Charging pile sampling inspection knowledge graph updating method and device, electronic equipment and storage medium

Abstract

The invention discloses a charging pile sampling inspection knowledge graph updating method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring multi-modal knowledge data related to charging pile sampling inspection from a plurality of data sources; wherein the multi-modal knowledge data comprises different levels of supervision announcement texts, enterprise sampling inspection report texts, field sampling inspection pictures and enterprise qualification information of sampling inspection units; mining candidate structured knowledge at least including key entities and mutual relations thereof from the multi-modal knowledge data; in response to target structured knowledge conflicting with the candidate structured knowledge in the charging pile sampling inspection knowledge graph, assigning credibility weights to the candidate structured knowledge and the target structured knowledge by using a preset credibility weight matrix; and in response to the fact that the credibility weight of the candidate structured knowledge is greater than the credibility weight of the target structured knowledge, changing the target structured knowledge with conflicts based on the candidate structured knowledge so as to update the charging pile sampling inspection knowledge graph.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of knowledge graph, and in particular to a charging pile sampling knowledge graph updating method and device, an electronic device and a storage medium. BACKGROUND

[0002] With the rapid development of the new energy vehicle industry, the product quality and operation safety of charging piles as key supporting infrastructure are crucial. The market supervision department protects the order of the charging pile market through continuous product quality supervision and sampling. In the industrial field, there have been attempts to build product quality knowledge graphs, but most focus on the integration and visualization of static knowledge, and their updating mechanism is cumbersome.

[0003] Therefore, there is an urgent need for a charging pile sampling knowledge graph and a charging pile sampling knowledge graph updating method. SUMMARY

[0004] The present application provides a charging pile sampling knowledge graph updating method and device, an electronic device, a storage medium and a computer program product.

[0005] According to an aspect of the present application, a charging pile sampling knowledge graph updating method is provided, comprising:

[0006] Obtaining multi-modal knowledge data related to charging pile sampling from multiple data sources; wherein the multi-modal knowledge data includes at least two of different levels of supervision announcement texts, enterprise sampling report texts, on-site sampling pictures or videos, and enterprise qualification information of the sampled unit related to charging pile sampling;

[0007] Mining candidate structured knowledge including at least key entities and their mutual relationships from the multi-modal knowledge data;

[0008] In response to the existence of target structured knowledge conflicting with the candidate structured knowledge in the charging pile sampling knowledge graph, a preset credibility weight matrix is used to give the candidate structured knowledge and the target structured knowledge credibility weights respectively;

[0009] In response to the credibility weight of the candidate structured knowledge being greater than the credibility weight of the target structured knowledge, the target structured knowledge with conflicts is changed based on the candidate structured knowledge to update the charging pile sampling knowledge graph.

[0010] According to another aspect of the present application, a charging pile sampling knowledge graph updating device is provided, comprising:

[0011] The data perception acquisition module is configured to acquire multi-modal knowledge data related to the charging pile sampling from a plurality of data sources; wherein the multi-modal knowledge data comprises at least two of different levels of supervision announcement texts, enterprise sampling report texts, on-site sampling pictures or videos, and enterprise qualification information of the sampled units related to the charging pile sampling;

[0012] The knowledge extraction module is configured to mine candidate structured knowledge including at least key entities and their mutual relationships from the multi-modal knowledge data;

[0013] The credibility confirmation module is configured to, in response to the existence of target structured knowledge conflicting with the candidate structured knowledge in the charging pile sampling knowledge graph, utilize a preset credibility weight matrix to assign credibility weights to the candidate structured knowledge and the target structured knowledge, respectively.

[0014] The first updating module is configured to, in response to the credibility weight of the candidate structured knowledge being greater than the credibility weight of the target structured knowledge, change the target structured knowledge in conflict based on the candidate structured knowledge, so as to update the charging pile sampling knowledge graph.

[0015] According to another aspect of the present application, an electronic device is provided, which comprises:

[0016] at least one processor; and

[0017] a memory connected in communication with the at least one processor; wherein

[0018] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to perform the charging pile sampling knowledge graph updating method of the embodiments of the present application.

[0019] According to another aspect of the present application, a computer readable storage medium is provided, which stores computer instructions for enabling a processor to perform the charging pile sampling knowledge graph updating method of the embodiments of the present application when executed by the processor.

[0020] According to another aspect of the present application, a computer program product is provided, which comprises a computer program for implementing the steps of the above method when executed by a processor.

[0021] The technical scheme of the embodiment of the present application forms a full-automatic technical closed loop of multi-source data acquisition, knowledge mining, conflict resolution and graph updating, and fundamentally overcomes inherent defects of the prior art, such as low automation degree, lagging update, no guarantee of knowledge consistency and weak processing capacity of unstructured data. The combination of multi-source data integration and domain adaptive knowledge mining ensures the comprehensiveness and accuracy of knowledge, solves the problem of insufficient professional term recognition accuracy of the prior art with single data source, and provides more abundant decision basis for supervision. The conflict resolution mechanism based on business rules and the incremental update mode improve the timeliness and reliability of the knowledge graph, avoid resource waste, and solve the problems of outdated graph content, chaotic logic and weak decision support capability in the prior art.

[0022] It should be understood that the content described in this part is not intended to identify key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0024] Figure 1 is a flowchart of the charging pile sampling knowledge graph updating method provided by the embodiment of the present application;

[0025] Figure 2 is a flowchart of another charging pile sampling knowledge graph updating method provided by the embodiment of the present application;

[0026] Figure 3 is a structural diagram of a charging pile sampling knowledge graph updating device provided by the embodiment of the present application;

[0027] Figure 4 is a structural diagram of an electronic device for implementing the charging pile sampling knowledge graph updating method of the embodiment of the present application. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the present application scheme, the technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the drawings of the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0029] Embodiment one

[0030] Figure 1 A flowchart of a charging pile sampling knowledge graph updating method provided by an embodiment of the present application, the embodiment can be applicable to the scene of constructing and updating a charging pile sampling knowledge graph, the method can be executed by a charging pile sampling knowledge graph updating device, the charging pile sampling knowledge graph updating device can be realized in the form of hardware and / or software, and the charging pile sampling knowledge graph updating device can be configured in an electronic device.

[0031] As shown in Figure 1 , the charging pile sampling knowledge graph updating method comprises:

[0032] S101, acquiring multi-modal knowledge data related to charging pile sampling from multiple data sources; wherein the multi-modal knowledge data comprises at least two of different levels of supervision announcement texts, enterprise sampling report texts, on-site sampling pictures or videos, and enterprise qualification information of the sampled units related to charging pile sampling.

