Road engineering low-carbon technology recommendation and carbon emission reduction prediction method based on knowledge graph

Abstract

This application discloses a knowledge graph-based method for recommending low-carbon technologies and predicting carbon emission reduction in road engineering. The method includes: acquiring target data of the target object and multi-source knowledge data related to low-carbon technologies in road engineering; constructing a knowledge graph of the low-carbon technology field in road engineering based on the multi-source knowledge data; determining the feature representation of the target object based on the target data and mapping it to an entity subgraph; determining the direct and indirect fit between the target object and candidate low-carbon technologies based on the knowledge graph to determine the fit of the candidate low-carbon technologies and determine the recommended low-carbon technologies; and predicting the carbon emission reduction corresponding to the recommended low-carbon technologies based on the knowledge graph. The method provided in this application can uniformly organize dispersed knowledge and combine it with the characteristics of the target object to achieve low-carbon technology fit analysis and carbon emission reduction benefit prediction.

Description

Technical Field

[0001] This application relates to the interdisciplinary fields of artificial intelligence knowledge engineering, road engineering, and low-carbon technology, and in particular to a knowledge graph-based method for recommending low-carbon technologies and predicting carbon emission reduction in road engineering. Background Technology

[0002] With the continuous advancement of dual-carbon goals, the demand for green, low-carbon, and intelligent construction in the highway engineering field is constantly increasing. The implementation of road construction, pavement treatment, and related engineering projects typically involves multiple stages, including material production and transportation, equipment investment, construction process organization, and adaptation to the site environment. Different technical solutions exhibit significant differences in resource consumption, energy efficiency, and carbon emission levels. Therefore, how to select an implementation plan that balances applicability and low-carbon benefits from a variety of available technical solutions, tailored to specific road engineering scenarios, has become a crucial issue in road engineering technology management and energy conservation and emission reduction decision-making.

[0003] In related technologies, the determination of low-carbon technology solutions for road engineering typically relies heavily on industry standards, engineering manuals, historical project experience, and the subjective judgment of technical personnel. For example, technical personnel compare and select different technologies such as cold recycling, warm-mix asphalt, thin-layer overlay, and micro-surfacing based on factors such as road segment defects, structural conditions, traffic load, climate, material supply conditions, and construction organization requirements, and estimate their applicability and carbon reduction effects based on experience. However, in existing technologies, this knowledge is usually scattered across standard documents, case studies, expert experience, and supplier documents, lacking a unified structured organization and relational expression method, making it difficult to systematically analyze the technology adaptation relationships under complex constraints. Furthermore, existing solutions mostly rely on subjective experience or static rules for technology selection, making it difficult to fully explore the potential correlations between road conditions, material systems, equipment capabilities, construction processes, and environmental factors, resulting in limited accuracy and consistency of the recommended results. In addition, existing technologies usually focus on carbon emission accounting for a single stage, making it difficult to target specific engineering projects and make targeted predictions of the life-cycle carbon reduction benefits of candidate technology solutions, and also making it difficult to provide clear and credible explanations for the recommended results.

[0004] Therefore, in determining low-carbon technology solutions for road engineering, the fragmentation of relevant knowledge, the complexity of technology applicability analysis, and the insufficient accuracy of carbon emission reduction benefit prediction have become urgent problems that need to be solved. Summary of the Invention

[0005] This application provides a knowledge graph-based method for recommending low-carbon technologies and predicting carbon emission reduction in road engineering, aiming to solve the problems of scattered knowledge, complex technology applicability analysis, and insufficient accuracy in predicting carbon emission reduction benefits in the determination of low-carbon technology solutions for road engineering.

[0006] Firstly, a knowledge graph-based method for recommending low-carbon technologies and predicting carbon emission reductions in road engineering is provided, the method comprising: Acquire target data for the target object, as well as multi-source knowledge data related to low-carbon technologies in road engineering; Based on the aforementioned multi-source knowledge data, a knowledge graph in the field of low-carbon technology for road engineering is constructed. Based on the target data, the feature representation of the target object is determined, and the feature representation is mapped to the entity subgraph corresponding to the target object in the knowledge graph; Based on the knowledge graph, the direct fit between the target object and the candidate low-carbon technology is determined, as well as the indirect fit between the target object and the candidate low-carbon technology through the intermediate entity of the knowledge graph. The suitability of candidate low-carbon technologies is determined based on the direct suitability and the indirect suitability. Based on the suitability of the candidate low-carbon technologies, the recommended low-carbon technologies are determined. Based on the knowledge graph, the carbon emission reduction corresponding to the recommended low-carbon technology is predicted, and the carbon emission reduction prediction result of the recommended low-carbon technology is obtained.

[0007] Optionally, in the above scheme, acquiring the target data of the target object and the multi-source knowledge data related to low-carbon technologies in road engineering includes: Obtain at least one of the following data for the target object: road structure data, defect data, construction status data, traffic load data, environmental data, material supply data, and historical engineering records; Obtain at least two of the following: industry standard data, technical manual data, historical case data, expert experience data, and supplier information related to low-carbon technologies in road engineering.

[0008] Optionally, in the above scheme, constructing a knowledge graph in the field of low-carbon technology for road engineering based on the multi-source knowledge data includes: Entity extraction and relation extraction are performed on the multi-source knowledge data to obtain technical entities, material entities, equipment entities, road structure entities, entities with defects or construction status, traffic condition entities, climate environment entities, and construction process entities. Establish target relationships between the extracted entities, including technology applicability relationships, material compatibility relationships, environmental limitation relationships, process dependence relationships, cost-related relationships, and / or carbon emission-related relationships; The extracted entities and the target relationships are fused to obtain the knowledge graph of the low-carbon technology field of road engineering.

[0009] Optionally, in the above scheme, the step of fusing the extracted entities and the target relationship to obtain the knowledge graph of the low-carbon technology field of road engineering includes: Align similar entities from different sources to identify target entities that point to the same object; Align and normalize synonymous target relationships from different sources to determine a unified target relationship; Perform deduplication and consistency processing on duplicate entities, duplicate target relationships, and conflicting data; Based on the target entities and target relationships after alignment, normalization, deduplication, and consistency processing, a knowledge graph for the low-carbon technology field of road engineering is constructed.

