A road behavior adaptive topology perception and application system and method

Abstract

The present application relates to the field of intelligent perception and application technology of road behavior, and specifically to a road behavior adaptive topological perception and application system and method: a behavior cluster perception module monitors road structural behavior parameters in real time; a data transmission and communication module implements differentiated data transmission strategies; a road section dynamic risk level assessment module calculates the road section risk index and level based on the road structural behavior parameters and the risk assessment model; a node adjustment and resource allocation module adjusts the activation state and sampling frequency of each sensor according to the road section risk level, and uses the constructed road system digital twin model to simulate and verify the implementation effect of the resource allocation plan under different road section risk levels, and adopts a multi-objective optimization algorithm to optimize the resource allocation plan; an application business module converts the risk level into specific business actions; the coordinated operation of the above modules realizes the flexible allocation of road behavior perception, communication and computing resources, and improves the resource utilization efficiency and system service performance.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent perception and application of road behavior, and in particular to a road behavior adaptive topological perception and application system and method. Background Art

[0002] As smart roads continue to develop, road state perception systems, as an important support for road traffic infrastructure monitoring and services that integrates advanced information and digital technologies such as the Internet of Things, big data, and artificial intelligence, deploy various sensors and monitoring equipment inside and outside the road, such as fiber grating strain sensors, earth pressure cells, cameras, radars, infrared detectors, etc., to collect real-time data on the structural state parameters of the road itself. Combined with high-performance communication transmission technology and advanced data processing and analysis algorithms, real-time perception and prediction of the road's state are achieved, which is of great significance for ensuring road safety and improving road service efficiency.

[0003] Traditional road performance perception systems typically utilize fixed sensor network deployments. For example, patent application publication number CN118933068A discloses a method and device for monitoring the status of power pipe groups in key sections of smart roads. This method utilizes a network of pipe pillow pressure sensors to achieve refined monitoring of the status of power pipe groups in key sections. However, this fixed sensor network layout can easily lead to redundant perception resources in conventional sections and insufficient monitoring capabilities for abnormal events in key sections. Furthermore, it is unable to dynamically adjust the activation logic and resource allocation of perception equipment based on road risks, restricting the efficiency of road safety warnings and emergency response. Ultimately, the current road performance perception system is unable to adapt to different road scenarios and the dynamic risks of actual road service, resulting in low monitoring efficiency and service effectiveness. Summary of the Invention

[0004] Aiming at the low monitoring efficiency and service effectiveness of the current road behavior perception system, a road behavior adaptive topology perception and application system and method are provided.

[0005] This application provides a road behavior adaptive topology perception and application system. The system includes:

[0006] A performance cluster perception module, configured to monitor the road structure performance parameters in real time through a heterogeneous perception network composed of multimodal sensors;

[0007] A road section dynamic risk level assessment module, configured to calculate a road section risk index based on the road structural performance parameters and a risk assessment model, and obtain a road section risk level based on the road section risk index;

[0008] A data transmission communication module, configured to use a layered communication protocol to provide low-latency transmission of abnormal events and low-power transmission of regular events, and to perform spatiotemporal alignment of multi-source heterogeneous data generated by the heterogeneous sensing network based on three-dimensional geographic grid coding and a precise time protocol;

[0009] The node adjustment and resource allocation module is used to adjust the activation state and sampling frequency of each sensor in the heterogeneous perception network according to the risk level of the road section; it is also used to use the constructed road system digital twin model to simulate and verify the implementation effect of various resource allocation schemes under different road section risk levels, and use a multi-objective optimization algorithm to obtain the optimal resource allocation scheme.

[0010] Furthermore, the road section dynamic risk level assessment module is also used to dynamically adjust the weights in the risk assessment model based on historical monitoring data and the LSTM-GRU hybrid network model; wherein, the LSTM-GRU hybrid network model includes an input layer, an LSTM hidden layer, a fully connected layer and an output layer, the input layer is used to receive the historical monitoring data, the LSTM hidden layer includes three layers of LSTM units, each layer of LSTM units includes 128 neurons, each neuron has a forget gate, an input gate and an output gate structure, the fully connected layer includes a first fully connected layer and a second fully connected layer, the first fully connected layer includes 64 neurons, the second fully connected layer includes 32 neurons, and the output layer is used to output environmental prediction parameters and weight adjustment coefficients for a period of time in the future.

[0011] Furthermore, during the training phase of the LSTM-GRU hybrid network model, the road section dynamic risk level assessment module uses the AdamW optimizer to implement adaptive learning rate adjustment and weight decay, and configures an early stopping mechanism to prevent overfitting. The loss function of the LSTM-GRU hybrid network is the Focal Loss loss function.

[0012] Furthermore, the LSTM-GRU hybrid network model assigns different weights to prediction results of different time scales through the attention mechanism to perform short-term, medium-term and long-term predictions, and the output of the LSTM-GRU hybrid network model also includes a confidence interval of the prediction result.