[0033] Among them, the multiple data sources refer to different channels that provide information related to charging pile sampling, including official websites of market supervision departments at all levels, enterprise internal network sharing directories, national enterprise credit information public system, CQC certification center and other authoritative agency platforms. The multi-modal knowledge data refers to charging pile sampling related data with multiple forms and different formats, which mainly includes text type (supervision announcement text, enterprise sampling report text), visual type (on-site sampling picture or on-site sampling video), and structured type (enterprise qualification information). The supervision announcement texts of different levels refer to the text type announcements (such as HTML pages or PDF documents) published by market supervision departments at different levels such as the state and the province, which are used to publicize the results of charging pile sampling. The enterprise sampling report text refers to the PDF format report text generated by the enterprise, which records the charging pile sampling process, detection data, and result determination. The on-site sampling picture refers to the JPG / PNG format static image taken on the spot of charging pile sampling, which contains key information such as the overall appearance of the charging pile, the nameplate, and the defect position. The enterprise qualification information of the sampled unit refers to the structured data certified by authoritative agencies, which is used to prove that the sampled enterprise is legally operating and the product is compliant, including enterprise registration information, product certification certificate information, and compliance test results.

[0034] In some embodiments, automated data acquisition technologies (including web data crawling, local file monitoring, and API data retrieval) can be used to integrate charging pile inspection data from different channels and in different formats. This enables unified acquisition and preliminary structured processing of multi-source heterogeneous data, providing comprehensive and original data support for subsequent knowledge mining. Specifically, a web crawler robot is deployed to periodically monitor the inspection announcement sections on the official websites of market supervision departments at all levels, crawling regulatory announcement texts (HTML pages or PDF documents) of different levels, and extracting core entity information from the text using the PyMuPDF tool. A configuration file system monitoring service monitors the enterprise's internal network shared directory, automatically triggering the data processing flow when new enterprise inspection report texts (PDF format), on-site inspection images (JPG / PNG format), or videos are stored. Through API interfaces, the enterprise qualification information of the inspected units is periodically retrieved from authoritative data sources such as the National Enterprise Credit Information Publicity System and the CQC Certification Center, directly converting it into structured fields as deep attribute layer data connecting regulatory data and internal enterprise data.

[0035] Understandably, this step addresses the shortcomings of relying on data from a single source and having limited data coverage, ensuring data comprehensiveness through the integration of multiple data sources. The use of automated collection and triggering mechanisms avoids the inefficiencies caused by manual intervention in existing technologies, improving the timeliness of data acquisition.

[0036] S102. Mine candidate structured knowledge from multimodal knowledge data, including at least key entities and their interrelationships.

[0037] Key entities refer to the core information carriers in the charging pile spot check scenario, including information units strongly related to the spot check results and product attributes, such as manufacturer, product model, non-conforming items, testing standards, rated power, and production batch. Interrelationships refer to the semantic associations between key entities, such as associations like "manufacturer-production-product model," "product model-non-conforming items-electrical clearance," and "product model-rated power-60kW." Candidate structured knowledge refers to structured knowledge that has been organized according to a fixed logical format (such as triples "entity-relationship-entity" or "entity-attribute-value") but has not yet been formally integrated into the charging pile spot check knowledge graph. It needs to undergo conflict detection and resolution before being included in the database. Knowledge mining refers to the process of extracting key entities and identifying relationships between entities from multimodal knowledge data using technologies such as natural language processing, computer vision, and cross-modal fusion, transforming unstructured / semi-structured data into structured knowledge.

[0038] In some embodiments, domain-adaptive intelligent information extraction technology (integrating natural language processing, computer vision, cross-modal fusion, etc.) can be used to accurately mine key entities and semantic relationships between entities from multimodal knowledge data, transforming unstructured / semi-structured data into standardized candidate structured knowledge, thereby realizing the transformation of data into knowledge.

[0039] In practice, candidate structured knowledge, including at least key entities and their relationships, is mined from multimodal knowledge data, including S1021-S1023:

[0040] S1021. For regulatory announcement texts or enterprise inspection report texts of different levels in multimodal knowledge data, extract text context semantic features from regulatory announcement texts or enterprise inspection report texts of different levels by using an enhanced semantic representation model; and perform named entity recognition and relation extraction on the text context semantic features by using an entity recognition model and a relation extraction model to obtain candidate structured knowledge.

[0041] Among them, the enhanced semantic representation model refers to a deep learning model that is pre-trained on charging pile sampling inspection texts (such as domain standards and historical announcements) and can accurately capture the semantic context of professional texts. Specifically, it is the ERNIE model (enhanced semantic representation model), which can effectively identify professional terms in fields such as creepage distance and load terminals. Text context semantic features refer to the feature information extracted from regulatory announcement texts and enterprise sampling inspection reports, reflecting word associations, sentence logic, and the context of professional terms in the text. It is the foundation for subsequent entity recognition and relation extraction. The entity recognition model refers to the model used to locate and identify key entities (such as manufacturers, product models, and non-conforming items) in the text. Specifically, it can be composed of BiLSTM-CRF layers (bidirectional long short-term memory network + conditional random field). By combining it with the ERNIE model, the accuracy of domain entity recognition is improved. The relation extraction model refers to the model used to directly identify the semantic relationships between key entities in the text. Specifically, it can be the TPLinker joint extraction model, which can avoid the error accumulation of traditional pipeline models (identifying entities first and then matching relations) and directly output "entity-relationship-entity" triples.

[0042] In practical implementation, text preprocessing can be performed first. This mainly involves formatting regulatory announcements and enterprise inspection reports from multimodal knowledge data. If the text is native PDF, PyMuPDF or pdfplumber libraries are used to parse and extract the plain text. If it is a scanned PDF, the PaddleOCR engine is first called for optical character recognition, and the position information of the text on the page (such as the paragraph position of "non-compliant items" in the report) is preserved to ensure text integrity. Further, semantic feature extraction is performed. This mainly involves inputting the preprocessed plain text into an ERNIE model optimized for the charging pile inspection domain. Through the model's multi-layer Transformer encoder, contextual semantic features such as "the association between the manufacturer and the product model" and "the correspondence between non-compliant items and the testing standards" are captured, providing accurate semantic support for entity recognition and relation extraction. Further entity recognition is performed, primarily by fine-tuning the ERNIE model's feature output layer with a BiLSTM-CRF layer. The BiLSTM network handles text sequence dependencies, and the CRF layer optimizes entity boundary prediction, accurately identifying key entities such as "Company A (manufacturer)," "Model A charging pile (product model)," "Electrical clearance (non-conforming item)," and "GB / T18487.1 (testing standard)." Finally, relation extraction is performed, mainly using the TPLinker joint extraction model. This directly extracts triples in the form of "entity-relationship-entity" from sentences containing key entities. For example, from "Model A charging piles manufactured by Company A have been found to have non-conforming electrical clearances," the remaining structured knowledge "(Company A, manufacturing, Model A charging pile)" and "(Model A charging pile, non-conforming item, electrical clearance)" are extracted.

[0043] For example, a provincial market supervision department issued a PDF-format charging pile spot check notice, which stated that "the B-type DC charging pile produced by Company B failed to meet the creepage distance requirements of GB / T20234.3-2015 standard during the spot check in November 2024."

[0044] Preprocessing stage: Parse the PDF announcement using PyMuPDF to extract the plain text content;

[0045] Semantic feature extraction: The ERNIE model, pre-trained in the text input domain, captures the semantic associations of “Company B - Model B DC charging pile - creepage distance - GB / T20234.3-2015”.

