Road continuous longitudinal slope section traffic safety knowledge graph construction method

Abstract

The invention belongs to the crossing field of traffic engineering and information technology, and discloses a highway continuous longitudinal slope section traffic safety knowledge graph construction method. According to the method, road section features, traffic flow, historical accidents and environmental data are integrated, and a multi-source data set is constructed through cleaning and standardization processing; defining a road section, an accident, an environment and a vehicle entity, and constructing a multi-level knowledge graph comprising a static topology layer, a dynamic attribute layer and a probabilistic reasoning layer; knowledge triples are extracted by using a natural language processing technology, and real-time risk assessment and early warning are realized by dynamically fusing association logic of gradient, environment and accident data in combination with a rule engine; a graph database is adopted to store the graph, and semantic query and dynamic updating are supported. The method breaks through the limitation of traditional single data source analysis, solves the problem of coupling risk assessment of dynamic environmental factors and static features under complex terrains through multi-source fusion and probabilistic reasoning mechanisms, and remarkably improves the traffic safety risk prediction precision.

Description

Technical Field

[0001] The present invention belongs to the cross - field of traffic engineering and information technology, and provides a method for constructing a traffic safety knowledge graph for continuous longitudinal slope sections of highways. Background Technique

[0002] With the rapid development of the highway traffic network, continuous longitudinal slope sections have become high - risk areas for traffic accidents due to their special terrain conditions (such as large slope changes, small curve radii, etc.). Traditional traffic safety analysis methods are mostly based on single data sources (such as historical accident statistics or static section parameters), lacking the comprehensive utilization of multi - source heterogeneous data (such as dynamic traffic flow, real - time environmental parameters). For example, existing technologies often evaluate risks through the linear relationship between slope and slope length, but ignore the coupled influence of dynamic environmental factors such as weather and road surface friction coefficient on the accident probability, resulting in insufficient prediction accuracy of the risk assessment model in actual scenarios.

[0003] In the field of knowledge graph technology, existing methods mostly focus on entity - relationship modeling of general traffic networks (such as road topology, vehicle trajectories), and do not design a dedicated knowledge representation system for continuous longitudinal slope sections. For example, some studies use the RDF model or property graph to store static relationships, but lack the ability to fuse real - time dynamic time - series data (such as weather state changes (t), friction coefficient μ(t)), and the modeling of the association relationships between entities mostly relies on deterministic rules, failing to introduce a probability inference mechanism to reflect the randomness of risk evolution.

[0004] Although recent studies have attempted to fuse multi - dimensional data to construct a risk graph (such as the patent CN 119399965 A of CCCC Highway Consultants Co., Ltd.), its model still has the following limitations:

[0005] 1) The association logic between static section features and dynamic environmental parameters is not clearly distinguished, resulting in limited reasoning ability of the rule engine;

[0006] 2) A multi - level knowledge model (such as a static topology layer, a dynamic attribute layer, a probability inference layer) is not established, making it difficult to achieve hierarchical mapping from data to knowledge;

[0007] 3) The real - time early warning function depends on fixed - threshold judgment and does not dynamically adjust the risk level by combining probability models such as Bayesian networks.

[0008] To address the above problems, this patent proposes a method for constructing a traffic safety knowledge graph for continuous longitudinal slope sections of highways. Through key technologies such as multi - source data fusion, hierarchical knowledge modeling, and dynamic probability inference, it solves the deficiencies of traditional methods in data integration and real - time risk assessment, and provides refined decision - making support for traffic safety management under complex terrain conditions. Summary of the Invention

[0009] The purpose of the present invention is to solve the problems in the traditional safety analysis method for continuous longitudinal slope sections of highways, including insufficient integration of multi-source heterogeneous data, lack of coupling risk assessment for dynamic environmental factors and static section characteristics, and low prediction accuracy caused by relying on fixed thresholds for real-time warning. It realizes refined traffic safety management by constructing a multi-level knowledge graph that integrates static topology, dynamic attributes, and probabilistic reasoning.

