Intelligent pavement maintenance scheme design method and device based on dynamic knowledge graph

Abstract

The present invention aims to provide an intelligent pavement maintenance plan design method based on a dynamic knowledge graph to solve the problem that the pavement maintenance plan design in the prior art relies on manual work and lacks dynamic adjustment capabilities. The present invention receives road infrastructure information, pavement detection conditions, pavement disease descriptions and grading and classification information, and constructs a dynamic knowledge graph based on the input entities and their relationships. The dynamic knowledge graph is used to output an applicable pavement maintenance plan template, including maintenance content, maintenance methods, disposal drawings, engineering quantity estimates, and engineering schedules. By introducing dynamic knowledge graphs and large model technology, the pavement condition is automatically analyzed and judged, a maintenance plan template with strong adaptability is generated, and the plan content is intelligently assigned and generated.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a method and device for designing an intelligent pavement maintenance plan based on a dynamic knowledge graph. Background Art

[0002] With the development of transportation and the continuous construction of roads, the total mileage of roads in my country has continued to increase, and road maintenance issues have received more and more attention. In particular, road surface maintenance is closely related to road traffic safety and road service life, and occupies an important position in infrastructure maintenance.

[0003] Traditional pavement maintenance mainly relies on manual inspections and repairs, which is inefficient and costly. In addition, the pavement maintenance requirements and processes for different types of roads are different, and targeted maintenance plans need to be formulated based on the different characteristics of the roads. Existing pavement maintenance plans usually rely on manual inspections and empirical judgments, which are not only time-consuming and labor-intensive, but also limited by the technical level of personnel. At present, there is no systematic system that can automatically generate maintenance plans. Summary of the invention

[0004] In view of the above problems, the present invention proposes an intelligent pavement maintenance plan design method and device based on dynamic knowledge graph. The present invention aims to provide an intelligent pavement maintenance plan design method based on dynamic knowledge graph to solve the problem that the pavement maintenance plan design in the prior art relies on manual work and lacks dynamic adjustment capabilities. By introducing dynamic knowledge graph and large model technology, the system can automatically analyze and judge the pavement condition, generate a maintenance plan template with strong adaptability, and intelligently assign and generate the plan content.

[0005] In order to solve the above technical problems, the present invention discloses the following technical solutions.

[0006] As a first aspect of the present invention, the present invention discloses an intelligent pavement maintenance scheme design method based on a dynamic knowledge graph, which is characterized by:

[0007] Receive road infrastructure information, road surface inspection conditions, road surface disease descriptions and classification information, and use them as entities in the knowledge graph, as well as entity attribute descriptions and preliminary entity relationship descriptions;

[0008] Based on the input entities and their mutual relations, a dynamic knowledge graph is constructed, and the entity types are automatically analyzed and determined by machine learning technology to establish relational links between entities; wherein, an association network between entities in the knowledge graph of road maintenance is constructed; and according to the entities and their preliminary entity relations, the dynamic knowledge graph is queried to assign values ​​to entities of known types therein, and the entity types of unlabeled entities are identified based on the association network of the dynamic knowledge graph through a first deep learning model; and the relational links between entities are predicted based on a feature representation matrix of relevant entity types through a second deep learning model;

[0009] Use dynamic knowledge graphs to output applicable pavement maintenance plan templates, including maintenance content, maintenance methods, disposal drawings, project quantity estimates, and project schedules.

[0010] Preferably, the road infrastructure information covers information on the basic structure and properties of the road; information on road surface detection conditions includes: road surface flatness, road surface friction status assessment, road surface gap status assessment; road surface disease description includes a detailed description of the hierarchical classification information of the disease location, disease type, severity and impact range of various diseases occurring on the road surface.

[0011] Preferably, the association network between the knowledge graph entities of road maintenance is represented as ;in, Represents the set of all entities, where That is, entities related to road maintenance; entity The corresponding entity type, A set of types representing the relationship between two entities in the association network between knowledge graph entities; a set of triples Representing Entities and There is a relationship between ,but is the union of the above triple sets.

