Photovoltaic project filing auditing method and system based on AI

Abstract

The invention discloses an AI-based photovoltaic project filing auditing method and system, and belongs to the technical field of intelligent operation and maintenance management, and the method comprises the steps: S1, carrying out the content recognition and semantic classification of a project material based on a multi-modal large model, obtaining the type of the project material, if the type of the project material meets a preset condition, executing S2, or else, indicating that the auditing is not passed; s2, constructing an ambiguous knowledge graph according to historical project materials and auditing conditions of the historical project materials; s3, taking the ambiguous knowledge graph as an ambiguous attribute acquisition basis, acquiring the attribute of the mid-value of the project material through a large language model, and performing consistency check on the attribute of the mid-value and the value of the project material to obtain a consistency check result; and S4, performing standard inspection on the project material based on the consistency inspection result, the mid-value of the project material and the attribute of the value, if the standard inspection is passed, indicating that the auditing is passed, otherwise, indicating that the auditing is not passed. The technical problem that the auditing efficiency and the auditing accuracy are difficult to consider in the prior art is solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance management technology, specifically to an AI-based method and system for reviewing and approving the filing of photovoltaic projects. Background Technology

[0002] As a crucial preliminary step in ensuring project compliance, the accuracy and efficiency of photovoltaic project registration and review directly determine the compliance, effectiveness, and implementation efficiency of subsequent project progress. Traditional review methods rely on manual operation, requiring reviewers to examine each applicant's submitted materials: first, verifying the completeness of documents, such as the registration application form, property ownership certificate, and qualification documents; second, checking cross-document consistency, such as comparing the content of the same core fields in different materials to ensure complete agreement; and third, verifying the format and standardization of information, such as checking whether the format, units, and expression of the submitted data meet the registration requirements. However, this manual review model suffers from low efficiency. To address these pain points, existing technologies introduce large language models to build automated review systems. These systems utilize natural language processing capabilities to intelligently parse, extract, and compare data content, significantly improving review efficiency. However, in actual registration materials, some field attributes are ambiguous. Even with semantic similarity analysis combined with contextual information, it is difficult to accurately determine their attributes. Therefore, relying solely on the semantic understanding capabilities of large language models is insufficient to accurately define the attribute attribution of such ambiguous fields, leading to frequent erroneous reviews. This shows that balancing review efficiency and accuracy is a technical challenge that existing technologies struggle to solve. Summary of the Invention

[0003] To address the technical challenge of balancing review efficiency and accuracy in existing technologies, this invention provides an AI-based photovoltaic project filing review method and system. By combining a multimodal large model with a large language model to review project materials, review efficiency is significantly improved. An ambiguous knowledge graph is constructed using historical project materials and their review status, and this graph serves as the basis for obtaining ambiguous attributes. The large language model is then used to extract the attributes from the project materials, thereby improving the accuracy of attribute determination. This solves the technical problem of balancing review efficiency and accuracy in existing technologies.

[0004] To address the aforementioned technical problems, this invention provides an AI-based method for reviewing and approving the filing of photovoltaic projects, comprising the following steps: S1: Based on the multimodal large model, perform content recognition and semantic classification on the project materials to obtain the project material type. If the project material type meets the preset conditions, then execute S2; otherwise, it means that the review has not passed. S2: Construct an ambiguous knowledge graph based on historical project materials and the review status of those materials; S3: Using the ambiguous knowledge graph as the basis for obtaining ambiguous attributes, the attributes of the project material median are obtained through the large language model, and the consistency test results are obtained by checking the consistency between the project material median and the attributes of the project material median. S4: Based on the consistency test results, the median value of the project materials, and the attributes of the median value of the project materials, conduct a standard inspection on the project materials. If the standard inspection passes, the audit is approved; otherwise, the audit is not approved.

[0005] Preferably, in S1, the step of obtaining the project material type by performing content recognition and semantic classification of project materials based on a multimodal large model includes: Obtain audit standards based on audit requirements; The project materials in various forms are converted into standardized text through a multimodal large model. The file name of the standardized text is matched with the preset project material type in the review standard to obtain the matching probability. The similarity probability is obtained based on the average semantic similarity between the text content in the standardized text and the preset project material type. The project material type is then determined based on the matching probability and the similarity probability.