[0010] To achieve the above purpose, the present invention adopts the following technical means:

[0011] The present invention provides a method for constructing a traffic safety knowledge graph for continuous longitudinal slope sections of highways, including the following steps:

[0012] Step 1, data collection and preprocessing: Integrate multi-source heterogeneous data, including section characteristic data, traffic flow data, historical accident data, and environmental data, and perform cleaning, de-duplication, and standardization processing on the data to output a standardized data set;

[0013] Step 2, knowledge representation and modeling: Define the entities, attributes, and relationships between entities of the knowledge graph, and construct a multi-level schema including section entities, accident entities, and environmental entities. The schema includes a static topology layer, a dynamic attribute layer, and a probabilistic reasoning layer;

[0014] Step 3, knowledge extraction and construction: Extract entity attribute values and relationships between entities from structured data, perform named entity recognition, relationship extraction, and event extraction on unstructured text data through natural language processing technology, integrate multi-source data, and solve entity reference ambiguity and conflicts to generate knowledge triples;

[0015] Step 4, knowledge reasoning and application: Implement dynamic risk assessment based on a preset rule engine. The rule engine integrates the association logic of slope characteristics, environmental parameters, and accident data to generate real-time risk warning signals;

[0016] Step 5, system deployment and update: Store the knowledge graph using a graph database and provide semantic query, risk warning, and graph dynamic update functions through a visualization interface.

[0017] Further, Step 1 includes the following steps:

[0018] Step 1.1 Multi-source data collection, collect four types of heterogeneous data for continuous longitudinal slope sections of highways:

[0019] Section characteristic data: Slope , continuous longitudinal slope length , curve radius , represents the section ID, denoted as a vector , where uphill >0, downhill < 0;

[0020] Traffic flow data: time - series traffic volume , average vehicle speed , which constitutes a time - series matrix ;

[0021] Accident data: accident type , timestamp , severity level {1, 2, 3}, which constitutes an event set ;

[0022] Environmental data: weather condition , road surface friction coefficient , visibility , which constitutes a time - series matrix ;

[0023] Step 1.2 Clean, de - duplicate, and standardize the data, and output a standardized data set:

[0024] .

[0025] Furthermore, the steps for defining the entities of the knowledge graph in Step 2 are as follows:

[0026] Define four types of core entities and their attributes, and use vector space and probability models for formal description:

[0027] Let the entity set of the knowledge graph be:

[0028]

[0029] Define the attribute definition set of all entities = , and the definition is as follows:

[0030] Section entity 's attribute set:

[0031]

[0032] Among them: represents the section ID, is the slope, is the elevation difference, is the horizontal distance, is the longitudinal slope continuous length, and R is the curve radius of the bend;

[0033] Accident entity 's attribute combination, representing accident characteristics, is defined as a multi - tuple:

[0034]

[0035] Among them: is the accident type code, is the accident timestamp, is the accident severity level, 1: Minor, 2: Moderate, 3: Major;

[0036] Environmental entity 's set of attributes:

[0037] = { , , }

[0038] Among them: is the weather condition, is the visibility, is the road surface friction coefficient, subject to 's conditional distribution:

[0039] =

[0040] Vehicle entity 's set of attributes, representing vehicle traffic characteristics, defined as a set of dynamic parameters:

[0041]

[0042] Among them: is the vehicle type , is the real-time vehicle speed, represents the vehicle acceleration, represents the braking performance coefficient, associated with .

[0043] The mapping relationship with Type is:

[0044] .

[0045] Furthermore, the associations between attributes and entities in step 2 specifically include the following steps:

[0046] Define the set of associations between entities as:

[0047]

[0048] Each relationship is defined as a weighted directed edge, and the weight represents the association strength or probability, specifically as follows:

[0049] Section - accident causal association:

[0050] Model the causal relationship between road section characteristics and accidents using a conditional probability model:

[0051]

[0052] where represents the road section feature vector, , represents the feature weight coefficient, corresponding to the slope , longitudinal slope length , and curve radius weights respectively, , , , and σ(⋅) represents the Sigmoid function;

[0053] Environment-accident risk probability association:

[0054] Describe the dynamic impact of environmental factors on accident risk and use Bayesian network modeling:

[0055]

[0056] represents the accident severity level, represents the indicator function, taking 1 in case of bad weather and 0 otherwise, , , represent the environmental factor weights;

[0057] Road section-environment coupling risk:

[0058] Define the interaction between slope and environmental factors and use a coupling risk index:

[0059]

[0060] : weather-related road surface friction coefficient, is the coupling coefficient.

[0061] Road section-environment-vehicle coupling risk:

[0062]

[0063] is the road surface friction coefficient provided by the environmental entity, where , , are all coupling coefficients.

