A Standardized Management Method for Power Industry Revenue Expansion Funds Based on OCR

By processing the pages and locating the fields of the power industry revenue expansion materials, generating OCR candidate values ​​and image evidence information, establishing confidence constraints, and performing local re-identification when there is a recognition deviation, the conflict problem caused by OCR recognition deviation in the power industry revenue expansion material processing is solved, and the accuracy and reliability of the recognition results are improved.

CN123118024APending Publication Date: 2026-11-17MARKETING SERVICE CENT OF STATE GRID HENAN ELECTRIC POWER CO +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611331134.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-31
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing methods for processing revenue in the power industry cannot accurately distinguish between apparent conflicts caused by OCR recognition bias and real conflicts caused by inconsistencies in the original content of the material, resulting in inaccurate recognition results.

Method used

By acquiring business data and receipt materials from power industry expansion orders, page processing and field region positioning are performed to generate OCR candidate values ​​and image evidence information. Confidence constraint relationships between image evidence nodes and business fact nodes are established, decorrelation processing is performed, and local re-identification is used to update OCR candidate values ​​when there is recognition deviation, ensuring the accuracy of the recognition results.

Benefits of technology

It enables the classification and judgment of OCR recognition deviations and material content conflicts, improves the reliability and accuracy of automatic verification of power industry revenue expansion materials, and reduces the need for manual review.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123118024A_ABST
    Figure CN123118024A_ABST
Patent Text Reader

Abstract

This invention relates to the field of power marketing data processing and discloses an OCR-based method for standardized management of power industry revenue collection materials. The method includes the following steps: processing the data collection materials by page layout and locating field regions, generating OCR candidate values ​​and image evidence information corresponding to each field region; establishing confidence constraint relationships between image evidence nodes and business fact nodes; determining business fact combinations and corresponding field recognition values; determining the field regions to be re-identified and re-identification paths based on the conflict constraint set; and classifying and determining conflicts induced by OCR recognition and genuine conflicts in the material content by reversing the location of the field regions to be re-identified based on whether the conflict is eliminated, the consistency of the image evidence information under different re-identification paths, and the independence between evidence source groups. By retaining OCR candidate values ​​supported by image evidence information and performing decorrelation processing on evidence from the same source field based on the field generation source, the method achieves classification and judgment of conflicts induced by OCR recognition and genuine conflicts in the material content.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of power marketing data processing technology, specifically to an OCR-based method for standardized management of power industry revenue expansion. Background Technology

[0002] With the digitalization of electricity marketing, business licenses, identity documents, authorization letters, property ownership certificates, and electricity application materials generated during the business expansion and installation process are gradually uploaded to the marketing system as scanned images or electronic documents. To achieve automated management of these materials, OCR technology is typically used to extract the entity name, document number, address, date, and business parameters from the materials, and then verify them against preset rules.

[0003] Existing methods for processing revenue collection in the power industry typically involve first locating field regions based on material type, obtaining field recognition results through OCR, and then verifying multiple revenue collection materials based on subject correspondence, time relationships, and material business relationships. This process usually uses the OCR output recognition results directly as the basis for subsequent verification. There is a lack of reverse feedback between OCR recognition and multi-material verification. However, when materials have blurred characters, obscured seals, or similar fonts, local OCR recognition deviations will be transmitted to the multi-material verification process. It is difficult to use the correlation between materials from different sources to correct OCR candidate values, thus failing to accurately distinguish between apparent conflicts caused by OCR recognition deviations and real conflicts caused by inconsistencies in the original content of the materials. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an OCR-based method for the standardized management of revenue expansion funds in the power industry. This method solves the problem that existing methods for processing revenue expansion funds in the power industry cannot accurately distinguish between apparent conflicts caused by OCR recognition bias and real conflicts caused by inconsistencies in the original content of the materials.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for standardized management of revenue expansion in the power industry based on OCR, comprising the following steps: Obtain business data and collection materials of the target power industry expansion order, perform page processing and field area positioning on the collection materials, and generate OCR candidate values ​​and image evidence information corresponding to each field area; Evidence source groups are divided according to the generation source of each field. Based on the OCR candidate value, the image evidence information and the business data, a confidence constraint relationship is established between the image evidence node and the business fact node, and the field evidence within the same evidence source group is decorrelated. Based on the decorrelated field evidence and the constraints of the power industry's revenue expansion, the combination of business facts and the corresponding field identification values ​​are determined. When there is no combination of business facts that satisfy the constraints of the power industry revenue expansion, extract the conflict constraint set, determine the field region to be re-identified and the re-identification path based on the conflict constraint set, and update the OCR candidate value of the corresponding field. Based on the updated OCR candidate values, the business fact combination is re-determined, and the data collection and management results are generated according to whether the conflict is eliminated, the consistency of the image evidence information under different re-identification paths, and the independence between evidence source groups.