[0012] Preferably, the first deep learning model is expressed as , using an RNN neural network architecture, where the input layer of the model receives the input, namely, the association network of the dynamic knowledge graph represented by the entities, entity types, entity relationships and their triple sets , is the parameter matrix of each hidden layer of the neural network of this model The parameter vector is composed of It is The unknown weight parameter matrix of the layer, , is the total number of hidden layers.

[0013] Preferably, the training of the first deep learning model includes: training the parameter vector of the first deep learning model Initialize to , that is, all parameter matrices of each hidden layer are initialized to , then for any entity in the association network of the dynamic knowledge graph , calculated by deep learning model in the first The feature representation obtained by the hidden layer , specifically:

[0014]

[0015] in is the activation function, For this entity Have a relationship The collection of other entities, is a predetermined constant, with a value of , Representing a collection The number of elements, Representing a collection The The entity in The feature representation output by the hidden layer is For this entity In the The feature representation output by the hidden layer; for entity , the first deep learning model obtains feature representation layer by layer until the last hidden layer, i.e., the Feature representation of the hidden layer ; The first deep learning model also includes a softmax classification layer as the first layer, connected to the last hidden layer, to obtain the feature representation As the input of the softmax classification layer, we get:

[0016]

[0017] The classification layer output is a K-dimensional vector, the vector Elements Indicates that according to the entity Corresponding Determine the entity Belong to The probability of the class type is:

[0018]

[0019] here ; Output from the classification layer , where the type with the highest probability is taken as the entity type; and then, by using sample training, optimizing the parameter vector of the first deep learning model , for the unknown weight parameter matrix of each hidden layer Assign values ​​until the loss function meets expectations.

[0020] Preferably, the second deep learning model , based on the relevant entity type feature representation matrix, predict the relationship links between entities; where, Represents the first deep learning model The matrix composed of the feature representations of the corresponding related entity types output by the last hidden layer is called the related entity type feature representation matrix. Represents the graph decoder network parameters.

[0021] Preferably, the second deep learning model includes a softmax layer for predicting relationship links between entities, and the softmax layer calculates the predicted probability of a relationship link between related entities, and the calculation formula is: , ,in, Represents the relationship The parameter matrix of Representing relationships The probability label of and Relationship Type When ,otherwise .

[0022] Preferably, based on the big model technology, specific maintenance plan item content is generated according to the entity information extracted from the dynamic knowledge graph, and the plan items of the pavement maintenance plan are assigned values ​​and corresponding content is generated.

[0023] Preferably, using big model technology to assign values ​​to solution items specifically includes: extracting relevant entity information from a dynamic knowledge graph; using big model technology to perform in-depth analysis and reasoning on the collected entity information; generating specific values ​​for each solution item based on the reasoning results of the big model; dynamically adjusting the generated solution content through a feedback mechanism, and continuously optimizing the generated solution items based on the latest pavement inspection data and construction feedback; verifying the generated solution to ensure that it meets industry standards and actual needs; after verification, the generated solution content is output in an operational format for reference and execution by the construction unit.

[0024] On the other hand, the present invention relates to an intelligent pavement maintenance scheme design system based on a dynamic knowledge graph, characterized in that it includes:

[0025] The pavement maintenance related information collection module receives road infrastructure information, pavement inspection conditions, pavement disease descriptions and classification information, and uses them as entities in the knowledge graph as well as entity attribute descriptions and preliminary entity relationship descriptions;

[0026] The dynamic knowledge graph module constructs a dynamic knowledge graph based on the input entities and their mutual relationships, and uses machine learning technology to automatically analyze and determine the entity types and establish relationship links between entities; wherein, an association network between entities in the knowledge graph of road maintenance is constructed; and according to the entities and their preliminary entity relationships, the dynamic knowledge graph is queried to assign values ​​to entities of known types therein, and the entity types of unlabeled entities are identified based on the association network of the dynamic knowledge graph through a first deep learning model; and the relationship links between entities are predicted based on the relevant entity type feature representation matrix through a second deep learning model;

[0027] The maintenance plan generation module uses a dynamic knowledge graph to output applicable pavement maintenance plan templates, including maintenance content, maintenance methods, disposal drawings, project quantity estimates, and project schedules.