[0010] Optionally, in the above scheme, determining the feature representation of the target object based on the target data and mapping the feature representation to the entity subgraph corresponding to the target object in the knowledge graph includes: Target features are extracted from the target data, including road structure features, defects features, construction status features, traffic load features, environmental features, material supply features and / or historical engineering record features; Based on the target features, determine the entity nodes corresponding to the target object; Based on the association between the target object and the corresponding entity node, an entity subgraph corresponding to the target object is constructed in the knowledge graph.

[0011] Optionally, in the above scheme, based on the knowledge graph, determining the direct fit between the target object and candidate low-carbon technologies includes: Vectorized representation learning is performed on the entities and target relationships in the knowledge graph to obtain the object embedding vector corresponding to the target object and the technology embedding vector corresponding to the candidate low-carbon technology; Based on the similarity between the object embedding vector and the technology embedding vector, the direct fit between the target object and the candidate low-carbon technology is determined.

[0012] Optionally, in the above scheme, the step of learning vectorized representations of entities and target relationships in the knowledge graph includes: A translational knowledge representation model is used to initially embed entities and target relationships in the knowledge graph; A graph attention network is used to aggregate the neighborhood information of nodes in the knowledge graph to obtain the updated node vector representation; Predict potential target relationships in the knowledge graph based on historical case data, and add potential target relationships that meet preset conditions to the knowledge graph.

[0013] Optionally, in the above scheme, based on the knowledge graph, determining the indirect fit between the target object and candidate low-carbon technologies via intermediate entities in the knowledge graph includes: Construct a preset meta-path set between the target object and the candidate low-carbon technology. The meta-paths in the preset meta-path set include paths formed through one of the intermediate entities: disease or construction status entity, material entity, climate environment entity, and construction process entity. Based on the preset meta-path set, the path similarity between the target object and the candidate low-carbon technology is calculated, and the indirect fit is determined. The determination of the suitability of candidate low-carbon technologies based on the direct suitability and the indirect suitability includes: The direct fit and the indirect fit are weighted and fused to obtain the fit of the candidate low-carbon technology.

[0014] Optionally, in the above scheme, the step of predicting the carbon emission reduction corresponding to the recommended low-carbon technology based on the knowledge graph to obtain the carbon emission reduction prediction result of the recommended low-carbon technology includes: Extract the object feature vector of the target object; Extract the technical feature vector of the recommended low-carbon technology; Based on the material nodes, equipment nodes, and construction process nodes associated with the recommended low-carbon technology in the knowledge graph, the graph association feature vector is determined. Based on the object feature vector, the technology feature vector, and the map association feature vector, the total life cycle carbon emission reduction corresponding to the recommended low-carbon technology is predicted, and the carbon emission reduction prediction result of the recommended low-carbon technology is obtained.

[0015] Secondly, a knowledge graph-based device for recommending low-carbon technologies and predicting carbon emission reduction in road engineering, the device comprising: The data acquisition module is used to acquire target data of the target object, as well as multi-source knowledge data related to low-carbon technologies in road engineering; The knowledge graph construction module is used to construct a knowledge graph in the field of low-carbon technology for road engineering based on the multi-source knowledge data. An object representation module is used to determine the feature representation of the target object based on the target data, and map the feature representation to the entity subgraph corresponding to the target object in the knowledge graph. The fit determination module is used to determine, based on the knowledge graph, the direct fit between the target object and the candidate low-carbon technology, and the indirect fit between the target object and the candidate low-carbon technology via intermediate entities in the knowledge graph, and to determine the fit of the candidate low-carbon technology based on the direct fit and the indirect fit. The determination module is used to determine the recommended low-carbon technology based on the suitability of the candidate low-carbon technologies; The prediction module is used to predict the carbon emission reduction corresponding to the recommended low-carbon technology based on the knowledge graph, and obtain the carbon emission reduction prediction result of the recommended low-carbon technology.

[0016] Compared with the prior art, this application has at least the following beneficial effects: This application, based on further analysis and research of existing technical problems, recognizes the issues of scattered knowledge, complex technical applicability analysis, and insufficient accuracy in predicting carbon emission reduction benefits in the determination of low-carbon technology solutions for road engineering. It addresses these problems by acquiring target data of the target object and multi-source knowledge data related to low-carbon technologies in road engineering, and constructing a knowledge graph of the low-carbon technology field based on this multi-source knowledge data. This firstly unifies and correlates the relevant knowledge originally scattered in industry standards, technical documents, case information, and experience information, thereby solving the problem of scattered and difficult-to-use related knowledge in existing technologies. Furthermore, this application determines the feature representation of the target object based on the target data and maps the feature representation to the entity subgraph corresponding to the target object in the knowledge graph. This allows information such as the road structure status, defects or construction status, traffic load conditions, environmental conditions, and material supply conditions of the target object to establish a correspondence with low-carbon technology knowledge in a graph-like manner. This ensures that the engineering conditions of the target object are no longer confined to isolated data but are incorporated into a correlated and analyzable knowledge structure. Based on this, this application further addresses these issues by... The knowledge graph determines the direct fit between the target object and the candidate low-carbon technology, as well as the indirect fit between the target object and the candidate low-carbon technology via intermediate entities in the knowledge graph. Furthermore, based on the direct and indirect fits, the fit of the candidate low-carbon technology is determined. This allows for simultaneous fit analysis of candidate low-carbon technologies from both direct and indirect association dimensions, thus solving the problem in existing technologies that rely solely on human experience or static rules, making it difficult to systematically analyze the applicability of technologies under complex multi-factor conditions. Moreover, after determining the recommended low-carbon technology, this application further predicts the carbon emission reduction corresponding to the recommended low-carbon technology based on the knowledge graph, obtaining the predicted carbon emission reduction result. This enables the provision of carbon emission reduction benefit assessment results corresponding to the recommended technology for specific target objects, thus solving the problem in existing technologies that struggle to predict the carbon emission reduction benefits of candidate technology solutions for specific engineering objects. Therefore, this application can achieve low-carbon technology recommendation and carbon emission reduction benefit prediction for target objects, addressing the problems of scattered related knowledge, complex technology applicability analysis, and insufficient accuracy in carbon emission reduction benefit prediction in the background technology. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a knowledge graph-based method for recommending low-carbon technologies and predicting carbon emission reduction in road engineering, provided as an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] In one embodiment, such as Figure 1 As shown, a knowledge graph-based method for recommending low-carbon technologies and predicting carbon emission reduction in road engineering is provided, including the following steps: Acquire target data for the target object, as well as multi-source knowledge data related to low-carbon technologies in road engineering; Based on the aforementioned multi-source knowledge data, a knowledge graph in the field of low-carbon technology for road engineering is constructed. Based on the target data, the feature representation of the target object is determined, and the feature representation is mapped to the entity subgraph corresponding to the target object in the knowledge graph; Based on the knowledge graph, the direct fit between the target object and the candidate low-carbon technology is determined, as well as the indirect fit between the target object and the candidate low-carbon technology through the intermediate entity of the knowledge graph. The suitability of candidate low-carbon technologies is determined based on the direct suitability and the indirect suitability. Based on the suitability of the candidate low-carbon technologies, the recommended low-carbon technologies are determined. Based on the knowledge graph, the carbon emission reduction corresponding to the recommended low-carbon technology is predicted, and the carbon emission reduction prediction result of the recommended low-carbon technology is obtained.

