A safety management system for highway maintenance

Abstract

The present invention relates to the technical field of highway maintenance management, and in particular to a highway maintenance safety management system. The system includes a road condition monitoring module, a traffic load assessment module, a traffic flow redistribution module, a maintenance timing determination module, and a maintenance effect verification module, which are used to monitor multi-layer road condition data of the highway, and perform table-deep correlation detection to obtain road surface abnormal point data; quantify the road surface loss risk of the road surface abnormal point data to obtain road surface loss early warning indicators; collect vehicle images and meteorological data; analyze vehicle model loss contribution based on vehicle images; construct an environmental impact coefficient matrix based on meteorological data; construct a dynamic weight model based on vehicle model loss contribution and environmental impact coefficient matrix to obtain a road surface loss dynamic weight model. The present invention improves the efficiency, benefit and intelligence level of highway maintenance management by organically integrating highway maintenance with traffic management.

Description

Technical Field

[0001] The present invention relates to the technical field of highway maintenance management, and in particular to a highway maintenance safety management system. Background Art

[0002] Highway maintenance management is a key link in the sustainable development of transportation infrastructure, involving the full life cycle management of inspection, evaluation, repair and maintenance of completed highways.

[0003] Traditional methods mainly rely on manual inspections or simple instrument detection, which can only detect road damage that has already appeared, such as obvious cracks and potholes. Early damage to the internal structure of the road (such as loose base and micro-crack expansion) cannot be discovered in time. This causes maintenance work to often lag behind the development of road diseases, forming a "passive response" mode. Once the road surface is obviously damaged, large-scale repairs are often required, which is not only costly, but also causes long-term traffic interruptions.

[0004] Traditional methods usually only count traffic loads based on traffic volume or simple vehicle classification (such as large vehicles and small vehicles), ignoring the huge differences in the actual wear and tear of road surfaces due to different vehicle types (such as axle weight, wheelbase, and tire pressure), different driving speeds, and different environmental conditions (such as temperature, humidity, and precipitation). This rough assessment method cannot accurately reflect the actual load on the road surface, resulting in a lack of targeted maintenance plans, and over-maintenance or under-maintenance may occur.

[0005] During road maintenance, traditional traffic control methods often use simple control measures, such as closing lanes, restricting traffic, etc. This method does not take into account the differences in actual travel needs of different users (such as commuting, emergency, freight, etc.), and takes the same control measures for all vehicles, which is likely to cause unnecessary traffic congestion and reduce the user's travel experience.

[0006] To sum up, the problems of passive response, inaccurate evaluation and extensive management and control in existing technologies need to be solved urgently. Summary of the invention

[0007] Based on this, it is necessary to provide a highway maintenance safety management system to solve at least one of the above technical problems.

[0008] To achieve the above purpose, a highway maintenance safety management system is provided, comprising the following modules: The road condition monitoring module is used to monitor the multi-layer road condition data of the highway to obtain the road surface vibration characteristic spectrum; perform surface-deep correlation detection on the road surface vibration characteristic spectrum to obtain the road surface abnormal point data; quantify the road surface loss risk of the road surface abnormal point data to obtain the road surface loss warning index; Traffic load assessment module, used to collect vehicle images and meteorological data; analyze vehicle model loss contribution based on vehicle images; construct an environmental impact coefficient matrix based on meteorological data; construct a dynamic weight model based on vehicle model loss contribution and environmental impact coefficient matrix to obtain a pavement loss dynamic weight model; perform section load sensitivity analysis based on the pavement loss dynamic weight model and pavement loss early warning indicators to obtain a section load sensitivity coefficient; The traffic flow redistribution module is used to predict the multi-path load impact according to the load sensitivity coefficient of the road section to obtain the path load impact data; generate demand-aware paths according to the path load impact data to obtain a demand-matching path set; and process the differentiated induction strategy according to the demand-matching path set to obtain a differentiated diversion plan; A maintenance timing determination module is used to determine the timing of road maintenance according to the differentiated diversion scheme and obtain a maintenance activity time window; The maintenance effect verification module is used to implement and optimize maintenance according to the maintenance activity time window to achieve highway maintenance management tasks.