[0064] Furthermore, in step 2, construct a multi-level model including road section entities, accident entities, and environmental entities. The model includes a static topology layer, a dynamic attribute layer, and a probability inference layer, and specifically includes the following steps:

[0065] Construct a multi - level model

[0066] Use a hierarchical graph structure to represent the knowledge graph model and define three types of hierarchies:

[0067] Static topology layer

[0068] Describe the fixed relationships between entities and use an adjacency matrix :

[0069]

[0070] Indicates that there is at least one;

[0071] Dynamic attribute layer

[0072] Use a time - series tensor to represent the change of entity attributes over time, is the number of entities, is the time step, is the attribute dimension;

[0073] Probabilistic reasoning layer

[0074] Integrate association rules and risk models and define the logic of the rule engine:

[0075] Risk judgment rule:

[0076]

[0077] Indicates the comprehensive risk index of the medium - risk threshold, indicates the comprehensive risk index of the high - risk threshold, represents the set of recent accident patterns, specifically, the same type of accidents have occurred within the time window.

[0078] Furthermore, step 2 also includes the steps for the scalability of the Schema:

[0079] Define the schema extension protocol:

[0080] Entity compatibility: The newly added entity needs to satisfy that it shares at least one common attribute with the existing entities. The specific attribute interface:

[0081]

[0082] represents the newly added entity, represents the union of all existing entity attributes, represents the intersection operation, that is, finding the common attributes, Indicates that the intersection is non-empty, i.e., there is at least one common attribute.

[0083] Relationship inheritance: Hierarchical extension is achieved through relationship generalization:

[0084]

[0085] The newly added default general relationship related_to indicates that there is a certain unrefined association between entities, allowing for gradual refinement.

[0086] Furthermore, step 2 also includes a formal verification step:

[0087] To ensure the logical consistency of the schema, constraint conditions are defined:

[0088] Causal closure: If , then there must be a reverse association , ensuring the two-way traceability of the knowledge graph, represents the forward relationship, represents the reverse relationship.

[0089] Furthermore, step 4 includes the following steps:

[0090] Step 4.1, Initialization of the rule engine

[0091] Load the preset association rules and risk models, and construct an extensible inference logic library by referring to the conditional probability, Bayesian network, and coupled risk index defined in Section 2.2:

[0092]

[0093] Step 4.2, Dynamic risk assessment

[0094] Input the real-time data stream , and perform multi-dimensional risk calculations:

[0095] Single-factor risk quantification

[0096] Slope accident probability:

[0097] represents the slope-related risk probability of the k-th type of accident under real-time data,

[0098] Environmental risk value:

[0099] Coupled risk coefficient:

[0100] Comprehensive risk model, define the weighted comprehensive risk index:

[0101] +

[0102] wherein , is the weight coefficient, satisfying , and is determined by fitting historical accident data;

[0103] Step 4.3, real-time risk warning, generating a three-level warning signal based on the risk threshold:

[0104]

[0105] represents the comprehensive risk index is the medium risk threshold of represents the comprehensive risk index is the high risk threshold of represents the set of recent accident patterns, specifically, the same type of accidents have occurred within the time window.

[0106] Since the present invention adopts the above technical means, it has the following beneficial effects:

[0107] 1. Through the multi-source heterogeneous data fusion and standardization processing technology, the problems of isolated data and inconsistent quality in the traditional method are solved, and the construction of a high-precision data foundation is realized, specifically as follows:

[0108] Technical means: Integrate four types of heterogeneous data including road section characteristics, traffic flow, accidents, and environment, and generate a unified data set through cleaning, deduplication, and standardization processing.

[0109] Technical problems: The traditional method relies on a single data source and cannot comprehensively consider dynamic environment and static road section characteristics, resulting in one-sided risk assessment.

[0110] Effect: The data coverage integrity is improved to four dimensions (gradient, traffic flow, accidents, environment), and the standardization error is reduced. For example, the road surface friction coefficient in the environmental data is dynamically calibrated according to the weather (rainy , sunny ) to ensure data reliability.

[0111] 2. Through the multi-level knowledge graph modeling technology, the problem of the disconnection between static and dynamic data is solved, and the spatio-temporal adaptability of risk reasoning is improved, specifically as follows:

[0112] Technical means: Construct a static topology layer, a dynamic attribute layer, and a probability reasoning layer, and define the weighted relationship between entities (such as the causal probability of road section - accident ).

[0113] Technical problems: The traditional graph lacks dynamic time series attributes and probability associations and cannot reflect real-time risk changes.