[0006] By adopting the above technical solution, OCR candidate values ​​supported by image evidence information are retained, and the evidence of the same source field is decorrelated based on the field generation source. This associates image evidence, business data, and power industry expansion revenue constraints in different revenue materials with business fact nodes. When the current OCR candidate values ​​cannot form a combination of business facts that satisfies the constraints, the area of ​​the field to be re-identified is located in reverse based on the conflict constraint set, and the corresponding re-identification path is selected according to the conflict characteristics to update the OCR candidate values ​​and re-determine the combination of business facts. This avoids the same source copied fields from forming duplicate support for the recognition results and enables cross-material constraints to participate in the correction of local OCR results. It realizes the classification and judgment of OCR recognition-induced conflicts and real conflicts in material content, solving the problem that existing power industry expansion revenue processing methods cannot accurately distinguish between apparent conflicts caused by OCR recognition deviations and real conflicts caused by inconsistencies in the original content of the material.

[0007] Preferably, the step of performing page processing and field region positioning on the received materials includes: The received materials are analyzed and page images are generated page by page. Orientation correction, distortion correction, interference area separation and page partitioning are performed to form a processed page. The actual material type is determined based on the declared material type of the received materials and the layout characteristics of the processing page; The field area is located by calling the field template associated with the actual material type, and the coordinate mapping relationship between the field area and the original page is recorded.

[0008] Preferably, the generation of OCR candidate values ​​and image evidence information corresponding to each field region includes: Perform OCR recognition on each field area and retain the character sequence and corresponding recognition confidence information; The character sequence is formatted and normalized based on the field type to form OCR candidate values; Establish a correspondence between the characters in the OCR candidate values ​​and the glyph positions in the field area to form glyph alignment information; Image support candidate values ​​are filtered based on the identification confidence information, the glyph alignment information, and the field region quality information, and the image support candidate values ​​and their coordinate information are used as image evidence information.

[0009] Preferably, the step of dividing the evidence source groups according to the generation source of each field includes: Read the material generation method, data entry path and upstream field identifier of each field, and classify fields with the same upstream field identifier or direct inheritance relationship into the same evidence source group; Fields with different upstream field identifiers and no inheritance relationship between generation paths are divided into different evidence source groups; Field evidence pointing to the same business fact node within the same evidence source group is merged to form the de-correlated field evidence corresponding to each evidence source group.

[0010] Preferably, establishing the confidence constraint relationship between the image evidence node and the business fact node includes: Image evidence nodes are constructed using the OCR candidate values ​​corresponding to the field regions, image evidence information, and evidence source group identifiers. The fields in the business data are associated with the OCR candidate values ​​according to the field semantics to form business fact nodes; Associate the image evidence node with the corresponding business fact node; Based on the evidence source group identifier, the source association of related field evidence is determined, and field evidence belonging to the same evidence source group is merged to form irrelevant field evidence. Establish field correspondence, temporal correlation, and material business correlation among the business fact nodes to form power industry revenue expansion constraints.

[0011] Preferably, determining the combination of business facts and corresponding field identification values ​​includes: Based on the evidence source group, read the irrelevant field evidence and generate candidate values ​​for each business fact node; The candidate values ​​corresponding to different business fact nodes are associated, and the combination of business facts that satisfy the constraint relationship is determined based on the power industry revenue expansion constraints. Based on the combination of business facts, field recognition values ​​supported by image evidence information are determined from the OCR candidate values ​​of the corresponding field region.

[0012] Preferably, the extraction of the conflict constraint set includes: Obtain the power industry revenue expansion constraints corresponding to the current business fact combination; The selection process cannot simultaneously satisfy the constraints of the current candidate values, resulting in a set of conflicting constraints. The corresponding field area is determined based on the business fact nodes and image evidence nodes associated with the conflict constraint set.

[0013] Preferably, determining the field region to be re-identified and the re-identification path based on the conflict constraint set includes: Determine the field regions associated with the set of conflicting constraints; Based on the constraint satisfaction status after removing the candidate constraint relationship of the corresponding field region, determine the field region to be re-identified; A conflict fingerprint is generated based on the image state features, identification anomaly features, and association constraint features of the field region to be re-identified, and a re-identification path is determined based on the conflict fingerprint.

[0014] Preferably, updating the OCR candidate value of the corresponding field includes: Configure the image processing parameters, recognition range, and OCR model of the field region to be re-identified according to the target re-identification path, and perform local re-identification; The candidate values ​​generated by local re-identification are used to establish an image position correspondence with the corresponding field region, and the candidate values ​​with the image position correspondence are written into the OCR candidate values ​​of the corresponding field to form the updated OCR candidate values.

[0015] Preferably, the generation of standardized management results for data collection includes: Record the conflict constraint set and image evidence information before and after local re-identification; When the updated OCR candidate values ​​can form a combination of business facts that satisfy the constraints of the power industry revenue expansion, and the corresponding field recognition values ​​are supported by image evidence information, a recognition correction status is generated. When the same set of conflict constraints still exists after using different re-identification paths, and the conflict fields belong to different evidence source groups, a material conflict state is generated. Based on the identified correction status and material conflict status, corresponding data collection and management results are generated.