[0009] Through multi-sensor collaboration, multi-level data collection and fusion, advanced signal processing and feature extraction technologies (such as FFT, CWT, wavelet packet decomposition, etc.), and surface-deep correlation analysis and machine learning methods (such as Apriori algorithm, Bayesian network) combined with the digital twin model of the pavement structure, the road condition monitoring module can comprehensively and accurately identify various early damages (such as cracks, potholes, subsidence, and voids) on the surface and deep layers of the pavement, and quantify them as abnormal point data of the pavement. Further, through the risk assessment of the abnormal point data, the pavement loss warning index is obtained. This enables the module to not only detect obvious pavement diseases, but also capture early damage and internal defects that are difficult to detect with the naked eye, achieving a comprehensive, in-depth, and refined perception of the pavement condition, providing timely and accurate data support for subsequent maintenance decisions, effectively avoiding the evolution of small diseases into major diseases, extending the service life of the pavement, and reducing the full life cycle cost. Through the deployment of high-definition cameras and weather stations, combined with deep learning image recognition technology (such as YOLO, ResNet), mechanical analysis models and normative empirical formulas, the traffic load assessment module can accurately identify various types of vehicles passing through the road section and calculate their contribution to the loss of the road surface. At the same time, by constructing an environmental impact coefficient matrix, the impact of different meteorological conditions on road performance is quantified. On this basis, the established pavement loss dynamic weight model comprehensively considers multiple factors such as vehicle type, vehicle speed, pavement stiffness, temperature, etc., and can evaluate the actual loss of vehicles to the road surface in real time and dynamically. Furthermore, the road section load sensitivity analysis is carried out in combination with the pavement loss warning index to obtain the road section load sensitivity coefficient. This enables the module to not only distinguish the different impacts of different vehicle types on the road surface, but also take into account the dynamic changes of environmental factors and road conditions, realize the refined and dynamic evaluation of traffic load, and provide a key basis for the reasonable allocation of traffic flow and scientific decision-making on road maintenance. Based on the load sensitivity coefficient of the road section, through multi-path load impact prediction, combined with the travel demand characteristics of vehicles (obtained through multi-source data fusion analysis), priority assessment (taking into account factors such as vehicle type, usage nature, real-time tasks, etc.), and multi-objective optimization algorithms (such as Dijkstra algorithm and genetic algorithm), the traffic flow redistribution module can generate personalized, demand-aware recommended paths for vehicles of different types and priorities. Through the differentiated induction strategy, the vehicles are guided to the appropriate path, the balanced distribution of the road network load is achieved, the traffic congestion is effectively alleviated, the road surface loss of high-load sections is reduced, and the overall traffic efficiency and safety of the road network are improved. At the same time, the right of way for emergency rescue vehicles and public transportation vehicles is prioritized, reflecting the people-oriented management concept. By analyzing the time series pattern of the traffic flow generated by the differentiated diversion scheme, combining meteorological data to evaluate the suitability of the maintenance environment, and considering the availability of maintenance resources (personnel, equipment, materials), the maintenance timing determination module can intelligently determine the optimal maintenance time window.Furthermore, combined with the road surface loss warning indicators, the maintenance priority and segmented maintenance plan are determined, and the maintenance time window is finely adjusted through optimization algorithms (such as genetic algorithms and rule-based expert systems). This enables the module to minimize the impact on traffic while meeting the resources and environmental conditions required for maintenance operations, realize the precise and efficient arrangement of maintenance operations, reduce maintenance costs, and improve the service level of roads. By comparing and collecting and analyzing the physical parameters of the road surface (deflection, flatness, structural depth, crack rate, etc.) and vehicle vibration response characteristics before and after maintenance, the maintenance effect verification module can objectively and quantitatively evaluate the effectiveness of maintenance measures. Based on the evaluation results, problems in the maintenance process (such as unqualified materials, improper construction technology, etc.) can be discovered in a timely manner, and feedback optimization can be carried out to form a closed-loop management of "detection-evaluation-maintenance-verification-feedback". This ensures the quality of maintenance operations, avoids ineffective maintenance or excessive maintenance, realizes continuous improvement of highway maintenance, and ensures the safety, durability and efficient operation of highways. Therefore, the present invention provides a safety management system for highway maintenance, which specifically solves the problems of passive response, inaccurate evaluation, extensive management and control existing in traditional methods, and realizes the transformation from "passive response" to "active prevention", from "empirical judgment" to "data-driven", and from "extensive management" to "fine management", thereby improving the efficiency, benefits and intelligence level of highway maintenance management. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a functional flow diagram of a highway maintenance safety management system; Figure 2 It is a functional flow diagram of the road condition monitoring module in the present invention.