[0114] Effect: The dynamic attribute layer uses a time-series tensor to support minute-level updates. The probability inference layer integrates a Bayesian network (such as ), improving the accuracy of risk prediction.

[0115] 3. By means of a dynamic rule engine and a comprehensive risk model, the problem of lag in static threshold warning is solved, and real-time risk classification and control are achieved, as follows:

[0116] Technical means: Integrate slope conditional probability, environmental Bayesian network, and coupled risk index to calculate the weighted risk .

[0117] Technical problem: Fixed thresholds cannot adapt to dynamic changes such as weather and traffic flow, resulting in a high false alarm rate.

[0118] Effect: The warning response time is shortened, and the high-risk identification rate is improved.

[0119] 4. By means of a graph database and visualization interface technology, the problems of low query efficiency and difficult update in traditional systems are solved, and the operation and maintenance efficiency is improved, as follows:

[0120] Technical means: Use a graph database such as Neo4j to store the knowledge graph, providing a semantic query interface and a dynamic update function.

[0121] Technical problem: Relational databases cannot efficiently handle complex entity associations.

[0122] Effect: The response time for complex queries is <200ms, supporting thousands of concurrent updates per second, and reducing the graph version iteration cycle from weekly to hourly.

[0123] 6. By means of a schema extension protocol and formal verification, the problem of poor knowledge graph compatibility is solved, ensuring the scalability and logical consistency of the system, as follows:

[0124] Technical means: Define entity compatibility conditions and causal closure constraints.

[0125] Technical problem: Adding new entities easily leads to data islands and logical conflicts.

[0126] Effect: Supports seamless access to new entities such as construction areas, and the passing rate of consistency verification for the extended graph is 100%. Description of the Drawings

[0127] Figure 1 This is a flow diagram of the present invention. Detailed Embodiments

[0128] The embodiments of the present invention will be described in detail below. Although the present invention will be described and illustrated in conjunction with some specific embodiments, it should be noted that the present invention is not limited to these embodiments only. On the contrary, any modifications or equivalent substitutions made to the present invention shall be covered by the scope of the claims of the present invention.

[0129] In addition, in order to better illustrate the present invention, numerous specific details are given in the following specific embodiments. Those skilled in the art will understand that the present invention can also be implemented without these specific details.

[0130] The present invention provides a method for constructing a traffic safety knowledge graph for continuous longitudinal slope sections of highways, including the following steps:

[0131] Data collection and preprocessing: Integrate multi-source heterogeneous data, including road section feature data, traffic flow data, historical accident data, and environmental data, and clean, deduplicate, and standardize the data;

[0132] Step 1.1, Multi-source data collection

[0133] Collect four types of heterogeneous data for continuous longitudinal slope sections of highways:

[0134] Road section feature data: Slope , Continuous longitudinal slope length , Curve radius , denoted as vector

[0135] (from the GIS system);

[0136] Traffic flow data: Time series traffic volume , Average vehicle speed , constituting a time series matrix (from the traffic monitoring system);

[0137] Accident data: Accident type (classification code), Timestamp , Severity level {1, 2, 3}, constituting an event set ;

[0138] Environmental data: Weather condition , Road surface friction coefficient (set according to rainy days, snowy days, sunny days), Visibility , constituting a time series matrix .

[0139] Step 1.2, Clean, deduplicate, and standardize the data, and output a standardized data set:

[0140] 。

[0141] Knowledge representation and modeling: Define the entities, attributes, and the associated relationships between entities in the traffic safety knowledge graph, and construct a multi-level schema that includes road section entities, accident entities, and environmental entities;

[0142] Step 2.1: Define four types of core entities and their attributes, and formalize them using vector space and probability models:

[0143] Let the entity set of the knowledge graph be:

[0144]

[0145] Define the attribute definition set of all entities = , and the definitions are as follows:

[0146] Road section entity The attribute set of:

[0147]

[0148] Among them: Indicates the road section ID, is the slope, is the elevation difference, is the horizontal distance, is the longitudinal slope continuous length, and R is the radius of curvature of the curve.

[0149] Accident entity The attribute combination, representing accident characteristics, is defined as a multi-tuple:

[0150]

[0151] Among them, is the accident type code (e.g., 1: rear-end collision, 2: rollover), is the accident timestamp, is the accident severity level (1: minor, 2: moderate, 3: major).