[0046] Entity recognition: The BiLSTM-CRF layer identifies "Company B (manufacturer)", "Model B DC charging pile (product model)", "creep distance (non-conforming item)" and "GB / T20234.3-2015 (testing standard)";

[0047] Relation extraction: The TPLinker model extracts the triples "(Company B, manufacturer, model B DC charging pile)", "(model B DC charging pile, non-conforming item, creepage distance)", and "(model B DC charging pile, violates standard, GB / T20234.3-2015)" to form candidate structured knowledge.

[0048] Understandably, this step combines a domain-pretrained ERNIE model with a BiLSTM-CRF layer to improve the accuracy of terminology recognition. The TPLinker joint extraction model directly extracts triples from the text, eliminating the need to first identify entities and then match relationships. This reduces the problem of entity recognition errors propagating to relationship extraction, thus improving the accuracy of relationship extraction. Whether it's national or provincial regulatory announcements or internal enterprise inspection reports, knowledge can be extracted through a unified model process, solving the problem of weak generalization ability in existing technologies that require separate rule design for different text formats.

[0049] S1022. For the obtained on-site inspection images, the on-site inspection images are identified by the target detection model to obtain the key component area of ​​the charging pile; the key component area of ​​the charging pile is subjected to text recognition to obtain the text recognition result of the key component area of ​​the charging pile; based on the text recognition result of the key component area of ​​the charging pile, the candidate structured knowledge is subjected to consistency verification, attribute authenticity verification and abnormal rationality verification.

[0050] The target detection model refers to a computer vision model used to locate specific target regions in an image, specifically the YOLOvX model. This model is pre-trained on the COCO dataset (a general image dataset) and a self-built charging pile image dataset (containing labeled data of the charging pile as a whole, nameplate, and defective parts), and can accurately locate key component regions of the charging pile. Key component regions of the charging pile refer to areas containing core information in the on-site inspection images, including the overall appearance area of ​​the charging pile, the nameplate area (recording product model, rated power, etc.), and defective parts (such as areas with damaged insulation components). These are the key targets for subsequent text recognition. Text recognition refers to the technology of extracting text information from specific regions of an image, and can identify text content from key component regions of the charging pile (especially the nameplate area). Consistency verification / attribute authenticity verification / anomaly rationality verification refer to three core logics used to verify the candidate structured knowledge extracted in step S1021 using the image recognition results. Consistency verification verifies the matching degree of entity information, attribute authenticity verification verifies the accuracy of attribute values, and anomaly rationality verification verifies the logical relevance of knowledge.

[0051] In practice, the first step is to locate the key component areas. This involves inputting on-site inspection images (JPG / PNG format) into a pre-trained YOLOvX object detection model. The model extracts features and predicts targets, outputting coordinate bounding boxes for key component areas such as the overall charging pile, nameplate, and defective parts, achieving precise location of these key areas. Next, text recognition is performed on the key areas. This involves cropping corresponding image segments from the key component areas located by YOLOvX (focusing on the nameplate area) and inputting them into the PaddleOCR engine. The engine then extracts text recognition results from the nameplate, including "product model (e.g., Model B DC charging pile)," "rated power (e.g., 60kW)," and "production batch (e.g., 202410)," through a two-step process of text detection and text recognition. Finally, three types of verification are performed: Consistency verification: The text recognition result is compared with the core entities (such as product model and manufacturer) in the candidate structured knowledge extracted by S1021. If the extracted "Model B" does not match the image recognition "Model B-2024", manual verification is prompted. After manual review and confirmation that there is an error in the model abbreviation in the announcement, the nameplate information is finally confirmed to be accurate, and the candidate structured knowledge is updated. Attribute authenticity verification: The attribute values ​​of the text recognition result (such as "Certification Expiration Date 202409" on the nameplate) are compared with the attributes in the candidate structured knowledge (such as "Certification Validity Period is 2025-09" extracted by the text). If there is a contradiction, further determination is made by combining authoritative certification data pulled by the API. Anomaly Reasonableness Verification: If the candidate structured knowledge extracts "(Model B charging pile, unqualified item, insulation resistance)", but YOLOvX does not detect any damage to the insulation components of the charging pile in the image, and the nameplate parameters recognized by PaddleOCR are consistent with the "standard value of insulation resistance" extracted from the text, anomaly reasonableness verification is triggered, prompting a check to see if there are any errors in the sampling process.

[0052] It should be noted that if the multimodal knowledge data includes on-site inspection videos, a preset number of on-site inspection images are extracted from the video using frame extraction technology, and then the process is carried out according to step S1022 above.

[0053] Understandably, combining YOLOvX with PaddleOCR extracts directly recorded information about physical entities (such as nameplates) from random on-site inspection images, providing visual verification for candidate structured knowledge and reducing errors from pure text extraction. Three verification mechanisms cover the core dimensions of knowledge reliability: consistency verification prevents entity information misalignment, attribute authenticity verification prevents incorrect attribute values, and anomaly rationality verification prevents logical contradictions. Manual intervention is only prompted when an anomaly is triggered during verification; otherwise, verification and correction are completed automatically, avoiding the inefficiency of manually checking text and images one by one.

[0054] S1023. Based on the structured enterprise qualification information of the sampled units, supplement the entity attributes of the candidate structured knowledge that has passed the verification to obtain the final candidate structured knowledge.

[0055] The structured enterprise qualification information refers to enterprise-related data retrieved from authoritative data sources such as the National Enterprise Credit Information Publicity System and CQC Certification Center via API interfaces. This data is organized in a fixed format and includes information such as the enterprise's registered address, registered capital, product certification certificate number, compliance testing report number, and production license validity period. Entity attribute supplementation involves associating key fields from the structured enterprise qualification information with corresponding entities in the candidate structured knowledge verified in S1022, thus completing the entity's attribute set. The final candidate structured knowledge, after extraction in S1021, verification in S1022, and attribute supplementation in S1023, possesses complete entity information, accurate relationships, and reliable attributes. It serves as the core input for subsequent conflict detection and graph updates.

[0056] In practical implementation, core entities such as manufacturing enterprises (e.g., Company B) and product models (e.g., model B-2024 charging pile) are extracted from the candidate structured knowledge that has passed verification in S1022. Using the company name, product model, and manufacturing enterprise as unique identifiers, the system matches the structured enterprise qualification information retrieved from the API (e.g., matching Company B's registered address, registered capital, and product certification certificate number). Attributes are added to the manufacturing enterprise entity, primarily by adding the matched registered address, registered capital, production license validity period, and compliance testing report number as new attributes to Company B, thus improving the entity's attribute set. Attributes are also added to the product entity by supplementing the product certification certificate number and certification standards corresponding to model B-2024 charging pile from the enterprise qualification information into the product entity's attributes. The attribute formats are standardized, storing the added attribute values ​​in a unified format (e.g., date format YYYY-MM-DD, certificate number retaining complete characters) to ensure consistency with the format of the candidate structured knowledge. Finally, the entities with the added attributes and the original relationships are integrated to form a complete structured knowledge that includes entity-attribute-value and entity-relationship-entity.