[0028] The present invention solves the problem that the existing road maintenance scheme design relies on manual work and lacks dynamic adjustment capabilities. By introducing dynamic knowledge graphs and large model technology, the system can automatically analyze and judge road conditions, generate highly adaptable maintenance scheme templates, and intelligently assign and generate scheme contents. This not only improves the efficiency and accuracy of scheme design, but also enables real-time adjustments based on actual conditions, thereby optimizing the effect of road maintenance and resource allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain and illustrate the present invention, but should not be construed as limiting the protection scope of the present invention.

[0030] Figure 1 It is a flow chart of an intelligent pavement maintenance scheme design method based on a dynamic knowledge graph disclosed in the present invention.

[0031] Figure 2 It is a structural diagram of an intelligent pavement maintenance plan design system based on a dynamic knowledge graph disclosed in the invention. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solution and advantages of the implementation of the present invention clearer, the technical solution in the embodiment of the present invention will be described in more detail below with reference to the accompanying drawings in the embodiment of the present invention.

[0033] It should be noted that: in the accompanying drawings, the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The described embodiments are part of the embodiments of the present invention, not all of the embodiments. In the absence of conflict, the embodiments in this application and the features in the embodiments can be combined with each other. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] The present invention aims to provide an intelligent pavement maintenance plan design method based on dynamic knowledge graph to solve the problem that the pavement maintenance plan design in the prior art relies on manual work and lacks dynamic adjustment capabilities. By introducing dynamic knowledge graph and large model technology, the system can automatically analyze and judge the pavement condition, generate a maintenance plan template with strong adaptability, and intelligently assign and generate the plan content. The specific steps of the design method can be found in Figure 1 , including the following steps.

[0035] Step S1: Receive road infrastructure information, road surface inspection conditions, road surface disease descriptions and classification information, and use them as entities of the knowledge graph as well as entity attribute descriptions and preliminary entity relationship descriptions.

[0036] Specifically, road infrastructure information covers the basic structure and properties of roads, such as the length, width, material type, construction year, and designed load capacity of roads. This information can be obtained by querying the database that records road planning and construction archives. Information on road surface detection conditions includes but is not limited to road surface flatness, road surface friction status assessment, and road surface gap status assessment. These data are usually collected by road surface detection equipment, which integrates multiple sensors such as road surface three-dimensional laser radar, and after preliminary data processing and cleaning, the above road surface detection status information is extracted through pattern recognition. Road surface disease description, this information involves a detailed description of various diseases that appear on the road surface, such as cracks, potholes, spalling, subsidence, etc. These descriptions include not only the location and type of disease, but also the classification information of its severity and scope of impact, so as to better manage and make decisions. The classification is usually based on the severity of the disease, while the classification is based on the type and cause of the disease. When converting the above information into knowledge graph entities, the data format needs to be standardized and consistent. Each entity contains relevant entity attribute descriptions. For example, if there is road subsidence, the road disease description is taken as an entity, and the hierarchical classification information of the disease location, disease type, severity and impact range is taken as the entity attribute description. The relationship between entities is taken as the entity relationship description, so as to facilitate effective storage and retrieval in the knowledge graph.

[0037] Step S2: Based on the input entities and their relationships, a dynamic knowledge graph is constructed, and machine learning technology is used to automatically analyze and determine entity types and establish relationship links between entities.

[0038] In this step, the various entities and their relationships provided in step S1 are converted into a structured dynamic knowledge graph. This process not only involves simple mapping of entities, but also requires the use of machine learning technology to automatically identify and classify entity types and establish relationship links between them, thereby forming a dynamic and scalable knowledge network.

[0039] Specifically, we first accept entity data, which includes road infrastructure information, road surface inspection conditions, road surface disease descriptions, and classification information. Each entity contains several entity attribute descriptions and possible preliminary relationships between entities. The task of this step is to further analyze and process these entities.

