A knowledge graph-based marine photovoltaic information management system

Abstract

This invention discloses a knowledge graph-based marine photovoltaic (PV) information management system, comprising: an information acquisition module for acquiring multi-source marine PV information and generating standard marine PV multi-source information; a graph construction module for generating a marine PV knowledge graph based on the standard marine PV multi-source information; a model construction module for constructing a marine PV box inference model; a query box generation module for generating a multi-dimensional adaptive query box for marine PV based on the marine PV knowledge graph; a path projection module for generating a path projection query box based on the multi-dimensional adaptive query box for marine PV; an evidence convergence module for generating an evidence convergence query box based on the standard marine PV multi-source information; and a result output module for generating marine PV information management results based on the evidence convergence query box. This invention improves the accuracy of marine PV information verification, anomaly identification, and electrical traceability.

Description

Technical Field

[0001] This invention relates to the field of operation and maintenance management of offshore photovoltaic power plants, and in particular to an offshore photovoltaic information management system based on knowledge graphs. Background Technology

[0002] Offshore photovoltaic projects typically involve multiple types of information, including design documents, equipment ledgers, spatial layout, operational status, environmental monitoring, construction and installation, testing and acceptance, operation and maintenance records, wiring ports, cable laying, electrical connections, and electrical testing. Existing information management methods often store data separately in ledger systems, operation and maintenance systems, testing record systems, and electrical data systems. While these methods can complete basic data registration and retrieval, the relationships between data from different sources are difficult to express uniformly. Cross-type information verification mainly relies on manual comparison, and when information updates are delayed, issues such as missing data, conflicts, and incomplete traceability can easily arise.

[0003] Existing knowledge graph technologies can structure and organize equipment, status, records, and connections. However, in the context of offshore photovoltaic information management, basic management information, on-site status information, process record information, and electrical installation information exhibit differences in source, time, space, and equipment number. Traditional knowledge graph queries and rule matching typically rely on fixed relationship paths for retrieval, making it difficult to dynamically adjust the query scope based on different status information and evidence consistency. This results in insufficient accuracy in identifying missing information, conflicting information, abnormal status information, abnormal acceptance criteria information, and electrical connection traceability information.

[0004] Therefore, how to provide a knowledge graph-based marine photovoltaic information management system is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a knowledge graph-based marine photovoltaic (PV) information management system. This invention generates a marine PV knowledge graph based on multi-source information from multiple sources, and constructs a Query2Box-based reasoning model. Through a state adjustment unit, a query box generation unit, a multi-relational path projection unit, a multi-source evidence convergence unit, and a box boundary correction unit, it correlates and matches basic management information, on-site status information, process record information, and electrical installation information to generate marine PV information management results. This system possesses advantages such as clear multi-source information correlation, dynamically adjustable query boxes, accurate evidence convergence, and strong electrical connection traceability.

[0006] A knowledge graph-based marine photovoltaic information management system according to an embodiment of the present invention includes: The information acquisition module is used to collect multi-source information of offshore photovoltaic systems, including basic management information, on-site status information, process record information and electrical installation information. It performs field unification, time alignment, equipment number normalization and invalid field removal to generate standard offshore photovoltaic multi-source information. The knowledge graph construction module is used to extract marine photovoltaic entities and relationships based on standard marine photovoltaic multi-source information and generate a marine photovoltaic knowledge graph. The model building module is used to build the Haiguang box inference model based on the Query2Box model, including the state adjustment unit, query box generation unit, multi-relation path projection unit, multi-source evidence convergence unit, and box boundary correction unit. The query box generation module is used to generate a state adjustment vector in the state adjustment unit based on the knowledge graph of marine photovoltaics. The state adjustment vector is then input into the query box generation unit to generate a multi-dimensional state adaptive query box for marine photovoltaics. The path projection module is used to input the multi-dimensional state adaptive query box of marine photovoltaic power into the multi-relation path projection unit, perform type projection, and generate the path projection query box. The evidence convergence module is used to generate evidence convergence parameters based on standard marine photovoltaic multi-source information, input multi-source evidence convergence units for convergence calculation, and generate evidence convergence query boxes; The results output module is used to input the evidence convergence query box into the box boundary correction unit, correct the center position and boundary range, generate a correction query box, and perform matching based on the correction query box to generate marine photovoltaic information management results.

[0007] Optionally, the basic management information includes design data information, equipment information, asset ledger information, and space layout information; the on-site status information includes operating status information and environmental monitoring information; the process record information includes construction and installation information, testing and acceptance information, and operation and maintenance record information; and the electrical installation information includes wiring port information, cable laying information, electrical connection information, and electrical testing information.

[0008] Optionally, the map construction module includes: Based on standard offshore photovoltaic multi-source information, the entity types are determined according to the information sources corresponding to basic management information, on-site status information, process record information and electrical installation information. Equipment entities, spatial entities, status entities, process entities and electrical installation entities are determined, and offshore photovoltaic entities are generated. Based on the multi-source information of offshore photovoltaic entities and standard offshore photovoltaic systems, basic management relationships are extracted from basic management information, on-site status relationships are extracted from on-site status information, process record relationships are extracted from process record information, and electrical installation relationships are extracted from electrical installation information. The offshore photovoltaic entity is written into the graph structure as an entity, and the basic management relationship, on-site status relationship, process record relationship and electrical installation relationship are written into the graph structure as relationship content to generate an offshore photovoltaic knowledge graph.

[0009] Optionally, the model building module includes: Call the original structure of the Query2Box model, including the box representation structure, relational projection structure, and box intersection structure; A query box generation unit is constructed based on the box representation structure. The center vector and the offset vector are retained to form the box representation structure of the query box, and the access position of the state adjustment vector is set. Based on the relation projection structure, a multi-relation path projection unit is constructed. The structure of projecting the query box according to the relations in the knowledge graph is retained, and the path type identification structure and path weight allocation structure corresponding to the basic management relationship, the field status relationship, the process record relationship and the electrical installation relationship are set. A multi-source evidence aggregation unit is constructed based on the box aggregation structure, retaining the structure of multiple query boxes for aggregation calculation, and setting the access position of evidence aggregation parameters; A state adjustment unit is built outside the original structure of the Query2Box model and set before the query box generation unit; a box boundary correction unit is built and set after the multi-source evidence convergence unit. The state adjustment unit, query box generation unit, multi-relation path projection unit, multi-source evidence convergence unit, and box boundary correction unit are connected in sequence to form the Haiguang box inference model.

[0010] Optionally, the query box generation module includes: Based on the knowledge graph of marine photovoltaics, the marine photovoltaic entities corresponding to the basic management information are associated with the basic management relationships, and the integrity of design data, consistency of equipment ledgers and spatial layout are extracted to generate the basic management status. The on-site status information is associated with the marine photovoltaic entities and the on-site status relationship. The operational stability, environmental monitoring effectiveness and alarm status are extracted to generate the on-site status. The process record information is associated with the marine photovoltaic entity and the process record relationship, and the construction completion status, inspection and acceptance status and operation and maintenance closed loop status are extracted to generate the process record status. The system associates the offshore photovoltaic entities corresponding to the electrical installation information with the electrical installation relationship, extracts the occupancy status of the wiring ports, the completion status of cable laying, the consistency status of electrical connections, and the qualified status of electrical testing, and generates the electrical installation status. Based on the basic management status, field status, process record status, and electrical installation status, the input status adjustment unit performs numerical encoding and vector concatenation to generate a status adjustment vector. The state adjustment vector is input into the query box generation unit, and the state adjustment vector is split into the center adjustment amount and the boundary adjustment amount. In the query box generation unit, an initial center vector and an initial offset vector are constructed based on the marine photovoltaic knowledge graph; In the query box generation unit, the center adjustment amount is converted into the position adjustment amount, which is then superimposed onto the initial center vector to generate the center vector. The boundary adjustment amount is converted into the boundary expansion amount and the boundary contraction amount, which are then superimposed onto the initial offset vector to generate the offset vector. The query box is then constructed to generate the marine photovoltaic multi-dimensional state adaptive query box.