[0057] Understandably, by supplementing the information with authoritative enterprise qualification information, the entities (enterprises, products) in the candidate structured knowledge not only include relevant information from random inspections, but also cover deeper attributes such as business qualifications and certification information, providing a more comprehensive basis for subsequent regulatory decisions (such as determining whether an enterprise has production qualifications).

[0058] S103. In response to the existence of target structured knowledge in the knowledge graph of charging pile spot checks that conflicts with candidate structured knowledge, a preset credibility weight matrix is ​​used to assign credibility weights to candidate structured knowledge and target structured knowledge respectively.

[0059] Among them, target structured knowledge refers to existing structured knowledge stored in the charging pile sampling knowledge graph that shares the same core association with newly mined candidate structured knowledge. Knowledge conflict refers to the contradiction between candidate structured knowledge and target structured knowledge regarding the same core association, which cannot be simultaneously valid. Examples include the same product entity (uniquely identified by product model and enterprise information) simultaneously existing in both qualified and unqualified states, or inconsistent rated power values ​​for the same product model. The preset credibility weight matrix is ​​the core technical carrier used to quantitatively assess the credibility of knowledge. The elements in the matrix are comprehensive credibility quantification values ​​(ranging from 0 to 1, with higher values ​​indicating stronger credibility) corresponding to specific attribute combinations (data source type + sampling domain + report standardization). It is constructed by integrating data source type scores, sampling domain attribute bonus scores, and report standardization bonus scores. Credibility weight refers to the quantitative score representing the reliability of knowledge, calculated based on the preset credibility weight matrix and the attribute combination corresponding to the knowledge. It serves as the core criterion for conflict resolution.

[0060] In practice, the candidate structured knowledge mined in step S102 is compared with the existing knowledge in the knowledge graph of charging pile sampling inspection. Based on the preset conflict rules (if the same product entity is uniquely identified by product model and enterprise information, and there are both qualified and unqualified states at the same time), it is determined whether there is conflicting target structured knowledge.

[0061] If a conflict exists, then follow these steps S1031-S1033 to assign credibility weights to the candidate structured knowledge and the target structured knowledge, specifically:

[0062] S1031. Analyze the candidate structured knowledge and the target structured knowledge separately to determine the data source type, sampling area, and reporting standardization of the enterprise self-inspection report corresponding to each candidate structured knowledge and the target structured knowledge.

[0063] The data source type of both candidate and target structured knowledge is extracted and determined from their metadata (such as data source identifiers and acquisition channel records). The priority of data source types is: national-level regulatory announcements > provincial-level regulatory announcements > compliant enterprise self-inspection reports > non-compliant enterprise self-inspection reports. For example, if candidate structured knowledge comes from a sampling inspection announcement issued by the State Administration for Market Regulation, its data source type is determined to be a national-level regulatory announcement; if target structured knowledge comes from a self-inspection report submitted by an enterprise without complete testing evidence, its data source type is determined to be a non-compliant enterprise self-inspection report. If the data source type is not directly indicated in the knowledge metadata, it is determined by tracing the data acquisition link in reverse: knowledge obtained from a web crawler to the official website of the State Administration for Market Regulation is determined to be a national-level regulatory announcement; knowledge obtained from internal enterprise document monitoring links without an authoritative institution's seal is determined to be an enterprise self-inspection report. Further verification is then conducted to check whether the report contains complete testing evidence, original records, and approval signatures to distinguish between compliant and non-compliant information.

[0064] Based on the sampling scenario information recorded in the knowledge (such as sampling location and sampling process description), the sampling domains of candidate and target structured knowledge are determined. Sampling domains are divided into three categories: production domain (sampling inspection of enterprise production lines), physical retail domain (sampling inspection of products sold in offline physical stores), and e-commerce domain (sampling inspection of products sold on e-commerce platforms). For example, if the knowledge includes the sampling location: Company B's production workshop, and the sampling process: finished product coming off the production line, it is determined to be in the production domain. If the knowledge does not have a direct scenario description, the determination is aided by the associated enterprise qualifications or sampling report number.

[0065] This judgment is applied only to knowledge whose data source type is a company self-inspection report. It involves verifying whether the report text contains three core elements: complete testing basis (e.g., citing GB / T18487.1-2015 standard), original testing records (e.g., specific testing data, signatures of testing personnel), and approval signature (e.g., signature of the company's quality manager). If all three elements are present, it is judged as a compliant company self-inspection report; if any element is missing, it is judged as a non-compliant company self-inspection report. For knowledge that is not a company self-inspection report (e.g., regulatory announcements), this attribute is marked as none, and no additional points for report compliance are involved in subsequent weighting calculations.

[0066] S1032. Based on the data source type, sampling field, and reporting standardization of the enterprise self-inspection report corresponding to the candidate structured knowledge and the target structured knowledge, respectively, determine the target elements that match the candidate structured knowledge and the target structured knowledge from the preset credibility weight matrix.

[0067] In this embodiment of the invention, the preset credibility weight matrix is ​​a three-dimensional matrix. The row dimension corresponds to the data source type (national regulatory announcement, provincial regulatory announcement, self-inspection report of compliant enterprises, self-inspection report of non-compliant enterprises), the column dimension corresponds to the sampling field (production field, distribution physical store field, distribution e-commerce field), and the third dimension corresponds to the standardization of the enterprise self-inspection report (compliant, non-compliant; non-self-inspection reports have no standardization by default). Each element of the matrix is ​​associated with a set of data source type score + sampling field attribute bonus score + report standardization bonus score. For example:

[0068] The element “National-level regulatory announcement + production sector + none” has the following associated scores: data source type score 0.9, sampling sector attribute bonus score 0.2, and report standardization bonus score 0; the element “Standardized enterprise self-inspection report + circulation e-commerce sector + standardized” has the following associated scores: data source type score 0.3, sampling sector attribute bonus score 0, and report standardization bonus score 0.1; the element “Non-standardized enterprise self-inspection report + circulation physical store sector + non-standardized” has the following associated scores: data source type score 0.1, sampling sector attribute bonus score 0.1, and report standardization bonus score 0.

[0069] Based on the above, for the candidate structured knowledge attribute combinations determined in S1031, a unique matching target element is located in the preset matrix. For example, if the candidate structured knowledge attribute is "Provincial Regulatory Announcement + Distribution Physical Store Sector + None", then the element corresponding to the row "Provincial Regulatory Announcement", column "Distribution Physical Store Sector", and third dimension "None" in the matrix is ​​found as the target element of this knowledge. The same operation is performed on the target structured knowledge: if the target structured knowledge attribute is "Non-compliant Enterprise Self-Inspection Report + Distribution E-commerce Sector + Non-compliant", then the element corresponding to "Non-compliant Enterprise Self-Inspection Report + Distribution E-commerce Sector + Non-compliant" in the matrix is ​​matched.

[0070] S1033. Based on the data source type score associated with each matched target element, the additional score of the sampling domain attribute, and the additional score of the report standardization, determine the credibility weights of the candidate structured knowledge and the target structured knowledge respectively.

