Power distribution network engineering project file review method, device and equipment

Abstract

The embodiment of the invention provides a power distribution network engineering project file review method, device and equipment, and the method comprises the steps: carrying out the modal fusion processing of heterogeneous content data in a target project file, and obtaining modal fusion data; constructing service structured data according to the modal fusion data; and performing business logic check processing according to the business structured data to obtain check result data. According to the technical scheme provided by the embodiment of the invention, full-automatic, high-efficiency and high-reliability project file review can be realized.

Description

Technical Field

[0001] This application relates to the field of project review, and in particular to a method, apparatus and equipment for reviewing project documents for power distribution network engineering projects. Background Technology

[0002] The review of power distribution network projects often involves multi-source heterogeneous documents. For example, in the review of reserve projects for rural power distribution networks, these documents include feasibility study reports in Word format, investment estimate tables in Excel format, budget documents in PDF format, scanned copies of approval opinions, design drawings, and signed or handwritten documents in image format. Current review work mainly relies on manual labor, which suffers from time-consuming manual parsing, difficulties in information extraction, and a lack of more reliable verification methods. Although technologies such as OCR (Optical Character Recognition), NLP (Natural Language Processing), and object detection have been widely researched with technological advancements, automated techniques for comprehensively parsing multi-source heterogeneous documents are still lacking for document review in the project review field. Summary of the Invention

[0003] The embodiments of this application provide a method, apparatus, and equipment for reviewing project documents for power distribution network engineering projects, which can achieve fully automated, highly efficient, and highly reliable project document review.

[0004] In a first aspect, embodiments of this application provide a method for reviewing project documents for power distribution networks, including:

[0005] Modal fusion processing is performed on the heterogeneous content data in the target project file to obtain modal fusion data;

[0006] Business structured data is constructed based on the modal fusion data;

[0007] The business logic is verified based on the structured business data to obtain the verification result data.

[0008] Optionally, the modal fusion processing of heterogeneous content data in the target project file to obtain modal fusion data includes:

[0009] Construct modal feature vectors for each of the heterogeneous content data;

[0010] The modal feature vectors are fused and mapped to obtain fused modal vectors;

[0011] Semantic alignment is performed on each of the fused modal vectors to generate the modal fusion data.

[0012] Optionally, constructing business structured data based on the modal fusion data includes:

[0013] Extract the target entity object from the modality fusion data;

[0014] Entity relationship identification is performed on the target entity object to obtain the target entity relationship;

[0015] The target entity relationship is stored as target structured data to obtain the business structured data.

[0016] Optionally, the step of identifying entity relationships in the target entity object to obtain target entity relationships includes:

[0017] Construct candidate pairing relationships for each of the target entity objects;

[0018] Context recognition and rule recognition are performed on the candidate pairing relationships to obtain the relationship recognition type and relationship recognition confidence of each candidate pairing relationship;

[0019] Based on the relationship identification type and the relationship identification confidence level, the target entity relationship is determined from the candidate pairing relationships.

[0020] Optionally, the step of performing business logic verification processing based on the structured business data to obtain verification result data includes:

[0021] Based on the structured business data, obtain the target verification parameters;

[0022] Obtain business verification rules that match the target verification parameters; wherein, the business verification rules include business logic rules and numerical calculation rules;

[0023] Based on the business verification rules, the target verification parameters are subjected to the business logic verification process to obtain the verification result data.

[0024] Optionally, after performing the business logic verification process on the target verification parameters based on the business verification rules to obtain the verification result data, the method further includes:

[0025] Based on the verification results, obtain the original abnormal data in the heterogeneous content data;

[0026] Based on the original abnormal data, the verification result data is reviewed.

[0027] If the verification results confirm that the verification results are accurate, structured review data is generated based on the verification results and the original abnormal data.

[0028] Optionally, after generating structured review data based on the verification result data and the original anomaly data, the method further includes:

[0029] The structured review data is then visualized to obtain visualized review data.

[0030] The visualized review data is exported to the target review interface, and then provided to the target user in the backend through the target review interface.

[0031] Secondly, embodiments of this application provide a power distribution network engineering project document review device, comprising:

[0032] The modality fusion module is used to perform modality fusion processing on heterogeneous content data in the target project file to obtain modality fused data;

[0033] The structured module is used to construct business structured data based on the modal fusion data;

[0034] The logic verification module is used to perform business logic verification processing based on the business structured data to obtain verification result data.