[0013] Furthermore, the node adjustment and resource allocation module is also used to calculate the contribution of each sensor in the heterogeneous perception network to risk assessment based on historical monitoring data, and use an integrated learning method to calculate the information gain of different sensor combinations to obtain a set of high-value perception nodes, and activate each node in the set of high-value perception nodes according to the real-time road section risk level.

[0014] Furthermore, the data transmission communication module is also used to obtain the timestamp message sent by each node in the heterogeneous perception network, and calculate the transmission delay of each node based on the system main clock source, perform time compensation on the timestamp message of each node according to the transmission delay, and use the compensated time as the data collection time label of each node.

[0015] Furthermore, the road behavior adaptive topology perception and application system also includes:

[0016] The application business module is used to map the road section risk level to specific business actions and conduct cross-departmental business closed-loop linkage.

[0017] In addition, this application also provides a method for adaptive topological perception and application of road properties. The method includes:

[0018] Step S1, obtaining road structural parameters monitored in real time by a heterogeneous sensing network composed of multimodal sensors;

[0019] Step S2, calculating a road section risk index based on the road structural performance parameters and the risk assessment model, and obtaining a road section risk level based on the road section risk index;

[0020] Step S3: using a layered communication protocol to transmit abnormal events with low latency and regular events with low power consumption, and performing spatiotemporal alignment of multi-source heterogeneous data generated by the heterogeneous sensing network based on three-dimensional geographic grid coding and precise time protocol;

[0021] In step S4, the activation state and sampling frequency of each sensor in the heterogeneous perception network are adjusted according to the risk level of the road section, and the implementation effect of various resource allocation schemes under different road section risk levels is simulated and verified using the constructed digital twin model of the road system, and the optimal resource allocation scheme is obtained using a multi-objective optimization algorithm.

[0022] Furthermore, the risk assessment model is expressed as:

[0023]

[0024] Where RI is the comprehensive risk index of the road section; G is the roadbed deformation rate, ranging from 0 to 1; A is the fatigue damage index of the asphalt layer, ranging from 0 to 1; B is the base layer void risk index, ranging from 0 to 1; α, β, and γ are all weight coefficients. The calculation formula for the roadbed deformation rate is:

[0025]

[0026] Where, is the cumulative settlement of the roadbed during the continuous monitoring period (mm), is the monitoring time interval (days), is the critical deformation rate threshold (mm / day);

[0027] The calculation formula of the asphalt layer fatigue damage index is:

[0028]

[0029] Where, Strain level The actual number of load actions under is the tensile strain at the bottom of the asphalt layer (με), is the elastic modulus of asphalt mixture (MPa), are experimental parameters related to the properties of asphalt materials;

[0030] The calculation formula of the grassroots void risk index is:

[0031]

[0032] Where, The bottom pressure of the base layer monitored by the pressure sensor array Exceeding the degassing pressure threshold Area (kPa), is the total monitoring area (m 2 ).

[0033] Furthermore, the steps of encoding the three-dimensional geographic grid include:

[0034] Step S301, dividing the road and surrounding environment into three-dimensional grids to form basic spatial units;

[0035] Step S302: adopting a hierarchical nested coding strategy, applying Geohash or H3 algorithm in the horizontal direction to generate plane coordinate codes, and superimposing Z-axis hierarchical codes in the vertical direction to obtain a unique geographic code for each basic spatial unit;

[0036] Step S303 : Mapping the physical locations of sensors in the heterogeneous sensing network to corresponding three-dimensional grid cells, and using their geocoding as spatial labels for data collected by the sensors.

[0037] The above-mentioned road behavior adaptive topological perception and application system and method provides a flexible scheduling mechanism for perception resources based on real-time risk levels, specifically including real-time monitoring of road structural behavior parameters through a behavior cluster perception module; a section dynamic risk level assessment module calculates the section risk index based on the road structural behavior parameters and the risk assessment model, and obtains the section risk level based on the section risk index; a node adjustment and resource allocation module adjusts the activation state and sampling frequency of each sensor in the heterogeneous perception network composed of multimodal sensors according to the section risk level, and uses the constructed road system digital twin model to simulate and verify the implementation effect of various resource allocation schemes under different section risk levels, and adopts a multi-objective optimization algorithm to obtain the optimal resource allocation scheme. This application realizes the flexible allocation of perception and computing resources according to the dynamic risk level of the section, while ensuring that key sections and abnormal events are fully perceived, avoiding resource waste in conventional sections, and significantly improving the efficiency of the road behavior perception and application system; at the same time, the data transmission and communication module performs spatiotemporal alignment of multi-source heterogeneous data, further improving the reliability of the monitoring data. This system can provide more accurate and efficient decision-making and services for traffic management, road maintenance, public travel, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A schematic diagram of the structure of a road behavior adaptive topology perception and application system in one embodiment;

[0039] Figure 2 A schematic diagram showing a comparison of the training accuracy of an LSTM-GRU hybrid model, an LSTM model, and a traditional time series model in one embodiment;

[0040] Figure 3 Schematic diagram of the overall comparison of model performance among the LSTM-GRU hybrid model, the LSTM model, and the traditional time series model in one embodiment. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solutions and advantages of this application clearer, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0042] Example 1

[0043] like Figure 1 As shown, this embodiment provides a road behavior adaptive topology perception and application system, including a behavior cluster perception module, a road section dynamic risk level assessment module, a data transmission and communication module, and a node adjustment and resource allocation module.