[0071] From the target elements matched by S1032, extract the data source type score (core basic score, with the highest weight), the sampling field attribute bonus score (scenario-differentiated bonus score), and the report standardization bonus score (only applicable to enterprise self-inspection reports, this item is 0 for non-self-inspection reports). The three are calculated by direct summation, that is: credibility weight = data source type score + sampling field attribute bonus score + report standardization bonus score.

[0072] Calculation of credibility weights for candidate structured knowledge:

[0073] Example: If the target element of the candidate structured knowledge matching is associated with the score of "data source type score 0.7 (provincial regulatory announcement) + sampling field attribute bonus score 0.1 (circulation physical store field) + report standardization bonus score 0 (non-self-inspection report)", then its credibility weight = 0.7 + 0.1 + 0 = 0.8.

[0074] Calculation of the credibility weight of target structured knowledge:

[0075] Example: If the target element association score of the target structured knowledge matching is "data source type score 0.1 (non-standard enterprise self-inspection report) + sampling field attribute bonus score 0 (circulation e-commerce field) + report standardization bonus score 0 (non-standard)", then its credibility weight = 0.1 + 0 + 0 = 0.1.

[0076] Recording and storing calculation results:

[0077] The calculated credibility weights of the candidate and target structured knowledge are associated with the corresponding knowledge ID, attribute combination, and target element information and stored in the update log library to facilitate the tracing of subsequent conflict adjudication basis and system closed-loop optimization.

[0078] Understandably, step S103 addresses the shortcomings of existing technologies, such as the lack of automated conflict detection mechanisms and the inefficiency caused by the need for manual conflict judgment. By implementing pre-defined rules, it achieves automatic conflict identification, thus improving the efficiency of conflict handling. Constructing a credibility weight matrix based on business rules solves the problem of existing technologies failing to deeply integrate business logic into conflict resolution and relying solely on algorithmic judgment, leading to a lack of guaranteed knowledge consistency. This makes the weight evaluation more aligned with the actual regulatory scenarios of charging pile spot checks. Using quantified weights as the basis for conflict determination avoids the subjective and arbitrary shortcomings of conflict resolution in existing technologies, improving the objectivity and accuracy of conflict handling and providing support for the credibility of the knowledge graph.

[0079] S104. In response to the fact that the credibility weight of the candidate structured knowledge is greater than that of the target structured knowledge, the target structured knowledge that conflicts with the candidate structured knowledge is modified based on the candidate structured knowledge to update the knowledge graph of the charging pile sampling inspection.

[0080] Among them, knowledge modification refers to replacing or modifying target structured knowledge in the knowledge graph of charging pile spot checks that conflicts with candidate structured knowledge, in order to eliminate knowledge contradictions and ensure the consistency of graph content.

[0081] In practice, the credibility weights of candidate structured knowledge and target structured knowledge are compared. If the credibility weight of candidate structured knowledge is higher, it is considered more reliable. Using the Neo4j graph database, the candidate structured knowledge deemed reliable is transformed into standardized Cypher operation statements. Conflicting target structured knowledge is modified. If the target structured knowledge is an entity attribute (such as product sampling status or enterprise risk level), the attribute values ​​of the corresponding nodes are directly modified. If the target structured knowledge is a relationship between entities (such as the association between a product and a non-conforming item), the semantic associations of the corresponding edges are adjusted.

[0082] The embodiments of this invention form a fully automated technical closed loop encompassing multi-source data acquisition, knowledge mining, conflict resolution, and graph updating. This fundamentally overcomes the inherent shortcomings of existing technologies, such as low automation, delayed updates, lack of knowledge consistency guarantees, and weak unstructured data processing capabilities. The combination of multi-source data integration and domain-adaptive knowledge mining ensures the comprehensiveness and accuracy of knowledge, solving the problems of single data sources and insufficient accuracy in identifying specialized terms in existing technologies, thus providing richer decision-making basis for regulation. The conflict resolution mechanism based on business rules and the incremental update mode improve the timeliness and credibility of the knowledge graph, avoid resource waste, and solve the problems of outdated graph content, logical confusion, and weak decision support capabilities in existing technologies.

[0083] Furthermore, in response to the absence of target structured knowledge conflicting with candidate structured knowledge in the charging pile sampling knowledge graph, corresponding nodes and edges are created in the charging pile sampling knowledge graph based on the entities and relationships in the candidate structured knowledge. Semantic connections are then established between the new entities and existing entities, resulting in an updated charging pile sampling knowledge graph. That is, when conflict detection (comparing candidate structured knowledge with existing knowledge in the charging pile sampling knowledge graph) confirms that there is no target structured knowledge conflicting with candidate structured knowledge in the graph, there is no need to execute a conflict resolution process. Instead, based on the entity and relationship information in the candidate structured knowledge, the new knowledge increment is directly integrated into the graph through the creation of nodes and edges and the establishment of semantic connections, achieving conflict-free graph updates and ultimately obtaining an updated graph containing the newly added knowledge.

[0084] Example 2

[0085] Figure 2 This invention provides a flowchart of a method for updating the knowledge graph of charging pile spot checks. See also... Figure 2 The method includes the following steps:

[0086] S201. Obtain multimodal knowledge data related to the random inspection of charging piles from multiple data sources.

[0087] The multimodal knowledge data includes at least two of the following: regulatory announcement texts of different levels related to the random inspection of charging piles, enterprise random inspection report texts, on-site random inspection pictures or videos, and enterprise qualification information of the inspected units.

[0088] S202. Mine candidate structured knowledge from multimodal knowledge data, including at least key entities and their interrelationships.

[0089] S203. In response to the existence of target structured knowledge in the knowledge graph of charging pile spot checks that conflicts with candidate structured knowledge, a preset credibility weight matrix is ​​used to assign credibility weights to candidate structured knowledge and target structured knowledge respectively.

[0090] S204. In response to the fact that the credibility weight of the candidate structured knowledge is the same as that of the target structured knowledge, obtain the publication time of the candidate structured knowledge and the target structured knowledge in their respective data sources.

[0091] Among them, "identical credibility weights" means that after calculation using a preset credibility weight matrix, the credibility weight values ​​of the candidate structured knowledge and the target structured knowledge are completely identical. In this case, the priority of knowledge cannot be determined solely by the weight size, and an additional judgment dimension needs to be introduced. The publication time of the data source refers to the time when the original data (such as regulatory announcements or corporate self-inspection reports) corresponding to the candidate structured knowledge and the target structured knowledge are officially published on their data source platform. It is a core indicator reflecting the timeliness of knowledge.

[0092] In practice, a targeted tracking mechanism for publication time is established for different types of data sources:

[0093] If the knowledge comes from regulatory announcement text (national / provincial level): extract the publication time field from the metadata of the regulatory announcement page crawled by the web crawler. If the page does not directly mark it, use the sampling completion time mentioned in the announcement text or the date in the announcement document number to help determine it.