[0006] Preferably, step S2: constructing an ambiguous knowledge graph based on historical project materials and the review status of those materials, including: The probability of the first attribute of the value in the historical project materials is obtained by using a large language model. The values ​​in the historical project materials are classified according to the probability of the first attribute, and the values ​​with ambiguous categories are marked as ambiguity points. Based on the values ​​and attributes of the uniquely attributed categories, and combined with the review status of historical project materials, the attributes of the ambiguous points are obtained. The ambiguous points are then associated with each other's attributes, and the frequency of association is recorded. Context nodes are obtained based on the context information of the ambiguity points, rule nodes are obtained based on the review standards obtained based on the review requirements, the attributes of the ambiguity points are used as attribute nodes, and the content in the historical project materials where the ambiguity points are located is bound to the attribute nodes. An ambiguous knowledge graph is obtained based on ambiguity points, context nodes, rule nodes, attribute nodes, and association frequency.

[0007] Preferably, the step of obtaining the ambiguous attributes based on the uniquely attributed value and the uniquely attributed value of the category, combined with the review status of historical project materials, includes: If the historical project materials are approved, the values ​​with unique categories will be matched with the attributes and review criteria of the values ​​with unique categories to obtain the attributes of the ambiguous points. If the materials for a historical project fail the review, the attributes of the points of ambiguity will be obtained based on the explanation of the failure and the review standards.

[0008] Preferably, in S3, the attribute of obtaining the value in the project material through a large language model, using an ambiguous knowledge graph as the basis for obtaining ambiguous attributes, includes: The probability of the second attribute of the value in the project material is obtained by using a large language model. If the probability of the second attribute meets the preset requirements, the attribute corresponding to the largest probability of the second attribute is taken as the attribute of the value in the project material. Otherwise, the value in the project material is matched with the ambiguous knowledge graph to locate the traversal node on the ambiguous knowledge graph. The traversal node is used as the starting point to traverse the ambiguous knowledge graph and obtain the traversal result. The attribute of the value in the project material is obtained based on the traversal result.

[0009] Preferably, in S3, the step of performing a consistency check on the attributes of the project material median and the project material median to obtain the consistency check result includes: Based on the semantic similarity between the attributes of the values ​​in the project materials and the preset attributes in the audit standards obtained based on the audit requirements, the values ​​in the project materials are located to the value set, and the values ​​in the project materials are matched with the value set to obtain the first consistency test result; Key-value pairs are constructed based on the median value of the project materials and the attributes of the median value of the project materials. The semantic similarity between the key-value pairs is obtained, and the second consistency test result is obtained based on the semantic similarity between the key-value pairs.

[0010] Preferably, step S4: Based on the consistency check results, the median value of the project materials, and the attributes of the median value of the project materials, a standardization check is performed on the project materials. If the standardization check passes, the audit is approved; otherwise, the audit fails. This includes: If the consistency check passes, then the key-value pairs undergo explicit order specification checks and explicit content integrity checks. If both pass, the review is approved. Otherwise, implicit specifications are obtained by embedding positive and negative examples of specifications in the prompts of the large language model. The key-value pairs are then subjected to implicit order specification checks and implicit content integrity checks. If both pass, the review is approved. Otherwise, the review is not approved.

[0011] Preferably, if the consistency check passes, the step of performing explicit order specification checks and explicit content integrity checks on the key-value pairs further includes: If both the first and second consistency tests pass, then the consistency test result is considered to have passed.

[0012] By adopting the above technical solution, the present invention has the following advantages: By combining a multimodal large model with a large language model to review project materials, the review efficiency is significantly improved. An ambiguous knowledge graph is constructed by combining historical project materials with the review status of historical project materials, and the ambiguous knowledge graph is used as the basis for obtaining ambiguous attributes. The attributes of the values ​​in the project materials are obtained through the large language model, thereby improving the accuracy of attribute determination and solving the technical problem that existing technologies cannot balance review efficiency and review accuracy. By extracting ambiguities from historical project materials and inversely deducing the attributes of these ambiguities based on the review process of those materials, a knowledge graph of ambiguity is constructed by combining context nodes, rule nodes, and association frequencies. By transforming historical ambiguity judgment experience into a structured knowledge graph of ambiguity, the large language model can accurately determine the attributes of ambiguities through node matching during subsequent attribute identification, thus significantly improving the accuracy of attribute determination. The tiered inspection strategy, which involves first explicitly inspecting project materials and then triggering implicit inspections as needed, significantly reduces unnecessary computing power consumption and further improves the accuracy of the review.