[0011] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0012] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0013] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0014] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0015] In the embodiment of the present invention, reference Figure 1 As shown, it is a functional flow diagram of the present invention applied to a highway maintenance safety management system. In this example, the highway maintenance safety management system includes the following modules: S1: Road condition monitoring module, used to monitor multi-layer road condition data of highways to obtain road vibration characteristic spectrum; perform surface-deep correlation detection on road vibration characteristic spectrum to obtain road abnormal point data; perform road loss risk quantification on road abnormal point data to obtain road loss warning indicators; In the embodiment of the present invention, first, by laying out piezoresistive pressure sensors (1.5-meter spacing along the wheel track line) and fiber grating vibration sensors (buried in the road base, 4.5-meter spacing) in key sections of the highway (based on historical damage data, AADT data, geological exploration reports, and determined using the hierarchical analysis method and k-means clustering algorithm), a section sensing network is constructed and a topological map is drawn. Then, according to the set sampling frequency (pressure sensor once every 3 minutes, vibration sensor once every 20 minutes), multi-layer road condition raw data (surface pressure value, three-axis acceleration time-history curve) is collected, and data verification, filtering (sliding average filtering, bandpass filtering) and synchronization processing (linear interpolation) are performed. Next, the vibration data is subjected to a fast Fourier transform (FFT) to extract the vibration spectrum characteristics (average amplitude spectrum, energy proportion of three frequency bands, peak frequency, spectral center of gravity) to obtain the road vibration characteristic spectrum. Afterwards, the road vibration characteristic spectrum is multi-scale decomposed using the continuous wavelet transform (CWT) to obtain a multi-scale vibration characteristic tensor. Combined with the digital twin model of the pavement structure (built based on the pavement engineering archive data, and different types of defects are preset in the model), a surface-deep correlation analysis is performed, and the measured vibration characteristic tensor is compared with the vibration characteristic tensor under different defect conditions simulated by the model for similarity, and the abnormal points of the pavement (time, location, abnormal type) are determined, and the Apriori algorithm is used to further explore the association rules between the surface vibration abnormality mode and the deep structural defects. Risk quantification considers the following three factors: ① Abnormal type: Different abnormal types have different effects on pavement loss and driving safety. For example, deep cracks have a higher risk than surface cracks, and potholes have a higher risk than cracks. A table of abnormal type risk coefficients can be established, for example: surface cracks: 1, deep cracks: 3, potholes: 5, subsidence: 8. This coefficient can be determined based on historical data or relevant specifications. ② Impact range: The larger the area affected by the abnormality, the higher the risk. In an embodiment, the area affected by the abnormality can be estimated through the topological map of the sensor network and the location information of the abnormal point. For example, if multiple adjacent sensors detect an abnormality, it means that the area affected by the abnormality is large. ③ Development speed: The faster the abnormality deteriorates, the higher the risk. By comparing the anomaly detection results at different time points, the development speed of the anomaly can be evaluated. For example, the rate of change of the anomaly probability can be calculated, or the change of the continuous wavelet transform coefficient can be compared. Combining the above three factors, the risk value of each anomaly point can be calculated: risk value = risk coefficient of anomaly type × impact range × development speed; for example, a deep crack has an impact area of ​​2 square meters and a development speed of 0.5 square meters per week, then the risk value = 3 × 2 × 0.5 = 3. Then, according to the size of the risk value, the anomaly points are divided into different risk levels, such as: low risk (risk value <1), medium risk (1≤risk value <5), high risk (risk value ≥5). This division standard can be adjusted according to actual conditions.Finally, the risk level, location, type and other information of each abnormal point are integrated to form a road loss warning indicator. The warning indicator can be displayed in the form of a table or map.

[0016] S2: Traffic load assessment module, used to collect vehicle images and meteorological data; analyze vehicle model loss contribution based on vehicle images; construct an environmental impact coefficient matrix based on meteorological data; construct a dynamic weight model based on vehicle model loss contribution and environmental impact coefficient matrix to obtain a pavement loss dynamic weight model; perform section load sensitivity analysis based on the pavement loss dynamic weight model and pavement loss early warning indicators to obtain a section load sensitivity coefficient; In the embodiment of the present invention, high-definition cameras (roadside gantry, 6.5 meters in height) and meteorological stations (roadside, 2.5 meters from the edge of the road surface, 9.5 meters in height) are installed at key sections to collect vehicle images and meteorological data (temperature, humidity, rainfall, wind speed, wind direction). For vehicle images, a deep learning-based target detection algorithm (such as YOLOv7) and a vehicle classification algorithm (such as ResNet-101) are used to identify the vehicle model (classified according to GB / T 3730.1-2001 standard), and the number of axles and wheelbase of the vehicle are measured. Combined with mechanical analysis (fourth power law, axle group equivalent conversion coefficient) and empirical formula (wheelbase correction coefficient), the loss contribution (equivalent axle number) of different vehicle models is calculated. For meteorological data, according to the "Highway Asphalt Pavement Design Specification" (JTGD50-2017) and related research literature, the attenuation coefficient of pavement material performance under different environmental conditions (temperature, humidity, precipitation level) is determined, and an environmental impact coefficient matrix is ​​constructed. Combined with the vehicle type loss contribution and environmental impact coefficient matrix, a dynamic weight model for pavement loss (WESAL=ESAL×(W×V×K×T)) is constructed using linear interpolation method and weighted calculation (modulus weight 0.65, tensile strength weight 0.35). According to the pavement loss warning indicators, high-risk sections are identified. These sections usually have a higher risk of damage or poor pavement conditions. For each section, its loss rate under different traffic load conditions is calculated. The pavement loss dynamic weight model can be used to calculate the loss rate. For example, the loss of the section under different traffic flows and different vehicle type compositions can be simulated. Calculate the load sensitivity coefficient of each section. The load sensitivity coefficient can be defined as the ratio of the change rate of the loss rate to the change rate of the traffic load. The larger the coefficient, the more sensitive the section is to the traffic load. Among them, load sensitivity coefficient = loss rate change rate / traffic load change rate. For example, if the traffic load of a section increases by 10%, resulting in a 5% increase in the loss rate, then the load sensitivity coefficient of the section = 5% / 10% = 0.5. Finally, the load sensitivity coefficient of each road section is stored for subsequent traffic flow redistribution.