[0020] In some implementations, the knowledge graph-based method for recommending low-carbon technologies and predicting carbon emission reductions in road engineering described in this embodiment can be executed by a software system deployed on a server, cloud platform, or local engineering decision-making terminal. This method first acquires target data for the target object and multi-source knowledge data related to low-carbon technologies in road engineering. The target object can be a road section, construction section, structural component, or other engineering object requiring low-carbon technology decision analysis, where a technical solution to be implemented is selected. Multi-source knowledge data supports the construction of the knowledge graph, and the target data characterizes the engineering and resource conditions of the object to be analyzed.

[0021] In this method, a knowledge graph for the low-carbon technology field of road engineering is constructed based on multi-source knowledge data. This knowledge graph is used to uniformly represent entities and their relationships related to low-carbon technology decisions in road engineering scenarios. The knowledge graph may include technical entities, material entities, equipment entities, road structure entities, entities related to defects or construction status, traffic condition entities, climate environment entities, and construction process entities, etc., with entities connected through target relationships. By transforming knowledge scattered in specifications, technical manuals, historical cases, expert experience, and supplier materials into a unified graph structure, a foundational knowledge carrier can be provided for subsequent feature mapping, fit calculation, and carbon emission reduction prediction.

[0022] In this method, a feature representation of the target object is determined based on the target data, and this feature representation is mapped to an entity subgraph corresponding to the target object in a knowledge graph. Here, the entity subgraph is not a single entity, but rather a local graph structure formed by a group of entity nodes related to the target object and the relationships between these entity nodes. For example, when the target object has specific road structure conditions, defect characteristics, traffic load levels, environmental conditions, and material supply conditions, a corresponding target object node can be created in the knowledge graph, and this target object node can be connected to multiple entity nodes representing the aforementioned characteristics, thereby forming the entity subgraph corresponding to the target object. The entity subgraph is used to represent a comprehensive engineering profile of the target object.

[0023] In this method, based on a knowledge graph, the direct fit between the target object and candidate low-carbon technologies, as well as the indirect fit between the target object and candidate low-carbon technologies via intermediate entities in the knowledge graph, are determined. Here, intermediate entities refer to bridging entity nodes located in the association path between the target object and candidate low-carbon technologies, such as entities related to disease or construction status, materials, climate, or construction processes. The direct fit reflects the strength of the direct association between the target object and candidate low-carbon technologies in the embedding space; the indirect fit reflects the strength of the multi-hop semantic association formed between the target object and candidate low-carbon technologies through intermediate entities. Based on the direct and indirect fits, the fit of candidate low-carbon technologies can be further determined, and low-carbon technology recommendations can be made accordingly.

[0024] In this method, after determining the recommended low-carbon technology, the carbon emission reduction corresponding to the recommended low-carbon technology is further predicted based on a knowledge graph, resulting in a predicted carbon emission reduction for the recommended low-carbon technology. This prediction process comprehensively considers the engineering conditions of the target object, the attributes of the recommended technology itself, and the association characteristics between the recommended technology and nodes such as materials, equipment, and construction processes in the knowledge graph, thereby outputting the carbon emission reduction of the recommended technology over its entire life cycle. The total life cycle carbon emission reduction can be expressed in tons / km, tons / project, relative emission reduction ratio, or other quantifiable forms.

[0025] The method described in this embodiment can unify multi-source heterogeneous knowledge into a knowledge graph in the field of low-carbon technology for road engineering, and map the target object to the corresponding entity subgraph. Based on this, the method combines direct and indirect fit to complete the recommendation of low-carbon technologies, and further predicts the carbon emission reduction of the recommended technologies, thereby achieving unified processing of low-carbon technology selection and carbon emission reduction benefit assessment for the target object.

[0026] In this embodiment, acquiring the target data of the target object and the multi-source knowledge data related to low-carbon technologies in road engineering includes: Obtain at least one of the following data for the target object: road structure data, defect data, construction status data, traffic load data, environmental data, material supply data, and historical engineering records; Obtain at least two of the following: industry standard data, technical manual data, historical case data, expert experience data, and supplier information related to low-carbon technologies in road engineering.

[0027] In some implementations, the target data of the target object described in this embodiment may originate from one or more of the following: road inspection equipment, construction monitoring equipment, inspection equipment, vehicle-mounted data acquisition systems, drone platforms, IoT sensors, manual data entry terminals, project management systems, or historical engineering databases. The road structure data may include structural layer type, layer thickness, strength parameters, deflection value, smoothness, compaction degree, etc.; the defect data may include defect types and their severity such as cracks, ruts, potholes, loosening, and spalling; the construction status data may include the current construction stage, on-site operation status, equipment operating status, and construction organization status; the traffic load data may include average daily traffic volume, proportion of heavy vehicles, and axle load level; the environmental data may include temperature, humidity, rainfall, frost depth, and wind speed; the material supply data may include material source, supply radius, transportation distance, and types of available materials; and the historical engineering records may include historical construction plans, maintenance records, quality inspection results, and lifespan performance.