[0040] Among them, in the entity analysis stage, machine learning algorithms are used to identify the specific types of entities. For example, natural language processing (NLP) technology is used to perform text analysis on the description of pavement defects to identify specific types and characteristics of defects. This can be achieved by training a supervised learning model, which learns how to identify and classify different types of defects through a large amount of labeled data. Next, the establishment of relationship links is achieved through graph algorithms. In the knowledge graph, the relationship between entities can be represented as edges in the graph. For each pair of related entities, the algorithm will determine whether there is a relationship and the specific type of relationship based on their attribute values ​​and business rules. For example, there may be an entity relationship between the "material type" entity in the road infrastructure information and the "pavement friction state assessment" entity in the pavement detection condition.

[0041] More specifically, the association network between entities in the knowledge graph of pavement maintenance is represented as ;in, Represents the set of all entities, where That is, entities related to road maintenance; entity The corresponding entity types include but are not limited to various types of road infrastructure information, road surface inspection conditions, road surface disease descriptions and classification information, namely, entity The entity type is ,in is a dimensional one-hot array, if the entity The type belongs to class type, the array Elements The value is 1, array The other elements of are set to 0. The value of is the total number of types covered by various aspects such as road infrastructure information, road surface inspection conditions, road surface disease descriptions, and classification information. For example, the types of road surface disease descriptions include cracks, potholes, spalling, subsidence, and other types of diseases. Represents the type set of relationships between two entities in the association network between knowledge graph entities, the relationship between entities, for example, crack disease-belongs to-crack type, repair measures-applicable to-crack type, detection condition A-identified-pothole disease, detection condition B-assessed as-surface wear, detection condition C-characterized-road surface flatness, disease image features-identified-disease area, etc. Use triple set Representing Entities and There is a relationship between ,but is the union of the above triple sets.

[0042] This step uses machine learning technology to automatically analyze and determine entity types and establish relationship links between entities. Query the existing dynamic knowledge graph separately. The entities that have been annotated in the dynamic knowledge graph can determine their types and the corresponding entities can be found according to their types. Assignment, that is, after determining the entity The type belongs to Class type, then One-hot array Elements The value is 1, array The other elements of are set to 0. If the entity is not labeled in the dynamic knowledge graph after query, the type of the entity is temporarily not labeled. Further, the entity type of the unlabeled entity is identified based on the association network of the dynamic knowledge graph through the first deep learning model. The first deep learning model is represented as , an RNN neural network architecture can be used, wherein the input layer of the model receives the input, that is, the association network of the dynamic knowledge graph represented by the entities, entity types, entity relationships and their triple sets. , is the parameter matrix of each hidden layer of the neural network of this model The parameter vector is composed of It is The unknown weight parameter matrix of the layer, , is the total number of hidden layers. The parameter vector of the deep learning model Initialize to , that is, all parameter matrices of each hidden layer are initialized to , then for any entity in the association network of the dynamic knowledge graph , calculated by deep learning model in the first The feature representation obtained by the hidden layer , specifically:

[0043]

[0044] in is the activation function, For this entity Have a relationship The collection of other entities, is a predetermined constant, which can generally be taken as , Representing a collection The number of elements, Representing a collection The The entity in The feature representation output by the hidden layer is similar. For this entity In the The feature representation output by the hidden layer. , the deep learning model obtains feature representation layer by layer until the last hidden layer, i.e. Feature representation of the hidden layer The deep learning model also includes a softmax classification layer as the first layer, connected to the last hidden layer, to obtain the feature representation As the input of the softmax classification layer, we get:

[0045]

[0046] The classification layer output is a K-dimensional vector, the vector Elements Indicates that according to the entity Corresponding Determine the entity Belong to The probability of the class type is:

[0047]

[0048] here . The output of the classification layer , where the type with the highest probability is taken as the entity Then, by using sample training, the parameter vector of the deep learning model is optimized. , that is, the unknown weight parameter matrix of each hidden layer Assign values ​​until the loss function meets expectations. After completing the training of the deep learning model, the model can recognize and output the types of entities other than the labeled entities for the various entities and the preliminary relationships between the entities provided in step S1.