[0017] S3: Traffic flow redistribution module, used to predict multi-path load impact according to the load sensitivity coefficient of the road section to obtain path load impact data; generate demand-aware paths according to the path load impact data to obtain a demand matching path set; perform differentiated induction strategy processing according to the demand matching path set to obtain a differentiated diversion plan; In the embodiment of the present invention, multiple feasible paths are calculated using the Dijkstra algorithm based on the road section load sensitivity coefficient and the vehicle OD information. Then, the load of each feasible path is calculated (path load = ∑ (road section inherent load i + ∑ (vehicle additional load ij)), where the vehicle additional load = vehicle dynamic weight × road section load sensitivity coefficient), and weighted according to the vehicle priority (determined based on factors such as vehicle type, usage nature, real-time travel purpose, time sensitivity, etc.) to obtain path load impact prediction data. Next, the travel demand characteristics of the vehicle are analyzed through multi-source data (vehicle registration information, historical driving trajectory data, real-time traffic information), a travel demand feature library is constructed, and traffic priority assessment rules are formulated in combination with traffic regulations and management goals to obtain a vehicle demand priority matrix. Then, combined with the vehicle demand priority matrix, vehicle load feature library (extracted based on path load impact prediction data and predefined vehicle classification system) and road network load sensitivity map (obtained through fine analysis based on priority management strategy set and path load impact prediction data), multi-objective optimization (Minimize: α×travel time+β×path load+γ×path length, α, β, γ adjusted according to vehicle priority) is performed to generate differentiated recommended paths for vehicles of different priorities and obtain a demand matching path set. Finally, based on the demand matching path set, a differentiated information release strategy is formulated to obtain a differentiated diversion solution.

[0018] S4: a maintenance timing determination module, used to determine the road maintenance timing according to the differentiated diversion scheme and obtain the maintenance activity time window; In the embodiment of the present invention, the traffic flow data generated by the differentiated diversion scheme is subjected to time series analysis (ARMA model or exponential smoothing model) to obtain a flow time series characteristic map (including historical traffic flow, predicted traffic flow, periodic pattern, peak / valley period). Then, according to the flow time series characteristic map and meteorological data (temperature, humidity, precipitation, wind speed, wind direction), combined with the highway maintenance technical specifications, the environmental suitability (a value between 0-1) of each time period is calculated to obtain an environmental suitability time series. Then, according to the environmental suitability time series, combined with the maintenance resource information (personnel, equipment, materials) obtained from the highway maintenance management department, the maintenance resources available in each time period are determined to obtain a resource availability configuration table. After that, the maintenance priority of each road section is determined according to the road loss warning indicators (road damage type, degree, location), and combined with the resource availability configuration table, a segmented maintenance plan is formulated (clarifying the maintenance operation type, planned maintenance time period, and required resources for each road section). Finally, the maintenance time window is optimized and adjusted by comprehensively considering the traffic time series characteristic map, environmental suitability time series, resource availability configuration table and segmented maintenance plan, using heuristic algorithms or rule-based expert systems to obtain the final maintenance activity time window.

[0019] S5: Maintenance effect verification module, used to implement and optimize maintenance according to the maintenance activity time window to achieve highway maintenance management tasks; In an embodiment of the present invention, at the start and end times of maintenance determined by the maintenance activity time window, a multifunctional road surface detection vehicle (with integrated laser deflection meter, ground penetrating radar, road surface roughness meter, and road surface image acquisition system) is used to collect road surface physical parameters (deflection value, IRI, structural depth, and crack rate) before and after maintenance, and calculate the recovery index of each parameter ([(parameter value after maintenance - parameter value before maintenance) / parameter value before maintenance]×100%). On the road section after maintenance, a vibration sensor network (piezoresistive pressure sensor and fiber grating vibration sensor) that is exactly the same as the initial arrangement of the road condition monitoring module is rearranged to collect vehicle vibration response data, and preprocessing (data verification, filtering) and feature extraction (FFT, frequency band energy ratio, peak frequency, spectral center of gravity) are performed. The change rate of vibration characteristic parameters before and after maintenance is compared and analyzed ([(parameter value after maintenance - parameter value before maintenance) / parameter value before maintenance]×100%). Based on the change rate of the vibration characteristic parameters and the recovery index of the road physical parameters, the maintenance effect (excellent, good, medium, poor) is comprehensively evaluated. If the effect is not ideal, the cause (material, process, degree of damage, drainage, etc.) is analyzed, and the maintenance plan is adjusted, and the road condition data is updated to form a closed-loop management.