[0044] The performance cluster perception module is used to monitor road structure performance parameters in real time through a heterogeneous perception network composed of multimodal sensors.

[0045] Among them, the road structural performance parameters include the vertical deformation rate and collapse rate of the roadbed, the mechanical response of the pavement structure layer, environmental parameters, and the road surface condition. The heterogeneous perception network composed of multimodal sensors includes different types of perception terminals, and each perception terminal monitors different road structural performance parameters in real time. For example, the perception terminal may include a distributed fiber grating sensor, a piezoelectric sensor array, a humidity and temperature sensor, and a high-resolution camera to form a monitoring system covering the road body and the surrounding environment. The distributed fiber grating sensor is used to collect the vertical deformation rate, settlement, and tensile strain of the asphalt layer at the bottom of the roadbed in real time. The piezoelectric sensor array is used to collect the mechanical response of the pavement structure layer in real time, including the tensile strain at the bottom of the asphalt layer and the pressure at the bottom of the base layer. The humidity and temperature sensor is used to monitor the environmental parameters such as temperature and humidity inside and outside the road. The high-resolution camera is used to visually monitor the road surface condition.

[0046] Furthermore, the deployment of sensing terminals is flexible and can be adjusted based on factors such as road grade, traffic volume, and historical risks. For example, distributed fiber Bragg grating sensors can be deployed in a grid-like configuration, forming a continuous monitoring baseline at 50-meter intervals along the longitudinal direction of the road. The wavelength drift characteristics of the Bragg grating can achieve millimeter-level deformation resolution, accurately capturing the vertical settlement rate of the roadbed and potential signs of collapse. Piezoelectric sensor arrays can deploy multiple sensing lines horizontally along the asphalt layer to collect real-time interlayer stress distribution and dynamic strain modulus changes under vehicle loads. Humidity and temperature sensor networks are deployed at 10-meter intervals on the shoulders and central median strips to continuously monitor the impact of temperature and humidity gradients on asphalt material properties. High-resolution cameras adopt a layered, three-dimensional deployment strategy, with a monitoring point set up every 1-2 kilometers along the longitudinal direction of the road. Camera groups with differentiated vertical angles are added at key nodes to form a comprehensive visual monitoring network covering road conditions and the surrounding environment. Key nodes are defined as those on sections with higher risk levels, such as those with long longitudinal slopes, fill sections, cut sections, partially fill and cut sections, sections with special soil subgrade, curved sections, bridge transition sections, and on / off ramps. The vertically differentiated camera system includes lenses with varying pitch angles, such as 30°, 60°, and 90°. The 30° lens is a low-angle lens capable of focusing on localized road surface details, such as crack width and other microscopic features; the 60° lens is a medium-angle lens capable of covering the entire road surface within a single lane; and the 90° lens is a vertically oriented lens capable of capturing the entire road surface. This multi-angle combination eliminates blind spots through perspective superposition, thereby improving monitoring accuracy.

[0047] The road section dynamic risk level assessment module is used to calculate the road section risk index based on the road structural parameters and the risk assessment model, and obtain the road section risk level based on the road section risk index.

[0048] The road section risk index can be the weighted average sum of the road structural parameters. In a preferred embodiment, the specific mathematical expression of the risk assessment model is shown in formula (1):

[0049] (1)

[0050] Where RI is the comprehensive risk index of the road section; G is the roadbed deformation rate, ranging from 0 to 1; A is the fatigue damage index of the asphalt layer, ranging from 0 to 1; B is the base layer void risk index, ranging from 0 to 1; α, β, and γ are all weight coefficients.

[0051] The roadbed deformation rate reflects the deformation speed of the roadbed under load and is an important indicator for evaluating roadbed stability. The roadbed deformation rate can be calculated based on the cumulative settlement of the roadbed during the monitoring period, the monitoring time interval, and the critical deformation rate threshold. The specific calculation formula is shown in formula (2):

[0052] (2)

[0053] Where, is the cumulative settlement of the roadbed during the continuous monitoring period (mm), is the monitoring time interval (days), is the critical deformation rate threshold (mm / day), usually 0.05~0.2 mm / day.

[0054] The fatigue damage index of the asphalt layer reflects the cumulative damage degree of the asphalt mixture under cyclic loading. It is calculated based on the actual number of load actions at the strain level, the tensile strain at the bottom of the asphalt layer, and the elastic modulus of the asphalt mixture. The specific calculation formula is shown in formula (3):

[0055] (3)

[0056] Where, Strain level The actual number of load actions under is the tensile strain at the bottom of the asphalt layer ( ), is the elastic modulus of asphalt mixture (MPa), are experimental parameters related to the properties of asphalt materials.

[0057] The base layer void risk index reflects the degree of separation between the base layer and the underlying structural layer, and is mainly calculated by monitoring the stress distribution characteristics between layers. The specific calculation formula is shown in formula (4):

[0058] (4)

[0059] Where, The bottom pressure of the base layer monitored by the pressure sensor array Exceeding the degassing pressure threshold Area (kPa), is the total monitoring area (m 2 ).