[0016] This invention provides an OCR-based method for the standardized management of revenue expansion in the power industry. It has the following beneficial effects: 1. This invention retains OCR candidate values ​​supported by image evidence information and performs decorrelation processing on evidence from the same source field based on the field generation source. It associates image evidence, business data, and power industry expansion revenue constraints from different data collection materials with business fact nodes. When the current OCR candidate values ​​cannot form a combination of business facts that satisfies the constraints, it reversely locates the field region to be re-identified based on the conflict constraint set and selects the corresponding re-identification path according to the conflict characteristics to update the OCR candidate values ​​and re-determine the combination of business facts. This avoids duplicate support from the same source field for the recognition result and enables cross-material constraints to participate in the correction of local OCR results, realizing the classification and determination of OCR recognition-induced conflicts and material content conflicts.

[0017] 2. By retaining candidate information and its original image evidence during the field recognition process, and combining the correlation between multiple data collection materials to cross-confirm the field content, this invention can reduce the transmission and amplification of local character misidentification in the subsequent verification process, so that the final determined field content has both image traceability and business consistency, thereby improving the reliability of the automatic verification results of data collection materials in the power industry.

[0018] 3. By distinguishing the source of field evidence and judging the persistence of conflict under various local re-identification conditions, this invention can avoid the formation of a false majority support from homologous copying information. At the same time, it can distinguish between anomalies that can be eliminated by OCR re-identification and anomalies caused by inconsistencies in the original material records, thereby improving the accuracy of anomaly location and reducing the repeated identification of irrelevant materials and unnecessary manual review. Attached Figure Description

[0019] Figure 1 The flowchart shows a method for standardized management of revenue expansion in the power industry based on OCR proposed in this invention. Figure 2 This is an architecture diagram of an OCR-based power industry revenue expansion and standardization management system proposed in an embodiment of the present invention; Figure 3 This is a flowchart of page processing and field region positioning proposed in an embodiment of the present invention; Figure 4 This is a flowchart illustrating the business fact combination inference proposed in this embodiment of the invention; Figure 5 This is a flowchart of the OCR candidate value update and result generation process proposed in an embodiment of the present invention. Detailed Implementation

[0020] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1: In the first embodiment of the present invention, the present invention provides an OCR-based method for the standardized management of revenue expansion in the power industry, such as... Figure 1 As shown, it includes the following steps: Please refer to the appendix. Figure 3 The system acquires business data and collection materials for the target power industry expansion order, performs page processing and field area positioning on the collection materials, and generates OCR candidate values ​​and image evidence information corresponding to each field area. Furthermore, page processing and field area positioning are performed on the received materials, including: The data collection materials are analyzed and page images are generated page by page. Orientation correction, distortion correction, interference area separation and page partitioning are performed to form a processed page. The actual material type is determined based on the declared material type of the received materials and the layout characteristics of the processing page; The field area is located by calling the field template associated with the actual material type, and the coordinate mapping relationship between the field area and the original page is recorded.

[0022] Furthermore, OCR candidate values ​​and image evidence information corresponding to each field region are generated, including: Perform OCR recognition on each field area and retain the character sequence and corresponding recognition confidence information; Based on the field type, the character sequence is formatted and normalized to form OCR candidate values; Establish a correspondence between the characters in the OCR candidate values ​​and the glyph positions in the field area to form glyph alignment information; Image support candidate values ​​are selected based on recognition confidence information, glyph alignment information, and field region quality information, and the image support candidate values ​​and their coordinate information are used as image evidence information.

[0023] Specifically, this step is used to convert unstructured data related to the target power industry expansion order into candidate field data with material category, page position, field semantics and image credibility basis, and to retain the traceability relationship from the field area to the original page, providing a unified input for subsequent evidence source grouping, business fact constraints and local re-identification of conflicting fields.

[0024] In this embodiment, the collected materials are parsed and a page image is generated. Page orientation correction, perspective distortion correction, background interference suppression, seal area separation, and page partitioning are performed sequentially. The coordinate transformation relationship before and after page processing is saved so that subsequent recognition results can be traced back to the original page. The material type is not directly adopted from the declared material type at the time of upload. Instead, the actual material type is determined by combining the page title, fixed text, table layout, seal position, and field distribution. When the declared material type is inconsistent with the page recognition result, the field template is called using the page recognition result, and the material is marked as a category pending confirmation.

[0025] In some embodiments, the business license template is used to locate the entity name, entity code and validity period, and the power of attorney template is used to locate the principal, agent and authorization period. After the field template is determined, the field area is located according to the field name position, table cell range and adjacent fixed text, and the material identification, page number and original page coordinates are recorded.

[0026] For each field region, multi-candidate OCR recognition is performed, retaining multiple character sequences and their recognition confidence information, instead of directly fixing the character sequence with the highest recognition probability. The character sequences are subjected to field format verification and normalization processing to form OCR candidate values. The ID number is verified for the number of characters and the check digit. The date field is converted to a unified date format. The main body name is cleared of spaces and the character format is unified.

[0027] The quality score for the field area was determined as follows ,in, Indicates the first Quality score for each field region Indicates sharpness, Indicates character completeness. Indicates the visibility of the seal after it is separated. This represents the regularity after distortion correction, with each parameter ranging from zero to one. Weights , , and The sum of them is one.