[0035] Thirdly, embodiments of this application provide an electronic device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the power distribution network engineering project document review method provided in embodiments of this application.

[0036] Fourthly, embodiments of this application provide a computer-readable storage medium, characterized in that it stores a computer program thereon, which, when executed in a computer, causes the computer to execute the power distribution network engineering project document review method provided in embodiments of this application.

[0037] The technical solution provided in this application embodiment performs modal fusion processing on heterogeneous content data in the target project file to obtain modal fusion data, and constructs business structured data accordingly. Then, it performs business logic verification processing based on the business structured data to obtain verification result data. This solves the technical problems of low efficiency and insufficient reliability caused by the need for manual processing of heterogeneous files in the prior art, and realizes fully automated, highly efficient and highly reliable project file review. Attached Figure Description

[0038] Figure 1 This is a flowchart of a method for reviewing power distribution network project documents provided in an embodiment of this application;

[0039] Figure 2 This is a structural block diagram of a power distribution network project document review device provided in an embodiment of this application;

[0040] Figure 3 This is a schematic diagram of an electronic device structure provided in an embodiment of this application. Detailed Implementation

[0041] The present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] Figure 1 This is a flowchart of a power distribution network project document review method provided in an embodiment of this application. The method can be executed by a project document review device, which can be implemented by software and / or hardware, and can be configured in an electronic device such as a computer.

[0043] like Figure 1 As shown, the technical solution provided in this application includes the following steps:

[0044] S110: Perform modal fusion processing on the heterogeneous content data in the target project file to obtain modal fusion data.

[0045] The target project documents can be power distribution network engineering documents that require project review. Heterogeneous content data can be all content in different file formats and data types included in the target project documents. Specifically, the specific content of the target project documents can be determined according to project needs, and preferably can include all documents required for project review. The heterogeneous content data in the target project documents can specifically include some or all of document data in various formats, tabular data, and image data including scanned copies, drawings, signatures, and handwritten content.

[0046] Furthermore, modality fusion processing can be an operation that processes heterogeneous content data of different formats and types into a unified data modality, thereby obtaining modality fusion data of a single data modality. The modality fusion data needs to include the content and features of the heterogeneous content data. For example, it can involve parsing the data format and performing semantic recognition on the heterogeneous content data, presenting the recognition results in a specific data format. Preferably, the modality fusion data can be a feature vector constructed based on the content and features of the heterogeneous content data.

[0047] In one optional implementation, modal fusion processing is performed on the heterogeneous content data in the target project file to obtain modal fusion data. This may include: constructing modal feature vectors for each heterogeneous content data; performing fusion mapping processing on the modal feature vectors to obtain fused modal vectors; and performing semantic alignment processing on each fused modal vector to generate modal fusion data.

[0048] The modality feature vector can be represented by the modality embedding vector of the content and features of each heterogeneous content data. Specifically, for document modal data that may be included in the heterogeneous content data, modality feature vectors can be constructed by analyzing its layout structure and semantic features. Optionally, the LayoutLMv3 model can be used to analyze the layout structure and semantic features of Word / PDF documents. For table modal data that may be included in the heterogeneous content data, modality feature vectors can be constructed by identifying the correspondence between table fields and labels. Optionally, the Table Transformer model can be used to identify the correspondence between table fields and labels. For image modal data such as scanned documents and design drawings that may be included in the heterogeneous content data, modality feature vectors can be constructed by extracting their visual features. Optionally, Swin can be used. The Transformer extracts visual features from scanned documents and design drawings. For signature modal data that may be included in heterogeneous content data, the authenticity of the signature can be verified through feature comparison, and a modal feature vector can be constructed based on this feature. Specifically, an improved YOLOv8 model can be used to detect the seal area, and the authenticity of the seal can be verified by comparing pHash with convolutional features. For handwritten content modal data that may be included in heterogeneous content data, the authenticity of the handwritten person can be identified through font feature comparison, and a modal feature vector can be constructed based on this feature. Specifically, a CNN-BiLSTM-CTC network can be used to identify handwritten signature content and compare the signer. Further optionally, constructing modal feature vectors can include unifying the output results of data processing of various heterogeneous content data into a modal vector representation, which may include file ID (Identifier), field labels, and vector features. Optionally, it can be stored in a preset intermediate feature cache layer.