[0013] This invention also provides an AI-based photovoltaic project filing and review system, applicable to the aforementioned AI-based photovoltaic project filing and review method, comprising: The project material type acquisition module is used to perform content recognition and semantic classification on project materials based on a multimodal large model to obtain the project material type and determine whether the project material type meets the preset conditions. When the preset conditions are met, the ambiguity knowledge graph acquisition module is activated. When the preset conditions are not met, a prompt indicating that the review has not been approved is output. The Ambiguity Knowledge Graph Acquisition Module is used to construct an ambiguity knowledge graph based on historical project materials and the review status of those materials. The consistency check result acquisition module is used to obtain the attributes of the project material median by using the ambiguous knowledge graph as the basis for obtaining ambiguous attributes, and to obtain the consistency check result by using the large language model to obtain the attributes of the project material median. The standardization inspection module is used to perform standardization inspection on project materials based on the consistency inspection results, the median value of project materials, and the attributes of the median value of project materials, and to determine whether the standardization inspection passes. When the inspection passes, an approval message is output; when the inspection fails, an approval message is output.

[0014] By adopting the above technical solution, the present invention has the following advantages: By combining a multimodal large model with a large language model to review project materials, the review efficiency is significantly improved. An ambiguous knowledge graph is constructed by combining historical project materials with the review status of historical project materials, and the ambiguous knowledge graph is used as the basis for obtaining ambiguous attributes. The attributes of the values ​​in the project materials are obtained through the large language model, thereby improving the accuracy of attribute determination and solving the technical problem that existing technologies cannot balance review efficiency and review accuracy.

[0015] The present invention also provides a computer device, including: a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of the AI-based photovoltaic project filing and review method. Attached Figure Description

[0016] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. The drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings.

[0017] Figure 1 This is a flowchart illustrating an AI-based photovoltaic project filing and review method according to the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only one preferred embodiment of this invention and are only used to explain this invention. They do not limit the scope of protection of this invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0019] Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the operations (or steps) as sequential processes, many of the operations (or steps) can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but it may also have additional steps not included in the figures; the process may correspond to a method, function, procedure, subroutine, subroutine, etc.

[0020] Example 1: like Figure 1As shown, an AI-based photovoltaic project filing and review method includes the following steps: S1: Based on the multimodal large model, perform content recognition and semantic classification on the project materials to obtain the project material type. If the project material type meets the preset conditions, then execute S2; otherwise, it means that the review has not passed.

[0021] In some embodiments, S1, the step of obtaining the project material type by performing content recognition and semantic classification on the project materials based on a multimodal large model includes: Obtain audit standards based on audit requirements; The project materials in various forms are converted into standardized text through a multimodal large model. The file name of the standardized text is matched with the preset project material type in the review standard to obtain the matching probability. The similarity probability is obtained based on the average semantic similarity between the text content in the standardized text and the preset project material type. The project material type is then determined based on the matching probability and the similarity probability.

[0022] Understandably, the audit standards are uniformly archived in the SOP document. The audit standards include preset project material types, preset attributes, naming conventions for key fields (such as the project name format should be "Project Company + District / County + Phase Number + Farmer Rooftop Distributed Photovoltaic Project"), and a set of field consistency checks (such as "Registration Area" and "Detailed Address of Construction Location" must match). The preset project material types are specifically a set of types consisting of the list of required document types corresponding to each province, city, district, and county. The preset attributes are specifically a set of attributes consisting of the attributes corresponding to each field in the list of required key fields of the document.

[0023] Multimodal large-scale models are cutting-edge models in the field of artificial intelligence that integrate the processing capabilities of multiple data types. They can simultaneously receive, understand, analyze, and generate data of different modalities such as text, images, voice, video, and tables. This breaks down the information barriers of traditional single-modal models (such as large language models that only process text, and visual models that only process images), achieving cross-modal semantic alignment and intelligent interaction. Understandably, photovoltaic project filing materials come in various formats, including PDF documents, scanned images, handwritten application forms, and Excel spreadsheets. Different formats make direct keyword matching and semantic analysis difficult. Multimodal large-scale models convert project materials of various formats into standardized text, eliminating verification barriers caused by format differences. This allows for the combination of the content in the project materials with review standards to determine the project material type. Furthermore, a dual-dimensional fusion judgment mechanism using filename matching probability and content similarity probability improves the accuracy of project material type determination. In this embodiment, the OCR capability integrated into the multimodal large-scale model specifically identifies text in images / PDFs; when the project material type matches a preset project material type, it indicates that the project material type meets the preset conditions.