[0011] Optionally, the path projection module includes: Based on the basic management relationships, field status relationships, process record relationships, and electrical installation relationships in the offshore photovoltaic knowledge graph, the source and direction of the relationships are identified, the basic management path, field status path, process record path, and electrical installation path are determined, and path type identifiers are configured for each. Based on the path type identifier corresponding to the basic management path, the equipment ownership relationship between design data information and equipment information, the asset correspondence relationship between equipment information and asset ledger information, and the spatial arrangement relationship between equipment information and spatial arrangement information are filtered from the basic management relationship to determine the projection parameters of the basic management path; Based on the path type identifier corresponding to the field status path, the state correspondence between the operation status information and the environmental monitoring information is filtered from the field status relationship to determine the field status path projection parameters. Based on the path type identifier corresponding to the process record path, the construction acceptance succession relationship between construction and installation information and inspection and acceptance information, and the acceptance and operation and maintenance succession relationship between inspection and acceptance information and operation and maintenance record information are filtered from the process record relationship to determine the process record path projection parameters; Based on the path type identifier corresponding to the electrical installation path, the port laying relationship between wiring port information and cable laying information, the laying connection relationship between cable laying information and electrical connection information, and the connection detection relationship between electrical connection information and electrical detection information are filtered from the electrical installation relationship to determine the electrical installation path projection parameters; Input the multi-dimensional state adaptive query box of marine photovoltaic system into the multi-relation path projection unit to analyze the center vector and offset vector in the multi-dimensional state adaptive query box of marine photovoltaic system. In the multi-relationship path projection unit, the center vector is projected according to the basic management path projection parameters to determine equipment affiliation, asset correspondence, and spatial layout direction, and the offset vector is adjusted to generate a basic management path query box. Project the center vector into the operating status and environmental monitoring direction according to the on-site status path projection parameters, and adjust the boundary of the offset vector to generate the on-site status path query box. Project the center vector into the construction, installation, inspection, acceptance, and operation and maintenance record directions according to the process record path projection parameters, and adjust the boundary of the offset vector to generate a process record path query box; Based on the electrical installation path projection parameters, the center vector is projected to the wiring ports, cable laying, electrical connections, and electrical testing directions, and the offset vector is adjusted to generate an electrical installation path query box. Based on the basic management path query box, the field status path query box, the process record path query box, and the electrical installation path query box, a convergence process is performed to generate a path projection query box.

[0012] Optionally, the evidence convergence module includes: Based on the design data, equipment information, asset ledger information and spatial layout information in the standard marine photovoltaic multi-source information, the information is encoded according to field type, equipment number, spatial location and information source to generate the center vector and offset vector corresponding to the basic management information, thus forming the basic management evidence box. Based on operational status information and environmental monitoring information, the data is encoded according to the collection time, spatial location, equipment number, operational status value, and environmental monitoring value to generate the center vector and offset vector corresponding to the field status information, thus forming a field status evidence box. Based on construction and installation information, testing and acceptance information, and operation and maintenance record information, the process record information is encoded according to equipment number, recording time, recording object, and recording result to generate the center vector and offset vector corresponding to the process record information, thus forming a process record evidence box. Based on the wiring port information, cable laying information, electrical connection information and electrical testing information, the electrical installation information is encoded according to the installation object, connection object, wiring port, laying path, electrical connection result and electrical testing result, and the center vector and offset vector corresponding to the electrical installation information are generated to form an electrical installation evidence box; Based on standard offshore photovoltaic multi-source information, the reliability of data sources, consistency of collection time, consistency of spatial location, consistency of equipment number, consistency of operating status, and consistency of electrical connection are calculated respectively to generate evidence convergence parameters; Input the path projection query box, basic management evidence box, on-site status evidence box, process record evidence box, electrical installation evidence box, and evidence intersection parameters into the multi-source evidence intersection unit, perform intersection calculations, and generate the evidence intersection query box.

[0013] Optionally, the result output module includes: Input the evidence intersection query box and evidence intersection parameters into the box boundary correction unit; Based on the reliability of data sources, consistency of collection time, consistency of spatial location, consistency of equipment number, consistency of operating status, and consistency of electrical connection, the difference between each parameter item and the preset qualified value is calculated in the box boundary correction unit, and weighted synthesis is performed to generate the center position correction amount and the boundary range correction amount, which are combined into the boundary correction amount. The boundary correction direction is determined based on the positional relationship between the evidence intersection query box, the basic management evidence box, the on-site status evidence box, the process record evidence box, and the electrical installation evidence box. When the electrical connection consistency is lower than the preset qualified value, the correction amount of the boundary range corresponding to the electrical installation relationship is reduced, and the data content of inconsistent connection objects in the electrical installation information is identified as electrical connection traceability information. The center position of the evidence intersection query box is corrected based on the center position correction amount, and the boundary range of the evidence intersection query box is corrected based on the boundary range correction amount to generate a corrected query box. Based on the calibration query box, the offshore photovoltaic entities, basic management relationships, on-site status relationships, process record relationships and electrical installation relationships are matched in the offshore photovoltaic knowledge graph to determine consistent information, missing information, conflicting information, abnormal status information and abnormal acceptance basis information. The system summarizes missing information, conflict information, abnormal status information, abnormal acceptance basis information, and electrical connection traceability information to generate marine photovoltaic information management results.

[0014] The beneficial effects of this invention are: First, this invention collects multi-source information on offshore photovoltaic (PV) systems and generates standardized offshore PV multi-source information. It unifies basic management information, on-site status information, process record information, and electrical installation information into an offshore PV knowledge graph. This allows design data, equipment information, asset ledger information, spatial layout information, operational status information, environmental monitoring information, construction and installation information, testing and acceptance information, operation and maintenance record information, wiring port information, cable laying information, electrical connection information, and electrical testing information to be expressed in a related manner within the same graph structure. This reduces the difficulty of information verification caused by scattered storage of offshore PV information, inconsistent fields, and inconsistent equipment numbers, and improves the integrity and traceability of multi-source information management.

[0015] Secondly, this invention constructs a marine photovoltaic box inference model based on the Query2Box model. Through a state adjustment unit, a query box generation unit, and a multi-relation path projection unit, the basic management state, field state, process record state, and electrical installation state can participate in the adjustment of the center vector and offset vector, generating a multi-dimensional state adaptive query box for marine photovoltaics. Then, it performs a type projection along the basic management path, field state path, process record path, and electrical installation path, so that the query box no longer relies solely on static query semantics, but can adaptively adjust with the state changes and relation path changes in the marine photovoltaic knowledge graph, improving the accuracy of identifying missing information, conflicting information, and abnormal state information.