[0049] Based on the clear entity type, the second deep learning model , based on the relevant entity type feature representation matrix, predict the relationship links between entities; where, Represents the first deep learning model The matrix composed of the feature representations of the corresponding related entity types output by the last hidden layer is called the related entity type feature representation matrix. represents the graph decoder network parameters. The second deep learning model includes a softmax layer for predicting the relationship link between entities. This layer calculates the predicted probability of a relationship link between related entities. The calculation formula is , ,in, Represents the relationship The parameter matrix of Representing relationships The probability label of and Relationship Type When ,otherwise . Perform sample training and use the gradient descent (Adam) algorithm to update the R-GNN network parameters of the second deep learning model , optimize the loss function until the loss function meets expectations. After completing the training of the second deep learning model, the model can identify and output the relationship link between any two entities for various entities and the initial and predicted relationships between entities.

[0050] Regarding the role of the above dynamic knowledge graph in pavement maintenance, for example, in the input stage, we obtained multiple entities about a road: such as "road width" is 15 meters, "maximum crack width" is 3 centimeters, "material type" is asphalt, etc. Based on the above dynamic knowledge graph, "crack" is automatically identified as a "disease entity" type based on these entities and their entity attribute descriptions. Next, through the establishment of relationship links between entities, if according to business rules, the "friction coefficient" of a road with "material type" asphalt may be affected by "crack width", the system will automatically create an "influence" relationship edge from "material type" to "friction coefficient" in the knowledge graph. Ultimately, the dynamic knowledge graph forms a network containing rich semantic information, covering various construction and disease information of the road, and provides a basis for in-depth analysis through the relationship between entities. This kind of graph can be used for subsequent pavement maintenance plan generation to ensure the scientificity and practicality of the plan.

[0051] Step S3: Use the dynamic knowledge graph to output an applicable pavement maintenance plan template, including maintenance content, maintenance methods, disposal drawings, engineering quantity estimation, engineering schedule, etc.

[0052] Based on the dynamic knowledge graph of pavement maintenance in step S2, the main function of this step is to extract relevant information from the dynamic knowledge graph and generate a pavement maintenance plan template suitable for the current road conditions. This module outputs a complete and specific maintenance plan by deeply analyzing the entities and their relationships in the dynamic knowledge graph and combining the reasoning ability of the large model. These plans include a detailed description of the maintenance content, specific maintenance methods, required disposal drawings, accurate estimates of the engineering quantity, and scheduling of the project.

[0053] Specifically, first, read all entities and relationships related to the current road in the dynamic knowledge graph. By analyzing these entities, it is possible to identify the main problems of the current road surface and the measures that need to be taken. For example, if the knowledge graph identifies the entity "crack" about road surface disease and its attribute description of "crack width" exceeds a certain threshold, this step will recommend the use of specific crack repair technology.

[0054] In order to generate specific maintenance methods, this step will combine historical data and industry best practices to select the most appropriate repair technology. For example, in the case of a large "crack width", it may be necessary to choose to use asphalt filling technology, and provide detailed operation steps and a list of required equipment. The generation of disposal drawings is another key function. This step will automatically generate corresponding construction drawings based on the identified disease type and location. These drawings can be used directly to guide on-site construction personnel to perform operations. The estimation of engineering quantities is achieved through statistical analysis of entity attributes in the knowledge graph. This step will calculate the quantity of materials required, estimated man-hours, etc. For project scheduling, this step will generate a reasonable construction plan based on the current road conditions, weather forecasts, and the availability of the construction team. The plan will list the start and end times of each task in detail to ensure a smooth construction process.

[0055] Moreover, in this step S3, the big model technology is combined to assign values ​​to the solution items and generate corresponding content according to the entity information. Using advanced big model technology, specific maintenance solution item content is generated according to the entity information extracted from the dynamic knowledge graph. In order to ensure the comprehensiveness and accuracy of the solution, the following processing steps S31-S34 are adopted to achieve reasonable assignment of values ​​and content generation for the solution items.

[0056] Step S31: Entity information collection and organization: Extract relevant entity information from the dynamic knowledge graph, including road infrastructure information, road surface inspection conditions, disease descriptions and their classification information. This information is organized and classified so that it can be processed in a targeted manner in subsequent steps.