[0024] S2: Construct an ambiguous knowledge graph based on historical project materials and the review status of those materials.

[0025] In some preferred embodiments, step S2: constructing an ambiguous knowledge graph based on historical project materials and their review status, includes: The probability of the first attribute of the value in the historical project materials is obtained by using a large language model. The values ​​in the historical project materials are classified according to the probability of the first attribute, and the values ​​with ambiguous categories are marked as ambiguity points. Based on the values ​​and attributes of the uniquely attributed categories, and combined with the review status of historical project materials, the attributes of the ambiguous points are obtained. The ambiguous points are then associated with each other's attributes, and the frequency of association is recorded. Context nodes are obtained based on the context information of the ambiguity points, rule nodes are obtained based on the review standards obtained based on the review requirements, the attributes of the ambiguity points are used as attribute nodes, and the content in the historical project materials where the ambiguity points are located is bound to the attribute nodes. An ambiguous knowledge graph is obtained based on ambiguity points, context nodes, rule nodes, attribute nodes, and association frequency.

[0026] In this embodiment, obtaining an ambiguous knowledge graph based on ambiguous points, context nodes, rule nodes, attribute nodes, and association frequency specifically refers to: treating ambiguous points as ambiguous nodes, connecting ambiguous nodes to context nodes, connecting context nodes to rule nodes, connecting rule nodes to attribute nodes, and using association frequency as the weight of the edge between rule nodes and attribute nodes. The review criteria also include geographic information, industry terminology, enterprise relationships, and filing rules. Taking geographic level as an example, if the project materials contain the information "Guangfeng Energy Co., Ltd.", it is difficult to determine whether Guangfeng is a county or a company name based solely on the literal meaning. In this case, comparing Guangfeng with geographic information reveals that Guangfeng County does not exist in the geographic information; therefore, it is identified as a company name. By setting rule nodes, traversal paths that do not meet the rules are eliminated, which not only improves the efficiency of attribute acquisition but also improves the accuracy of attributes.

[0027] Specifically, the attributes for obtaining ambiguity points based on the unique category value and the attribute with a unique category value, combined with the review status of historical project materials, include: If the historical project materials are approved, the values ​​with unique categories will be matched with the attributes and review criteria of the values ​​with unique categories to obtain the attributes of the ambiguous points. If the materials for a historical project fail the review, the attributes of the points of ambiguity will be obtained based on the explanation of the failure and the review standards.

[0028] Understandably, when the difference between the two largest probabilities of the first attribute is greater than a preset difference, the value is uniquely assigned to a category; otherwise, it is ambiguously assigned. In this embodiment, if the historical project materials pass the review, it means that the value in the historical project materials and the attribute of the value in the historical project materials meet the review criteria. Therefore, after matching, the attribute of the ambiguous point can be deduced based on the matching result. If the historical project materials fail the review, it means that the value in the historical project materials and the attribute of the value in the historical project materials do not meet the review criteria. Therefore, the content that does not meet the criteria is first located based on the explanation of the failure, and then the attribute of the ambiguous point is deduced based on the review criteria.

[0029] S3: Using the ambiguous knowledge graph as the basis for obtaining ambiguous attributes, the attributes of the project material median are obtained through the large language model, and the consistency test results are obtained by checking the consistency between the project material median and the attributes of the project material median.

[0030] In one embodiment, S3, the step of using an ambiguous knowledge graph as the basis for obtaining ambiguous attributes and acquiring the attributes of values ​​in project materials through a large language model includes: The probability of the second attribute of the value in the project material is obtained by using a large language model. If the probability of the second attribute meets the preset requirements, the attribute corresponding to the largest probability of the second attribute is taken as the attribute of the value in the project material. Otherwise, the value in the project material is matched with the ambiguous knowledge graph to locate the traversal node on the ambiguous knowledge graph. The traversal node is used as the starting point to traverse the ambiguous knowledge graph and obtain the traversal result. The attribute of the value in the project material is obtained based on the traversal result.