[0094] If the knowledge comes from the enterprise's random inspection report text: extract the creation time (i.e., the report generation time, which is regarded as the release time) from the report PDF attributes obtained by the file monitoring module, or extract it from the report issuance date field in the report text;

[0095] If the knowledge comes from on-site inspection images: extract the shooting time from the EXIF ​​information of the image file (considered as the release time of the on-site inspection data), or extract the upload time from the inspection record associated with the image;

[0096] If the knowledge comes from enterprise qualification information: directly obtain the information update time field from the structured data pulled from the API.

[0097] The published times traced are standardized and uniformly converted to the format "YYYY-MM-DDHH:MM:SS" to avoid time comparison errors caused by format differences. If the published time of a certain knowledge cannot be directly traced (e.g., no metadata, no text annotation), its published time is set as the data source acquisition time (e.g., web crawler crawling time, file listening trigger time) by default and marked as a non-original published time, which will only be used as a fallback basis when adjusting the weight in the future.

[0098] S205. Based on the order of publication, assign different credibility weights to the candidate structured knowledge and the target structured knowledge respectively.

[0099] Assigning different credibility weights again means adjusting the scores based on the time relationship and timeliness on the basis of the original same weights, so that the weights of the candidate knowledge and the target knowledge are different, and the conflict can be adjudicated by the weight.

[0100] In practice, if the release time of the candidate structured knowledge is later than that of the target structured knowledge, a timeliness adjustment score is added to the credibility weight of the target structured knowledge, while the credibility weight of the candidate structured knowledge remains unchanged. If the release time of the candidate structured knowledge is earlier than that of the target structured knowledge, a timeliness adjustment score is added to the credibility weight of the candidate structured knowledge, while the credibility weight of the target structured knowledge remains unchanged.

[0101] Understandably, adjusting the timeliness of the score breaks the deadlock in weighting, clearly distinguishes the priority of conflict knowledge, avoids the inefficiency caused by manual intervention in adjudication, and ensures an automated closed loop in the conflict resolution process.

[0102] S206. In response to the fact that the credibility weight of the candidate structured knowledge is greater than that of the target structured knowledge, the target structured knowledge that conflicts with the candidate structured knowledge is modified based on the candidate structured knowledge in order to update the knowledge graph of the charging pile sampling inspection.

[0103] S207. Monitor the status changes of the knowledge graph of the charging pile spot check, and identify risks through any of the following methods: rule judgment, model calculation and graph mining, to obtain the risk identification results.

[0104] Monitoring the state changes of the knowledge graph for charging pile spot checks refers to real-time tracking of the addition, modification, and deletion of entity attributes and relationships in the graph, and recording information such as the time, content, and related knowledge of the changes to ensure that no state changes are missed. Rule determination / model calculation / graph mining refers to three core risk identification methods: rule determination performs immediate identification based on preset thresholds, model calculation performs predictive identification based on machine learning algorithms, and graph mining performs deep association identification based on graph algorithms.

[0105] In practice, a real-time monitoring module is deployed, which is linked with the operation logs of the graph database (Neo4j). When the graph performs operations such as adding nodes / edges, modifying entity attributes, or deleting relationships, the module automatically captures the change events and records information such as the change type (add / modify / delete), the entities / relationships involved, the values ​​before and after the change, and the change time.

[0106] Three risk identification methods are implemented:

[0107] Rule-based judgment (real-time identification): Risk identification is triggered based on preset thresholds. For example, when a company's non-conforming records increase from 2 to 5 within 30 days (exceeding the threshold of 3), the company's quality risk surge identification result is triggered; when the number of products associated with a non-conforming item (such as creepage distance) increases from 10 to 20 (exceeding the threshold of 15), the common defect diffusion identification result is triggered; when the non-conforming rate of charging piles in a certain area increases from 5% to 12% (exceeding the threshold of 10%), the regional quality focus identification result is triggered.

[0108] Model calculation (predictive identification): Based on machine learning algorithms, historical data is analyzed. For example, an LSTM time series prediction model is used to analyze the trend of a company's non-conformance rate over the past 6 months. If the non-conformance rate is predicted to exceed 15% in the next month, the company will be identified as a high-risk enterprise. A logistic regression risk probability model is used to calculate the probability of quality risk based on multi-dimensional features such as the company's production license status, historical non-conformance frequency, and sampling areas. If the probability exceeds 80%, the company will be identified as a high-risk enterprise.

[0109] Graph mining (deep identification): Based on graph algorithms, we can mine associated risks. For example, we can use community discovery algorithms to identify high-risk enterprise clusters in the graph (such as multiple related supply chain companies having non-compliance issues), triggering the identification results of associated risk networks; and use path analysis algorithms to quickly locate all charging pile products that use this component when a problem is detected in a key component (such as a charging module), triggering the identification results of supply chain risk tracing.

[0110] Understandably, this step addresses the limitation of existing knowledge graphs being used only for passive querying. Through state monitoring and risk identification, it enables the graph to proactively perceive risks, upgrading it from a data storage tool to a regulatory decision support tool, thus meeting the needs of market regulators for proactive risk prevention. The three identification methods cover different risk scenarios, avoiding the limitations of a single identification method and improving the comprehensiveness of risk identification.

[0111] S208. In response to triggering an early warning based on the risk identification results, generate a structured early warning report containing an early warning title, risk subject, risk level, evidence chain, and handling recommendations, and push it to the enterprise's regulatory personnel according to the preset prompt method.

[0112] This step primarily involves automatically generating standardized, evidence-chain-supported early warning reports based on risk identification results. These reports are then pushed to relevant regulatory personnel through multiple channels, achieving automated integration of risk identification, early warning generation, and information dissemination. This ensures that regulatory personnel can promptly obtain risk information and take appropriate action. The preset notification methods include:

[0113] Regulatory system messages: Pop-up messages are pushed to the regulatory platform accounts of enterprise regulators. Clicking on the message will directly open the structured early warning report;

[0114] Email: Automatically sends warning reports to the regulator's preset email address, with a downloadable PDF version of the report attached to the email;

[0115] Visual dashboard: Warning information is updated in real time on the large visual screen in the monitoring center, and is marked with red (high), yellow (medium), and blue (low) according to the risk level, which facilitates global monitoring.

[0116] S209. Record key information during each update of the charging pile sampling knowledge graph through system logs; the key information includes at least complete conflict resolution cases and their adjudication basis, and a difference analysis between the information extraction results and the final adopted knowledge; based on the recorded system logs, extract training samples for optimizing the training of the entity recognition model, relation extraction model, and object detection model.

[0117] In some embodiments, training sample extraction and generation includes: screening records with complete information and annotation basis from system logs, and removing log entries with missing information (such as conflict resolution results) or abnormalities (such as model calculation errors); labeling based on the final correct results in the logs. Specifically, entity recognition model samples include extracting regulatory announcement text fragments / nameplate areas of inspection images (raw data) + correctly identified entities (such as manufacturers, product models) (annotation results) + extracting error types (such as missed identification, incorrect identification) (error analysis); relation extraction model samples include extracting sentence text (raw data) + correctly extracted triples (annotation results) + relation matching errors (such as mismatched relation types) (error analysis); target detection model samples include extracting key areas of inspection images + correctly located coordinate boxes + identified text content (annotation results) + positioning errors (such as coordinate offset) (error analysis); finally, the annotated samples are converted into a format acceptable to the model and stored in the sample library.