[0060] The road section risk level can be divided into low risk, medium risk and high risk. In a preferred embodiment, when the comprehensive risk index of a road section is less than 0.3, the road section risk level is low risk; when the comprehensive risk index of a road section is greater than or equal to 0.3 and less than 0.6, the road section risk level is medium risk; when the comprehensive risk index of a road section is greater than or equal to 0.6, the road section risk level is high risk.

[0061] Furthermore, the road section dynamic risk level assessment module of this embodiment is also used to dynamically adjust the weight coefficients in the risk assessment model based on historical monitoring data and the LSTM-GRU hybrid network model. Specifically, the LSTM-GRU hybrid network model includes an input layer, an LSTM hidden layer, a fully connected layer and an output layer. The input layer is used to receive historical monitoring data. Specifically, the historical monitoring data can be multidimensional time series data including historical temperature and humidity data, historical rainfall and historical traffic load data. The LSTM hidden layer includes three layers of LSTM units, each layer of LSTM units includes 128 neurons, and each neuron has a forget gate, input gate and output gate structure. The fully connected layer includes a first fully connected layer and a second fully connected layer. The first fully connected layer includes 64 neurons, and the second fully connected layer includes 32 neurons. The first fully connected layer and the second fully connected layer both use the ReLU activation function. The output layer is used to output environmental prediction parameters and weight adjustment coefficients for a period of time in the future, for example, outputting environmental prediction parameters and weight adjustment coefficients for the next 24 hours.

[0062] Furthermore, the LSTM-GRU hybrid network model of this embodiment uses an attention mechanism to assign different weights to prediction results at different time scales, enabling short-term, medium-term, and long-term predictions. Specifically, short-term predictions (1-3 hours) are achieved by capturing high-frequency dynamic features based on the spatiotemporal attention mechanism. A periodic pattern encoder is constructed using GRU units to achieve medium-term predictions (24 hours), and long-term predictions (7 days) are achieved through the LSTM hidden layer. Each branch output is dynamically weighted through adaptive fusion gating, with the weight coefficients generated in real time based on the covariance matrix of historical prediction errors, forming a collaborative optimization mechanism for short-term, medium-term, and long-term prediction results.

[0063] In terms of training optimization, the LSTM-GRU hybrid network model uses Focal Loss to improve the cross-entropy loss function. This modulates factors to enhance attention to unusual traffic events, and uses the AdamW optimizer (weight decay coefficient λ = 1e-5) to achieve adaptive parameter updates. To prevent overfitting, the model also employs an early stopping strategy, terminating training after 10 rounds of validation set loss without improvement. Regular updates to the model parameters ensure the model's continued adaptability to dynamic traffic environments. To address the sparsity of traffic scenario data, the LSTM-GRU hybrid network model employs a progressive transfer learning strategy. Specifically, parameters are first initialized on historical road segment data, followed by a feature alignment layer to eliminate spatial and temporal differences. Subsequently, progressive transfer learning is implemented on new road segment data to fine-tune the model parameters. Figure 2 The comparison between the accuracy of the LSTM-GRU hybrid network model after adopting this training optimization strategy and the traditional time series model and LSTM model during the training process is shown. As can be seen from the figure, the accuracy of the LSTM-GRU hybrid network model during the training process is higher. In addition, Figure 3 This figure shows an overall performance comparison of the LSTM-GRU hybrid network model with traditional time series models and LSTM models. The figure shows that the LSTM-GRU hybrid network model performs better overall than the other models, and that the LSTM-GRU hybrid network model improves prediction accuracy by approximately 18% compared to traditional time series models.

[0064] In addition, the output of the LSTM-GRU hybrid network model also includes the confidence interval of the prediction results, which provides an uncertainty measure for risk assessment and further improves the reliability of the prediction results.

[0065] The data transmission communication module is used to transmit abnormal events with low latency and routine events with low power consumption using a layered communication protocol. The layered communication protocol includes a high-speed communication protocol and a low-power wide area network protocol. The high-speed communication protocol includes, but is not limited to, the V2X or 5G-Uu protocol, and the low-power wide area network protocol includes, but is not limited to, the LoRa or NB-IoT protocol. Specifically, different transmission methods are selected based on the importance and timeliness of the data. For example, urgent data such as abnormal events (such as severe deformation alarms) is transmitted with low latency using a high-speed communication protocol, while routine events (such as road structural parameters monitored daily) are transmitted with low power consumption using a low-power wide area network protocol to reduce energy consumption.