[0049] Furthermore, the fusion mapping process can be an operation that merges the representations of modal feature vectors from different modalities into a joint representation, enabling the representations of the same semantic entity in different modalities to transfer information to each other. Fusion mapping of modal feature vectors yields fused modal vectors. Optionally, a cross-modal encoder can be used to perform fusion mapping on the modal feature vectors. Semantic alignment can be an operation in the fusion representation space that groups together fields from different modalities that are semantically similar or identical, performing semantic alignment on each fused modal vector to generate modal fusion data. Optionally, contrastive learning can be used to achieve semantic alignment of the same indicator, such as investment amount or measure rate, across different documents; cross-document attention mechanisms can also be combined to improve alignment accuracy. Further, optionally, the modal fusion data can be a semantic alignment table, which can record the source document, semantic similarity, and consistency results for each core field.

[0050] S120: Construct business structured data based on modal fusion data.

[0051] Among them, business structured data can be data containing modal fusion data and describing the specific structure of the business logic of power distribution network engineering projects, preferably a knowledge graph.

[0052] In this embodiment, after mapping heterogeneous content data of different modalities to modal fusion data with unified modality, the heterogeneous content data of different modalities can be semantically understood in a unified manner, and the overall content of the target project file can be analyzed in a unified manner. Therefore, the business structured data constructed based on modal fusion data can truly reflect the business logic of the complete power distribution network project.

[0053] In one optional implementation, constructing business structured data based on modal fusion data may include: extracting target entity objects from the modal fusion data; identifying entity relationships of the target entity objects to obtain target entity relationships; and storing the target entity relationships as target structured data to obtain business structured data.

[0054] The target entity objects can include specific concepts and / or terms from the power distribution network project, such as investment amount, measure rate, lag coefficient, power outage plan, signature type, and design unit. Entity relationship identification can be the operation of obtaining the relationships between various target entity objects. Target entity relationships can be the relationships between various target entity objects. Target structured data can be data with a preset specific structure.

[0055] Specifically, by uniformly analyzing the same semantic content across different expressions and modalities in the modality fusion data, target entity objects can be extracted. Entity relationship identification can be performed based on the descriptions in the modality fusion data, determining the target entity relationships between various target entity objects. Optionally, entity relationship identification can be automatically extracted based on rules such as domain knowledge, unit rules, or format rules, or it can use learning-driven methods, such as BERT (Bidirectional Encoder Representations from Transformers) combined with side-by-side classification or prediction for automatic extraction. Based on a pre-defined specific structure, the target entity relationships are stored as target structured data, serving as business structured data. Specifically, this can be achieved by writing the target entity relationships into RDF (Resource Description Framework) triples and performing consistency checks using known rules, such as checking for circular dependencies or missing key relationships.

[0056] In an optional implementation, after obtaining the structured business data, the system may further include chain-like reasoning and anomaly detection using a GNN (Graph Neural Network). Chain-like reasoning, for example, involves deriving the "number of households experiencing power outages" from two relationships: "power outage plan - affected transformer areas" and "transformer area - number of users," through graph propagation. When missing data or semantic ambiguity prevents rules from covering certain data, the GNN can provide possible values ​​or consistency scores based on neighborhood information and offer interpretable multi-hop reasoning paths. Anomaly detection involves propagating the inconsistency score along relationships if a node is inconsistent with its neighboring nodes, such as investment amount and measure rate, or adjustment coefficient, helping to locate the root cause node. Further optionally, this can be encoded as edge weights in the graph and used for soft reasoning with the GNN, simulating expert judgment on complex situations based on multi-hop relationships and empirical weights.

[0057] In one optional implementation, identifying the target entity relationships by performing entity relationship identification on the target entity objects may include: constructing candidate pairing relationships for each target entity object; performing context identification and rule identification on the candidate pairing relationships to obtain the relationship identification type and relationship identification confidence of each candidate pairing relationship; and determining the target entity relationship from the candidate pairing relationships based on the relationship identification type and relationship identification confidence.

[0058] Here, candidate pairing relationships can be determined directly based on the position of each target entity object within the overall semantics of the modal fusion data. Context recognition can be an operation that identifies relationships by combining the contextual semantics of the position of each target entity object. Rule recognition can be an operation that identifies relationships based on explicit rule expressions associated with each target entity object, such as using relevant formulas. Relationship recognition type can be the type of relationship determined, optionally including adjustment relationships, inclusion relationships, and causal relationships. Relationship recognition confidence can be data describing the reliability of the identified candidate pairing relationships.