[0057] Step S32: Big model processing and reasoning: Use big model technology to conduct in-depth analysis and reasoning on the collected entity information. The big model can understand complex contexts and technical details through its extensive data-driven capabilities. The Transformer model trained by pavement maintenance is trained with specific industry data sets and is specially adjusted to suit pavement maintenance scenarios. The model can more accurately predict and generate solution content.

[0058] Step S33: Solution item content generation: Based on the reasoning results of the large model, specific values ​​are generated for each solution item; for example, for the "maintenance content" item, the module will generate a detailed description, including the parts that need to be repaired, the repair methods, etc.; in the "maintenance methods" item, the module combines historical data to generate specific operating steps and recommended technical means. The generated solution content is dynamically adjusted through the feedback mechanism. The system continuously optimizes the generated solution items based on the latest pavement inspection data and construction feedback. This dynamic adjustment mechanism ensures the timeliness and practicality of the generated content.

[0059] Step S34, verification and output of generated results: Verify the generated solution to ensure that it meets industry standards and actual needs. After verification, the generated solution content is output in an operable format for reference and execution by the construction unit.

[0060] For example, in a road maintenance project in a certain city, the latest road surface temperature and humidity data from sensors are received. After preprocessing these data, the module identifies "high temperature" as a new entity and associates it with the existing "road cracking" problem; through analysis, it is found that high temperature may aggravate the development of cracking, so the relationship between "high temperature" and "aggravated cracking" is updated in the knowledge graph. Subsequently, the large model is retrained so that it can take into account the impact of high temperature factors on the road surface when generating maintenance plans. Through this dynamic update and optimization mechanism, the automatically generated maintenance plan can reflect the latest environmental changes in a timely manner. For example, it is recommended to use materials with better heat resistance for repairs during high temperatures and adjust the construction time to avoid temperature peaks. This process not only improves the scientificity and feasibility of the plan, but also provides more precise guidance for construction units.

[0061] See also Figure 2 The present invention discloses an intelligent pavement maintenance plan design system based on a dynamic knowledge graph, which specifically includes: a pavement maintenance related information collection module, a dynamic knowledge graph module and a maintenance plan generation module.

[0062] The pavement maintenance related information collection module receives road infrastructure information, pavement detection conditions, pavement disease descriptions and classification information, and uses them as entities of the knowledge graph as well as entity attribute descriptions and preliminary entity relationship descriptions;

[0063] The dynamic knowledge graph module constructs a dynamic knowledge graph based on the input entities and their mutual relationships, and uses machine learning technology to automatically analyze and determine the entity types and establish relationship links between entities; wherein, an association network between entities in the knowledge graph of road maintenance is constructed; and according to the entities and their preliminary entity relationships, the dynamic knowledge graph is queried to assign values ​​to entities of known types therein, and the entity types of unlabeled entities are identified based on the association network of the dynamic knowledge graph through a first deep learning model; and the relationship links between entities are predicted based on the relevant entity type feature representation matrix through a second deep learning model;

[0064] The maintenance plan generation module uses a dynamic knowledge graph to output applicable pavement maintenance plan templates, including maintenance content, maintenance methods, disposal drawings, project quantity estimates, and project schedules.