[0020] As an example of the present invention, refer to Figure 2 As shown, Figure 1 Functional flow diagram of the road condition monitoring module: S11: Identify key sections of the highway and arrange sensors to obtain a topological map of the sensor network of the section; In the embodiment of the present invention, the pavement damage records, annual average daily traffic volume (AADT) data and geological exploration reports of the target highway in the past three years are retrieved. The pavement damage records are accurate to the 100-meter pile number, and the damage type (cracks, potholes, subsidence) and repair date are recorded. The AADT data is refined to the vehicle type (small passenger car, large passenger car, light truck, heavy truck). The geological exploration report contains information on soil type, groundwater level and foundation bearing capacity along the line. A road section risk scoring matrix is ​​constructed, with the 100-meter pile number as the basic unit, the horizontal direction is the risk factor (historical damage frequency, AADT, foundation bearing capacity), and the vertical direction is each 100-meter pile number. The analytic hierarchy process (AHP) is used to determine the weight of each risk factor: the historical damage frequency weight is 0.6, the AADT weight is 0.3, and the foundation bearing capacity weight is 0.1. The weighted sum is used to obtain the comprehensive risk score of each 100-meter pile number. The k-means clustering algorithm is used to divide all 100-meter pile numbers into three levels: high risk, medium risk and low risk. High-risk sections are selected as priority sensor deployment areas. In high-risk sections, piezoresistive pressure sensors (to measure the pressure distribution on the road surface) and fiber Bragg grating vibration sensors (to measure the vibration response inside the road surface) are used. Pressure sensors are laid out along the wheel track line at intervals of 1.5 meters to form a longitudinal pressure monitoring belt. Vibration sensors are buried in the road base at intervals of 4.5 meters (25 cm deep from the surface) to form a deep vibration monitoring network. All sensors are networked through the Zigbee wireless communication protocol, and data is collected to the edge computing unit on the side of the road. Draw a topological diagram of the road section sensor network, which clearly indicates the model, number, installation location (coordinates and depth) and communication connection relationship of each sensor. The time synchronization of all sensors is completed through the GPS module.

[0021] S12: Perform multi-level data collection according to the road section sensor network topology map to obtain multi-level road condition raw data; In the embodiment of the present invention, the data acquisition parameters of the sensor are configured according to the topological diagram of the road section sensor network. The sampling frequency of the piezoresistive pressure sensor is set to once every 3 minutes, and the sampling frequency of the fiber grating vibration sensor is set to once every 20 minutes. The data collected by the pressure sensor is the surface pressure value of the road surface (unit: kPa), which reflects the stress distribution of the road surface under the action of the vehicle load. The data collected by the vibration sensor is the acceleration time history curve of the road surface in three orthogonal directions (X, Y, and Z axes) (unit: g, gravity acceleration). The received raw data is verified to remove invalid data that exceeds the sensor range or communication interruption. The sliding average filter algorithm is applied to the pressure data for noise reduction processing to filter out high-frequency noise interference (the filter window size is set to 7 sampling points). For vibration data, bandpass filtering (cutoff frequencies are 0.5Hz and 250Hz) is used to remove low-frequency drift and high-frequency noise. To ensure data synchronization, the pressure data is linearly interpolated based on the sampling time of the vibration sensor to generate a multi-layer road condition raw data set that is completely aligned with the vibration data timestamp. The dataset is stored in binary file format, each file contains all sensor data for one hour and is named with a timestamp.

[0022] S13: extracting vibration spectrum features from the multi-layer road condition raw data to obtain a road surface vibration characteristic spectrum; In an embodiment of the present invention, a fast Fourier transform (FFT) is performed on the vibration data (three-axis acceleration time history curve) in the original data set of multi-layer road conditions. A section of vibration data collected every 20 minutes is divided into multiple overlapping windows (the window length is 4096 sampling points, and the overlap rate is 25%). A Hanning window function is applied to each window for windowing processing to reduce spectrum leakage. FFT calculation is performed on each windowed data segment to obtain the amplitude spectrum of the window. The amplitude spectra of all windows are averaged to obtain the average amplitude spectrum of the 20-minute period. The frequency range is divided into three frequency bands: low frequency band (0.5-25Hz, reflecting the overall structural response of the road surface), medium frequency band (25-75Hz, reflecting the state of the road base and surface layer), and high frequency band (75-250Hz, reflecting the micro cracks of the road surface). The sum of the energy in each frequency band is calculated and normalized (divided by the total energy) to obtain the energy proportion of the three frequency bands. Calculate the peak frequency (frequency corresponding to the maximum energy) and the spectral center of gravity (weighted average of frequencies, with the weight being the amplitude) of the amplitude spectrum. Combine the average amplitude spectrum, energy proportion of the three frequency bands, peak frequency, and spectral center of gravity of each 20-minute period into a feature vector. Arrange the feature vectors of all time periods in chronological order to form a road vibration feature spectrum. The feature spectrum is stored in matrix form, with each row representing a time point and each column representing a vibration feature parameter.