[0066] Furthermore, to address the spatiotemporal alignment of multi-source heterogeneous data, the data transmission and communication module is also used to perform spatiotemporal alignment of multi-source heterogeneous data generated by a heterogeneous sensing network composed of multimodal sensors based on 3D geo-grid coding and a precision time protocol. Multi-source heterogeneous data refers to the different types of data generated by various sensors after pre-processing the collected road structural parameters. Pre-processing can include data deduplication, format conversion, and outlier processing. The specific steps of 3D geo-grid coding include: first, a 3D grid is created for the road and surrounding environment to form basic spatial units (such as cubes or hexagonal columns). Then, a hierarchical nested coding strategy is employed, applying the Geohash or H3 algorithm horizontally to generate planar coordinate codes. Z-axis hierarchical codes are superimposed vertically to obtain unique geocodes for each basic spatial unit. Next, the physical locations of the sensors are mapped to the corresponding 3D grid units, and the geocodes of the 3D grid units are used as spatial labels for the data collected by the sensors. Three-dimensional geographic grid coding uses the hierarchical structure characteristics of the coding to achieve rapid retrieval and association of spatial data of different ranges and granularities. Based on the same or adjacent coding areas, it performs spatial aggregation and fusion analysis on multi-source sensor data, effectively solving the spatial alignment problem of multi-source heterogeneous data and realizing spatial association of multi-source sensor data.

[0067] The data transmission and communication module is also used to obtain the timestamp messages sent by each node in the heterogeneous perception network composed of multimodal sensors, and calculate the transmission delay of each node based on the system master clock source, and perform time compensation on the timestamp messages of each node according to the transmission delay, and use the compensated time as the data collection time label of each node.

[0068] Specifically, the system builds a master-slave clock architecture based on the Precision Time Protocol (PTP). It deploys a master clock source as the global time reference, with each sensing terminal acting as a slave clock node. This system achieves nanosecond-level synchronization accuracy through a bidirectional timestamp exchange mechanism. The specific workflow is as follows: Each node in the heterogeneous sensing network composed of multimodal sensors continuously sends timestamp messages carrying local clock information to the data transmission and communication module. The data transmission and communication module calculates the transmission delay of each node based on the system master clock source and offset-compensates the timestamps of each node based on the real-time measured transmission delay. This generates a corrected unified time reference, which serves as the time tag for multi-source heterogeneous data to ensure data temporal consistency. Furthermore, the system supports an automatic recovery mechanism within the network environment to ensure long-term operational stability. This precise time synchronization mechanism ensures the accuracy of temporal correlation analysis of multi-source heterogeneous data and is an important foundation for multi-source data fusion.

[0069] The data transmission and communication module also supports an edge-cloud collaborative computing architecture, flexibly adjusting the allocation of computing tasks based on data processing needs. Specifically, when processing time-sensitive data, such as generating an alarm when a road structural state parameter exceeds a threshold, data processing can be performed on the sensor side. When processing data with higher computational complexity, the sensor uploads the monitored road structural state parameters to the cloud for relevant calculations. For example, the data transmission and communication module can send the monitored road structural state parameters to the cloud to calculate the risk level of a road section. The cloud side also deploys a risk assessment model to further improve monitoring efficiency.

[0070] The node adjustment and resource allocation module is used to adjust the activation status and sampling frequency of each sensor in the heterogeneous perception network composed of multimodal sensors based on the road section risk level. For example, when the road section risk level is low, some sensors (approximately 20%) are activated and the sampling frequency is reduced. When the road section risk level is medium, most sensors (approximately 70%) are activated and the sampling frequency is adjusted appropriately. When the road section risk level is high, all sensors (approximately 100%) are activated, the sampling frequency is increased, and more edge computing resources are allocated.

[0071] The node adjustment and resource allocation module is also used to use the constructed digital twin model of the road system to simulate and verify the implementation effects of various resource allocation schemes under different road section risk levels, and use a multi-objective optimization algorithm to obtain the optimal resource allocation scheme.

[0072] The road system digital twin model consists of a geometric model, a physical model, and a behavioral model. The geometric model accurately reproduces the road structure. The physical model simulates the road's material properties. The behavioral model simulates traffic flow and environmental effects. The model parameters of the road system digital twin are inverted from measured data, establishing a mapping between the real world and the digital model. Simultaneously, it supports simulations at both the macro-network level and the micro-structure level, enabling cross-scale analysis.

[0073] Specifically, the digital twin model simulates the effects of various resource allocation schemes under different road risk levels and employs a multi-objective optimization algorithm to find the optimal resource allocation solution that satisfies multiple constraints. Constraints include, but are not limited to, monitoring coverage, data accuracy, energy consumption, and response time. By establishing a comprehensive evaluation metric encompassing these factors, the team aims to find the optimal balance between monitoring quality and resource consumption.

[0074] The node adjustment and resource allocation module also collects actual performance data after implementing the optimal resource allocation plan and optimizes the parameters of the road system digital twin model based on this data. The specific implementation process is as follows: the node adjustment and resource allocation module assesses the current risk status every five minutes, calculates the section risk index for each road section, activates / deactivates nodes on each section based on the section risk index, and generates the optimal resource allocation plan based on the road system digital twin model and a multi-objective optimization algorithm. Afterwards, the module collects actual performance data and updates the optimization model parameters.

[0075] Furthermore, the node adjustment and resource allocation module is also used to evaluate the sensitivity of resource configuration to the accuracy of risk assessment and determine key resource points.