[0028] In some implementations, multi-source knowledge data related to low-carbon technologies in road engineering can come from at least two of the following: industry standard data, technical manual data, historical case data, expert experience data, and supplier information. Industry standard data can be used to provide the application boundaries of the technology, construction conditions, and standard requirements; technical manual data can be used to provide technical principles, equipment configurations, process flows, and material parameters; historical case data can be used to provide technology usage scenarios and effect records in existing projects; expert experience data can be used to supplement tacit knowledge and judgment logic for complex scenarios; and supplier information can be used to provide material, equipment, and process implementation conditions. In this example embodiment, the multi-source knowledge sources may include 3 industry standards, 2 technical manuals, interviews with 50 experts, 200 historical cases, and information from 10 suppliers.

[0029] In some implementations, the target data and multi-source knowledge data can undergo preprocessing after collection, including format standardization, data cleaning, missing value imputation, outlier removal, and time alignment, to facilitate subsequent knowledge extraction and feature mapping. Structured data can be directly normalized, while semi-structured and unstructured text can be processed through sentence segmentation, word segmentation, entity recognition, and semantic tagging.

[0030] This embodiment can obtain data representing the actual conditions of the project from the target object side, and obtain multi-source knowledge data covering technical rules, case experience and resource constraints from the knowledge side, thereby providing a data foundation for subsequent knowledge graph construction, entity subgraph mapping, technology adaptability calculation and carbon emission reduction prediction.

[0031] In this embodiment, constructing a knowledge graph in the field of low-carbon road engineering technology based on the multi-source knowledge data includes: Entity extraction and relation extraction are performed on the multi-source knowledge data to obtain technical entities, material entities, equipment entities, road structure entities, entities with defects or construction status, traffic condition entities, climate environment entities, and construction process entities. Establish target relationships between the extracted entities, including technology applicability relationships, material compatibility relationships, environmental limitation relationships, process dependence relationships, cost-related relationships, and / or carbon emission-related relationships; The extracted entities and the target relationships are fused to obtain the knowledge graph of the low-carbon technology field of road engineering.

[0032] In some implementations, the "construction of a knowledge graph in the field of low-carbon road engineering technology based on multi-source knowledge data" described in this embodiment may include three main steps: entity extraction, relation extraction, and graph fusion. Entity extraction is used to identify key objects in different data sources; relation extraction is used to identify semantic relationships between different objects; and graph fusion is used to unify data from different sources into a complete graph structure.

[0033] In entity extraction, different extraction methods can be used for different types of data sources. For structured data, entity types can be directly determined based on field names, parameter labels, and preset mapping tables. For unstructured text data such as technical manuals, historical cases, and expert interviews, a combination of rule-based and deep learning methods can be used for entity extraction. For example, named entity recognition models can be used to identify technical entities, material entities, equipment entities, road structure entities, defect or construction status entities, traffic condition entities, climate environment entities, and construction process entities from the text. In some implementations, BERT-like language models can be used for entity recognition, combined with manual review for calibration.

[0034] In terms of relationship extraction, target relationships between the entities can be identified. These target relationships can include at least technology applicability relationships, material compatibility relationships, environmental limitation relationships, process dependence relationships, cost-related relationships, and carbon emission-related relationships. For example, if a technology faces implementation limitations in low-temperature environments, an environmental limitation relationship can be established between the technology entity and the climate environment entity; if a technology relies on specific renewable material supply conditions, a material compatibility relationship or process dependence relationship can be established between the technology entity and the material entity.

[0035] In some implementations, the extracted entities and target relationships are fused to form a knowledge graph in the field of low-carbon road engineering. In this example, the graph may contain thousands of entity nodes and tens of thousands of relationship edges to support subsequent graph representation learning and graph reasoning computation.

[0036] By constructing knowledge graphs, technical, engineering, and resource knowledge that were originally scattered across multiple data sources can be uniformly expressed, thereby forming a structured knowledge base suitable for recommending low-carbon technologies and predicting carbon emission reduction in road engineering.

[0037] In this embodiment, the process of fusing the extracted entities and the target relationships to obtain the knowledge graph of the low-carbon technology field of road engineering includes: Align similar entities from different sources to identify target entities that point to the same object; Align and normalize synonymous target relationships from different sources to determine a unified target relationship; Perform deduplication and consistency processing on duplicate entities, duplicate target relationships, and conflicting data; Based on the target entities and target relationships after alignment, normalization, deduplication, and consistency processing, a knowledge graph for the low-carbon technology field of road engineering is constructed.

[0038] In some implementations, the fusion process described in this embodiment may further include entity alignment, target relationship alignment and normalization, deduplication, and consistency processing. Entity alignment refers to the unified identification of multiple entities pointing to the same object from different sources. For example, descriptions with the same core meaning, such as "warm-mix asphalt technology," "warm-mix construction plan," and "low-temperature asphalt mixing process," are grouped into a unified technical entity. Target relationship alignment and normalization refers to the transformation of semantically similar or synonymous relationship expressions from different sources into a unified relationship type. For example, "applicable to," "can be used for," and "recommended for" are uniformly standardized into a technology applicability relationship.

[0039] In some implementations, deduplication can be used to remove duplicate entities and duplicate target relationships; consistency processing can be used to resolve data conflicts between different data sources. For example, when a technical manual and a historical case describe the applicable boundaries of the same technology inconsistently, the final retention result can be determined through priority rules, confidence assessment, expert review, or weighted voting. The processed target entities and target relationships are uniformly organized into a knowledge graph of the low-carbon technology field of road engineering, thereby ensuring semantic consistency and clear structure within the graph.