[0016] Furthermore, this invention utilizes a multi-source evidence convergence unit and a box boundary correction unit to perform convergence calculations on basic management evidence boxes, field status evidence boxes, process record evidence boxes, electrical installation evidence boxes, and evidence convergence parameters. It also corrects the center position and boundary range of the evidence convergence query box to generate a correction query box. This allows for matching under the constraints of data source credibility, collection time consistency, spatial location consistency, equipment number consistency, operating status consistency, and electrical connection consistency. This improves the reliability of judging abnormal information and electrical connection traceability information based on acceptance criteria and reduces misjudgments caused by time deviations, spatial deviations, and inconsistencies in connection objects among multi-source information. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of a knowledge graph-based marine photovoltaic information management method proposed in this invention; Figure 2 This is a schematic diagram of the construction process of the sea light box inference model proposed in this invention; Figure 3 This is a schematic diagram of the multi-source evidence convergence and box boundary correction processing flow proposed in this invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0019] refer to Figures 1-3 A knowledge graph-based marine photovoltaic information management system includes: The information acquisition module is used to collect multi-source information of offshore photovoltaic systems. This information includes basic management information, on-site status information, process record information, and electrical installation information. The module performs field unification, time alignment, equipment numbering normalization, and invalid field removal on the multi-source information of offshore photovoltaic systems to generate standard multi-source information of offshore photovoltaic systems. Among them, basic management information includes design data information, equipment information, asset ledger information, and spatial layout information; on-site status information includes operational status information and environmental monitoring information; process record information includes construction and installation information, testing and acceptance information, and operation and maintenance records; and electrical installation information includes wiring port information, cable laying information, electrical connection information, and electrical testing information. The graph construction module is used to extract offshore photovoltaic entities from basic management information, field status information, process record information and electrical installation information based on standard offshore photovoltaic multi-source information, and to extract basic management relationships, field status relationships, process record relationships and electrical installation relationships based on standard offshore photovoltaic multi-source information. The offshore photovoltaic entities, basic management relationships, field status relationships, process record relationships and electrical installation relationships are written into the graph structure to generate an offshore photovoltaic knowledge graph. The model building module is used to build the Haiguang Box inference model based on the Query2Box model. The Haiguang Box inference model includes a state adjustment unit, a query box generation unit, a multi-relation path projection unit, a multi-source evidence intersection unit, and a box boundary correction unit. The state adjustment unit, query box generation unit, multi-relation path projection unit, multi-source evidence intersection unit, and box boundary correction unit are connected in sequence. The query box generation module is used to extract basic management status, field status, process record status and electrical installation status based on the offshore photovoltaic knowledge graph. The basic management status, field status, process record status and electrical installation status are input into the status adjustment unit to generate a status adjustment vector. The status adjustment vector is then input into the query box generation unit. The center vector and offset vector are adjusted through the status adjustment vector to generate a multi-dimensional adaptive query box for offshore photovoltaic status. The path projection module is used to construct basic management paths, field status paths, process record paths, and electrical installation paths based on the basic management relationships, field status relationships, process record relationships, and electrical installation relationships in the offshore photovoltaic knowledge graph. The offshore photovoltaic multi-dimensional status adaptive query box is input into the multi-relationship path projection unit, and the basic management path, field status path, process record path, and electrical installation path are subjected to type-based projection to generate path projection query boxes. The evidence convergence module is used to generate basic management evidence boxes, field status evidence boxes, process record evidence boxes, and electrical installation evidence boxes based on standard offshore photovoltaic multi-source information and according to the box representation method of the query box generation unit. It also generates evidence convergence parameters based on standard offshore photovoltaic multi-source information. The path projection query box, basic management evidence box, field status evidence box, process record evidence box, electrical installation evidence box, and evidence convergence parameters are input into the multi-source evidence convergence unit for convergence calculation to generate evidence convergence query boxes. The evidence convergence parameters include data source credibility, acquisition time consistency, spatial location consistency, equipment number consistency, operating status consistency, and electrical connection consistency. The result output module is used to input the evidence intersection query box into the box boundary correction unit, correct the center position and boundary range of the evidence intersection query box, generate a correction query box, and match the marine photovoltaic entities, basic management relationships, on-site status relationships, process record relationships and electrical installation relationships in the marine photovoltaic knowledge graph based on the correction query box, and generate marine photovoltaic information management results. The marine photovoltaic information management results include missing information, conflict information, abnormal status information, abnormal acceptance basis information and electrical connection traceability information.

[0020] In this embodiment, basic management information includes design data information, equipment information, asset ledger information, and spatial layout information; on-site status information includes operational status information and environmental monitoring information; process record information includes construction and installation information, testing and acceptance information, and operation and maintenance record information; and electrical installation information includes wiring port information, cable laying information, electrical connection information, and electrical testing information.

[0021] In this embodiment, the map construction module includes: Based on standard offshore photovoltaic multi-source information, entity types are determined according to the information sources corresponding to basic management information, on-site status information, process record information, and electrical installation information. Data content with equipment numbers in design data, equipment information, asset ledger information, and wiring port information is determined as equipment entities; data content with spatial location in spatial layout information is determined as spatial entities; data content with status description in operation status information and environmental monitoring information is determined as status entities; data content with process records in construction and installation information, testing and acceptance information, and operation and maintenance record information is determined as process entities; and data content with installation objects, connection objects, and testing objects in wiring port information, cable laying information, electrical connection information, and electrical testing information is determined as electrical installation entities, thus generating offshore photovoltaic entities. Based on the multi-source information of marine photovoltaic entities and standard marine photovoltaic systems, basic management relationships are extracted from data content with the same equipment number, array number, and spatial location among design data information, equipment information, asset ledger information, and spatial layout information. The basic management relationships express the attribution, configuration, and layout relationships among design data information, equipment information, asset ledger information, and spatial layout information. Based on offshore photovoltaic entities and standard offshore photovoltaic multi-source information, the field status relationship is extracted from data content with the same equipment number, the same acquisition time and the same spatial location between the operation status information and the environmental monitoring information. The field status relationship expresses the state correspondence between the operation status information and the environmental monitoring information. Based on the multi-source information of offshore photovoltaic entities and standard offshore photovoltaic systems, process record relationships are extracted from data content that have the same equipment number, the same construction batch, the same test number, and the same operation and maintenance record number among construction and installation information, inspection and acceptance information, and operation and maintenance record information. The process record relationships express the process succession relationship between construction and installation information, inspection and acceptance information, and operation and maintenance record information. Based on the multi-source information of offshore photovoltaic entities and standard offshore photovoltaic systems, electrical installation relationships are extracted from data content that share the same equipment number, same wiring port number, same cable number, and same test number among wiring port information, cable laying information, electrical connection information, and electrical testing information. The electrical installation relationships express the connection relationships, laying relationships, testing relationships, and acceptance correspondences among wiring port information, cable laying information, electrical connection information, and electrical testing information. The offshore photovoltaic entity is written into the graph structure as an entity, and the basic management relationship, on-site status relationship, process record relationship and electrical installation relationship are written into the graph structure as relationship content to generate an offshore photovoltaic knowledge graph.