[0076] Resource allocation involves activating and dormant different sensors within a heterogeneous perception network comprised of multimodal sensors. By controlling the activation and dormancy of these sensors, the magnitude of changes in the road section risk index is observed. When the magnitude of these changes exceeds a certain threshold, the activated sensors in this resource allocation are identified as having a significant impact on risk assessment accuracy and are designated as key resource points.

[0077] Furthermore, this embodiment also optimizes the activation logic of the sensor. Specifically, the node adjustment and resource allocation module is also used to calculate the contribution of each sensor in the heterogeneous perception network composed of multimodal sensors to risk assessment based on historical monitoring data, and use an integrated learning method to calculate the information gain of different sensor combinations to obtain a set of high-value perception nodes, and activate each node in the high-value perception node set according to the real-time road section risk level.

[0078] Historical monitoring data includes data from various sensors under different environmental conditions and road risk level scenarios. The node adjustment and resource allocation module uses sensitivity analysis to deeply mine this historical monitoring data, identifying the contribution of different sensor data to the road risk level calculation. This module then assesses the uniqueness and importance of the information provided by each sensor in the risk assessment process. Furthermore, the node adjustment and resource allocation module uses the random forest algorithm to calculate the information gain of different sensor combinations. The random forest algorithm improves overall classification or regression accuracy by constructing multiple decision tree models and integrating their prediction results. Specifically, the random forest algorithm uses different sensor combinations as input features and the calculated road risk level as the target variable. It trains multiple decision tree models to simulate the impact of different sensor combinations on the road risk level calculation. During training, the algorithm automatically learns the weight of each sensor combination and the degree to which each combination improves risk assessment accuracy, i.e., the information gain. This ensemble learning-based calculation method leverages the complementarity of multiple models, effectively avoiding the potential overfitting or underfitting of a single model, resulting in more robust and accurate information gain assessment results for sensor combinations.

[0079] Based on the contribution evaluation and information gain calculation results described above, the node adjustment and resource allocation module further explores the distribution patterns of contributions of different sensor combinations to risk assessment. By setting appropriate thresholds or optimization objectives (such as maximizing information gain or minimizing perception cost), the node adjustment and resource allocation module can screen out high-value perception nodes from all possible sensor combinations that significantly contribute to risk assessment. These high-value perception nodes not only possess high data quality and accuracy but also provide unique and complementary information in different risk scenarios, helping to improve the comprehensiveness and reliability of risk assessments.

[0080] Finally, based on the dynamic changes in real-time road section risk levels, the node adjustment and resource allocation module flexibly activates individual nodes in the high-value perception node set. For example, when the module detects an increase in the risk level of a road section, it immediately activates the corresponding perception nodes, increasing the perception density and accuracy in that area to obtain more detailed and timely risk information. Conversely, when the risk level decreases, the module appropriately shuts down some non-high-value perception nodes to conserve perception resources and reduce network energy consumption. This risk-based dynamic node activation strategy not only improves the perception network's responsiveness and adaptability, but also effectively extends the network's overall service life, providing a strong guarantee for the stable operation of heterogeneous perception networks composed of multimodal sensors in complex environments.

[0081] Furthermore, the road behavior adaptive topology perception and application system also includes:

[0082] The application business module is used to map the road section risk level to specific business actions and conduct cross-departmental business closed-loop linkage. Specifically, different levels of lane speed limits or closures are implemented according to the road section risk level, dynamic route planning and risk warnings are pushed through the navigation application, and a maintenance priority list containing risk type, location coordinates, and urgency is generated. The list is automatically synchronized to the municipal maintenance system and a work order is dispatched. When the road section risk level exceeds the threshold, a cross-departmental collaborative response mechanism is automatically triggered. In addition, the application business module provides a standardized API interface to support data interoperability with heterogeneous systems such as traffic management and municipal management, ensuring the timely flow of information and efficient business linkage.

[0083] The adaptive road behavior topology perception and application system of this embodiment utilizes: first, a dynamic risk-driven resource allocation mechanism that dynamically adjusts sensor activation status and acquisition frequency based on risk level, achieving "on-demand perception," ensuring high-quality monitoring of key road sections while avoiding resource waste on regular sections. Second, a multi-source heterogeneous data fusion method that utilizes three-dimensional geographic grid encoding and precise time protocols to address the spatiotemporal alignment of multi-source heterogeneous data and improve the accuracy of the risk assessment model. Third, an intelligent weight adjustment algorithm that introduces an LSTM network to dynamically adjust the weights of the risk assessment model, effectively adapting to changing environmental conditions and improving risk prediction accuracy. Fourth, a cross-departmental business linkage framework establishes a mapping between risk level and business actions, enabling rapid response from abnormal events to emergency disposal, significantly improving road safety management. Through the collaborative design and innovative operation of various modules, the system flexibly allocates perception and computing resources based on road section risk level, ensuring sufficient perception of key sections and abnormal events while avoiding resource waste on regular sections. This significantly improves the overall performance of the intelligent road behavior perception and application system, providing more accurate and efficient support and services for traffic management, road maintenance, and public transportation.