[0059] Specifically, after identifying each target entity, candidate pairings can be constructed for entities within the project based on page and / or table column proximity and time / / or paragraph association. For example, each "investment amount" entity can be paired with its nearest "adjustment coefficient" entity. Further, contextual recognition is performed on the candidate pairings. Preferably, this involves using a Transformer with positional encoding or a graph neural network as input to the entity pairs in the candidate pairings, outputting a relationship identification type of adjustment, inclusion, or causation, and providing the corresponding relationship identification confidence level. Rule recognition is then performed on the candidate pairings. Preferably, relevant formulas are identified, outputting a relationship identification type of adjustment, and extracting the coefficient values ​​from the formulas. The relationship identification confidence level is a fully confident value. For example, if "amount × (1 + adjustment coefficient)" is explicitly expressed, it can be directly determined that "amount" and "adjustment coefficient" have an adjustment relationship. Therefore, based on the relationship identification confidence levels obtained from contextual recognition and rule recognition, the target entity relationships can be identified from a sufficiently certain pool of candidate pairings.

[0060] In an optional implementation, after performing context recognition and rule recognition on candidate pairings to obtain the relationship recognition type and relationship recognition confidence of each candidate pairing, the method may further include: automatically extracting candidate pairings whose relationship recognition confidence is lower than a preset confidence threshold, marking the automatically extracted candidate pairings and providing them to the manual confirmation interface to obtain the manual confirmation result, and using the corresponding candidate pairings and the manual confirmation result for learning subsequent entity relationship recognition.

[0061] S130: Perform business logic verification processing based on the business structured data to obtain the verification result data.

[0062] Specifically, business logic verification processing can be an operation that verifies the correctness and rationality of the business logic of a power distribution network project described by structured business data. The verification result data can be data describing the results of the business logic verification processing.

[0063] In this embodiment, business logic verification processing can be performed on structured business data according to preset rules and / or methods. The purpose is to determine whether the content of the target project document described by the structured business data conforms to real-world logic, which is reflected in the verification result data. This verification result data can then provide a basis for subsequent project review. Specifically, the verification can be based on matching rules for each entity concept and / or term in the structured business data. These rules can be pre-set or automatically extracted from historical data, and can include logical rules and / or numerical calculation rules. In an optional implementation, a model can be built and trained using expert review of historical data to achieve the fusion and dynamic optimization of rules and experience.

[0064] In one optional implementation, performing business logic verification processing on the business structured data to obtain verification result data may include: obtaining target verification parameters based on the business structured data; obtaining business verification rules that match the target verification parameters; and performing business logic verification processing on the target verification parameters based on the business verification rules to obtain verification result data.

[0065] The target verification parameter can be a specific data value in the structured business data that needs to be verified. The business verification rule can be the rule that the target verification parameter needs to conform to. The business verification rule includes business logic rules and numerical calculation rules.

[0066] Specifically, business logic rules can include rules determined based on the actual logical relationships between various concepts and / or terms in a distribution network project. For example, these could be business logic rules that limit the logic of power outage plans during distribution network construction based on actual conditions. Specifically, the number of outage customers could equal the total number of outage customers in each transformer area, and the outage area should cover the construction area. If the area or quantity does not match, it is considered abnormal. Rules can also include verification rules for the necessary format and content of project documents. For example, each document could be required to include at least the design unit's seal and the reviewer's signature; otherwise, it would be marked as non-compliant. Numerical calculation rules are rules determined based on the numerical calculation relationships between various numerical data in a distribution network project. For example, these could be numerical calculation rules between investment amounts, adjustment coefficients, and measure rates determined by calculation formulas.

[0067] In one optional implementation, the target verification parameters may include investment amount parameters, adjustment coefficient parameters, and measure fee rate parameters. The investment amount parameter can be a value identified from the document, for example, extracted from a table or text and then standardized. The adjustment coefficient parameter can be derived from budget sheets, technical specifications, and quota attachments. If not explicitly stated in the document, it can be inferred from project templates or historical data, for example, through GNN inference or by selecting a default value. The measure fee rate parameter may be directly given in the document and obtained through identification, or it may be a preset compliance rate threshold, such as from industry standards, local regulations, or company experience values. During verification, the identified value can be compared with the preset standard.

[0068] Therefore, by performing business logic verification according to business verification rules, verification result data can be obtained. In an optional implementation, the verification result data can be a logic inference object, which includes the verification conclusion, exception type, and reasoning basis.