[0028] Align each character in the OCR candidate value with its corresponding glyph position within the field area to obtain the glyph alignment degree. ,in, Indicates the first The first field in the field area The glyph alignment of each OCR candidate value This indicates the number of characters in the OCR candidate value. Indicates the first Normalized deviation between the position of each candidate character and the center of the corresponding image glyph The allowable deviation scale for character position is indicated, preferably 0.15.

[0029] The image evidence score of the OCR candidate value was determined as follows: ,in, Indicates the score of image evidence. This indicates the identification confidence information. Indicates the validity of the field format. , , and This indicates the corresponding weights; in this embodiment, the four weights are preferably 0.35, 0.3, 0.2, and 0.15 respectively; when Not less than 0.72 and When the value is not lower than 0.65, the corresponding OCR candidate value is determined as the image support candidate value, and its recognition confidence information, glyph alignment information, field region quality information and original page coordinates are used together as image evidence information.

[0030] Evidence source groups are divided according to the generation source of each field. Based on OCR candidate values, image evidence information and business data, a confidence constraint relationship is established between image evidence nodes and business fact nodes. Field evidence within the same evidence source group is decorrelated. Furthermore, evidence source groups are categorized based on the generation source of each field, including: Read the material generation method, data entry path and upstream field identifier of each field, and classify fields with the same upstream field identifier or direct inheritance relationship into the same evidence source group; Fields with different upstream field identifiers and no inheritance relationship between generation paths are divided into different evidence source groups; Field evidence pointing to the same business fact node within the same evidence source group is merged to form the de-correlated field evidence corresponding to each evidence source group.

[0031] Furthermore, establish confidence constraint relationships between image evidence nodes and business fact nodes, including: Image evidence nodes are constructed using the OCR candidate values ​​corresponding to the field regions, image evidence information, and evidence source group identifiers. Associate the fields in the business data with the OCR candidate values ​​according to the field semantics to form business fact nodes; Associate the image evidence nodes with the corresponding business fact nodes; Based on the evidence source group identifier, the source association of related field evidence is determined, and field evidence belonging to the same evidence source group is merged to form irrelevant field evidence. Establish field correspondence, time sequence association, and material business association between business fact nodes to form a constraint on revenue expansion in the power industry.

[0032] Specifically, this step is used to further identify the generation dependency relationship between the evidence in each field based on the aforementioned OCR candidate values ​​and image evidence information, and to map the local recognition content in the material page to work order-level business facts; by distinguishing between independent source evidence and homologous copy evidence, it avoids the same business data being repeatedly reinforced by multiple generated materials, and at the same time provides a unified association structure for subsequent determination of business fact combination and extraction of conflict constraints.

[0033] In this embodiment, the material generation method, data entry path, and upstream field identifier corresponding to each field are read. The material generation method is used to characterize whether the field originates from the original material scanning, manual entry, automatic generation by the system, or transmission through an external interface. The data entry path is used to characterize the transmission process of the field between various business links. The upstream field identifier is used to indicate the original data field directly referenced by the current field.

[0034] The source correlation between any two fields is determined by the following formula. ,in, Representation field With fields The degree of source correlation between them This indicates whether the upstream field identifiers of the two fields are the same. This indicates the degree of overlap in the data entry paths of the two fields. This indicates whether there is a direct inheritance relationship between the two fields. , and These represent the weights of the corresponding source features; in this embodiment, The preferred value is 0.5. The preferred value is 0.3. Preferably 0.2; when Not lower than the source grouping threshold At that time, the two fields were assigned to the same evidence source group. The preferred value is 0.7.

[0035] The name of the work order entity is automatically imported into the electricity application form and the electricity supply contract. The three fields have the same upstream field identifier, so they are classified into the same evidence source group. The entity name in the business license comes from an independent scanned image and does not have the above-mentioned inheritance relationship, so it is classified into another evidence source group.

[0036] Image evidence nodes are constructed using the OCR candidate values, image evidence information, and evidence source group identifiers corresponding to each field region. The image evidence nodes are then associated with the corresponding business fact nodes according to the field semantics. The business fact nodes are used to represent the subject identity, electricity address, authorization period, applied capacity, and other business facts to be confirmed in the target power industry expansion order.

[0037] For business fact nodes candidate values Evidence source group The relevant support score provided is determined according to the following formula. ,in, Indicates the source of evidence group For candidate values Support score, Indicates source group Middle point refers to business fact nodes The set of image evidence nodes, The aforementioned first The first field in the field area Image evidence scoring of OCR candidate values, This represents the corresponding OCR candidate value. This indicates the semantic matching degree between OCR candidate values ​​and business fact candidate values. The semantic matching weight is preferably 0.4; the field evidence within the same source group is merged using the maximum support score, so that the same source copy field forms only one valid support.

[0038] The overall support for candidate values ​​of business facts from different evidence sources is determined by the following formula. ,in, Represents business fact nodes candidate values Overall support received This represents the set of source groups that provide evidence to this business fact node. Indicates the reliability of the source group; in this embodiment, the original scanned material corresponds to... The preferred option is 1, corresponding to the external authoritative interface. The preferred value is 0.9, corresponding to manually entered data. The preferred value is 0.7, corresponding to the system inherited data. The preferred value is 0.6.