[0022] In this embodiment, the model building module includes: The original structure of the Query2Box model is called. The original structure includes a box representation structure, a relation projection structure, and a box intersection structure. The box representation structure uses a center vector and an offset vector to form a query box. The relation projection structure projects the query box onto the relation in the knowledge graph. The box intersection structure performs intersection calculations on multiple query boxes. A query box generation unit is constructed based on the box representation structure. The center vector and offset vector are retained in the query box generation unit to form the box representation structure of the query box. The state adjustment vector access position is set in the query box generation unit to form a query box generation unit that can receive the state adjustment vector and adjust the center vector and offset vector. Based on the relation projection structure, a multi-relation path projection unit is constructed. The structure of projecting the query box according to the relationship in the knowledge graph is retained in the multi-relation path projection unit. The path type identification structure and path weight allocation structure corresponding to the basic management relationship, the field status relationship, the process record relationship and the electrical installation relationship are set in the multi-relation path projection unit, forming a multi-relation path projection unit that can perform subtype projection according to different relationship types. A multi-source evidence convergence unit is constructed based on the box convergence structure. In the multi-source evidence convergence unit, multiple query boxes are retained for convergence calculation. Evidence convergence parameter access positions are set in the multi-source evidence convergence unit to form a multi-source evidence convergence unit that can receive evidence convergence parameters and participate in convergence calculation. A state adjustment unit is built outside the original structure of the Query2Box model. The state adjustment unit is placed before the query box generation unit, and the output of the state adjustment unit is connected to the state adjustment vector access position of the query box generation unit to form a data connection relationship between the state adjustment unit and the query box generation unit. A box boundary correction unit is built outside the original structure of the Query2Box model. The box boundary correction unit is placed after the multi-source evidence intersection unit, and the output of the multi-source evidence intersection unit is connected to the input of the box boundary correction unit to form a data connection relationship between the multi-source evidence intersection unit and the box boundary correction unit. The state adjustment unit, query box generation unit, multi-relation path projection unit, multi-source evidence convergence unit, and box boundary correction unit are connected in sequence to form the Haiguang box inference model. During the training of the marine photovoltaic box inference model, historical marine photovoltaic multi-source information, historical marine photovoltaic knowledge graph, historical manual verification records, and historical information management results are collected and merged according to the same equipment number, the same spatial location, the same collection time, and the same electrical connection relationship to form a marine photovoltaic box inference training sample set. The historical manual verification records include manually confirmed missing information, conflict information, abnormal status information, abnormal acceptance basis information, and electrical connection traceability information. Based on the marine photovoltaic box inference training sample set, standard marine photovoltaic multi-source information for training is generated, and a marine photovoltaic knowledge graph for training is constructed. Basic management status, field status, process record status, and electrical installation status are extracted from the multi-source information of standard offshore photovoltaic systems used for training, serving as training inputs for the status adjustment unit. Basic management relationships, field status relationships, process record relationships, and electrical installation relationships are extracted from the knowledge graph of offshore photovoltaic systems used for training, serving as training inputs for the multi-relationship path projection unit. Basic management evidence boxes, field status evidence boxes, process record evidence boxes, electrical installation evidence boxes, and evidence intersection parameters are generated from the multi-source information of standard offshore photovoltaic systems used for training, serving as training inputs for the multi-source evidence intersection unit. Missing information, conflicting information, abnormal status information, abnormal acceptance basis information, and electrical connection traceability information confirmed in historical manual verification records are used as training labels for the offshore photovoltaic box inference model. During training, the state adjustment unit generates a state adjustment vector based on the basic management state, on-site state, process record state, and electrical installation state. The query box generation unit generates a training query box based on the state adjustment vector. The multi-relation path projection unit performs a subtype projection on the training query box based on the basic management relationship, on-site state relationship, process record relationship, and electrical installation relationship to generate a training path projection query box. The multi-source evidence convergence unit generates a training evidence convergence query box based on the training path projection query box, basic management evidence box, on-site state evidence box, process record evidence box, electrical installation evidence box, and evidence convergence parameters. The box boundary correction unit corrects the center position and boundary range of the training evidence convergence query box to generate a training correction query box. Based on the training correction query box, it completes the matching in the training marine photovoltaic knowledge graph to obtain the training information management result. The loss is calculated based on the difference between the training information management results and the training labels. For missing and conflicting information, binary cross-entropy loss is used to calculate the difference between the predicted probability of the corresponding category in the training information management results and the confirmation result of the corresponding category in the training labels. For abnormal status information and abnormal acceptance basis information, multi-class cross-entropy loss is used to calculate the difference between the distribution of abnormal categories in the training information management results and the manually confirmed categories in the training labels. For electrical connection traceability information, sorting loss is used to calculate the order difference between the sorting result of the electrical connection traceability path matched by the training correction query box and the manually confirmed electrical connection traceability path in the training labels. For the center position and boundary range of the training correction query box, box boundary constraint loss is used to calculate the boundary deviation generated when the training correction query box covers the correct marine photovoltaic entity and excludes the incorrect marine photovoltaic entity. The total loss of the marine photovoltaic box inference model is generated by weighted summing in the order of missing information loss, conflicting information loss, abnormal status information loss, abnormal acceptance basis information loss, electrical connection traceability information loss, and box boundary constraint loss. The Haiguangbox inference model uses the AdamW optimizer for parameter updates, with an initial learning rate of 0.0005, a batch size of 24, a maximum training epoch of 80, and a weight decay coefficient of 0.0001. Training samples are divided into training, validation, and test sets, with the training set comprising 70%, the validation set 15%, and the test set 15%. After each training epoch, the total loss of the Haiguangbox inference model is calculated on the validation set. The model is considered convergent when the total loss on the validation set decreases by less than 0.00005 for eight consecutive training epochs. Similarly, convergence is considered achieved when the total loss on the validation set stops decreasing after 80 training epochs. After training, the model parameters of the state adjustment unit, query box generation unit, multi-relation path projection unit, multi-source evidence intersection unit, and box boundary correction unit corresponding to the minimum total loss on the validation set are saved, resulting in the trained Haiguangbox inference model.

[0023] In this embodiment, the query box generation module includes: Based on the knowledge graph of marine photovoltaics, the marine photovoltaic entities corresponding to the basic management information are associated with the basic management relationships, and the integrity of design data, consistency of equipment ledgers and spatial layout are extracted to generate the basic management status. Based on the knowledge graph of marine photovoltaics, the marine photovoltaic entities corresponding to the on-site status information are associated with the on-site status relationship, and the operational stability, environmental monitoring effectiveness and alarm status are extracted to generate the on-site status. Based on the knowledge graph of offshore photovoltaics, the offshore photovoltaic entities corresponding to the process record information are associated with the process record relationship, and the construction completion status, inspection and acceptance status and operation and maintenance closed loop status are extracted to generate the process record status. Based on the knowledge graph of offshore photovoltaics, the offshore photovoltaic entities corresponding to electrical installation information are associated with the electrical installation relationship, and the occupancy status of wiring ports, the completion status of cable laying, the consistency status of electrical connection and the qualified status of electrical inspection are extracted to generate the electrical installation status; Based on the basic management status, field status, process record status, and electrical installation status, the data is input to the status adjustment unit. The status adjustment unit then performs numerical encoding and vector concatenation on the basic management status, field status, process record status, and electrical installation status to generate a status adjustment vector. The state adjustment vector is input into the query box generation unit, which then splits the state adjustment vector into a center adjustment amount and a boundary adjustment amount. The center adjustment amount corresponds to the position adjustment content of each dimension of the center vector, and the boundary adjustment amount corresponds to the boundary adjustment content of each dimension of the offset vector. In the query box generation unit, an embedded representation is established based on entities, basic management relationships, field status relationships, process record relationships, and electrical installation relationships in the offshore photovoltaic knowledge graph. Entities are mapped to entity embedding vectors, and basic management relationships, field status relationships, process record relationships, and electrical installation relationships are mapped to relationship embedding vectors respectively. Entity embedding vectors and relationship embedding vectors use the same vector dimension. Based on the basic management status, field status, process record status, and electrical installation status, entities participating in query box generation are determined, and the entity embedding vectors corresponding to the entities participating in query box generation are used as the initial representation. According to the connection order of basic management relationships, field status relationships, process record relationships, and electrical installation relationships in the offshore photovoltaic knowledge graph, the initial representation is sequentially projected onto the corresponding relationship embedding vectors to obtain the relationship path end representation. The relationship path end representation is used as the initial center vector. Based on the entity embedding vectors corresponding to the entities participating in query box generation, the entity distribution range in each vector dimension is calculated, the entity distribution range is converted into a non-negative boundary range, and the non-negative boundary range is used as the initial offset vector. In the query box generation unit, the center adjustment amount is converted into a position adjustment amount with the same dimension as the initial center vector. The position adjustment amount is then added to the initial center vector item by item according to the vector dimension to generate the center vector. The boundary adjustment amount is converted into a boundary expansion amount and a boundary contraction amount with the same dimension as the initial offset vector. The boundary expansion amount and the boundary contraction amount are respectively subjected to non-negativity processing. The non-negativity processed boundary expansion amount is added to the initial offset vector. The non-negativity processed boundary contraction amount is subtracted from the initial offset vector. Vector dimensions that are less than the lower boundary limit after subtraction are replaced according to the lower boundary limit to generate the offset vector. Based on the center vector and offset vector, a query box is constructed to generate a multi-dimensional state adaptive query box for marine photovoltaic systems.