[0069] In an optional implementation, after performing business logic verification processing on the target verification parameters based on business verification rules to obtain verification result data, the method may further include: obtaining the original abnormal data in the heterogeneous content data based on the verification result data; performing verification result review on the verification result data based on the original abnormal data; and generating structured review data based on the verification result data and the original abnormal data if the verification result data is confirmed to be accurate through the verification result review.

[0070] The original abnormal data can be related heterogeneous content data described as abnormal in the verification result data. Verification result review can be an operation that verifies the accuracy of the verification result data by checking the content of the original abnormal data. Structured review data can be data with a preset specific format and structure used to record verification result data and original abnormal data.

[0071] Specifically, when the verification results are obtained based on structured business data, to verify whether anomalies in the verification results actually exist in the original content of the target project file, the original anomaly data from the heterogeneous content data can be obtained and the verification results reviewed. If the verification results are confirmed to be accurate, indicating that anomalies do exist, then the verification results and the original anomaly data can be recorded together in the structured review data.

[0072] In one optional implementation, the verification results are reviewed based on the original abnormal data. Specifically, this may include: comparing the numerical consistency of fields in multiple documents; detecting issues such as missing signatures, incomplete signatures, and forged seals; and identifying issues such as misaligned rows in tables, blurred scans, and missing pages in documents based on the Deep SVDD (Deep Support Vector Data Description) model.

[0073] In one optional implementation, the structured review data can be a structured validation record, including field names, issue types, confidence levels, and location information. The field names can be the field names within the heterogeneous content data of the target project file.

[0074] In an optional implementation, after generating structured review data based on the verification result data and the original abnormal data, the process may further include: visualizing the structured review data to obtain visualized review data; and exporting the visualized review data to the target review interface to provide it to the target user in the background through the target review interface.

[0075] The visualization process involves presenting structured review data in a visual format. Visualized review data refers to data that visualizes the content of structured review data. The target review interface is a pre-defined data transmission interface used to present data to the target user in the backend. The target user in the backend can be any user who can view the visualized review data using any electronic device; preferably, based on the automated analysis and organization of the target project documents in the visualized review data, it can be directly provided to review experts. Therefore, the target user in the backend can be an expert who needs to conduct project reviews based on power distribution network project documents.

[0076] In one optional implementation, the visualized review data may include review comment forms, error reports, and statistical reports. The review comment forms can be used to present key verification conclusions and recommendations; error reports can be used to present summarized problems and their impact; and statistical reports can be used to present indicators such as statistical consistency pass rate and signature completeness rate. Optionally, the data can be exported and pushed to the target review interface in Word and / or PDF format.

[0077] The technical solution provided in this application embodiment performs modal fusion processing on heterogeneous content data in the target project file to obtain modal fusion data, and constructs business structured data accordingly. Then, it performs business logic verification processing based on the business structured data to obtain verification result data. This solves the technical problems of low efficiency and insufficient reliability caused by the need for manual processing of heterogeneous files in the prior art, and realizes fully automated, highly efficient and highly reliable project file review.

[0078] Figure 2 This is a structural block diagram of a power distribution network engineering project document review device provided in an embodiment of this application, such as... Figure 2 As shown, the device includes:

[0079] Modality fusion module 210 is used to perform modality fusion processing on heterogeneous content data in the target project file to obtain modality fusion data;

[0080] The structuring module 220 is used to construct business structured data based on the modal fusion data;

[0081] The logic verification module 230 is used to perform business logic verification processing based on the business structured data to obtain verification result data.

[0082] In one optional implementation, the modality fusion module 210 can be specifically used to: construct modality feature vectors for each of the heterogeneous content data; perform fusion mapping processing on the modality feature vectors to obtain fused modality vectors; and perform semantic alignment processing on each of the fused modality vectors to generate the modality fusion data.

[0083] In one optional implementation, the structuring module 220 may include: an entity extraction unit for extracting target entity objects from the modality fusion data; a relationship identification unit for identifying entity relationships in the target entity objects to obtain target entity relationships; and a structured storage unit for storing the target entity relationships as target structured data to obtain the business structured data.

[0084] In one optional implementation, the relationship identification unit may be specifically used to: construct candidate pairing relationships for each of the target entity objects; perform context identification and rule identification on the candidate pairing relationships to obtain the relationship identification type and relationship identification confidence level of each candidate pairing relationship; and determine the target entity relationship in the candidate pairing relationships based on the relationship identification type and the relationship identification confidence level.