[0152] Environmental entity The attribute set of:

[0153] = { , , }

[0154] is the weather condition, is the road surface friction coefficient, which follows The conditional distribution of:

[0155] =

[0156] is visibility (unit: meter).

[0157] Vehicle entity The set of attributes, representing vehicle passing characteristics, is defined as a set of dynamic parameters:

[0158]

[0159] Among them is the vehicle type, is the real-time vehicle speed;

[0160] is the acceleration, according to the speed-time conversion

[0161] Step 2.2. Define the association relationship between entities and construct a probability association model

[0162] Define the set of association relationships between entities as:

[0163]

[0164] Each relationship is defined as a weighted directed edge, and the weight represents the association strength or probability, specifically as follows:

[0165] 1. Road section - accident causal association ( )

[0166] Model the causal relationship between road section characteristics and accident occurrence, and use a conditional probability model:

[0167]

[0168] Symbol explanation:

[0169] : Road section feature vector (slope, longitudinal slope length, curve radius)

[0170] : Feature weight coefficient (such as indicates that an increase in slope will increase the accident probability)

[0171] : Accident type (such as rear-end collision, rollover, etc.)

[0172] : is the activation function.

[0173] 2. Environment - accident risk probability association (environmental impact)

[0174] Describe the dynamic impact of environmental factors on accident risk and use Bayesian network modeling:

[0175]

[0176] Symbol Explanation:

[0177] : Accident severity level

[0178] : Indicator function (takes 1 in bad weather and 0 otherwise)

[0179] : Weight of environmental factors (needs to be fitted with historical data through existing conventional means);

[0180] 3. Road section - environment coupling risk (triggering risk)

[0181] Define the interaction between slope and environmental factors, and adopt a coupling risk index. The larger the value, the higher the risk:

[0182]

[0183] Symbol Explanation:

[0184] : Road surface friction coefficient related to weather

[0185] : Coupling coefficient (determined by regression analysis of accident data).

[0186] Road section - environment - vehicle coupling risk:

[0187]

[0188] The road surface friction coefficient provided for the environmental entity, where , , are all coupling coefficients, and this formula is a dimensionless risk index.

[0189] Step 2.3. Construct a multi - level schema

[0190] Use a hierarchical graph structure to represent the knowledge graph schema and define three types of hierarchies:

[0191] 1. Static topology layer

[0192] Describe the fixed relationships between entities and use an adjacency matrix :

[0193]

[0194] Indicates that if there is at least one relationship belongs to the relationship set and this relationship points from entity to entity , then the condition holds.

[0195] Example: Indicates that there is an association between the road section and the accident.

[0196] 2. Dynamic attribute layer

[0197] Use a time series tensor to represent the change of entity attributes over time:

[0198] : The number of entities

[0199] : The time step

[0200] : The attribute dimension (such as the road section entity , corresponding to )

[0201] 3. Probability inference layer

[0202] Integrate association rules and risk models to define the rule engine logic:

[0203] Risk determination rule:

[0204]

[0205] Represents the medium risk threshold of the comprehensive risk index , Represents the high risk threshold of the comprehensive risk index , Represents the set of recent accident patterns, specifically that the same type of accident has occurred within the time window.

[0206] Step 2.4, Scalability design of the Schema

[0207] To support future new entities (such as traffic signs, construction areas), define a schema extension protocol:

[0208] Entity compatibility: The new entity needs to satisfy sharing at least one common attribute with the existing entities. The specific attribute interface:

[0209]

[0210] This formula is a compatibility rule for knowledge graph schema expansion, which requires that new entities must be associated with existing entities through at least one common attribute. This rule ensures the data relevance, the coherence of reasoning logic, and the maintainability of the overall schema.

[0211] : The set of attributes of the new entity (such as "construction area").

[0212] For example:

[0213] Represents the union of all existing entity attributes.

[0214] Suppose the existing entities include "road section" and "accident":

[0215]

[0216]

[0217] Then the union is: .

[0218] : The intersection operation, that is, finding the common attributes.

[0219] If the attributes of the new entity are , then the intersection with the union of existing attributes is .

[0220] : The intersection is not empty, that is, there is at least one common attribute.

[0221] Ensure data relevance through this constraint:

[0222] If the new entity has no relation with the existing attributes (such as , then it cannot be associated with the existing entity (such as "road section") through the common attribute (such as "gradient"), resulting in data islands in the knowledge graph.

[0223] Support knowledge reasoning:

[0224] The common attribute is the basis for establishing logical relationships between entities. For example, through the "gradient" attribute, the impact of the "construction area" on the safety of the "road section" can be inferred.