[0040] In some implementations, the merged knowledge graph can also retain additional attributes such as data source identifier, timestamp, confidence level, and version number, so as to facilitate subsequent tracing, updating, and continuous learning.

[0041] The fusion processing method described in this embodiment can improve the consistency, standardization, and computability of entity and target relationships in the knowledge graph, thereby providing a semantically unified graph foundation for subsequent fit calculation and carbon emission reduction prediction.

[0042] In this embodiment, determining the feature representation of the target object based on the target data and mapping the feature representation to the entity subgraph corresponding to the target object in the knowledge graph includes: Target features are extracted from the target data, including road structure features, defects features, construction status features, traffic load features, environmental features, material supply features and / or historical engineering record features; Based on the target features, determine the entity nodes corresponding to the target object; Based on the association between the target object and the corresponding entity node, an entity subgraph corresponding to the target object is constructed in the knowledge graph.

[0043] In some implementations, the "determining the feature representation of the target object based on target data" described in this embodiment can first extract target features from the target data. The target features may include one or more of the following: road structure features, defect features, construction status features, traffic load features, environmental features, material supply features, and historical engineering record features. For numerical features, standardization, discretization, or binning can be used; for categorical features, label encoding or one-hot encoding can be used; for textual features, vectorized representation or semantic label representation can be used.

[0044] In some implementations, entity nodes corresponding to the target object are determined based on the extracted target features. For example, when the deflection value of the target object is within a certain preset range, it can be mapped to the corresponding road structure state entity; when the proportion of heavy vehicles exceeds a preset threshold, it can be mapped to the corresponding traffic condition entity; when the ambient temperature, frost depth, or humidity meets specific conditions, it can be mapped to the corresponding climate environment entity; when there is a nearby recycled material supply point or the material transportation distance is lower than a preset value, it can be mapped to a material supply-related entity. In this embodiment, the PCI, heavy vehicle proportion, climate zone, and accessibility of recycled materials of the target road segment are mapped to entities such as "PCI=70-80", "heavy vehicle proportion > 25%", and "mild climate zone".

[0045] In some implementations, an entity subgraph corresponding to the target object is constructed in the knowledge graph based on the association between the target object and its corresponding entity nodes. Here, the entity subgraph is a local graph structure formed around the target object, containing the target object node, multiple entity nodes associated with it, and corresponding connections. The entity subgraph is used to characterize the comprehensive feature state of the target object in the graph space and provides input for subsequent calculations of direct and indirect fitness.

[0046] The feature representation and entity subgraph construction method described in this embodiment can map the multidimensional engineering conditions of the target object into a local structural representation in the knowledge graph, thereby enabling the target object to participate in subsequent technology matching and carbon emission reduction prediction in the form of graph semantics.

[0047] In this embodiment, determining the direct fit between the target object and candidate low-carbon technologies based on the knowledge graph includes: Vectorized representation learning is performed on the entities and target relationships in the knowledge graph to obtain the object embedding vector corresponding to the target object and the technology embedding vector corresponding to the candidate low-carbon technology; Based on the similarity between the object embedding vector and the technology embedding vector, the direct fit between the target object and the candidate low-carbon technology is determined.

[0048] In some implementations, the direct fit degree described in this embodiment is used to characterize the degree of direct association between the target object and the candidate low-carbon technology. To calculate the direct fit degree, vectorized representation learning can be performed on the entities and target relationships in the knowledge graph to obtain the object embedding vector corresponding to the target object and the technology embedding vector corresponding to the candidate low-carbon technology. The object embedding vector reflects the comprehensive semantic position of the target object in the knowledge graph, and the technology embedding vector reflects the comprehensive semantic position of the candidate low-carbon technology in the knowledge graph.

[0049] In some implementations, direct fit can be determined by calculating the similarity between the object embedding vector and the technology embedding vector. This similarity can be measured using cosine similarity, dot product similarity, inverse Euclidean distance, or other similarity metrics. A higher direct fit indicates a greater degree of direct matching between the target object and the candidate low-carbon technology in the knowledge graph embedding space.

[0050] In some implementations, this embodiment describes the direct path as calculating the graph convolutional network embedding similarity between the target object node and each technical node, which can essentially be classified as the direct fit calculation process between the target object and the technical nodes.

[0051] The direct fit calculation method described above can quantify the strength of the direct correlation between the target object and the candidate low-carbon technology in the embedding space, thereby providing a direct matching basis for technology fit calculation.

[0052] In some implementations, the direct fit degree described in this embodiment is used to reflect the degree of direct association between the target object and the candidate low-carbon technology. Specifically, it is implemented as follows: first, vectorized representation learning is performed on the entities and target relationships in the knowledge graph to obtain the object embedding vector corresponding to the target object and the technology embedding vector corresponding to the candidate low-carbon technology; then, based on the similarity between the object embedding vector and the technology embedding vector, the direct fit degree between the target object and the candidate low-carbon technology is determined.

[0053] In some implementations, the object embedding vector reflects the semantic representation of the target object entity subgraph in the knowledge graph embedding space, and the technology embedding vector reflects the semantic representation of the candidate low-carbon technology entity in the knowledge graph embedding space. A higher direct fit indicates a higher degree of direct matching between the target object and the candidate low-carbon technology in the semantic space.

[0054] In this embodiment, the step of learning vectorized representations of entities and target relationships in the knowledge graph includes: A translational knowledge representation model is used to initially embed entities and target relationships in the knowledge graph; A graph attention network is used to aggregate the neighborhood information of nodes in the knowledge graph to obtain the updated node vector representation; Predict potential target relationships in the knowledge graph based on historical case data, and add potential target relationships that meet preset conditions to the knowledge graph.

[0055] In some implementations, the vectorized representation learning described in this embodiment can employ a translational knowledge representation model to initially embed entities and target relationships in the knowledge graph. For example, the TransE model can be used to map entities and relationships to a unified low-dimensional vector space, ensuring that entities and relationships satisfy preset translation constraints, thereby obtaining initial vector representations.