[0031] In this embodiment, when the difference between the two largest probabilities of the second attribute is greater than a preset difference, it indicates that the probability of the second attribute meets the preset requirement. The values ​​in the project materials are matched with the ambiguous knowledge graph to locate the traversal node on the ambiguous knowledge graph. The traversal node is used as the starting point to traverse the ambiguous knowledge graph and obtain the traversal result. Based on the traversal result, the attributes of the values ​​in the project materials are obtained. Specifically, the values ​​in the project materials are matched with the ambiguous nodes in the ambiguous knowledge graph to locate the target ambiguous node. The traversal direction is obtained based on the context information of the values ​​in the project materials, and the ambiguous knowledge graph is traversed starting from the target ambiguous node. When a rule node is reached, the values ​​in the project materials are matched with... The rules of each rule node are compared. If a match is successful, the traversal continues in that direction. If multiple attribute nodes are located after the traversal, the content of the project material containing the value is matched with the content of the historical project material bound to the attribute node, in descending order of association frequency. If the matching degree is greater than the preset matching degree, the attribute of the value in the project material is the attribute corresponding to that attribute node; otherwise, the content of the project material containing the value is matched with the content of the historical project material bound to another attribute node. The preset matching degree can be flexibly set according to user needs. Through a multi-level link of ambiguous node matching, context-oriented traversal, rule node verification, and historical content association matching, the problem of difficulty in accurately judging attributes due to the ambiguity of field attributes, even with semantic similarity analysis combined with contextual information, is overcome, thus significantly improving the accuracy of attribute determination.

[0032] In S3, the step of performing a consistency check on the median value of the project materials and obtaining the consistency check result includes: Based on the semantic similarity between the attributes of the values ​​in the project materials and the preset attributes in the audit standards obtained based on the audit requirements, the values ​​in the project materials are located to the value set, and the values ​​in the project materials are matched with the value set to obtain the first consistency test result; Key-value pairs are constructed based on the median value of the project materials and the attributes of the median value of the project materials. The semantic similarity between the key-value pairs is obtained, and the second consistency test result is obtained based on the semantic similarity between the key-value pairs.

[0033] Specifically, a large language model is introduced to calculate the semantic similarity between preset attributes in the review standards (i.e., preset attributes in the review system, such as "agent name") and attribute descriptions in project materials (such as "agent name" or "agent unit"), thereby achieving accurate matching across naming expressions. It is important to clarify that each preset attribute in the system corresponds to a set of values ​​consisting of multiple specific values. If the first consistency check passes, it indicates that there is a value in the preset attribute value set that matches a value in the project materials, meaning that the system's preset field and the field in the project materials have completed the matching verification. The second consistency check result is obtained through the semantic similarity between key-value pairs. When the second consistency check result passes, it indicates that the cross-validation of the same content within the project materials has been successful.

[0034] S4: Based on the consistency test results, the median value of the project materials, and the attributes of the median value of the project materials, conduct a standard inspection on the project materials. If the standard inspection passes, the audit is approved; otherwise, the audit is not approved.

[0035] In some embodiments, step S4: performing a specification inspection on the project materials based on the consistency check results, the median value of the project materials, and the attributes of the median value of the project materials; if the specification inspection passes, the audit is approved; otherwise, the audit fails, including: If the consistency check passes, then the key-value pairs undergo explicit order specification checks and explicit content integrity checks. If both pass, the review is approved. Otherwise, implicit specifications are obtained by embedding positive and negative examples of specifications in the prompts of the large language model. The key-value pairs are then subjected to implicit order specification checks and implicit content integrity checks. If both pass, the review is approved. Otherwise, the review is not approved.

[0036] Specifically, if the consistency check passes, the explicit order specification check and explicit content integrity check for the key-value pairs also include: If both the first and second consistency tests pass, then the consistency test result is considered to have passed.

[0037] In this embodiment, for fields with ambiguous expressions or distorted formats (such as "Chint Aneng Changfeng Phase II Photovoltaic"), the large language model can perform intent recognition and standardized restoration, improving its fault tolerance. Faced with abbreviations like "Fengtai New Energy Changfeng Phase II Photovoltaic," which omit keywords such as "project" and "distributed," general rules are difficult to cover. By embedding typical positive and negative examples in the large language model's prompts, the model can be guided to learn implicit specifications. Three to five standard and non-standard examples are provided in the prompt, such as: Standard: "Fengtai New Energy Changfeng County Phase II Rural Rooftop Distributed Photovoltaic Project" → Compliant; Variant: "Changfeng Phase II Photovoltaic" → Missing elements such as "project company" and "farmer rooftops" → Non-compliant; Ambiguous: "Anhui Changfeng Phase 2" → Can be standardized to "Chint Solar Anhui Changfeng County Phase II..." → Compliant; Through context learning, the large language model can not only identify explicit compliant texts, but also restore the intent and standardize situations such as abbreviations, mixed use of numbers / Chinese characters ("Phase 2" vs "Phase II"), and reversed order ("Changfeng Phase II"). This significantly improves the recall rate. Finally, the parsed results are compared with the specification template, and fields that do not meet the requirements are automatically marked as "not passed". By explicitly checking the project materials and triggering the gradient checking strategy of implicit checking as needed, not only is the unnecessary computing power consumption greatly reduced, but the accuracy of the review is further improved.