[0225] Maintain schema consistency:

[0226] Avoid chaotic schema caused by random attribute expansion and ensure that new entities can be integrated into the existing framework.

[0227] Actual example

[0228] Legal expansion:

[0229] Add a new entity "Traffic Sign" with attributes {Type, Location, Associated Road Section}.

[0230] Common Attribute: Associated Road Section (The "Road Section ID" of the road section entity can be mapped to "Associated Road Section").

[0231] Condition to be met: There is a common attribute "Associated Road Section" associated with the "Road Section" entity.

[0232] Relationship inheritance: Achieve hierarchical expansion through relationship generalization:

[0233]

[0234] The newly added default general relationship related_to indicates that there is a certain unspecified association between entities, allowing for gradual refinement.

[0235] While retaining the original relationships, by adding the default relationship related_to, a "placeholder" - like association mechanism is provided for the newly added entity. This not only ensures the stability of the system but also reserves space for future refined modeling.

[0236] Significance of relationship inheritance:

[0237] Support progressive expansion:

[0238] When adding a new entity (such as "Construction Area"), it may not be possible to immediately define its specific relationship with existing entities (such as "Road Section") (such as blocks_traffic or increases_risk). In this case, first establish a preliminary association with related_to and then gradually refine it later.

[0239] Avoid rigid relationship definition:

[0240] If it is mandatory to pre - define specific relationships for all newly added entities, it may lead to over - design or frequent schema modifications. related_to provides a flexible intermediate state.

[0241] Maintain the consistency of the knowledge graph:

[0242] Even if the newly added entity has not been fully integrated into the business logic, through related_to, its basic association with other entities in the graph can be ensured, avoiding data silos.

[0243] Specific role of related_to:

[0244] Scenario Application Example Significance Initial Association of New Entities Construction Area related_to Road Section Indicates that there is an association between the construction area and a certain road section, but the specific impact type (such as closure, traffic restriction) needs to be refined. Generalized Association of Multiple Entities Traffic Sign related_to Road Section, Meteorological Sensor related_to Environment Unified management of various unclassified association relationships. Conflict Resolution Transition The new entity "Temporary Control" is associated with the accident entity through related_to, and will be replaced with triggers later Maintain a temporary association when the data conflict is not resolved.

[0245] Collaboration with the original relationship

[0246] The original relationship (such as has_accident) is a strongly semantic and highly weighted association that directly supports reasoning (such as the causal relationship between a road section and an accident).

[0247] related_to is a weakly semantic and low-weighted association that only indicates the existence of a connection and requires further verification or refinement.

[0248] Example:

[0249] If the newly added entity "animal migration" is associated with "road section" through related_to, the potential relationship between it and accidents can be discovered through data analysis, and finally related_to can be upgraded to "causing accidents".

[0250] Step 2.5, Formal verification

[0251] To ensure the logical consistency of the schema, define the following constraints:

[0252] Causal closure: If then there must exist a reverse association to ensure the two-way traceability of the knowledge graph.

[0253] Forward relationship:

[0254] For example: "A road section has had an accident ", from a data perspective, starting from the road section, associate its historical accident records.

[0255] Reverse relationship:

[0256] For example: "An accident occurred on a road section ", more specifically, accident A (rollover) occurred on road section X (gradient 5%, longitudinal slope length 2 km), from a data perspective, starting from the accident, locate its occurrence location.

[0257] Final output

[0258] The knowledge graph schema can be represented as a quadruple:

[0259]

[0260] is the above set of constraint conditions to ensure the logical and physical consistency of the data.

[0261] Step 3, Knowledge extraction and construction: From structured data Extract entity attribute values and relationships between entities. Through natural language processing techniques, perform named entity recognition, relationship extraction, and event extraction on unstructured text data, fuse multi-source data, and resolve entity reference ambiguities and conflicts to generate knowledge triples;

[0262] Step 4, Knowledge Reasoning and Application: Implement dynamic risk assessment based on a preset rule engine, and the rule engine integrates the correlation logic of slope features, environmental parameters, and accident data;

[0263] Step 4.1, Rule Engine Initialization

[0264] Load preset association rules and risk models, reference the conditional probabilities, Bayesian networks, and coupled risk indices defined in Section 2.2, and construct an extensible inference logic library:

[0265]

[0266] Among them:

[0267] is the conditional probability of slope features and accidents;