[0056] In some implementations, a graph attention network can be further employed to aggregate the neighborhood information of nodes in the knowledge graph, resulting in an updated node vector representation. Through the graph attention network, different weights can be adaptively assigned based on the varying importance of adjacent nodes and edges, thereby enhancing the embedding representation's ability to represent higher-order relations and local structural features.

[0057] In some implementations, potential target relationships in the knowledge graph can be predicted based on historical case data, and potential target relationships that meet preset conditions can be added to the knowledge graph. These preset conditions may include a prediction confidence threshold, expert approval, and no conflict with existing relationships. By adding potential target relationships, the completeness of the knowledge graph and its subsequent reasoning capabilities can be improved.

[0058] By using the vectorized representation learning and the supplementary method of potential target relationship, the accuracy of entity representation in knowledge graphs and the integrity of knowledge graph structure can be improved, thereby increasing the reliability of direct fit calculation and subsequent recommendation analysis.

[0059] In this embodiment, based on the knowledge graph, determining the indirect fit between the target object and candidate low-carbon technologies via intermediate entities in the knowledge graph includes: Construct a preset meta-path set between the target object and the candidate low-carbon technology. The meta-paths in the preset meta-path set include paths formed through one of the intermediate entities: disease or construction status entity, material entity, climate environment entity, and construction process entity. Based on the preset meta-path set, the path similarity between the target object and the candidate low-carbon technology is calculated, and the indirect fit is determined. The determination of the suitability of candidate low-carbon technologies based on the direct suitability and the indirect suitability includes: The direct fit and the indirect fit are weighted and fused to obtain the fit of the candidate low-carbon technology.

[0060] In some implementations, the indirect fit described in this embodiment refers to the strength of the path association between the target object and candidate low-carbon technologies via intermediate entities in the knowledge graph. Here, intermediate entities are not equivalent to entity subgraphs. An entity subgraph is the local graph structure corresponding to the target object, while an intermediate entity is a single entity node or a type of entity node that acts as a bridge in the path. An entity subgraph can contain multiple intermediate entities, which can be used to construct multiple meta-paths from the target object to candidate low-carbon technologies.

[0061] In some implementations, a pre-defined set of meta-paths can be constructed between the target object and candidate low-carbon technologies. These meta-paths may include, but are not limited to, "target object—disease or construction status entity—candidate low-carbon technology," "target object—material entity—candidate low-carbon technology," "target object—climate and environment entity—candidate low-carbon technology," and "target object—construction process entity—candidate low-carbon technology," etc. Through this pre-defined set of meta-paths, the multi-hop semantic relationships between the target object and candidate low-carbon technologies can be explicitly described.

[0062] In some implementations, based on a preset set of meta-paths, the path similarity between the target object and candidate low-carbon technologies is calculated to determine the indirect fit. Path similarity can be calculated based on path hit rate, path weight, path length, node importance, edge weight, or similarity metrics such as PathSim. The comprehensive fit can be calculated using a weighted fusion of direct fit and meta-path similarity, where meta-path similarity corresponds to indirect fit.

[0063] In some implementations, the direct fit and the indirect fit can be weighted and fused to obtain the fit of the candidate low-carbon technology. The weights can be preset fixed values ​​or learnable parameters obtained through training, in order to balance the contributions of direct matching information and indirect semantic path information in the final fit.

[0064] Through the aforementioned indirect fit calculation and fusion method, the intermediate entities and meta-path structures in the knowledge graph can be used to mine multi-hop semantic associations between the target object and candidate low-carbon technologies, thereby enabling the fit calculation to take into account both direct matching relationships and indirect association relationships.

[0065] In some implementations, the indirect fit degree described in this embodiment refers to the strength of the multi-hop semantic association formed between the target object and candidate low-carbon technologies via intermediate entities in the knowledge graph. Here, intermediate entities are bridging entity nodes in the path, such as entities related to disease or construction status, materials, climate, and construction processes; while the entity subgraph is a local graph structure corresponding to the target object, and the two are not the same. An entity subgraph can contain multiple intermediate entities, and the calculation of the indirect fit degree utilizes these intermediate entities to construct a meta-path between the target object and candidate low-carbon technologies.

[0066] In this embodiment, a method for recommending low-carbon technologies and predicting carbon emission reduction in road engineering based on knowledge graphs is provided, including the following steps: Multi-source knowledge fusion construction steps: Integrate highway maintenance industry standards, expert experience databases, historical project case data, and material supplier technical manuals to construct a knowledge graph for the low-carbon maintenance field. Graph nodes include: maintenance technology, material system, equipment type, road structure, disease type, traffic conditions, climate environment, and construction process; graph edges represent the relationships between nodes, including: technology applicability, material compatibility, environmental constraints, process dependence, cost correlation, and carbon emission impact; Graph Representation Learning and Enhancement Steps: A hybrid representation learning method combining TransE and graph attention networks is adopted to learn low-dimensional dense vector representations for entities and relations in the knowledge graph, and potential relation edges are automatically discovered and supplemented through graph neural networks based on historical case data; Target road segment feature extraction and representation steps: Collect multi-dimensional feature data of the target road segment through road inspection vehicles, drone inspections and IoT sensors, including road damage status, structural strength, material properties, traffic load spectrum, environmental temperature and humidity and historical maintenance records, and map the feature data into entity subgraphs in the knowledge graph; Adaptability calculation and intelligent recommendation steps: Design a dual-path matching model based on graph convolutional networks and attention mechanisms. One path calculates the direct adaptability between the target road segment features and the technology nodes, while the other path calculates the indirect adaptability through intermediate entities (such as materials and environment). The combined output includes a list of recommended technologies and an adaptability score. The steps for accurately predicting carbon emission reduction benefits are as follows: Establish a maintenance technology-carbon emission impact prediction model. This model is based on the correlation strength between technology nodes and material, equipment, and process nodes in the knowledge graph. It predicts the total life cycle carbon emission reduction after adopting the recommended technology and provides the prediction confidence interval and analysis of key influencing factors.

[0067] The knowledge graph is stored in the Neo4j graph database, containing no fewer than 5,000 entity nodes and 20,000 relation edges, and supports SPARQL queries and subgraph retrieval.