[0065] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. An intelligent pavement maintenance scheme design method based on dynamic knowledge graph, characterized in that: The following steps are involved: Receive road infrastructure information, road surface inspection conditions, road surface disease descriptions and classification information, and use them as entities in the knowledge graph, as well as entity attribute descriptions and preliminary entity relationship descriptions; Based on the input entities and their mutual relations, a dynamic knowledge graph is constructed, and the entity types are automatically analyzed and determined by machine learning technology to establish relational links between entities; wherein, an association network between entities in the knowledge graph of road maintenance is constructed; and according to the entities and their preliminary entity relations, the dynamic knowledge graph is queried to assign values ​​to entities of known types therein, and the entity types of unlabeled entities are identified based on the association network of the dynamic knowledge graph through a first deep learning model; and the relational links between entities are predicted based on a feature representation matrix of relevant entity types through a second deep learning model; Use dynamic knowledge graphs to output applicable pavement maintenance plan templates, including maintenance content, maintenance methods, disposal drawings, engineering quantity estimates, and engineering schedules; The road infrastructure information includes information on the basic structure and properties of the road; the road surface detection information includes: road surface flatness, road surface friction status assessment, road surface gap status assessment; the road surface disease description includes a detailed description of the location, type, severity and impact range of various diseases on the road surface; The association network between entities in the knowledge graph of pavement maintenance is represented as ;in, Represents the set of all entities, where That is, entities related to road maintenance; entity The corresponding entity type, A set of types representing the relationship between two entities in the association network between knowledge graph entities; a set of triples Representing Entities and There is a relationship between ,but is the union of the above triple sets; The first deep learning model is represented as , using an RNN neural network architecture, where the input layer of the model receives the input, namely, the association network of the dynamic knowledge graph represented by the entities, entity types, entity relationships and their triple sets , is the parameter matrix of each hidden layer of the neural network of this model The parameter vector is composed of It is The unknown weight parameter matrix of the layer, , is the total number of hidden layers; The training of the first deep learning model includes: training the parameter vector of the first deep learning model Initialize to , that is, all parameter matrices of each hidden layer are initialized to , then for any entity in the association network of the dynamic knowledge graph , calculated by deep learning model in the first The feature representation obtained by the hidden layer , specifically: in is the activation function, For this entity Have a relationship The collection of other entities, is a predetermined constant, with a value of , Representing a collection The number of elements of Representing a collection The The entity in The feature representation output by the hidden layer is For this entity In the The feature representation output by the hidden layer; for entity , the first deep learning model obtains feature representation layer by layer until the last hidden layer, i.e., the Feature representation of the hidden layer ; The first deep learning model also includes a softmax classification layer as the first layer, connected to the last hidden layer, to obtain the feature representation As the input of the softmax classification layer, we get: The classification layer output is a K-dimensional vector, the vector Elements Indicates that according to the entity Corresponding Determine the entity Belong to The probability of the class type is: here ; Output from the classification layer , where the type with the highest probability is taken as the entity type; and then, by using sample training, optimizing the parameter vector of the first deep learning model , for the unknown weight parameter matrix of each hidden layer Assign values ​​until the loss function meets expectations; The second deep learning model , based on the relevant entity type feature representation matrix, predict the relationship links between entities; where, Represents the first deep learning model The matrix composed of the feature representations of the corresponding related entity types output by the last hidden layer is called the related entity type feature representation matrix. represents the graph decoder network parameters; The second deep learning model includes a softmax layer for predicting relationship links between entities. The softmax layer calculates the predicted probability of a relationship link between related entities. The calculation formula is: , ,in, Represents the relationship The parameter matrix of Representing relationships The probability label of and Relationship Type When ,otherwise ; Based on the big model technology, specific maintenance plan item content is generated according to the entity information extracted from the dynamic knowledge graph, and the plan items of the pavement maintenance plan are assigned values ​​and corresponding content is generated; The use of big model technology to assign values ​​to solution items specifically includes: extracting relevant entity information from the dynamic knowledge graph; using big model technology to conduct in-depth analysis and reasoning on the collected entity information; generating specific values ​​for each solution item based on the reasoning results of the big model; dynamically adjusting the generated solution content through a feedback mechanism, and continuously optimizing the generated solution items based on the latest pavement inspection data and construction feedback; verifying the generated solution to ensure that it meets industry standards and actual needs; after verification, the generated solution content is output in an operational format for reference and execution by the construction unit.