[0085] In one optional implementation, the logic verification module 230 can be specifically used to: obtain target verification parameters based on the business structured data; obtain business verification rules that match the target verification parameters; wherein the business verification rules include business logic rules and numerical calculation rules; and perform the business logic verification processing on the target verification parameters based on the business verification rules to obtain the verification result data.

[0086] In one optional implementation, the logic verification module 230 can also be used to: obtain the original abnormal data in the heterogeneous content data based on the verification result data; perform verification result review on the verification result data based on the original abnormal data; and generate structured review data based on the verification result data and the original abnormal data if the verification result review determines that the verification result data is accurate.

[0087] In an optional implementation, the logic verification module 230 can also be used to: perform visualization processing on the structured review data to obtain visualized review data; and export the visualized review data to the target review interface to provide it to the target user in the background through the target review interface.

[0088] like Figure 3 As shown in the figure, this application provides an electronic device, including a processor 111, a communication interface 112, a memory 113, and a communication bus 114, wherein the processor 111, the communication interface 112, and the memory 113 communicate with each other through the communication bus 114.

[0089] Memory 113 is used to store computer programs;

[0090] In one embodiment of this application, when the processor 111 executes the program stored in the memory 113, it implements the power distribution network engineering project document review method provided in any of the foregoing method embodiments, including:

[0091] Modal fusion processing is performed on the heterogeneous content data in the target project file to obtain modal fusion data;

[0092] Business structured data is constructed based on the modal fusion data;

[0093] The business logic is verified based on the structured business data to obtain the verification result data.

[0094] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the method provided in any of the foregoing method embodiments.

[0095] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0096] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0097] The above embodiments are merely illustrative examples and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this application.

Claims

1. A method for reviewing project documents for power distribution network engineering projects, characterized in that, include: Modal fusion processing is performed on the heterogeneous content data in the target project file to obtain modal fusion data; Business structured data is constructed based on the modal fusion data; The business logic is verified based on the structured business data to obtain the verification result data.

2. The method according to claim 1, characterized in that, The modal fusion processing of heterogeneous content data in the target project file to obtain modal fusion data includes: Construct modal feature vectors for each of the heterogeneous content data; The modal feature vectors are fused and mapped to obtain fused modal vectors; Semantic alignment is performed on each of the fused modal vectors to generate the modal fusion data.

3. The method according to claim 1, characterized in that, The step of constructing business structured data based on the modality fusion data includes: Extract the target entity object from the modality fusion data; Entity relationship identification is performed on the target entity object to obtain the target entity relationship; The target entity relationship is stored as target structured data to obtain the business structured data.

4. The method according to claim 3, characterized in that, The step of identifying entity relationships in the target entity object to obtain target entity relationships includes: Construct candidate pairing relationships for each of the target entity objects; Context recognition and rule recognition are performed on the candidate pairing relationships to obtain the relationship recognition type and relationship recognition confidence of each candidate pairing relationship; Based on the relationship identification type and the relationship identification confidence level, the target entity relationship is determined from the candidate pairing relationships.

5. The method according to claim 1, characterized in that, The step of performing business logic verification processing based on the structured business data to obtain verification result data includes: Based on the structured business data, obtain the target verification parameters; Obtain business verification rules that match the target verification parameters; wherein, the business verification rules include business logic rules and numerical calculation rules; Based on the business verification rules, the target verification parameters are subjected to the business logic verification process to obtain the verification result data.

6. The method according to claim 5, characterized in that, After performing the business logic verification process on the target verification parameters based on the business verification rules to obtain the verification result data, the method further includes: Based on the verification results, obtain the original abnormal data in the heterogeneous content data; Based on the original abnormal data, the verification result data is reviewed. If the verification results confirm that the verification results are accurate, structured review data is generated based on the verification results and the original abnormal data.

7. The method according to claim 6, characterized in that, After generating structured review data based on the verification result data and the original anomaly data, the process further includes: The structured review data is then visualized to obtain visualized review data. The visualized review data is exported to the target review interface, and then provided to the target user in the background through the target review interface.

8. A document review device for power distribution network engineering projects, characterized in that, include: The modality fusion module is used to perform modality fusion processing on heterogeneous content data in the target project file to obtain modality fused data; A structured module is used to construct business structured data based on the modal fusion data; The logic verification module is used to perform business logic verification processing based on the business structured data to obtain verification result data.

9. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the power distribution network engineering project document review method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores a computer program, which, when executed in a computer, causes the computer to perform the power distribution network project document review method as described in any one of claims 1-7.

Images