[0038] Example 2: This example also provides a photovoltaic project filing review system based on AI, applicable to the described photovoltaic project filing review method based on AI, including: A project material type acquisition module, used to identify the content and perform semantic classification of project materials based on a multi-modal large model to obtain the project material type, and determine whether the project material type meets the preset conditions. When it is determined that the preset conditions are met, it prompts the ambiguous knowledge graph acquisition module to run. When it is determined that the preset conditions are not met, it outputs a review not passed prompt; An ambiguous knowledge graph acquisition module, used to construct an ambiguous knowledge graph based on historical project materials and the review situations of historical project materials; A consistency check result acquisition module, used to take the ambiguous knowledge graph as the basis for obtaining ambiguous attributes, obtain the attributes of the values in the project materials through a large language model, and perform a consistency check on the values in the project materials and the attributes of the values in the project materials to obtain the consistency check result; A specification check module, used to perform a specification check on the project materials based on the consistency check result, the values in the project materials, and the attributes of the values in the project materials, and determine whether the specification check passes. When it is determined that the check passes, it outputs a review passed prompt. When it is determined that the check does not pass, it outputs a review not passed prompt.

[0039] Example 3: This example also provides a computer device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the computer device runs, the processor communicates with the memory through the bus, and the processor executes the machine-readable instructions to perform the steps of the described photovoltaic project filing review method based on AI.

[0040] The specific embodiments described above are preferred embodiments of the AI-based photovoltaic project filing and review method and system of the present invention, and are not intended to limit the specific scope of the present invention. The scope of the present invention includes but is not limited to the specific embodiments described above. All equivalent changes made in accordance with the shape and structure of the present invention are within the protection scope of the present invention.

Claims

1. An AI-based method for reviewing and approving the filing of photovoltaic projects, characterized in that, Includes the following steps: S1: Based on the multimodal large model, perform content recognition and semantic classification on the project materials to obtain the project material type. If the project material type meets the preset conditions, then execute S2; otherwise, it means that the review has not passed. S2: Construct an ambiguous knowledge graph based on historical project materials and the review status of those materials; S3: Using the ambiguous knowledge graph as the basis for obtaining ambiguous attributes, the attributes of the project material median are obtained through the large language model, and the consistency test results are obtained by checking the consistency between the project material median and the attributes of the project material median. S4: Based on the consistency test results, the median value of the project materials, and the attributes of the median value of the project materials, conduct a standard inspection on the project materials. If the standard inspection passes, the audit is approved; otherwise, the audit is not approved.

2. The method for reviewing and approving photovoltaic project filings based on AI according to claim 1, characterized in that, In S1, the step of obtaining the project material type by performing content recognition and semantic classification of project materials based on a multimodal large model includes: Obtain audit standards based on audit requirements; The project materials in various forms are converted into standardized text through a multimodal large model. The file name of the standardized text is matched with the preset project material type in the review standard to obtain the matching probability. The similarity probability is obtained based on the average semantic similarity between the text content in the standardized text and the preset project material type. The project material type is then determined based on the matching probability and the similarity probability.

3. The method for reviewing and approving photovoltaic project filings based on AI according to claim 1, characterized in that, S2: Constructing an ambiguous knowledge graph based on historical project materials and their review status, including: The probability of the first attribute of the value in the historical project materials is obtained by using a large language model. The values ​​in the historical project materials are classified according to the probability of the first attribute, and the values ​​with ambiguous categories are marked as ambiguity points. Based on the values ​​and attributes of the uniquely attributed categories, and combined with the review status of historical project materials, the attributes of the ambiguous points are obtained. The ambiguous points are then associated with each other's attributes, and the frequency of association is recorded. Context nodes are obtained based on the context information of the ambiguity points, rule nodes are obtained based on the review standards obtained based on the review requirements, the attributes of the ambiguity points are used as attribute nodes, and the content in the historical project materials where the ambiguity points are located is bound to the attribute nodes. An ambiguous knowledge graph is obtained based on ambiguity points, context nodes, rule nodes, attribute nodes, and association frequency.