[0024] In this embodiment, the path projection module includes: Based on the basic management relationships, field status relationships, process record relationships, and electrical installation relationships in the offshore photovoltaic knowledge graph, the source and direction of the relationships are identified. The relationship chain formed by the connection of basic management relationships is identified as the basic management path, the relationship chain formed by the connection of field status relationships is identified as the field status path, the relationship chain formed by the connection of process record relationships is identified as the process record path, and the relationship chain formed by the connection of electrical installation relationships is identified as the electrical installation path. Path type identifiers are configured for the basic management path, field status path, process record path, and electrical installation path, respectively. Based on the path type identifier corresponding to the basic management path, the equipment ownership relationship between design data information and equipment information, the asset correspondence relationship between equipment information and asset ledger information, and the spatial arrangement relationship between equipment information and spatial arrangement information are selected from the basic management relationships. The equipment ownership relationship, asset correspondence relationship, and spatial arrangement relationship are encoded according to the relationship type, starting entity type, ending entity type, and connection direction, respectively, to obtain a relationship encoding vector consistent with the dimension of the center vector. The relationship encoding vectors corresponding to the equipment ownership relationship, the asset correspondence relationship, and the spatial arrangement relationship are linearly mapped according to the connection order in the basic management path to obtain the center projection content used to adjust the projection direction of the center vector. Based on the field matching completeness among the design data information, equipment information, asset ledger information, and spatial arrangement information, the boundary projection content used to adjust the boundary range of the offset vector is obtained. The center projection content and the boundary projection content are determined as the projection parameters of the basic management path. Based on the path type identifier corresponding to the field status path, the state correspondence between operational status information and environmental monitoring information is filtered from the field status relationships. The state correspondence is encoded according to the relationship type, starting entity type, ending entity type, and influence direction to obtain a relationship encoding vector consistent with the dimension of the center vector. The relationship encoding vector corresponding to the state correspondence is linearly mapped according to the influence direction of environmental monitoring information pointing to operational status information to obtain the center projection content used to adjust the center vector projection direction. Based on the consistency of acquisition time, spatial location, and degree of correspondence of state changes between operational status information and environmental monitoring information, the boundary projection content used to adjust the boundary range of the offset vector is obtained. The center projection content and the boundary projection content are determined as the field status path projection parameters. Based on the path type identifier corresponding to the process record path, the construction acceptance succession relationship between construction and installation information and inspection and acceptance information, and the acceptance and maintenance succession relationship between inspection and acceptance information and maintenance record information are filtered from the process record relationships. These relationships are then encoded according to relationship type, starting entity type, ending entity type, and record order to obtain a relationship encoding vector consistent with the central vector dimension. The relationship encoding vectors corresponding to the construction acceptance succession relationship and the acceptance and maintenance succession relationship are linearly mapped according to the record order of construction and installation information, inspection and acceptance information, and maintenance record information to obtain the central projection content used to adjust the projection direction of the central vector. Based on the object consistency, record time continuity, and acceptance result closure degree among construction and installation information, inspection and acceptance information, and maintenance record information, the boundary projection content used to adjust the boundary range of the offset vector is obtained. The central projection content and the boundary projection content are determined as the process record path projection parameters. Based on the path type identifier corresponding to the electrical installation path, the following relationships are selected from the electrical installation relationships: port laying relationships between wiring port information and cable laying information, laying connection relationships between cable laying information and electrical connection information, and connection detection relationships between electrical connection information and electrical detection information. These relationships are then encoded according to relationship type, starting entity type, ending entity type, and connection direction to obtain a relationship encoding vector consistent with the dimension of the center vector. The relationship encoding vectors corresponding to port laying relationships, laying connection relationships, and connection detection relationships are linearly mapped according to the connection directions of wiring port information, cable laying information, electrical connection information, and electrical detection information to obtain the center projection content used to adjust the center vector projection direction. Based on the port occupancy consistency, cable path continuity, connection object consistency, and electrical detection qualification status among wiring port information, cable laying information, electrical connection information, and electrical detection information, the boundary projection content used to adjust the boundary range of the offset vector is obtained. The center projection content and boundary projection content are then determined as the electrical installation path projection parameters. The multi-dimensional state adaptive query box of offshore photovoltaic is input into the multi-relation path projection unit. The multi-relation path projection unit parses the center vector and offset vector in the multi-dimensional state adaptive query box of offshore photovoltaic, and calls the basic management path projection parameters, field status path projection parameters, process record path projection parameters and electrical installation path projection parameters according to the path type identifier. In the multi-relationship path projection unit, the center vector is projected according to the basic management path projection parameters to determine equipment affiliation, asset correspondence, and spatial layout direction, and the offset vector is adjusted according to the basic management path projection parameters to generate a basic management path query box. In the multi-relation path projection unit, the center vector is projected with the operating status and environmental monitoring direction according to the field status path projection parameters, and the offset vector is adjusted according to the field status path projection parameters to generate the field status path query box. In the multi-relation path projection unit, the center vector is projected according to the process record path projection parameters to the construction, installation, inspection and acceptance and operation and maintenance record directions, and the offset vector is adjusted according to the process record path projection parameters to generate the process record path query box. In the multi-relation path projection unit, the center vector is projected according to the electrical installation path projection parameters to the wiring port, cable laying, electrical connection and electrical detection direction, and the offset vector is adjusted according to the electrical installation path projection parameters to generate an electrical installation path query box; Based on the basic management path query box, the field status path query box, the process record path query box, and the electrical installation path query box, a merging process is performed. The center positions of the four types of path query boxes are merged according to the path type identifier, and the boundary ranges of the four types of path query boxes are merged according to the path type identifier to generate a path projection query box.

[0025] In this embodiment, the evidence convergence module includes: Based on standard marine photovoltaic multi-source information, in the query box generation unit, the evidence box is determined to be composed of a center vector and an offset vector. The center vector is determined by the data encoding position corresponding to the same information type, and the offset vector is determined by the data encoding range corresponding to the same information type. The center vector represents the evidence position of the corresponding information type in the embedding space, and the offset vector represents the evidence coverage range of the corresponding information type in the embedding space. Based on the design data, equipment information, asset ledger information and spatial layout information in the standard offshore photovoltaic multi-source information, the information is encoded according to field type, equipment number, spatial location and information source to generate the center vector and offset vector corresponding to the basic management information, and the center vector and offset vector corresponding to the basic management information constitute the basic management evidence box. Based on the operational status information and environmental monitoring information in the standard marine photovoltaic multi-source information, the data is encoded according to the collection time, spatial location, equipment number, operational status value and environmental monitoring value to generate the center vector and offset vector corresponding to the field status information, and the field status evidence box is formed by the center vector and offset vector corresponding to the field status information. Based on the construction and installation information, testing and acceptance information, and operation and maintenance record information in the standard offshore photovoltaic multi-source information, the information is encoded according to the equipment number, recording time, recording object, and recording result to generate the center vector and offset vector corresponding to the process record information. The center vector and offset vector corresponding to the process record information constitute the process record evidence box. Based on the wiring port information, cable laying information, electrical connection information and electrical testing information in the standard offshore photovoltaic multi-source information, the installation object, connection object, wiring port, laying path, electrical connection result and electrical testing result are encoded to generate the center vector and offset vector corresponding to the electrical installation information, and the center vector and offset vector corresponding to the electrical installation information constitute the electrical installation evidence box. Based on standard offshore photovoltaic multi-source information, this study calculates data source reliability, acquisition time consistency, spatial location consistency, equipment number consistency, operational status consistency, and electrical connection consistency. Data source reliability is used as a weighting parameter for evidence boxes participating in the intersection calculation; acquisition time consistency is used as a time offset control parameter between different evidence boxes; spatial location consistency is used as a spatial offset control parameter between different evidence boxes; equipment number consistency is used as a device entity matching control parameter; operational status consistency is used as a status deviation control parameter between on-site status evidence boxes and path projection query boxes; and electrical connection consistency is used as a connection deviation control parameter between electrical installation evidence boxes and path projection query boxes. These parameters are then used to calculate the data source reliability, acquisition time consistency, spatial location consistency, equipment number consistency, and operational status consistency. The fixed sequence of evidence convergence parameters for electrical connection consistency is used to generate evidence. Among them, the data source credibility is determined by the completeness of the information source record and the credibility of the test record; the acquisition time consistency is determined by the acquisition time difference between basic management information, on-site status information, process record information and electrical installation information; the spatial location consistency is determined by the spatial location difference between basic management information, on-site status information, process record information and electrical installation information; the equipment number consistency is determined by the matching result of equipment number in basic management information, on-site status information, process record information and electrical installation information; the operating status consistency is determined by the degree of correspondence between the operating status information and the environmental monitoring information; and the electrical connection consistency is determined by the matching result of the connection object between the wiring port information, cable laying information, electrical connection information and electrical test information. Input the path projection query box, basic management evidence box, field status evidence box, process record evidence box, electrical installation evidence box, and evidence intersection parameters into the multi-source evidence intersection unit. Determine the weights of the basic management evidence box, field status evidence box, process record evidence box, and electrical installation evidence box in the intersection calculation according to the reliability of the data source. Adjust the center vector offset between each evidence box according to the consistency of collection time and spatial location. Adjust the offset vector boundary between each evidence box according to the consistency of equipment number, operating status, and electrical connection. In the multi-source evidence convergence unit, the path projection query box is first converged with the basic management evidence box to obtain the baseline convergence box. The baseline convergence box is then converged with the field status evidence box to obtain the status convergence box. The status convergence box is then converged with the process record evidence box to obtain the process convergence box. Finally, the process convergence box is converged with the electrical installation evidence box to generate the evidence convergence query box.