[0068] In the dual-path matching model, the formula for calculating the comprehensive fit score S(r,t) between the target road segment rr and the maintenance technology tt is as follows: ; where h r and h t, r, and t are the node embedding vectors learned by the graph convolutional network for road segment r and technology t, respectively; cosine is the cosine similarity function, which measures the strength of direct association; P is a set of predefined meta-paths (e.g., "road segment-disease-applicable technology"); PathSim(r,t|p) is the similarity between road segment r and technology t calculated based on meta-path p; λ is a learnable weight parameter that balances direct and indirect associations, 0≤λ≤1.

[0069] In this embodiment, the step of predicting the carbon emission reduction corresponding to the recommended low-carbon technology based on the knowledge graph, and obtaining the predicted carbon emission reduction result of the recommended low-carbon technology, includes: Extract the object feature vector of the target object; Extract the technical feature vector of the recommended low-carbon technology; Based on the material nodes, equipment nodes, and construction process nodes associated with the recommended low-carbon technology in the knowledge graph, the graph association feature vector is determined. Based on the object feature vector, the technology feature vector, and the map association feature vector, the total life cycle carbon emission reduction corresponding to the recommended low-carbon technology is predicted, and the carbon emission reduction prediction result of the recommended low-carbon technology is obtained.

[0070] In some implementations, the carbon emission reduction prediction described in this embodiment can be performed after the recommended low-carbon technologies are determined. First, the object feature vector of the target object is extracted. The object feature vector may be derived from the structural state, defects or construction status, traffic conditions, environmental conditions, material supply conditions, and other characteristics of the target object. Second, the technical feature vector of the recommended low-carbon technologies is extracted. The technical feature vector may be derived from the technology type, material requirements, equipment requirements, process parameters, applicable boundaries, and other characteristics.

[0071] In some implementations, a graph association feature vector can be determined based on material nodes, equipment nodes, and construction process nodes associated with the recommended low-carbon technology in the knowledge graph. This graph association feature vector characterizes the strength of the association between the recommended technology and resource conditions, implementation conditions, and process conditions within the graph. In some implementations, a graph attention network can be used to aggregate the aforementioned associated nodes to obtain the graph association feature vector. This embodiment can employ a hybrid prediction architecture combining a graph attention network and a gradient boosting decision tree to predict life-cycle carbon emission reductions.

[0072] In some implementations, the life-cycle carbon emission reduction corresponding to recommended low-carbon technologies is predicted based on object feature vectors, technology feature vectors, and graph association feature vectors. The prediction model can be implemented using a regression model, an ensemble learning model, or a combination of graph neural networks and tree models. For example, the three vectors can be concatenated and input into a GBDT model to output the corresponding carbon emission reduction prediction result.

[0073] By using the above-mentioned carbon emission reduction prediction methods, we can further provide the full life cycle carbon emission reduction prediction results corresponding to the target object based on the recommended low-carbon technologies, thereby providing a quantifiable carbon emission reduction assessment basis for the selection of engineering technologies.

[0074] In some implementations, the carbon emission reduction prediction described in this embodiment involves quantitatively calculating the life-cycle carbon emission reduction of the recommended low-carbon technology relative to the target scenario after the recommended low-carbon technology has been determined. Specifically, the object feature vector of the target object can be extracted first, then the technical feature vector of the recommended low-carbon technology can be extracted, and further, based on the material nodes, equipment nodes, and construction process nodes associated with the recommended low-carbon technology in the knowledge graph, the graph association feature vector can be determined.

[0075] The carbon emission reduction benefit accurate prediction model described in this embodiment adopts a hybrid architecture of graph attention network and gradient boosting decision tree. The expression for its predicted life-cycle carbon emission reduction ΔC is as follows: .

[0076] in, (r) is the feature vector of road segment rr; ψ(t) is the feature vector of technology t; σ(r,t) is the associated feature vector obtained by aggregating technology t through graph attention network after interacting with the nodes of related attributes (such as materials and processes) of road segment r in the knowledge graph; Represents vector concatenation; GBDT( ) is a gradient boosting decision tree model.

[0077] This embodiment also includes a recommendation explanation generation step: using the attention weights in the graph attention network, the knowledge graph substructure (nodes and edges) that contributes the most to the fit score is located, and a natural language explanation is generated based on this to explain the reasons for the recommendation and the main sources of emission reduction.

[0078] In one embodiment, a knowledge graph-based system for recommending low-carbon technologies and predicting carbon emission reduction in road engineering is provided, comprising: The data acquisition module is used to acquire target data of the target object, as well as multi-source knowledge data related to low-carbon technologies in road engineering; The knowledge graph construction module is used to construct a knowledge graph in the field of low-carbon technology for road engineering based on the multi-source knowledge data. An object representation module is used to determine the feature representation of the target object based on the target data, and map the feature representation to the entity subgraph corresponding to the target object in the knowledge graph. The fit determination module is used to determine, based on the knowledge graph, the direct fit between the target object and the candidate low-carbon technology, and the indirect fit between the target object and the candidate low-carbon technology via intermediate entities in the knowledge graph, and to determine the fit of the candidate low-carbon technology based on the direct fit and the indirect fit. The determination module is used to determine the recommended low-carbon technology based on the suitability of the candidate low-carbon technologies; The prediction module is used to predict the carbon emission reduction corresponding to the recommended low-carbon technology based on the knowledge graph, and obtain the carbon emission reduction prediction result of the recommended low-carbon technology.

[0079] The specific implementation details of each module can be found in the above description of the limitations of knowledge graph-based road engineering low-carbon technology recommendation and carbon emission reduction prediction methods, and will not be repeated here.