[0118] This invention improves the conflict resolution system, ensuring the credibility of the knowledge graph. S204-S205 addresses the challenge of conflict adjudication in scenarios with identical weights by supplementing weights with timeliness-based adjustments. Combined with the weight matrix in S203, a four-dimensional conflict resolution system is formed, encompassing data source, sampling domain, report standardization, and timeliness. This completely overcomes the shortcomings of existing technologies, such as single-dimensional conflict resolution and the need for manual intervention, ensuring the consistency and reliability of the knowledge graph. S207-S208 transforms the static graph into a dynamic risk perception tool. Through three types of risk identification and structured report delivery, it addresses the problem that existing graphs are only used for storage and cannot support real-time regulatory decision-making. This allows the graph to directly serve risk warning and handling, improving the intelligence level of charging pile quality supervision. S209 provides data support for core model optimization through log recording and sample extraction, enabling the system to continuously improve knowledge extraction accuracy and risk identification accuracy with actual operation. This solves the shortcomings of existing systems with fixed performance and inability to adapt to new scenarios, ensuring the system meets the long-term development needs of the charging pile sampling inspection field.

[0119] Example 3

[0120] Figure 3 This is a schematic diagram of a charging pile spot check knowledge graph updating device provided in an embodiment of the present invention. This device can execute any of the charging pile spot check knowledge graph updating methods of the present invention. For example... Figure 3 As shown, the knowledge graph update device for charging pile spot checks includes:

[0121] The data perception and acquisition module 301 is used to acquire multimodal knowledge data related to the random inspection of charging piles from multiple data sources; wherein, the multimodal knowledge data includes at least two of the following: regulatory announcement texts of different levels related to the random inspection of charging piles, enterprise random inspection report texts, on-site random inspection pictures or videos, and enterprise qualification information of the inspected unit.

[0122] The knowledge extraction module 302 is used to mine candidate structured knowledge from multimodal knowledge data, including at least key entities and their interrelationships.

[0123] The credibility confirmation module 303 is used to respond to the existence of target structured knowledge that conflicts with candidate structured knowledge in the knowledge graph of charging pile sampling inspection, and to assign credibility weights to candidate structured knowledge and target structured knowledge respectively using a preset credibility weight matrix.

[0124] The first update module 304 is used to respond to the situation where the credibility weight of the candidate structured knowledge is greater than that of the target structured knowledge. Based on the candidate structured knowledge, the module modifies the conflicting target structured knowledge to update the knowledge graph of the charging pile sampling inspection.

[0125] In some embodiments, each element in the preset credibility weight matrix is ​​associated with a specific combination of attributes, including data source type, sampling area, and report standardization, as well as the data source type score, sampling area attribute bonus score, and report standardization bonus score in the specific attribute combination;

[0126] In assigning credibility weights to candidate structured knowledge and target structured knowledge respectively using a preset credibility weight matrix, the credibility confirmation module 303 is specifically used for:

[0127] The candidate structured knowledge and the target structured knowledge are analyzed separately to determine the data source type, sampling area, and reporting standardization of the enterprise self-inspection report corresponding to each candidate structured knowledge and the target structured knowledge.

[0128] Based on the data source type, sampling field, and reporting standardization of the enterprise self-inspection report corresponding to the candidate structured knowledge and the target structured knowledge, target elements that match the candidate structured knowledge and the target structured knowledge respectively are determined from the preset credibility weight matrix.

[0129] Based on the data source type score associated with each matched target element, the additional score for the sampling domain attribute, and the additional score for the report standardization, the credibility weights of the candidate structured knowledge and the target structured knowledge are determined respectively.

[0130] In some embodiments, it also includes:

[0131] The timeliness acquisition module is used to obtain the publication time of the candidate structured knowledge and the target structured knowledge in their respective data sources when the credibility weight of the candidate structured knowledge is the same as that of the target structured knowledge.

[0132] The secondary weighting module is used to assign different credibility weights to the candidate structured knowledge and the target structured knowledge based on their publication time.

[0133] In some embodiments, regarding the mining of candidate structured knowledge from multimodal knowledge data, including at least key entities and their interrelationships, the knowledge extraction module 302 is specifically used for:

[0134] For regulatory announcement texts or enterprise inspection report texts at different levels in multimodal knowledge data, an enhanced semantic representation model is used to extract text context semantic features from the texts at different levels. Named entity recognition and relation extraction models are used to extract the text context semantic features to obtain candidate structured knowledge.

[0135] For the acquired on-site inspection images, the on-site inspection images are identified using a target detection model to obtain the key component areas of the charging pile; text recognition is performed on the key component areas of the charging pile to obtain the text recognition results of the key component areas of the charging pile; based on the text recognition results of the key component areas of the charging pile, consistency verification, attribute authenticity verification, and anomaly rationality verification are performed on the candidate structured knowledge.

[0136] Based on the structured enterprise qualification information of the sampled units, the entity attributes of the candidate structured knowledge that has passed the verification are supplemented to obtain the final candidate structured knowledge.

[0137] In some embodiments, it also includes:

[0138] The second update module is used to respond to the fact that there is no target structured knowledge in the charging pile sampling knowledge graph that conflicts with the candidate structured knowledge. Then, based on the entities and relationships in the candidate structured knowledge, it creates corresponding nodes and edges in the charging pile sampling knowledge graph and establishes semantic connections between the new entities and the existing entities to obtain the updated charging pile sampling knowledge graph.

[0139] In some embodiments, it also includes:

[0140] The status monitoring module is used to monitor the status changes of the knowledge graph of charging pile spot checks, and to identify risks through any of the following methods: rule judgment, model calculation and graph mining, and obtain the risk identification results.

[0141] The early warning module is used to respond to early warnings triggered based on risk identification results, generate a structured early warning report containing an early warning title, risk subject, risk level, evidence chain, and handling suggestions, and push it to enterprise regulatory personnel according to preset prompts.

[0142] In some embodiments, it also includes:

[0143] The logging module is used to record key information during each update of the charging pile sampling knowledge graph through system logs. The key information includes at least the complete case of conflict resolution and its adjudication basis, and the difference analysis between the information extraction results and the final adopted knowledge.

[0144] The data mining module is used to extract training samples from the recorded system logs to optimize the training of entity recognition models, relation extraction models, and object detection models.

[0145] The charging pile sampling inspection knowledge graph update device provided in this embodiment of the invention can execute the charging pile sampling inspection knowledge graph update method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0146] According to embodiments of the present invention, the present invention also provides an electronic device, a readable storage medium, and a computer program product.

[0147] Example 4

[0148] Figure 4 A schematic diagram of the structure of an electronic device 10 that can be used to implement embodiments of the present invention is shown. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the invention described and / or claimed herein.

[0149] like Figure 4 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory 12 or a random access memory 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the read-only memory 12 or loaded from storage unit 18 into the random access memory 13. The random access memory 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, read-only memory 12, and random access memory 13 are interconnected via a bus 14. An input / output interface 15 is also connected to the bus 14.