[0084] Example 2

[0085] This embodiment provides a method for adaptive topology perception and application of road behavior, which specifically includes the following steps:

[0086] Step S1: Acquire road structural parameters monitored in real time by a heterogeneous sensing network composed of multimodal sensors.

[0087] Step S2: Calculate the road section risk index based on the road structural parameters and the risk assessment model, and obtain the road section risk level based on the road section risk index.

[0088] In step S3, a layered communication protocol is used to transmit abnormal events with low latency and regular events with low power consumption, and multi-source heterogeneous data generated by the heterogeneous perception network are spatiotemporally aligned based on three-dimensional geographic grid coding and precise time protocol.

[0089] In step S4, the activation state and sampling frequency of each sensor in the heterogeneous perception network are adjusted according to the road section risk level. The constructed digital twin model of the road system is used to simulate and verify the implementation effects of various resource allocation schemes under different road section risk levels, and a multi-objective optimization algorithm is used to obtain the optimal resource allocation scheme.

[0090] Furthermore, the road behavior adaptive topology perception and application method also includes:

[0091] Step S5: Dynamically adjust the weights in the risk assessment model based on historical monitoring data and the LSTM-GRU hybrid network model.

[0092] The LSTM-GRU hybrid network model consists of an input layer, an LSTM hidden layer, a fully connected layer, and an output layer. The input layer receives historical monitoring data. The LSTM hidden layer consists of three layers of LSTM units, each with 128 neurons, each equipped with a forget gate, an input gate, and an output gate. The fully connected layers include a first fully connected layer (64 neurons) and a second fully connected layer (32 neurons). The output layer outputs environmental prediction parameters and weight adjustment coefficients for the future period. Furthermore, during the training phase of the LSTM-GRU hybrid network model, the method uses the AdamW optimizer for adaptive learning rate adjustment and weight decay, and implements an early stopping mechanism to prevent overfitting. The loss function of the LSTM-GRU hybrid network is the Focal Loss function. Specifically, the LSTM-GRU hybrid network model uses an attention mechanism to assign different weights to prediction results at different time scales, enabling short-term, medium-term, and long-term predictions. The output of the LSTM-GRU hybrid network also includes confidence intervals for the prediction results.

[0093] Furthermore, the road behavior adaptive topology perception and application method also includes:

[0094] Step S6: Calculate the contribution of each sensor in the heterogeneous perception network to risk assessment based on historical monitoring data, and use an integrated learning method to calculate the information gain of different sensor combinations to obtain a set of high-value perception nodes. Activate each node in the set of high-value perception nodes according to the real-time road section risk level.

[0095] Furthermore, the road behavior adaptive topology perception and application method also includes:

[0096] Step S7: Evaluate the sensitivity of resource allocation to risk assessment accuracy and determine key resource points.

[0097] Furthermore, the road behavior adaptive topology perception and application method also includes:

[0098] Step S8, obtain the timestamp message sent by each node in the heterogeneous perception network, and calculate the transmission delay of each node based on the system master clock source, perform time compensation on the timestamp message of each node according to the transmission delay, and use the compensated time as the data collection time tag of each node.

[0099] Furthermore, the road behavior adaptive topology perception and application method also includes:

[0100] Step S9: Map the road section risk level to specific business actions to perform cross-departmental business closed-loop linkage.

[0101] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned 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.

[0102] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A road behavior adaptive topology perception and application system, characterized by: The system comprises: A performance cluster perception module, configured to monitor the road structure performance parameters in real time through a heterogeneous perception network composed of multimodal sensors; A road section dynamic risk level assessment module, configured to calculate a road section risk index based on the road structural performance parameters and a risk assessment model, and obtain a road section risk level based on the road section risk index; The risk assessment model is expressed as: ； Where RI is the comprehensive risk index of the road section; G is the roadbed deformation rate, ranging from 0 to 1; A is the fatigue damage index of the asphalt layer, ranging from 0 to 1; B is the base layer void risk index, ranging from 0 to 1; α, β, and γ are all weight coefficients. The calculation formula for the roadbed deformation rate is: ； Where, is the cumulative settlement of the roadbed during the continuous monitoring period, is the monitoring time interval, is the critical deformation rate threshold; The calculation formula of the asphalt layer fatigue damage index is: ； Where, Strain level The actual number of load actions under is the tensile strain at the bottom of the asphalt layer, is the elastic modulus of asphalt mixture, are experimental parameters related to the properties of asphalt materials; The calculation formula of the grassroots void risk index is: ； Where, The bottom pressure of the base layer monitored by the pressure sensor array Exceeding the degassing pressure threshold The area, is the total monitoring area; A data transmission communication module, configured to use a layered communication protocol to provide low-latency transmission of abnormal events and low-power transmission of regular events, and to perform spatiotemporal alignment of multi-source heterogeneous data generated by the heterogeneous sensing network based on three-dimensional geographic grid coding and a precise time protocol; The node adjustment and resource allocation module is used to adjust the activation state and sampling frequency of each sensor in the heterogeneous perception network according to the risk level of the road section; and is used to simulate and verify the implementation effect of various resource allocation schemes under different road section risk levels through the constructed road system digital twin model, and then use a multi-objective optimization algorithm to obtain the optimal resource allocation scheme.