[0039] Establish relationships between business fact nodes based on subject field correspondence, date sequence, document validity coverage, and material-business type association, and configure each relationship as a constraint for power industry revenue expansion. Thus, image evidence nodes provide candidate content supported by the original page, business fact nodes carry work order-level facts to be confirmed, and evidence source groups control evidence independence, providing a foundation for subsequently determining business fact combinations and identifying conflicts between multiple materials.

[0040] Please refer to the appendix. Figure 4 Based on the dereferential field evidence and the constraints of the power industry's revenue expansion, the combination of business facts and the corresponding field identification values ​​are determined. Furthermore, determine the combination of business facts and corresponding field identification values, including: Based on the evidence source group, read the irrelevant field evidence and generate candidate values ​​for each business fact node; The candidate values ​​corresponding to different business fact nodes are associated, and the combination of business facts that satisfy the constraint relationship is determined based on the power industry revenue expansion constraints. Based on the combination of business facts, the field recognition value supported by image evidence information is determined from the OCR candidate values ​​of the corresponding field region.

[0041] Specifically, this step is used to aggregate the derelevant field evidence formed by the aforementioned evidence source groups to the corresponding business fact node, and to determine the work order-level fact state that can jointly meet the power industry revenue expansion constraints within the scope of image evidence support, avoiding the direct use of the highest confidence OCR result of a single field.

[0042] In this embodiment, for the first For each business fact node, the OCR candidate values ​​and business data values ​​supported by each evidence source group are standardized and merged to form a candidate value set. The candidate combinations of all business fact nodes are represented as follows: ,in, Indicates the first Candidate values ​​for each business fact node, This indicates the number of business fact nodes.

[0043] The joint consensus score of each candidate combination is determined as follows: ,in, This represents the joint consensus score of the candidate combinations. This indicates the overall support level of the aforementioned evidence, excluding relevant fields, for the candidate values. Indicates the first The degree to which the capital constraints of the power industry expansion project are met. This indicates the value of the business fact involved in the constraint. Indicates constraint weights. Indicates the number of constraints; in this embodiment, the subject corresponds to the constraint. The preferred value is 1, the preferred timing constraint is 0.8, and the preferred material business association constraint is 0.7.

[0044] In one possible implementation, only candidate combinations that meet the mandatory requirements of document format, subject association, and validity period are retained, and then determined. ,in, This represents the set of candidate combinations that satisfy the mandatory constraints. This indicates a combination of business facts.

[0045] Based on the combination of business facts, from the first Determine the field identification value in each field area. ,in, Indicates the score of image evidence. This indicates the degree of matching between the OCR candidate value and the corresponding business fact value. The preferred value is 0.4. The selected field identification value must be supported by image evidence information; otherwise, it is marked as an unconfirmed field. When the set If the value is empty, proceed to the subsequent step of extracting the conflict constraint set.

[0046] Please see the appendix Figure 5 When there is no combination of business facts that satisfy the constraints of the power industry's revenue expansion, extract the conflict constraint set, determine the field region to be re-identified and the re-identification path based on the conflict constraint set, and update the OCR candidate value of the corresponding field. Furthermore, extract the set of conflicting constraints, including: Obtain the power industry revenue expansion constraints corresponding to the current business fact combination; The selection process cannot simultaneously satisfy the constraints of the current candidate values, resulting in a set of conflicting constraints. The corresponding field area is determined based on the business fact nodes and image evidence nodes associated with the conflict constraint set.

[0047] Furthermore, the region of fields to be re-identified and the re-identification path are determined based on the conflict constraint set, including: Determine the field range associated with the set of conflicting constraints; Based on the constraint satisfaction status after removing the candidate constraint relationship of the corresponding field region, determine the field region to be re-identified; A conflict fingerprint is generated based on the image state features, anomaly features, and association constraint features of the field region to be re-identified, and a re-identification path is determined based on the conflict fingerprint.

[0048] Furthermore, update the OCR candidate values ​​for the corresponding fields, including: Configure the image processing parameters, recognition range, and OCR model of the field region to be re-identified according to the target re-identification path, and perform local re-identification; The candidate values ​​generated by local re-identification are associated with the corresponding field regions to establish an image position correspondence. The candidate values ​​with the image position correspondence are then written into the OCR candidate values ​​of the corresponding fields to form updated OCR candidate values.

[0049] Specifically, this step is used to reverse trace the local constraints and field evidence that cause the unsatisfiable state when the aforementioned business fact candidate values ​​cannot form a compliant combination, and to select the corresponding image processing method using the conflict state, thereby expanding the OCR candidate range of the original field area and providing new image evidence for re-determining the business fact combination.

[0050] In this embodiment, all power industry revenue expansion constraints corresponding to the current candidate values ​​are obtained, and constraints that cannot be simultaneously satisfied are determined as the initial conflict set. Each item in the initial conflict set is removed and verified, and constraints that remain unsatisfiable after removal are deleted, until any constraint in the resulting constraint set can form a feasible candidate combination after removal, thus obtaining the conflict constraint set. Generally, a conflict constraint set is associated with at least two business fact nodes and can be traced back to the corresponding image evidence node, material page number, and field area through the business fact node.