[0150] Multiple components in electronic device 10 are connected to input / output interface 15, including: input unit 16; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disks, optical disks, etc.; and communication unit 19, such as network interface cards, modems, wireless transceivers, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0151] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, central processing units, graphics processing units, various special-purpose artificial intelligence computing chips, various processors running machine learning model algorithms, digital signal processors, and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as performing a charging pile sampling knowledge graph update method.

[0152] In some embodiments, the charging pile sampling knowledge graph update method can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via read-only memory 12 and / or communication unit 19. When the computer program is loaded into random access memory 13 and executed by processor 11, one or more steps of the charging pile sampling knowledge graph update method described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the charging pile sampling knowledge graph update method by any other suitable means (e.g., by means of firmware).

[0153] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays, application-specific integrated circuits (ASICs), application-specific standard products (ASICs), systems-on-a-chip (SoCs), complex programmable logic devices, computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0154] Computer programs used to implement the methods of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to the processor of a general-purpose computer, a special-purpose computer, or other programmable charging pile sampling knowledge graph update device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The computer programs can be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0155] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory, optical fibers, portable compact disk read-only memory, optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0156] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device or liquid crystal display for displaying information to a user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with a user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0157] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet. The computing system can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having client-server relationships with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, a host product within the cloud computing service system, addressing the shortcomings of traditional physical hosts and virtual private servers, such as high management difficulty and weak business scalability.

[0158] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0159] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for updating a knowledge graph for random inspections of charging piles, characterized in that, include: Multimodal knowledge data related to the random inspection of charging piles is obtained from multiple data sources; wherein, the multimodal knowledge data includes at least two of the following: regulatory announcement texts of different levels related to the random inspection of charging piles, enterprise random inspection report texts, on-site random inspection pictures or videos, and enterprise qualification information of the inspected units; Mine candidate structured knowledge, including at least key entities and their relationships, from the multimodal knowledge data; In response to the existence of target structured knowledge in the knowledge graph of charging pile spot checks that conflicts with the candidate structured knowledge, a preset credibility weight matrix is ​​used to assign credibility weights to the candidate structured knowledge and the target structured knowledge respectively. In response to the fact that the credibility weight of the candidate structured knowledge is greater than that of the target structured knowledge, the target structured knowledge that conflicts with the candidate structured knowledge is modified based on the candidate structured knowledge to update the knowledge graph of the charging pile sampling inspection.

2. The method according to claim 1, characterized in that, Each element in the preset credibility weight matrix is ​​associated with a specific attribute combination including data source type, sampling area, and report standardization, as well as the data source type score, sampling area attribute bonus score, and report standardization bonus score in the specific attribute combination; The step of assigning credibility weights to the candidate structured knowledge and the target structured knowledge using a preset credibility weight matrix includes: The candidate structured knowledge and the target structured knowledge are analyzed separately to determine the data source type, sampling area, and reporting standardization of the enterprise self-inspection report corresponding to each of the candidate structured knowledge and the target structured knowledge. Based on the data source type, sampling field, and reporting standardization of the enterprise self-inspection report corresponding to the candidate structured knowledge and the target structured knowledge, target elements matching the candidate structured knowledge and the target structured knowledge are determined from the preset credibility weight matrix respectively. Based on the data source type score associated with each matched target element, the additional score for the sampling domain attribute, and the additional score for the report standardization, the credibility weights of the candidate structured knowledge and the target structured knowledge are determined respectively.

3. The method according to claim 1 or 2, characterized in that, Also includes: In response to the fact that the credibility weight of the candidate structured knowledge is the same as that of the target structured knowledge, the publication time of the candidate structured knowledge and the target structured knowledge in their respective data sources is obtained; Based on the relationship between the release times, different credibility weights are assigned to the candidate structured knowledge and the target structured knowledge respectively.

4. The method according to claim 1, characterized in that, Mining candidate structured knowledge from the multimodal knowledge data, including at least key entities and their relationships, includes: For regulatory announcement texts or enterprise inspection report texts at different levels in multimodal knowledge data, an enhanced semantic representation model is used to extract text context semantic features from the regulatory announcement texts or enterprise inspection report texts at different levels; named entity recognition model and relation extraction model are used to perform named entity recognition and relation extraction on the text context semantic features to obtain candidate structured knowledge. For the acquired on-site inspection images, the on-site inspection images are identified using a target detection model to obtain the key component areas of the charging pile; text recognition is performed on the key component areas of the charging pile to obtain the text recognition results of the key component areas of the charging pile; based on the text recognition results of the key component areas of the charging pile, the candidate structured knowledge is subjected to consistency verification, attribute authenticity verification, and anomaly rationality verification. Based on the structured enterprise qualification information of the sampled units, the entity attributes of the candidate structured knowledge that has passed the verification are supplemented to obtain the final candidate structured knowledge.

5. The method according to claim 1, characterized in that, Also includes: In response to the absence of target structured knowledge conflicting with candidate structured knowledge in the charging pile sampling knowledge graph, corresponding nodes and edges are created in the charging pile sampling knowledge graph based on the entities and relationships in the candidate structured knowledge, and semantic connections are established between the new entities and existing entities to obtain the updated charging pile sampling knowledge graph.

6. The method according to claim 1, characterized in that, Also includes: Monitor the status changes of the knowledge graph of the charging pile spot check, and identify risks through any of the following methods: rule judgment, model calculation and graph mining, to obtain risk identification results; In response to the triggering of an alert based on the risk identification results, a structured alert report is generated, which includes an alert title, the risk subject, the risk level, the chain of evidence, and disposal recommendations. The report is then pushed to the enterprise's regulatory personnel according to the preset prompts.

7. The method according to claim 1, characterized in that, Also includes: The system logs record key information during each update of the knowledge graph for charging pile spot checks; the key information includes at least complete conflict resolution cases and their adjudication basis, and a difference analysis between the information extraction results and the final adopted knowledge. Based on the recorded system logs, training samples are extracted for optimizing the entity recognition model, relation extraction model, and object detection model.

8. A knowledge graph update device for charging pile spot checks, characterized in that, include: The data perception and acquisition module is used to acquire multimodal knowledge data related to the random inspection of charging piles from multiple data sources; wherein, the multimodal knowledge data includes at least two of the following: regulatory announcement texts of different levels related to the random inspection of charging piles, enterprise random inspection report texts, on-site random inspection pictures or videos, and enterprise qualification information of the inspected units. The knowledge extraction module is used to mine candidate structured knowledge from the multimodal knowledge data, including at least key entities and their interrelationships. The credibility confirmation module is used to respond to the existence of target structured knowledge in the knowledge graph of charging pile spot checks that conflicts with the candidate structured knowledge, and to assign credibility weights to the candidate structured knowledge and the target structured knowledge respectively using a preset credibility weight matrix. The update module is used to respond to the fact that the credibility weight of the candidate structured knowledge is greater than the credibility weight of the target structured knowledge, and to modify the conflicting target structured knowledge based on the candidate structured knowledge, so as to update the knowledge graph of the charging pile sampling inspection.

9. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the method of any one of claims 1-7.

Images