[0268] is the environmental coupling risk index;

[0269] is the vehicle dynamics equation;

[0270] Step 4.2, Dynamic Risk Assessment

[0271] Input real-time data stream , perform multi-dimensional risk calculations:

[0272] Among them:

[0273] 1. Single-factor risk quantification

[0274] Slope accident probability:

[0275] represents the slope-related risk probability of the kth type of accident under real-time data, and σ(⋅) represents the Sigmoid function;

[0276] Environmental risk value:

[0277] Coupled risk coefficient:

[0278] Comprehensive risk model, define the weighted comprehensive risk index:

[0279] +

[0280] Among them is the weight coefficient, , satisfying , which is determined by fitting historical accident data;

[0281] Step 4.3, Real-time risk warning

[0282] Generate three-level warning signals based on the risk threshold:

[0283]

[0284] represents the comprehensive risk index is the medium risk threshold of represents the comprehensive risk index is the high risk threshold of represents the set of recent accident patterns, specifically, the same type of accidents have occurred within the time window;

[0285] The triggering actions include:

[0286] High: Issue a speed limit instruction and activate the variable message sign;

[0287] Medium: Generate a maintenance inspection work order;

[0288] Low: Update the risk status attribute of the knowledge graph;

[0289] System deployment and update: Use a graph database to store the knowledge graph and provide semantic query, risk warning, and graph dynamic update functions through a visualization interface.

Claims

1. A method for constructing a traffic safety knowledge graph for continuous longitudinal slope sections of highways, characterized in that, It includes the following steps: Step 1, Data collection and preprocessing: Integrate multi-source heterogeneous data, including road section feature data, traffic flow data, historical accident data, and environmental data, and perform cleaning, deduplication, and standardization processing on the data, and output a standardized data set; Step 2, Knowledge representation and modeling: Define the entities, attributes, and relationships between entities of the knowledge graph, and construct a multi-level schema including road section entities, accident entities, and environmental entities. The schema includes a static topology layer, a dynamic attribute layer, and a probability inference layer; Step 3, Knowledge extraction and construction: Extract entity attribute values and relationships between entities from structured data, perform named entity recognition, relationship extraction, and event extraction on unstructured text data through natural language processing technology, fuse multi-source data, and resolve entity reference ambiguities and conflicts to generate knowledge triples; Step 4, Knowledge reasoning and application: Implement dynamic risk assessment based on a preset rule engine. The rule engine integrates the association logic of slope features, environmental parameters, and accident data to generate real-time risk warning signals; Step 5, System deployment and update: Store the knowledge graph using a graph database, and provide semantic query, risk warning, and graph dynamic update functions through a visualization interface.

2. The construction method of a traffic safety knowledge graph for a continuous longitudinal slope section of a highway according to claim 1, wherein, Step 1 includes the following steps: Step 1.1 Multi-source data collection Collect four types of heterogeneous data for continuous longitudinal slope sections of highways: Section feature data: slope 、Continuous longitudinal slope length 、Radius of curve , denotes the section ID, denoted as a vector , where uphill > 0, downhill < 0; Traffic flow data: Time series traffic volume , average vehicle speed , forming a time series matrix ; Accident data: accident type , timestamp , severity level {1, 2, 3}, forming an event set ; Environmental data: weather condition , road surface friction coefficient , visibility , which form a time series matrix ; Step 1.2 Perform cleaning, deduplication, and standardization processing on the data, and output a standardized data set: 。 3. A method for constructing a traffic safety knowledge graph for continuous longitudinal slope sections of highways according to claim 1, characterized in that: The specific steps for defining the entities of the knowledge graph in Step 2 include the following: Define four types of core entities and their attributes, and perform formal description using a vector space and a probability model: Let the entity set of the knowledge graph be: The set of attribute definitions that define all entities = , are defined as follows: Section entity Attribute set of: Wherein: represents the road section ID, is the slope, is the elevation difference, is the horizontal distance, is the continuous longitudinal slope length, and R is the radius of curvature of the curve; Accident entity Combined with the attributes, representing accident characteristics, defined as a multi-tuple: Wherein: is the accident type code, is the accident timestamp, is the accident severity level, 1: minor, 2: moderate, 3: major; Environmental entity Attribute set of: ={ ， ， } Wherein: is the weather condition, is the visibility, is the road surface friction coefficient, subject to the conditional distribution of: = Vehicle entity The set of attributes, representing vehicle passing characteristics, is defined as a set of dynamic parameters: Wherein: is the vehicle type , is the real-time vehicle speed, represents the vehicle acceleration, represents the braking performance coefficient, which is associated with ; The mapping relationship with Type is: 。 4. A method for constructing a traffic safety knowledge graph for continuous longitudinal slope sections of highways according to claim 1, characterized in that: The specific steps for the attributes and relationships between entities in Step 2 include the following: Define the set of relationships between entities as: Each relationship is defined as a weighted directed edge, and the weight represents the association strength or probability, specifically as follows: Road section - accident causal association: Model the causal relationship between road section features and accidents using a conditional probability model: Among them represents the road segment feature vector, , represents the feature weight coefficient, corresponding to the slope , longitudinal slope length , curve radius weights respectively, , , , and σ(⋅) represents the Sigmoid function; Environment - accident risk probability association: Describe the dynamic impact of environmental factors on accident risk, and use a Bayesian network for modeling: Indicates the severity level of the accident, Indicates the indicator function, taking 1 in bad weather and 0 otherwise, , , Indicates the weight of environmental factors; Road section - environment coupling risk: Define the interaction between slope and environmental factors, and use a coupling risk index: : Road surface friction coefficient related to weather, is the coupling coefficient; Road section - environment - vehicle coupling risk: The road surface friction coefficient provided for the environmental entity, where , , are all coupling coefficients.