2. An intelligent pavement maintenance scheme design system based on dynamic knowledge graph, characterized in that: include: The pavement maintenance related information collection module receives road infrastructure information, pavement inspection conditions, pavement disease descriptions and classification information, and uses them as entities in the knowledge graph as well as entity attribute descriptions and preliminary entity relationship descriptions; The dynamic knowledge graph module constructs a dynamic knowledge graph based on the input entities and their mutual relationships, and uses machine learning technology to automatically analyze and determine the entity types and establish relationship links between entities; wherein, an association network between entities in the knowledge graph of road maintenance is constructed; and according to the entities and their preliminary entity relationships, the dynamic knowledge graph is queried to assign values ​​to entities of known types therein, and the entity types of unlabeled entities are identified based on the association network of the dynamic knowledge graph through a first deep learning model; and the relationship links between entities are predicted based on the relevant entity type feature representation matrix through a second deep learning model; The maintenance plan generation module uses the dynamic knowledge graph to output applicable pavement maintenance plan templates, including maintenance content, maintenance methods, disposal drawings, engineering quantity estimates, and engineering schedules; The road infrastructure information includes information on the basic structure and properties of the road; the road surface detection information includes: road surface flatness, road surface friction status assessment, road surface gap status assessment; the road surface disease description includes a detailed description of the location, type, severity and impact range of various diseases on the road surface; The association network between entities in the knowledge graph of pavement maintenance is represented as ;in, Represents the set of all entities, where That is, entities related to road maintenance; entity The corresponding entity type, A set of types representing the relationship between two entities in the association network between knowledge graph entities; a set of triples Representing Entities and There is a relationship between ,but is the union of the above triple sets; The first deep learning model is represented as , using an RNN neural network architecture, where the input layer of the model receives the input, namely, the association network of the dynamic knowledge graph represented by the entities, entity types, entity relationships and their triple sets , is the parameter matrix of each hidden layer of the neural network of this model The parameter vector is composed of It is The unknown weight parameter matrix of the layer, , is the total number of hidden layers; The training of the first deep learning model includes: training the parameter vector of the first deep learning model Initialize to , that is, all parameter matrices of each hidden layer are initialized to , then for any entity in the association network of the dynamic knowledge graph , calculated by deep learning model in the first The feature representation obtained by the hidden layer , specifically: in is the activation function, For this entity Have a relationship The collection of other entities, is a predetermined constant, with a value of , Representing a collection The number of elements, Representing a collection The The entity in The feature representation output by the hidden layer is For this entity In the The feature representation output by the hidden layer; for entity , the first deep learning model obtains feature representation layer by layer until the last hidden layer, i.e., the Feature representation of the hidden layer ; The first deep learning model also includes a softmax classification layer as the first layer, connected to the last hidden layer, to obtain the feature representation As the input of the softmax classification layer, we get: The classification layer output is a K-dimensional vector, the vector Elements Indicates that according to the entity Corresponding Determine the entity Belong to The probability of the class type is: here ; Output from the classification layer , where the type with the highest probability is taken as the entity type; and then, by using sample training, optimizing the parameter vector of the first deep learning model , for the unknown weight parameter matrix of each hidden layer Assign values ​​until the loss function meets expectations; The second deep learning model , based on the relevant entity type feature representation matrix, predict the relationship links between entities; where, Represents the first deep learning model The matrix composed of the feature representations of the corresponding related entity types output by the last hidden layer is called the related entity type feature representation matrix. represents the graph decoder network parameters; The second deep learning model includes a softmax layer for predicting relationship links between entities. The softmax layer calculates the predicted probability of a relationship link between related entities. The calculation formula is: , ,in, Represents the relationship The parameter matrix of Representing relationships The probability label of and Relationship Type When ,otherwise ; Based on the big model technology, specific maintenance plan item content is generated according to the entity information extracted from the dynamic knowledge graph, and the plan items of the pavement maintenance plan are assigned values ​​and corresponding content is generated; The use of big model technology to assign values ​​to solution items specifically includes: extracting relevant entity information from the dynamic knowledge graph; using big model technology to conduct in-depth analysis and reasoning on the collected entity information; generating specific values ​​for each solution item based on the reasoning results of the big model; dynamically adjusting the generated solution content through a feedback mechanism, and continuously optimizing the generated solution items based on the latest pavement inspection data and construction feedback; verifying the generated solution to ensure that it meets industry standards and actual needs; after verification, the generated solution content is output in an operational format for reference and execution by the construction unit.

Images