[0026] In this embodiment, the result output module includes: The evidence intersection query box and evidence intersection parameter input box boundary correction unit are used to extract the center position and boundary range from the evidence intersection query box, and the data source credibility, collection time consistency, spatial location consistency, equipment number consistency, operating status consistency and electrical connection consistency are extracted from the evidence intersection parameters. Based on data source reliability, acquisition time consistency, spatial location consistency, equipment number consistency, operating status consistency, and electrical connection consistency, the difference between each parameter and the preset qualified value is calculated in the box boundary correction unit. The difference corresponding to data source reliability is converted into a source reliability correction value; the difference corresponding to acquisition time consistency is converted into a time consistency correction value; the difference corresponding to spatial location consistency is converted into a spatial consistency correction value; the difference corresponding to equipment number consistency is converted into a number consistency correction value; the difference corresponding to operating status consistency is converted into a status consistency correction value; and the difference corresponding to electrical connection consistency is converted into a connection consistency correction value. Based on the source reliability correction value, time consistency correction value, spatial consistency correction value, number consistency correction value, and status... The consistency correction value and the connection consistency correction value are used to weight and synthesize the center position offset direction of the evidence intersection query box to generate the center position correction amount. Based on the source credibility correction value, time consistency correction value, spatial consistency correction value, number consistency correction value, status consistency correction value and connection consistency correction value, the boundary range shrinkage amplitude of the evidence intersection query box is weighted and synthesized to generate the boundary range correction amount. The center position correction amount and the boundary range correction amount are combined into the boundary correction amount. When there are parameters in the data source credibility, collection time consistency, spatial location consistency, equipment number consistency, operating status consistency and electrical connection consistency that are lower than the preset qualified value, the boundary range shrinkage amplitude is increased according to the number of parameters that are lower than the preset qualified value, and the matching threshold is increased simultaneously. The boundary correction direction is determined based on the positional relationship between the evidence convergence and query box, the basic management evidence box, the field status evidence box, the process record evidence box, and the electrical installation evidence box. When the center positions of the basic management evidence box, the field status evidence box, the process record evidence box, and the electrical installation evidence box are all within the boundary range of the evidence convergence and query box, and the boundary ranges of the basic management evidence box, the field status evidence box, the process record evidence box, and the electrical installation evidence box all overlap with the boundary range of the evidence convergence and query box, the boundary correction direction is determined as the boundary expansion direction. When there is an evidence box among the basic management evidence box, the field status evidence box, the process record evidence box, and the electrical installation evidence box whose center position is outside the boundary range of the evidence convergence and query box, the boundary correction direction is determined as the boundary preservation direction. When the electrical connection consistency is lower than the preset qualified value, the correction amount of the boundary range corresponding to the electrical installation relationship is reduced, and the data content of inconsistent connection objects in the electrical installation information is identified as electrical connection traceability information. The center position of the evidence intersection query box is corrected based on the center position correction amount, the boundary range of the evidence intersection query box is corrected based on the boundary range correction amount, and the matching conditions of the correction query box are configured according to the matching threshold to generate the correction query box. Based on the calibration query box, the offshore photovoltaic entities, basic management relationships, on-site status relationships, process record relationships and electrical installation relationships are matched in the offshore photovoltaic knowledge graph. The offshore photovoltaic entities, basic management relationships, on-site status relationships, process record relationships and electrical installation relationships that fall into the calibration query box and meet the matching threshold are identified as consistent information. Based on the matching results of the correction query box in the offshore photovoltaic knowledge graph, the data content that does not match the corresponding offshore photovoltaic entity, basic management relationship, field status relationship, process record relationship and electrical installation relationship is identified as missing information, and the data content that is outside the boundary range of the correction query box but has a corresponding relationship with the path projection query box is identified as conflict information. Based on the matching results of the calibration query box in the marine photovoltaic knowledge graph, data content with the same equipment number but inconsistent status content in the field status relationship and process record relationship is identified as status abnormal information, and data content with the same equipment number but inconsistent acceptance content and electrical installation content in the inspection and acceptance information and electrical installation information is identified as acceptance basis abnormal information. The system summarizes missing information, conflict information, abnormal status information, abnormal acceptance basis information, and electrical connection traceability information to generate marine photovoltaic information management results.

[0027] Example 1: To verify the feasibility of this invention in practice, it was applied to a marine photovoltaic (PV) information management scenario. In this scenario, marine PV projects generate a large amount of scattered data during construction, acceptance, and operation management. The data includes basic management information, on-site status information, process record information, and electrical installation information. Basic management information suffers from inconsistencies between design documents and equipment ledgers; on-site status information shows a lack of effective correspondence between operational status information and environmental monitoring information; process record information exhibits incomplete connections between construction and installation information, testing and acceptance information, and operation and maintenance records; and electrical installation information suffers from unclear traceability relationships between wiring port information, cable laying information, electrical connection information, and electrical testing information. Traditional management methods primarily rely on manual ledger verification and field retrieval, which can identify obvious missing items but struggles to identify conflicting content across information types. This is especially problematic when there are indirect correlations between equipment numbers, spatial layouts, wiring ports, and electrical testing results, leading to delayed anomaly location and incomplete acceptance criteria.

[0028] In this scenario, a total of 128,460 pieces of multi-source information on offshore photovoltaic systems were collected, including 23,680 pieces of basic management information, 41,850 pieces of on-site status information, 31,270 pieces of process record information, and 31,660 pieces of electrical installation information. After performing field unification, time alignment, equipment number normalization, and invalid field removal on the multi-source information of offshore photovoltaic systems, standard offshore photovoltaic multi-source information was formed. After processing, 4,186 records with invalid fields were removed, 27,140 records had their equipment numbers normalized, and 38,920 records had their timestamps aligned, forming a data foundation that can be used for subsequent map construction and box inference processing.

[0029] During application, offshore photovoltaic (PV) entities and relationships are extracted based on standard multi-source information to form an offshore PV knowledge graph. The PV entities cover equipment, space, status, process, and electrical installation entities. The relationships cover basic management, on-site status, process record, and electrical installation relationships. The offshore PV knowledge graph contains 18,436 entities and 52,680 relationships, with electrical installation relationships accounting for a significant proportion, providing clearer data associations for electrical connection traceability and acceptance verification.