5. A method for constructing a traffic safety knowledge graph for continuous longitudinal slope sections of highways according to claim 1, characterized in that: In Step 2, construct a multi-level schema including road section entities, accident entities, and environmental entities. The schema includes a static topology layer, a dynamic attribute layer, and a probability inference layer. The specific steps include the following: Construct a multi-level schema Use a hierarchical graph structure to represent the knowledge graph schema, and define three types of levels: Static topology layer Describe the fixed relationships between entities using an adjacency matrix : Indicates that there is at least one; Dynamic attribute layer Use a temporal tensor to represent the change of entity attributes over time, where is the number of entities, is the time step, and is the attribute dimension; Probability inference layer Integrate association rules and risk models, and define the rule engine logic: Risk determination rules: Represents the comprehensive risk index of the medium risk threshold, Represents the comprehensive risk index of the high risk threshold, Represents the set of recent accident patterns, specifically, the same type of accidents occurred within the time window.

6. A method for constructing a traffic safety knowledge graph for continuous longitudinal slope sections of highways according to claim 1, characterized in that: Step 2 also includes the step of Schema extensibility: Define the schema extension protocol: Entity Compatibility: New Entities It is required to share at least one common attribute with the existing entities. The specific attribute interface is as follows: Indicates a newly added entity, Represents the union of all existing entity attributes, Indicates an intersection operation, that is, finding common attributes, Indicates that the intersection is not empty, that is, there is at least one common attribute; Relationship inheritance: Achieve hierarchical extension through relationship generalization: The newly added default general relationship related_to indicates that there is a certain unrefined association between entities, allowing gradual refinement.

7. A method for constructing a traffic safety knowledge graph for continuous longitudinal slope sections of highways according to claim 1, characterized in that: Step 2 also includes the step of formal verification: To ensure the logical consistency of the schema, define constraint conditions: Causal Closure: If , then there must be a reverse association to ensure the bidirectional traceability of the knowledge graph. represents a forward relationship, represents a reverse relationship.

8. A method for constructing a traffic safety knowledge graph for continuous longitudinal slope sections of highways according to claim 1, characterized in that: Step 4 includes the following steps: Step 4.1, Rule engine initialization Load the preset association rules and risk models, reference the conditional probability, Bayesian network, and coupled risk index defined in Section 2.2, and construct an extensible inference logic library: Step 4.2, Dynamic risk assessment Input real-time data stream , perform multi-dimensional risk calculation: Single-factor risk quantification Slope accident probability: Indicates the slope-related risk probability of the k-th type of accident under real-time data, Environmental risk value: Coupling risk coefficient: Comprehensive risk model, define the weighted comprehensive risk index: + wherein , is a weight coefficient, satisfying , and is determined by fitting historical accident data; Step 4.3, Real-time risk warning, generate three-level warning signals based on the risk threshold: represents the comprehensive risk index of the medium risk threshold represents the comprehensive risk index of the high risk threshold represents the set of recent accident patterns, specifically, the same type of accidents occurred within the time window.

Images