[0051] The minimum violation cost of the current set of conflicting constraints is determined as follows: ,in, This represents the lowest conflict cost that cannot be eliminated within the current range of OCR candidate values. Indicates participation in the The value of the business fact constrained by a given condition is set to 1 if it satisfies the constraint, and 0 if it does not. Indicates the first The cost of violating the constraint; in this embodiment, the subject identity constraint corresponds to The preferred value is 1, the preferred timing constraint is 0.8, and the preferred material business association constraint is 0.7.

[0052] The first association of conflict constraint sets For each field region, temporarily remove the existing OCR candidate value restrictions and allow the corresponding business fact node to be in a pending confirmation state, and redetermine the minimum conflict cost. The conflict reduction amount for this field area is determined as follows: ,in, Indicates expansion of the first The potential contribution of each field region's OCR candidate value to eliminating the current conflict; Larger field regions are prioritized as fields to be re-identified; if the business license subject code is inconsistent with the application form subject code, and removing the candidate restriction of the corresponding character region of the business license can restore the subject constraint to a satiable state, then local re-identification of that character region is prioritized.

[0053] Generate conflict fingerprints based on the clarity, occlusion status, character segmentation status, candidate character confusion status, and association constraint type of the field region to be re-identified; The adaptation score of the duplicate identification path is determined as follows: ,in, Indicates the first The re-identification path for the first The adaptation score for each field range Indicates the image state fit. Indicates the degree of fit in identifying abnormalities. Indicates the fitness of the associated constraints. , and The corresponding weights are indicated; in this embodiment, the three weights are preferably 0.4, 0.35 and 0.25 respectively, and the path with the highest adaptation score is selected as the target re-identification path.

[0054] Seal occlusion corresponds to the seal color gamut separation path, number and letter confusion corresponds to the fine-grained character recognition path, field truncation corresponds to the expanded recognition range path, and table misalignment corresponds to the cell repositioning path. Image enhancement parameters, recognition range, and OCR model are configured according to the target re-recognition path, and only the field region to be re-recognized is processed.

[0055] The new candidate values ​​obtained from local re-identification must establish a correspondence with the character positions in the original image. The updated OCR candidate set is then determined as follows. ,in, This represents the updated set of OCR candidate values. This represents the original set of OCR candidate values. This represents the set of candidate values ​​generated by local re-identification and having corresponding image positions; the updated OCR candidate values ​​are returned to the aforementioned business fact combination determination step to form a closed-loop processing between constraint detection and local re-identification.

[0056] Based on the updated OCR candidate values, the business fact combination is redefined, and the data collection and management results are generated according to whether the conflict is eliminated, the consistency of image evidence information under different re-identification paths, and the independence between evidence source groups.

[0057] Furthermore, the results of standardized management of capital collection are generated, including: Record the conflict constraint set and image evidence information before and after local re-identification; When the updated OCR candidate values ​​can form a combination of business facts that meet the constraints of the power industry's revenue expansion, and the corresponding field recognition values ​​are supported by image evidence information, a recognition correction status is generated. When the same set of conflict constraints still exists after using different re-identification paths, and the conflict fields belong to different evidence source groups, a material conflict state is generated. Based on the identification and correction status and the material conflict status, corresponding data collection and management results are generated.

[0058] Specifically, this step is used to perform a work order-level joint verification on the new image evidence formed by local re-identification, and to determine the nature of the conflict based on the constraint recovery, the stability of the recognition result and the independence of the evidence source, thereby avoiding directly identifying correctable OCR misidentification as errors in the received materials.

[0059] In this embodiment, the updated OCR candidate values ​​are re-input into the aforementioned confidence constraint relationship to update the candidate values ​​and image evidence scores of the corresponding image evidence nodes, and to redetermine the feasible business fact combination set. When the set is not empty, the business fact combination with the highest joint consistency score is selected as the updated business fact combination, and the field recognition values, conflict constraint set, material page number and field coordinates before and after re-recognition are recorded.

[0060] Specifically, the The consistency of results for each field under different re-identification paths was determined as follows: ,in, Indicates the consistency of the re-identification results. This indicates the number of re-identification paths that have been executed. Indicates the first The field values ​​output by the weight recognition path. This represents the candidate field values ​​in the output values ​​of each path. This means that if the field values ​​are consistent, one is taken, and if they are inconsistent, zero is taken; in this embodiment, The preferred value is three, and the result consistency threshold is... The preferred value is 0.8.

[0061] When the updated combination of business facts satisfies all constraints on the expansion of revenue in the power industry, the changed field identification value is supported by image evidence information, and the confidence level of the corresponding business fact is not lower than the aforementioned confirmation confidence threshold. When the error occurs, a recognition correction status is generated. Under normal circumstances, this status indicates that the original conflict was caused by an error in the selection of OCR candidate values, and the original field value, the corrected field value, and the constraint relationship supporting the correction are saved.