[0030] A marine photovoltaic (PV) box inference model is constructed based on a knowledge graph. This model retains the box representation structure, relation projection structure, and box intersection structure from the Query2Box model, and includes a state adjustment unit, a query box generation unit, a multi-relation path projection unit, a multi-source evidence intersection unit, and a box boundary correction unit. The state adjustment unit generates a state adjustment vector based on the basic management status, field status, process record status, and electrical installation status. The query box generation unit generates a multi-dimensional adaptive query box for marine PV status based on the state adjustment vector. The multi-relation path projection unit performs segmented projection along the basic management path, field status path, process record path, and electrical installation path. The multi-source evidence intersection unit performs intersection calculations based on evidence intersection parameters. The box boundary correction unit corrects the center position and boundary range of the evidence intersection query box. Through this process, basic management information, field status information, process record information, and electrical installation information no longer participate in retrieval as isolated fields, but instead participate in the generation of information management results through relation paths, evidence boxes, and boundary correction.

[0031] To verify the effectiveness, the traditional ledger verification method, the ordinary knowledge graph retrieval method, and the method of this invention were compared on the same batch of standard offshore photovoltaic multi-source information. The traditional ledger verification method uses manual rules and field consistency checks, the ordinary knowledge graph retrieval method uses entity relationship retrieval and fixed rule matching, and the method of this invention uses an offshore photovoltaic knowledge graph, a marine photovoltaic box reasoning model, a marine photovoltaic multi-dimensional state adaptive query box, a path projection query box, an evidence convergence query box, and a correction query box to generate offshore photovoltaic information management results. In the test data, manual review confirmed 312 missing information entries, 286 conflicting information entries, 241 abnormal status information entries, 198 abnormal acceptance basis information entries, and 354 electrical connection traceability information entries.

[0032] Table 1 Comparison of Information Management Effectiveness of Offshore Photovoltaic Projects

[0033] As shown in Table 1, the method of this invention outperforms traditional ledger verification methods and ordinary knowledge graph retrieval methods in all indicators. The missing information identification rate is increased from 76.8% in the traditional ledger verification method to 90.6%. The performance improvement is due to the fact that this invention does not only perform integrity checks on a single field, but also establishes associations between basic management information, on-site status information, process record information, and electrical installation information through a marine photovoltaic knowledge graph. When a certain equipment entity lacks corresponding spatial layout information, inspection and acceptance information, or electrical connection information, the missing content can be found through relational content and query box matching results.

[0034] The conflict information identification rate increased from 82.1% for ordinary knowledge graph retrieval methods to 89.4%. This performance improvement is due to the fact that ordinary knowledge graph retrieval methods primarily rely on fixed relationship queries, which can detect conflicts within direct relationships but have limited ability to handle cross-path conflicts. The method of this invention uses a categorized projection approach, encompassing basic management paths, field status paths, process record paths, and electrical installation paths. This allows information corresponding to different relationship types to be included in separate path projection query boxes, and then unified through an evidence convergence query box, thereby improving the ability to identify cross-information type conflicts.

[0035] The anomaly status information identification rate reached 88.2%, an improvement of 18.5 percentage points compared to traditional ledger verification methods. This improvement stems from the processing of the status correspondence between on-site status information and environmental monitoring information. Traditional methods typically only determine whether operational status information exceeds a fixed threshold, making it difficult to combine environmental monitoring information to determine whether status changes are reasonable. The method of this invention inputs both the on-site status evidence box and the path projection query box into a multi-source evidence convergence unit, enabling the establishment of a correspondence between operational status information, environmental monitoring information, and equipment entities, thus resulting in more stable identification of anomaly status information.

[0036] The identification rate of abnormal information in the acceptance criteria reached 86.8%. In offshore photovoltaic projects, some abnormalities in the acceptance criteria do not come from the lack of single inspection and acceptance information, but from inconsistencies between construction and installation information, inspection and acceptance information, operation and maintenance record information, and electrical installation information. The method of this invention uses process record relationships and electrical installation relationships to participate in the evidence convergence, which can identify the inconsistencies between inspection and acceptance information and electrical installation information. Therefore, it is more suitable for the verification of acceptance criteria than simply searching for inspection and acceptance information.

[0037] The accuracy rate of electrical connection traceability reached 91.2%. This performance improvement is due to the need to simultaneously combine wiring port information, cable laying information, electrical connection information, and electrical testing information. Traditional ledger verification methods are easily affected by inconsistent equipment numbers and scattered connection records. While ordinary knowledge graph retrieval methods can establish connections, they do not adequately utilize electrical connection consistency and evidence intersection parameters. This invention's method corrects the evidence intersection query box through an electrical installation evidence box, an electrical connection consistency correction unit, and a box boundary correction unit. This reduces mismatches caused by excessively wide boundaries and improves the accuracy of electrical connection traceability information.

[0038] The average processing time per batch was reduced from 46.5 minutes using the traditional ledger verification method to 21.4 minutes. This is primarily because, after constructing the standard multi-source information for marine photovoltaic systems, the method of this invention can perform batch matching through the marine photovoltaic knowledge graph and the marine photovoltaic box inference model, eliminating the need for manual verification of each ledger field. Compared to ordinary knowledge graph retrieval methods, although the method of this invention adds state adjustment, fractal projection, evidence convergence, and box boundary correction processing, it reduces a significant amount of repetitive queries and manual verification work, thus still reducing the overall processing time.

[0039] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A knowledge graph-based marine photovoltaic information management system, characterized in that, include: The information acquisition module is used to collect multi-source information of offshore photovoltaic systems, including basic management information, on-site status information, process record information and electrical installation information. It performs field unification, time alignment, equipment number normalization and invalid field removal to generate standard offshore photovoltaic multi-source information. The knowledge graph construction module is used to extract marine photovoltaic entities and relationships based on standard marine photovoltaic multi-source information and generate a marine photovoltaic knowledge graph. The model building module is used to build the Haiguang box inference model based on the Query2Box model, including the state adjustment unit, query box generation unit, multi-relation path projection unit, multi-source evidence convergence unit, and box boundary correction unit. The query box generation module is used to generate a state adjustment vector in the state adjustment unit based on the knowledge graph of marine photovoltaics. The state adjustment vector is then input into the query box generation unit to generate a multi-dimensional state adaptive query box for marine photovoltaics. The path projection module is used to input the multi-dimensional state adaptive query box of marine photovoltaic power into the multi-relation path projection unit, perform type projection, and generate the path projection query box. The evidence convergence module is used to generate evidence convergence parameters based on standard marine photovoltaic multi-source information, input multi-source evidence convergence units for convergence calculation, and generate evidence convergence query boxes; The results output module is used to input the evidence convergence query box into the box boundary correction unit, correct the center position and boundary range, generate a correction query box, and perform matching based on the correction query box to generate marine photovoltaic information management results.

2. The marine photovoltaic information management system based on knowledge graphs according to claim 1, characterized in that, The basic management information includes design data, equipment information, asset ledger information, and spatial layout information; the on-site status information includes operational status information and environmental monitoring information; the process record information includes construction and installation information, testing and acceptance information, and operation and maintenance record information; and the electrical installation information includes wiring port information, cable laying information, electrical connection information, and electrical testing information.

3. The marine photovoltaic information management system based on knowledge graphs according to claim 1, characterized in that, The map construction module includes: Based on standard offshore photovoltaic multi-source information, the entity types are determined according to the information sources corresponding to basic management information, on-site status information, process record information and electrical installation information. Equipment entities, spatial entities, status entities, process entities and electrical installation entities are determined, and offshore photovoltaic entities are generated. Based on the multi-source information of offshore photovoltaic entities and standard offshore photovoltaic systems, basic management relationships are extracted from basic management information, on-site status relationships are extracted from on-site status information, process record relationships are extracted from process record information, and electrical installation relationships are extracted from electrical installation information. The offshore photovoltaic entity is written into the graph structure as an entity, and the basic management relationship, on-site status relationship, process record relationship and electrical installation relationship are written into the graph structure as relationship content to generate an offshore photovoltaic knowledge graph.