4. The AI-based photovoltaic project filing and review method according to claim 3, characterized in that, The attributes for obtaining ambiguity points based on the unique category and the attribute of the unique category, combined with the review status of historical project materials, include: If the historical project materials are approved, the values ​​with unique categories will be matched with the attributes and review criteria of the values ​​with unique categories to obtain the attributes of the ambiguous points. If the materials for a historical project fail the review, the attributes of the points of ambiguity will be obtained based on the explanation of the failure and the review standards.

5. The method for reviewing and approving photovoltaic project filings based on AI according to claim 1, characterized in that, In S3, the attribute that uses an ambiguous knowledge graph as the basis for obtaining ambiguous attributes and obtains the values ​​in the project materials through a large language model includes: The probability of the second attribute of the value in the project material is obtained by using a large language model. If the probability of the second attribute meets the preset requirements, the attribute corresponding to the largest probability of the second attribute is taken as the attribute of the value in the project material. Otherwise, the value in the project material is matched with the ambiguous knowledge graph to locate the traversal node on the ambiguous knowledge graph. The traversal node is used as the starting point to traverse the ambiguous knowledge graph and obtain the traversal result. The attribute of the value in the project material is obtained based on the traversal result.

6. The method for reviewing and approving photovoltaic project filings based on AI according to claim 1, characterized in that, In S3, the step of performing a consistency check on the median value of the project materials and obtaining the consistency check result includes: Based on the semantic similarity between the attributes of the values ​​in the project materials and the preset attributes in the audit standards obtained based on the audit requirements, the values ​​in the project materials are located to the value set, and the values ​​in the project materials are matched with the value set to obtain the first consistency test result; Key-value pairs are constructed based on the median value of the project materials and the attributes of the median value of the project materials. The semantic similarity between the key-value pairs is obtained, and the second consistency test result is obtained based on the semantic similarity between the key-value pairs.

7. The method for reviewing and approving photovoltaic project filing based on AI according to claim 6, characterized in that, S4: Based on the consistency check results, the median value of the project materials, and the attributes of the median value of the project materials, a standard check is performed on the project materials. If the standard check passes, the audit is approved; otherwise, the audit fails. This includes: If the consistency check passes, then the key-value pairs undergo explicit order specification checks and explicit content integrity checks. If both pass, the review is approved. Otherwise, implicit specifications are obtained by embedding positive and negative examples of specifications in the prompts of the large language model. The key-value pairs are then subjected to implicit order specification checks and implicit content integrity checks. If both pass, the review is approved. Otherwise, the review is not approved.

8. The method for reviewing and approving photovoltaic project filing based on AI according to claim 7, characterized in that, If the consistency check passes, the explicit order specification check and explicit content integrity check for the key-value pairs also include: If both the first and second consistency tests pass, then the consistency test result is considered to have passed.

9. An AI-based photovoltaic project filing and review system, applicable to the AI-based photovoltaic project filing and review method described in any one of claims 1-8, characterized in that, include: The project material type acquisition module is used to perform content recognition and semantic classification on project materials based on a multimodal large model to obtain the project material type and determine whether the project material type meets the preset conditions. When the preset conditions are met, the ambiguity knowledge graph acquisition module is activated. When the preset conditions are not met, a prompt indicating that the review has not been approved is output. The Ambiguity Knowledge Graph Acquisition Module is used to construct an ambiguity knowledge graph based on historical project materials and the review status of those materials. The consistency check result acquisition module is used to obtain the attributes of the project material median by using the ambiguous knowledge graph as the basis for obtaining ambiguous attributes, and to obtain the consistency check result by using the large language model to obtain the attributes of the project material median. The standardization inspection module is used to perform standardization inspection on project materials based on the consistency inspection results, the median value of project materials, and the attributes of the median value of project materials, and to determine whether the standardization inspection passes. When the inspection passes, an approval message is output; when the inspection fails, an approval message is output.

10. A computer device, comprising: The computer device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of the AI-based photovoltaic project filing and review method as described in any one of claims 1-8.

Images