[0062] If the business license main code obtains a field value consistent with the work order main code through both the seal separation path and the fine-grained character recognition path, thus eliminating the original main conflict constraint, then the field is marked as recognition correction status.

[0063] For the remaining conflicts, define the number of independent sources supporting the conflict. ,in, Indicates conflicting business fact nodes The corresponding number of independent evidence source groups, This represents a set of evidence source groups that support conflicting field values. In this embodiment, the threshold for the number of independent sources is preferably two.

[0064] When the consistency of field results under different re-identification paths is not lower than The same set of conflicting constraints persists, and If the condition is not less than two, a material conflict status is generated; if the conditions for generating an identification and correction status or a material conflict status are not met, a pending review mark is formed; finally, the field status, image evidence information, evidence source group and conflict constraint relationship are written into the target power industry expansion order to form the data collection and management results.

[0065] Example 2: In a second embodiment of the present invention, the present invention provides an OCR-based power industry revenue expansion and standardization management system, such as... Figure 2 As shown, it includes the following modules: Document Recognition Module: Used to acquire business data and receipt materials of target power industry expansion orders, perform page processing and field area positioning on the receipt materials, and generate OCR candidate values ​​and image evidence information corresponding to each field area; Source modeling module: It is used to divide the evidence source groups according to the generation source of each field, establish the confidence constraint relationship between image evidence nodes and business fact nodes based on OCR candidate values, image evidence information and business data, and perform decorrelation processing on field evidence within the same evidence source group; Fact Inference Module: Used to determine the combination of business facts and corresponding field identification values ​​based on the decorrelated field evidence and the power industry revenue expansion constraints; Conflict Recognition Module: When there is no combination of business facts that satisfy the power industry revenue expansion constraints, it extracts the conflict constraint set, determines the field region to be re-identified and the re-identification path based on the conflict constraint set, and updates the OCR candidate value of the corresponding field. The results management module is used to redetermine the combination of business facts based on the updated OCR candidate values, and generate data collection and management results according to whether the conflict is eliminated, the consistency of image evidence information under different re-identification paths, and the independence between evidence source groups.

[0066] In the power industry expansion application process, staff need to upload business licenses, identity documents, authorization letters, property ownership certificates, and electricity application materials to the marketing system. Due to the diverse sources of these materials—some scanned from original documents, others automatically generated from work order fields—this process is prone to issues such as page tilting, seal obstruction, character confusion, and duplicate support of the same source fields. When a single field experiences OCR recognition errors, it can lead to apparent conflicts in the subject, address, or time relationships between different materials. Conversely, uploading incorrect materials can cause conflicts in the actual content. Existing systems struggle to accurately distinguish these situations, easily resulting in erroneous interception or increased manual verification. To address these problems, this invention employs an OCR-based power industry expansion application management system, the architecture of which is as follows: Figure 2 As shown. The specific implementation process of this system is as follows: First, the data recognition module acquires the target power industry expansion order and its receipt materials, performs page correction, layout analysis and field area positioning on each material, and generates OCR candidate values ​​and image evidence information with the original page position correspondence. Subsequently, the source modeling module reads the material generation method, data entry path and upstream field identifier of the field, divides the fields with the same source into the same evidence source group, and merges the field evidence pointing to the same business fact within the same evidence source group, and establishes the relationship between image evidence nodes, business fact nodes and power industry revenue expansion constraints. Next, the fact inference module generates candidate values ​​for each business fact node based on the decorrelated field evidence, determines the combination of business facts that meet the constraints through subject correspondence, temporal correlation and material business correlation, and determines the field identification value supported by image evidence information from the corresponding field area; Then, when there is no combination of business facts that satisfy the constraints of the power industry's revenue expansion, the conflict re-identification module extracts the set of conflict constraints, traces the associated field regions, determines the field regions to be re-identified based on the constraint status after the candidate constraints are lifted, and selects the local image processing method and OCR model based on the conflict fingerprint to expand the OCR candidate values ​​of the corresponding fields. Finally, the results management module redetermines the business fact combination based on the updated OCR candidate values. When the conflict is eliminated by the new candidate values ​​supported by image evidence information, a recognition correction status is generated. When the conflict persists under different re-recognition paths and the sources of conflict evidence are independent of each other, a material conflict status is generated, and the field status, material location and constraint relationship are written into the data collection and management results.

[0067] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for standardized management of revenue expansion in the power industry based on OCR, characterized in that, Includes the following steps: Obtain business data and collection materials of the target power industry expansion order, perform page processing and field area positioning on the collection materials, and generate OCR candidate values ​​and image evidence information corresponding to each field area; Evidence source groups are divided according to the generation source of each field. Based on the OCR candidate value, the image evidence information and the business data, a confidence constraint relationship is established between the image evidence node and the business fact node. Field evidence within the same evidence source group is decorrelated. Based on the decorrelated field evidence and the constraints of the power industry's revenue expansion, the combination of business facts and the corresponding field identification values ​​are determined. When there is no combination of business facts that satisfy the constraints of the power industry revenue expansion, extract the conflict constraint set, determine the field region to be re-identified and the re-identification path based on the conflict constraint set, and update the OCR candidate value of the corresponding field. Based on the updated OCR candidate values, the business fact combination is re-determined, and the data collection and management results are generated according to whether the conflict is eliminated, the consistency of the image evidence information under different re-identification paths, and the independence between evidence source groups.