[0023] S14: performing pavement layer detection on the pavement vibration characteristic spectrum to obtain pavement abnormal point data; In an embodiment of the present invention, pavement stratification detection is performed based on the pavement vibration characteristic spectrum and combined with the digital twin model of the pavement structure. First, the pavement vibration characteristic spectrum is decomposed at multiple scales using continuous wavelet transform (CWT). Morlet wavelet is selected as the mother wavelet. The scale parameter range is set to 1 to 64, and 32 scales are generated at logarithmic intervals. For each feature in the pavement vibration characteristic spectrum (for example, the energy proportion of a certain frequency band), CWT is performed separately to obtain a wavelet coefficient sequence. The wavelet coefficient sequences of 32 scales are arranged in order from small to large scale to form a two-dimensional matrix (scale-time). The above operation is performed on all features in the vibration characteristic spectrum to obtain multiple two-dimensional matrices. These two-dimensional matrices are stacked along the third dimension (feature dimension) to form a three-dimensional tensor (scale-time-feature). Then, combined with the constructed digital twin model of the pavement structure (the model is constructed through pavement engineering archive data, including surface layer, base layer, subbase layer and roadbed soil, and the material parameters of each layer are set according to actual conditions), surface-deep layer correlation analysis is performed. Different types of defects (cracks in the surface layer, voids in the base layer, and roadbed settlement) are preset in the digital twin model, and different defect locations and sizes are set. The finite element method is used to perform transient dynamic analysis to simulate the vibration response of the pavement structure under vehicle load. The vibration response (acceleration time history curve) of the key points on the pavement surface and inside under different defect conditions is recorded. The simulated vibration response data is processed by FFT and CWT to obtain the vibration characteristics and multi-scale characteristic tensors of the simulated data. The vibration characteristic tensors measured on site are compared and analyzed with the vibration characteristic tensors under different defect conditions simulated by the digital twin model. The similarity between the two is calculated (for example, using cosine similarity or Euclidean distance). A similarity threshold is set (for example, 0.85). If the similarity between the measured vibration characteristic tensor and the simulated vibration characteristic tensor under a certain defect condition exceeds the threshold, it is determined that the road section has a corresponding type of defect. All time points determined to be abnormal and their corresponding road section locations (determined by the sensor network topology map) and the inferred defect types are recorded. The information of these abnormal points (time, location, abnormal type) is stored as road surface abnormal point data.

[0024] Preferably, the road surface delamination detection includes: Perform multi-scale vibration feature extraction on the road vibration characteristic spectrum to obtain a multi-scale vibration feature tensor; Self-supervised abnormal pattern recognition is performed on multi-scale vibration feature tensors to obtain a road surface abnormality probability distribution map; According to the road surface anomaly probability distribution diagram, surface-deep association rules are mined to obtain the association data between the pavement structure layers; The damage evolution is predicted based on the inter-layer correlation data of the pavement structure and the pavement anomaly probability distribution map to obtain the pavement anomaly point data.

[0025] In an embodiment of the present invention, continuous wavelet transform (CWT) is used to perform multi-scale decomposition on the road vibration characteristic spectrum (from the road condition monitoring module). The complex Morlet wavelet is selected as the mother wavelet, and its expression is: ψ(t)=(1 / √(πfb))×exp(2iπfct)×exp(-t² / fb), where fb is the bandwidth parameter and fc is the wavelet center frequency, which are set to 1 and 1 respectively. The scale parameter range is set to 1 to 128, and 64 scales are generated at logarithmic intervals. The scale sequence is: 1, 1.41, 2, 2.83,..., 128. For each feature in the road vibration characteristic spectrum (for example, the energy proportion of the 1-25Hz frequency band), CWT is performed separately. The calculation formula of CWT is: W(a, b)=∫f(t)×conj(ψ((tb) / a))dt, where f(t) is the input vibration characteristic signal (time series), ψ(t) is the complex Morlet wavelet function, a is the scale parameter, b is the time translation parameter, conj(ψ) represents the complex conjugate, and W(a, b) is the wavelet coefficient. For each scale a, a wavelet coefficient sequence W(a, b) is calculated, and the length of the sequence is the same as the time axis length of the original vibration characteristic spectrum. The wavelet coefficient sequences of 64 scales are arranged in order from small to large scale to form a two-dimensional matrix (scale-time). The above operation is performed on all features of the vibration characteristic spectrum (energy proportion of low frequency band, medium frequency band, high frequency band, peak frequency, spectral center of gravity) to obtain multiple two-dimensional matrices. These two-dimensional matrices are stacked along the third dimension (feature dimension) to form a three-dimensional tensor (scale-time-feature). This three-dimensional tensor is the multi-scale vibration feature tensor, which is stored in the HDF5 file format.