[0080] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A method for recommending low-carbon technologies and predicting carbon emission reduction in road engineering based on knowledge graphs, characterized in that, The method includes: Acquire target data for the target object, as well as multi-source knowledge data related to low-carbon technologies in road engineering; Based on the aforementioned multi-source knowledge data, a knowledge graph in the field of low-carbon technology for road engineering is constructed. Based on the target data, the feature representation of the target object is determined, and the feature representation is mapped to the entity subgraph corresponding to the target object in the knowledge graph; Based on the knowledge graph, the direct fit between the target object and the candidate low-carbon technology is determined, as well as the indirect fit between the target object and the candidate low-carbon technology through the intermediate entity of the knowledge graph. The suitability of candidate low-carbon technologies is determined based on the direct suitability and the indirect suitability. Based on the suitability of the candidate low-carbon technologies, the recommended low-carbon technologies are determined. Based on the knowledge graph, the carbon emission reduction corresponding to the recommended low-carbon technology is predicted, and the carbon emission reduction prediction result of the recommended low-carbon technology is obtained.

2. The method according to claim 1, characterized in that, The acquisition of target data of the target object and multi-source knowledge data related to low-carbon technologies in road engineering includes: Obtain at least one of the following data for the target object: road structure data, defect data, construction status data, traffic load data, environmental data, material supply data, and historical engineering records; Obtain at least two of the following: industry standard data, technical manual data, historical case data, expert experience data, and supplier information related to low-carbon technologies in road engineering.

3. The method according to claim 1, characterized in that, The construction of a knowledge graph in the field of low-carbon technology for road engineering based on the multi-source knowledge data includes: Entity extraction and relation extraction are performed on the multi-source knowledge data to obtain technical entities, material entities, equipment entities, road structure entities, entities with defects or construction status, traffic condition entities, climate environment entities, and construction process entities. Establish target relationships between the extracted entities, including technology applicability relationships, material compatibility relationships, environmental limitation relationships, process dependence relationships, cost-related relationships, and / or carbon emission-related relationships; The extracted entities and the target relationships are fused to obtain the knowledge graph of the low-carbon technology field of road engineering.

4. The method according to claim 3, characterized in that, The process of fusing the extracted entities and the target relationships to obtain the knowledge graph of the low-carbon technology field of road engineering includes: Align similar entities from different sources to identify target entities that point to the same object; Align and normalize synonymous target relationships from different sources to determine a unified target relationship; Perform deduplication and consistency processing on duplicate entities, duplicate target relationships, and conflicting data; Based on the target entities and target relationships after alignment, normalization, deduplication, and consistency processing, a knowledge graph for the low-carbon technology field of road engineering is constructed.

5. The method according to claim 1, characterized in that, The step of determining the feature representation of the target object based on the target data and mapping the feature representation to the entity subgraph corresponding to the target object in the knowledge graph includes: Target features are extracted from the target data, including road structure features, defects features, construction status features, traffic load features, environmental features, material supply features and / or historical engineering record features; Based on the target features, determine the entity nodes corresponding to the target object; Based on the association between the target object and the corresponding entity node, an entity subgraph corresponding to the target object is constructed in the knowledge graph.

6. The method according to claim 1, characterized in that, Based on the knowledge graph, the direct fit between the target object and candidate low-carbon technologies is determined, including: Vectorized representation learning is performed on the entities and target relationships in the knowledge graph to obtain the object embedding vector corresponding to the target object and the technology embedding vector corresponding to the candidate low-carbon technology; Based on the similarity between the object embedding vector and the technology embedding vector, the direct fit between the target object and the candidate low-carbon technology is determined.

7. The method according to claim 6, characterized in that, The step of learning vectorized representations of entities and target relationships in the knowledge graph includes: A translational knowledge representation model is used to initially embed entities and target relationships in the knowledge graph; A graph attention network is used to aggregate the neighborhood information of nodes in the knowledge graph to obtain the updated node vector representation; Predict potential target relationships in the knowledge graph based on historical case data, and add potential target relationships that meet preset conditions to the knowledge graph.

8. The method according to claim 1, characterized in that, Based on the knowledge graph, the indirect fit between the target object and candidate low-carbon technologies via intermediate entities in the knowledge graph is determined, including: Construct a preset meta-path set between the target object and the candidate low-carbon technology. The meta-paths in the preset meta-path set include paths formed through one of the intermediate entities: disease or construction status entity, material entity, climate environment entity, and construction process entity. Based on the preset meta-path set, the path similarity between the target object and the candidate low-carbon technology is calculated, and the indirect fit is determined. The determination of the suitability of candidate low-carbon technologies based on the direct suitability and the indirect suitability includes: The direct fit and the indirect fit are weighted and fused to obtain the fit of the candidate low-carbon technology.

9. The method according to claim 1, characterized in that, The step of predicting the carbon emission reduction corresponding to the recommended low-carbon technology based on the knowledge graph, and obtaining the predicted carbon emission reduction result of the recommended low-carbon technology, includes: Extract the object feature vector of the target object; Extract the technical feature vector of the recommended low-carbon technology; Based on the material nodes, equipment nodes, and construction process nodes associated with the recommended low-carbon technology in the knowledge graph, the graph association feature vector is determined. Based on the object feature vector, the technology feature vector, and the map association feature vector, the total life cycle carbon emission reduction corresponding to the recommended low-carbon technology is predicted, and the carbon emission reduction prediction result of the recommended low-carbon technology is obtained.

10. A knowledge graph-based system for recommending low-carbon technologies and predicting carbon emission reduction in road engineering, characterized in that, The system includes: The data acquisition module is used to acquire target data of the target object, as well as multi-source knowledge data related to low-carbon technologies in road engineering; The knowledge graph construction module is used to construct a knowledge graph in the field of low-carbon technology for road engineering based on the multi-source knowledge data. An object representation module is used to determine the feature representation of the target object based on the target data, and map the feature representation to the entity subgraph corresponding to the target object in the knowledge graph. The fit determination module is used to determine, based on the knowledge graph, the direct fit between the target object and the candidate low-carbon technology, and the indirect fit between the target object and the candidate low-carbon technology via intermediate entities in the knowledge graph, and to determine the fit of the candidate low-carbon technology based on the direct fit and the indirect fit. The determination module is used to determine the recommended low-carbon technology based on the suitability of the candidate low-carbon technologies; The prediction module is used to predict the carbon emission reduction corresponding to the recommended low-carbon technology based on the knowledge graph, and obtain the carbon emission reduction prediction result of the recommended low-carbon technology.

Images