2. The road behavior adaptive topology perception and application system according to claim 1 is characterized in that: The road section dynamic risk level assessment module is also used to dynamically adjust the weights in the risk assessment model based on historical monitoring data and the LSTM-GRU hybrid network model; wherein, the LSTM-GRU hybrid network model includes an input layer, an LSTM hidden layer, a fully connected layer and an output layer, the input layer is used to receive the historical monitoring data, the LSTM hidden layer includes three layers of LSTM units, each layer of LSTM units includes 128 neurons, each neuron has a forget gate, an input gate and an output gate structure, the fully connected layer includes a first fully connected layer and a second fully connected layer, the first fully connected layer includes 64 neurons, the second fully connected layer includes 32 neurons, and the output layer is used to output environmental prediction parameters and weight adjustment coefficients for a period of time in the future.

3. The road behavior adaptive topology perception and application system according to claim 2, characterized in that: During the LSTM-GRU hybrid network model training phase, the road section dynamic risk level assessment module uses the AdamW optimizer to implement adaptive learning rate adjustment and weight decay, and configures an early stopping mechanism to prevent overfitting. The loss function of the LSTM-GRU hybrid network is the Focal Loss loss function.

4. The road behavior adaptive topology perception and application system according to claim 2 or 3, characterized in that: The LSTM-GRU hybrid network model assigns different weights to prediction results of different time scales through the attention mechanism to perform short-term, medium-term and long-term predictions, and the output of the LSTM-GRU hybrid network model also includes a confidence interval of the prediction result.

5. The road behavior adaptive topology perception and application system according to claim 1 is characterized in that: The node adjustment and resource allocation module is also used to calculate the contribution of each sensor in the heterogeneous perception network to risk assessment based on historical monitoring data, and use an integrated learning method to calculate the information gain of different sensor combinations to obtain a set of high-value perception nodes, and activate each node in the set of high-value perception nodes according to the real-time road section risk level.

6. The road behavior adaptive topology perception and application system according to claim 1, characterized in that: The data transmission communication module is also used to obtain the timestamp message sent by each node in the heterogeneous perception network, and calculate the transmission delay of each node based on the system main clock source, perform time compensation on the timestamp message of each node according to the transmission delay, and use the compensated time as the data collection time label of each node.

7. The road behavior adaptive topology perception and application system according to claim 1, characterized in that: The system further comprises: The application business module is used to map the road section risk level to specific business actions and conduct cross-departmental business closed-loop linkage.

8. A road behavior adaptive topology perception and application method, characterized in that: The method comprises: Step S1, obtaining road structural parameters monitored in real time by a heterogeneous sensing network composed of multimodal sensors; Step S2: Calculate the road section risk index based on the road structural parameters and the risk assessment model, and obtain the road section risk level based on the road section risk index; the expression of the risk assessment model is: ； Where RI is the comprehensive risk index of the road section; G is the roadbed deformation rate, ranging from 0 to 1; A is the fatigue damage index of the asphalt layer, ranging from 0 to 1; B is the base layer void risk index, ranging from 0 to 1; α, β, and γ are all weight coefficients. The calculation formula for the roadbed deformation rate is: ； Where, is the cumulative settlement of the roadbed during the continuous monitoring period, is the monitoring time interval, is the critical deformation rate threshold; The calculation formula of the asphalt layer fatigue damage index is: ； Where, Strain level The actual number of load actions under is the tensile strain at the bottom of the asphalt layer, is the elastic modulus of asphalt mixture, are experimental parameters related to the properties of asphalt materials; The calculation formula of the grassroots void risk index is: ； Where, The bottom pressure of the base layer monitored by the pressure sensor array Exceeding the degassing pressure threshold The area, is the total monitoring area; Step S3: using a layered communication protocol to transmit abnormal events with low latency and regular events with low power consumption, and performing spatiotemporal alignment of multi-source heterogeneous data generated by the heterogeneous sensing network based on three-dimensional geographic grid coding and precise time protocol; In step S4, the activation state and sampling frequency of each sensor in the heterogeneous perception network are adjusted according to the risk level of the road section, and the implementation effect of various resource allocation schemes under different road section risk levels is simulated and verified using the constructed digital twin model of the road system, and the optimal resource allocation scheme is obtained using a multi-objective optimization algorithm.

9. The road behavior adaptive topology perception and application method according to claim 8, characterized in that: The steps of 3D geographic grid encoding include: Step S301, dividing the road and surrounding environment into three-dimensional grids to form basic spatial units; Step S302: adopting a hierarchical nested coding strategy, applying Geohash or H3 algorithm in the horizontal direction to generate plane coordinate codes, and superimposing Z-axis hierarchical codes in the vertical direction to obtain a unique geographic code for each basic spatial unit; Step S303 : Mapping the physical locations of the sensors in the heterogeneous sensing network to corresponding three-dimensional grid cells, and using the geographic codes of the three-dimensional grid cells as spatial labels for the data collected by the sensors.

Images