4. The marine photovoltaic information management system based on knowledge graphs according to claim 1, characterized in that, The model building module includes: Call the original structure of the Query2Box model, including the box representation structure, relational projection structure, and box intersection structure; A query box generation unit is constructed based on the box representation structure. The center vector and the offset vector are retained to form the box representation structure of the query box, and the access position of the state adjustment vector is set. Based on the relation projection structure, a multi-relation path projection unit is constructed. The structure of projecting the query box according to the relations in the knowledge graph is retained, and the path type identification structure and path weight allocation structure corresponding to the basic management relationship, the field status relationship, the process record relationship and the electrical installation relationship are set. A multi-source evidence aggregation unit is constructed based on the box aggregation structure, retaining the structure of multiple query boxes for aggregation calculation, and setting the access position of evidence aggregation parameters; A state adjustment unit is built outside the original structure of the Query2Box model and set before the query box generation unit; a box boundary correction unit is built and set after the multi-source evidence convergence unit. The state adjustment unit, query box generation unit, multi-relation path projection unit, multi-source evidence convergence unit, and box boundary correction unit are connected in sequence to form the Haiguang box inference model.

5. A knowledge graph-based marine photovoltaic information management system according to claim 1, characterized in that, The query box generation module includes: Based on the knowledge graph of marine photovoltaics, the marine photovoltaic entities corresponding to the basic management information are associated with the basic management relationships, and the integrity of design data, consistency of equipment ledgers and spatial layout are extracted to generate the basic management status. The on-site status information is associated with the marine photovoltaic entities and the on-site status relationship. The operational stability, environmental monitoring effectiveness and alarm status are extracted to generate the on-site status. The process record information is associated with the marine photovoltaic entity and the process record relationship, and the construction completion status, inspection and acceptance status and operation and maintenance closed loop status are extracted to generate the process record status. The system associates the offshore photovoltaic entities corresponding to the electrical installation information with the electrical installation relationship, extracts the occupancy status of the wiring ports, the completion status of cable laying, the consistency status of electrical connections, and the qualified status of electrical testing, and generates the electrical installation status. Based on the basic management status, field status, process record status, and electrical installation status, the input status adjustment unit performs numerical encoding and vector concatenation to generate a status adjustment vector. The state adjustment vector is input into the query box generation unit, and the state adjustment vector is split into the center adjustment amount and the boundary adjustment amount. In the query box generation unit, an initial center vector and an initial offset vector are constructed based on the marine photovoltaic knowledge graph; In the query box generation unit, the center adjustment amount is converted into the position adjustment amount, which is then superimposed onto the initial center vector to generate the center vector. The boundary adjustment amount is converted into the boundary expansion amount and the boundary contraction amount, which are then superimposed onto the initial offset vector to generate the offset vector. The query box is then constructed to generate the marine photovoltaic multi-dimensional state adaptive query box.

6. A knowledge graph-based marine photovoltaic information management system according to claim 1, characterized in that, The path projection module includes: Based on the basic management relationships, field status relationships, process record relationships, and electrical installation relationships in the offshore photovoltaic knowledge graph, the source and direction of the relationships are identified, the basic management path, field status path, process record path, and electrical installation path are determined, and path type identifiers are configured for each. Based on the path type identifier corresponding to the basic management path, the equipment ownership relationship between design data information and equipment information, the asset correspondence relationship between equipment information and asset ledger information, and the spatial arrangement relationship between equipment information and spatial arrangement information are filtered from the basic management relationship to determine the projection parameters of the basic management path; Based on the path type identifier corresponding to the field status path, the state correspondence between the operation status information and the environmental monitoring information is filtered from the field status relationship to determine the field status path projection parameters. Based on the path type identifier corresponding to the process record path, the construction acceptance succession relationship between construction and installation information and inspection and acceptance information, and the acceptance and operation and maintenance succession relationship between inspection and acceptance information and operation and maintenance record information are filtered from the process record relationship to determine the process record path projection parameters; Based on the path type identifier corresponding to the electrical installation path, the port laying relationship between wiring port information and cable laying information, the laying connection relationship between cable laying information and electrical connection information, and the connection detection relationship between electrical connection information and electrical detection information are filtered from the electrical installation relationship to determine the electrical installation path projection parameters; Input the multi-dimensional state adaptive query box of marine photovoltaic system into the multi-relation path projection unit to analyze the center vector and offset vector in the multi-dimensional state adaptive query box of marine photovoltaic system. In the multi-relationship path projection unit, the center vector is projected according to the basic management path projection parameters to determine equipment affiliation, asset correspondence, and spatial layout direction, and the offset vector is adjusted to generate a basic management path query box. Project the center vector into the operating status and environmental monitoring direction according to the on-site status path projection parameters, and adjust the boundary of the offset vector to generate the on-site status path query box. Project the center vector into the construction, installation, inspection, acceptance, and operation and maintenance record directions according to the process record path projection parameters, and adjust the boundary of the offset vector to generate a process record path query box; Based on the electrical installation path projection parameters, the center vector is projected to the wiring ports, cable laying, electrical connections, and electrical testing directions, and the offset vector is adjusted to generate an electrical installation path query box. Based on the basic management path query box, the field status path query box, the process record path query box, and the electrical installation path query box, a convergence process is performed to generate a path projection query box.

7. A knowledge graph-based marine photovoltaic information management system according to claim 1, characterized in that, The evidence exchange module includes: Based on the design data, equipment information, asset ledger information and spatial layout information in the standard marine photovoltaic multi-source information, the information is encoded according to field type, equipment number, spatial location and information source to generate the center vector and offset vector corresponding to the basic management information, thus forming the basic management evidence box. Based on operational status information and environmental monitoring information, the data is encoded according to the collection time, spatial location, equipment number, operational status value, and environmental monitoring value to generate the center vector and offset vector corresponding to the field status information, thus forming a field status evidence box. Based on construction and installation information, testing and acceptance information, and operation and maintenance record information, the process record information is encoded according to equipment number, recording time, recording object, and recording result to generate the center vector and offset vector corresponding to the process record information, thus forming a process record evidence box. Based on the wiring port information, cable laying information, electrical connection information and electrical testing information, the electrical installation information is encoded according to the installation object, connection object, wiring port, laying path, electrical connection result and electrical testing result, and the center vector and offset vector corresponding to the electrical installation information are generated to form an electrical installation evidence box; Based on standard offshore photovoltaic multi-source information, the reliability of data sources, consistency of collection time, consistency of spatial location, consistency of equipment number, consistency of operating status, and consistency of electrical connection are calculated respectively to generate evidence convergence parameters; Input the path projection query box, basic management evidence box, on-site status evidence box, process record evidence box, electrical installation evidence box, and evidence intersection parameters into the multi-source evidence intersection unit, perform intersection calculations, and generate the evidence intersection query box.

8. A knowledge graph-based marine photovoltaic information management system according to claim 1, characterized in that, The result output module includes: Input the evidence intersection query box and evidence intersection parameters into the box boundary correction unit; Based on the reliability of data sources, consistency of collection time, consistency of spatial location, consistency of equipment number, consistency of operating status, and consistency of electrical connection, the difference between each parameter item and the preset qualified value is calculated in the box boundary correction unit, and weighted synthesis is performed to generate the center position correction amount and the boundary range correction amount, which are combined into the boundary correction amount. The boundary correction direction is determined based on the positional relationship between the evidence intersection query box, the basic management evidence box, the on-site status evidence box, the process record evidence box, and the electrical installation evidence box. When the electrical connection consistency is lower than the preset qualified value, the correction amount of the boundary range corresponding to the electrical installation relationship is reduced, and the data content of inconsistent connection objects in the electrical installation information is identified as electrical connection traceability information. The center position of the evidence intersection query box is corrected based on the center position correction amount, and the boundary range of the evidence intersection query box is corrected based on the boundary range correction amount to generate a corrected query box. Based on the calibration query box, the offshore photovoltaic entities, basic management relationships, on-site status relationships, process record relationships and electrical installation relationships are matched in the offshore photovoltaic knowledge graph to determine consistent information, missing information, conflicting information, abnormal status information and abnormal acceptance basis information. The system summarizes missing information, conflict information, abnormal status information, abnormal acceptance basis information, and electrical connection traceability information to generate marine photovoltaic information management results.

Images