[0026] Based on the multi-scale vibration feature tensor, its statistical features are calculated as the basis for abnormal pattern recognition. For each time point in the multi-scale vibration feature tensor, its mean, standard deviation, skewness, and kurtosis are calculated at all scales and features. These statistical features are combined into a feature vector. Then the Mahalanobis distance of the feature vector at each time point is calculated. The calculation formula of the Mahalanobis distance is: D(x)=√((x-μ)ᵀΣ⁻¹(x-μ)), where x is the feature vector of the time point to be calculated, μ is the mean vector of the feature vectors of all time points, and Σ is the covariance matrix of the feature vectors of all time points. The Mahalanobis distance reflects the degree of deviation of the feature vector at each time point from the overall feature vector distribution. The larger the Mahalanobis distance, the more abnormal the vibration feature at that time point. The Mahalanobis distances of all time points are normalized and mapped to between 0 and 1. The normalization formula is: P(t)=(D(t)-Dmin) / (Dmax-Dmin), where D(t) is the Mahalanobis distance at time t, Dmin and Dmax are the minimum and maximum values ​​of the Mahalanobis distance at all time points, and P(t) is the normalized abnormal probability. The abnormal probabilities of all time points are arranged in chronological order to form a one-dimensional vector (time-abnormal probability). The vector is visualized in the form of a heat map, with the horizontal axis representing time and the vertical axis representing the abnormal probability, and the color depth representing the probability size (red represents high probability, blue represents low probability). This heat map is the road abnormality probability distribution map, stored in PNG format.

[0027] Combined with the pavement engineering archive data and the digital twin model of the pavement structure, the correlation between the high abnormal probability area in the pavement abnormality probability distribution map and the internal structural defects of the pavement is analyzed. First, the completion drawings, construction records, material inspection reports and geological survey reports of the target section are obtained from the highway construction unit. Based on these data, the finite element method is used to construct a three-dimensional solid model of the multi-layer structure of the pavement. From top to bottom, the model is asphalt surface layer (fine-grained asphalt concrete AC-13, thickness 4cm; medium-grained asphalt concrete AC-20, thickness 6cm; coarse-grained asphalt concrete AC-25, thickness 8cm), cement-stabilized crushed stone base layer (thickness 35cm), graded crushed stone subbase layer (thickness 20cm), and roadbed soil (layered modeling according to the geological survey report, the thickness is 3 times the thickness of the pavement structure). The structural layers are set to be in complete continuous contact. The corresponding mechanical properties parameters are assigned to the materials of each structural layer. The viscoelastic constitutive model is used for the asphalt mixture, the Mohr-Coulomb elastoplastic model is used for cement-stabilized crushed stone and graded crushed stone, and the Duncan-Chang model is used for the roadbed soil. Different types of defects (surface cracks, base layer voids, roadbed settlement) are preset in the model, and different defect locations and sizes are set. The finite element method is used for transient dynamic analysis to simulate the vibration response of the pavement structure under vehicle load. The vibration response (acceleration time history curve) of the key points on the pavement surface and inside under different defect conditions is recorded. The simulated vibration response data is processed by FFT and CWT to obtain the pavement abnormality probability distribution map of the simulated data. The abnormal probability distribution map of the simulated data is analyzed in association with the defect type, location and size in the finite element model. The Apriori algorithm is used to mine frequent item sets to find the association rules between the surface vibration abnormality mode (the characteristics of the abnormal probability distribution map) and the deep structural defects. For example, a rule is expressed as: "If the proportion of vibration energy in the 15-25Hz frequency band is continuously high and the abnormal probability exceeds 0.92, there is an 85% confidence level indicating that there is void in the pavement base layer, and the void range is between 50cm-100cm". The association rules mined are stored as a rule base. This rule base is the association data between the pavement structure layers, which is stored in XML file format.

[0028] Based on the pavement structure interlayer correlation data and the pavement anomaly probability distribution map, combined with the historical pavement performance attenuation data, the future development trend of pavement damage is inferred. First, the pavement anomaly probability distribution map is divided into multiple windows by time (one window every 72 hours). For each window, the average anomaly probability is calculated, and the defect type, location and severity inferred in the time period are extracted from the pavement structure interlayer correlation data. The average anomaly probability and the inference result are combined into a feature vector. Then, the historical pavement performance database is queried to obtain the performance attenuation data of similar sections (same structure type, similar traffic volume, similar climatic conditions) under similar defect conditions. The performance attenuation data includes: pavement roughness (IRI), pavement deflection value, pavement skid resistance (BPN), etc. The historical performance attenuation data is fitted using the time series analysis method to obtain the performance attenuation curve. According to the defect status and performance indicators of the current section, the best matching curve is found from the historical performance attenuation curve. This curve is used as the prediction curve for the future performance attenuation of the current section. According to the prediction curve, the changes in pavement performance indicators in the future (such as the next 30 days) are predicted. Set the threshold of the pavement performance index (such as IRI exceeding 4m / km, deflection value exceeding 0.6mm). If the predicted performance index exceeds the threshold, it is determined that the road section will be abnormal in the future. Record the time of the abnormality (determined by the prediction curve), the location (determined by the inference result of the inter-layer correlation data of the pavement structure), and the predicted abnormality type (such as crack expansion and increased base debonding). Store this information as pavement abnormality point data in JSON format.

