structured processing on the feasibility file of the project to be reviewed.

6. The system according to claim 1, wherein: The audit result output module is specifically used to generate an audit report according to a pre-configured report generation template based on the audit results corresponding to the first audit type of each target and the audit results corresponding to the second audit type of each target.

7. The system according to claim 1, characterized in that The data extraction model is trained as follows: Obtaining first sample data extraction prompts and first standard audit data corresponding to each of the first audit types for any first sample project feasibility file in the data extraction sample set; Acquire the first expected audit data corresponding to each first audit type respectively based on the first sample data extraction prompts respectively corresponding to each first audit type through the basic data extraction model; Based on the first standard audit data and the first expected audit data, the basic data extraction model is trained to obtain a trained data extraction model.

8. The system of claim 1, wherein: An audit model that supports any of the first audit types is trained as follows: Obtain a second sample data extraction prompt corresponding to any second sample engineering feasibility file in the first audit type sample set and its corresponding first standard audit result; wherein the second sample data extraction prompt includes the second sample engineering feasibility file and the first instruction template of the first audit type; Acquire the second expected audit data of the first audit type from the second sample engineering feasibility file through the basic data extraction model based on the second sample data extraction prompt; Determine, by means of a basic audit model, a first expected audit result of the second sample engineering feasibility file for the first audit type based on the audit prompt corresponding to the second sample engineering feasibility file; wherein the audit prompt corresponding to the second sample engineering feasibility file includes the second expected audit data and a second instruction template corresponding to the first audit type; Based on the first expected audit result and the first standard audit result, the parameters in the basic data extraction model and the basic audit model are jointly adjusted to obtain a trained data extraction model and audit model that support the first audit type.

9. The system according to claim 3, characterized in that An audit model that supports either second audit type is trained as follows: Obtaining a sample audit prompt corresponding to any third sample engineering feasibility file in the second audit type sample set and its corresponding second standard audit result; wherein the sample audit prompt includes the sample engineering feasibility file and the third instruction template of the second audit type; Determine, by means of the basic audit model and based on the sample audit prompt, a second expected audit result of the sample engineering feasibility document in the second audit type; Based on the second expected audit result and the second standard audit result, the basic audit model is trained to obtain a trained audit model that supports the second audit type.

10. A method for reviewing the safety compliance of engineering feasibility documents, characterized in that: The method comprises: Obtaining the feasibility documents of the project to be reviewed and its corresponding target review information; wherein the target review information includes the first review type of each target; Acquire the first instruction templates corresponding to the first audit types of the targets from the pre-saved first instruction templates corresponding to the first audit types; Acquire the audit data corresponding to the first audit type of each target through a pre-trained data extraction model based on the data extraction prompts corresponding to the first audit type of each target; wherein the data extraction prompts include the first instruction templates corresponding to the first audit type of each target and the feasibility file of the project to be audited; For each target first audit type, obtain the second instruction template of the target first audit type from the pre-saved second instruction templates corresponding to each first audit type; determine the audit result of the feasibility document of the project to be audited on the target first audit type based on the audit prompt corresponding to the target first audit type through the pre-trained audit model supporting the target first audit type; wherein the audit prompt corresponding to the target first audit type includes the second instruction template of the target first audit type and the audit data of the target first audit type; Output the audit results corresponding to the first audit type of each target.

Images