2. The method for standardized management of power industry revenue expansion based on OCR according to claim 1, characterized in that: The page processing and field area positioning of the received materials include: The received materials are analyzed and page images are generated page by page. Orientation correction, distortion correction, interference area separation and page partitioning are performed to form a processed page. The actual material type is determined based on the declared material type of the received materials and the layout characteristics of the processing page; The field area is located by calling the field template associated with the actual material type, and the coordinate mapping relationship between the field area and the original page is recorded.

3. The method for standardized management of power industry revenue expansion funds based on OCR according to claim 2, characterized in that: The generation of OCR candidate values ​​and image evidence information corresponding to each field region includes: Perform OCR recognition on each field area and retain the character sequence and corresponding recognition confidence information; The character sequence is formatted and normalized based on the field type to form OCR candidate values; Establish a correspondence between the characters in the OCR candidate values ​​and the glyph positions in the field area to form glyph alignment information; Image support candidate values ​​are filtered based on the identification confidence information, the glyph alignment information, and the field region quality information, and the image support candidate values ​​and their coordinate information are used as image evidence information.

4. The method for standardized management of power industry revenue expansion funds based on OCR according to claim 1, characterized in that: The process of grouping evidence sources based on the generation source of each field includes: Read the material generation method, data entry path and upstream field identifier of each field, and classify fields with the same upstream field identifier or direct inheritance relationship into the same evidence source group; Fields with different upstream field identifiers and no inheritance relationship between generation paths are divided into different evidence source groups; Field evidence pointing to the same business fact node within the same evidence source group is merged to form the de-correlated field evidence corresponding to each evidence source group.

5. The method for standardized management of power industry revenue expansion funds based on OCR according to claim 4, characterized in that: The establishment of the confidence constraint relationship between image evidence nodes and business fact nodes includes: Image evidence nodes are constructed using the OCR candidate values ​​corresponding to the field regions, image evidence information, and evidence source group identifiers. The fields in the business data are associated with the OCR candidate values ​​according to the field semantics to form business fact nodes; Associate the image evidence node with the corresponding business fact node; Based on the evidence source group identifier, the source association of related field evidence is determined, and field evidence belonging to the same evidence source group is merged to form irrelevant field evidence. Establish field correspondence, temporal correlation, and material business correlation among the business fact nodes to form power industry revenue expansion constraints.

6. The method for standardized management of power industry revenue expansion funds based on OCR according to claim 5, characterized in that: The determination of the combination of business facts and corresponding field identification values ​​includes: Based on the evidence source group, read the irrelevant field evidence and generate candidate values ​​for each business fact node; The candidate values ​​corresponding to different business fact nodes are associated, and the combination of business facts that satisfy the constraint relationship is determined based on the power industry revenue expansion constraints. Based on the combination of business facts, field recognition values ​​supported by image evidence information are determined from the OCR candidate values ​​of the corresponding field region.

7. A method for standardized management of power industry revenue expansion funds based on OCR as described in claim 6, characterized in that: The extraction of the conflict constraint set includes: Obtain the power industry revenue expansion constraints corresponding to the current business fact combination; The selection process cannot simultaneously satisfy the constraints of the current candidate values, resulting in a set of conflicting constraints. The corresponding field area is determined based on the business fact nodes and image evidence nodes associated with the conflict constraint set.

8. A method for standardized management of power industry revenue expansion funds based on OCR as described in claim 7, characterized in that: The step of determining the field region to be re-identified and the re-identification path based on the conflict constraint set includes: Determine the field regions associated with the set of conflicting constraints; Based on the constraint satisfaction status after removing the candidate constraint relationship of the corresponding field region, determine the field region to be re-identified; A conflict fingerprint is generated based on the image state features, identification anomaly features, and association constraint features of the field region to be re-identified, and a re-identification path is determined based on the conflict fingerprint.

9. A method for standardized management of power industry revenue expansion funds based on OCR as described in claim 8, characterized in that: The updating of the corresponding field's OCR candidate value includes: Configure the image processing parameters, recognition range, and OCR model of the field region to be re-identified according to the target re-identification path, and perform local re-identification; The candidate values ​​generated by local re-identification are used to establish an image position correspondence with the corresponding field region, and the candidate values ​​with the image position correspondence are written into the OCR candidate values ​​of the corresponding field to form the updated OCR candidate values.

10. A method for standardized management of power industry revenue expansion funds based on OCR as described in claim 9, characterized in that: The generated data collection and management results include: Record the conflict constraint set and image evidence information before and after local re-identification; When the updated OCR candidate values ​​can form a combination of business facts that satisfy the constraints of the power industry revenue expansion, and the corresponding field recognition values ​​are supported by image evidence information, a recognition correction status is generated. When the same set of conflict constraints still exists after using different re-identification paths, and the conflict fields belong to different evidence source groups, a material conflict state is generated. Based on the identified correction status and material conflict status, corresponding data collection and management results are generated.