[0029] Preferably, the table-deep association rule mining includes: Obtain pavement engineering archive data; construct a digital twin of the pavement multi-layer structure based on the pavement engineering archive data to obtain a digital twin model of the pavement structure; Design the physical constraint architecture based on the digital twin model of the pavement structure to obtain the physical constraint data; The damage-vibration transfer function is derived based on the physical constraint data and the digital twin model of the pavement structure to obtain a damage-vibration transfer function library; Multi-layer Bayesian network reasoning is performed based on the damage vibration transfer function library and the road surface abnormality probability distribution map to obtain the multi-layer state probability map of the road surface; The surface-deep symptom association knowledge is extracted from the multi-layer state probability map of the pavement to obtain the correlation data between the pavement structure layers.

[0030] In an embodiment of the present invention, complete engineering archive data of the target road section (for example, the section from K100+000 to K101+000 of the G30 Expressway) is obtained from the highway construction unit or management department. The data include: completion drawings (pavement structure design drawings, including plan views, longitudinal sections, cross-sections, material types and thickness markings of each structural layer, and drawing scale 1:500), construction records (construction date of each structural layer, construction process parameters, such as compaction, paving temperature, and batch number of materials used), material inspection reports (mechanical performance parameters of each structural layer material, such as elastic modulus, Poisson's ratio, compressive strength, tensile strength, and dynamic modulus test results of asphalt mixtures, triaxial test results of cement-stabilized macadam), geological survey reports (types of roadbed soil, physical and mechanical properties, such as water content, liquid limit, plastic limit, compression modulus, and groundwater level information). Based on the engineering archive data, a three-dimensional solid model of the multi-layer structure of the pavement is constructed using the finite element method. The geometric dimensions of the model are determined according to the completion drawings. The model is constructed from top to bottom: asphalt surface layer (fine-grained asphalt concrete AC-13, thickness 40mm; medium-grained asphalt concrete AC-20, thickness 60mm; coarse-grained asphalt concrete AC-25, thickness 80mm), cement-stabilized crushed stone base (thickness 350mm), graded crushed stone subbase (thickness 200mm), and subgrade soil (layered modeling according to the geological survey report, the thickness is 3 times the thickness of the pavement structure, for example, here it is 2370mm). The structural layers are set to be in complete continuous contact. The corresponding mechanical properties are assigned to the materials of each structural layer. The elastic modulus of the asphalt mixture is converted to the values ​​at different temperatures based on the dynamic modulus test results in the material test report by the time-temperature equivalence principle. The elastic modulus and Poisson's ratio of cement-stabilized crushed stone and graded crushed stone are determined according to the triaxial test results in the material test report. The parameters of the subgrade soil are determined according to the geotechnical test results in the geological survey report. The longitudinal length of the model along the road is 100m, and the transverse width is the actual width of the road (for example, two-way four-lane plus hard shoulders on both sides, the total width is 2×3.75m+2×3.5m+2×0.75m=16m). Hexahedral solid units are used for meshing, and the mesh size is 0.1m in the surface layer, 0.2m in the base layer and subbase layer, and 0.5m in the roadbed soil.

[0031] Based on the digital twin model of the pavement structure, the geometry, material and boundary condition information of the model are extracted as physical constraint data. Specifically, it includes: the node coordinates, unit type and connection relationship of the model, the material parameters of each structural layer (elastic modulus, Poisson's ratio, density), and boundary conditions (the position of fixed constraints, the direction of displacement constraints). These data are organized into a table. Each row of the table represents a node or unit, and each column represents an attribute (such as node coordinates, unit type, material parameters, boundary conditions). For boundary conditions, different constraint types are represented by coding. For example, the code "111" means that all three directions are fixed, and "110" means that the X and Y directions are fixed, and the Z direction is free.

[0032] In the digital twin model of the pavement structure, damage of different types, locations and sizes is introduced. The damage types include: surface cracks (longitudinal cracks, transverse cracks, cracks, crack depths of 20mm, 40mm, 60mm, respectively, and crack widths of 2mm, 4mm, 6mm, respectively), base layer voids (void lengths of 500mm, 1000mm, 1500mm, respectively, and void widths of half the width of the pavement, respectively), and roadbed settlement (settlement depths of 20mm, 40mm, 60mm, respectively, and settlement ranges are circular, with diameters of 1000mm, 2000mm, and 3000mm, respectively). For each damage condition, transient dynamic analysis is performed using the finite element method. A moving load of a standard axle load (BZZ-100, double-wheeled single axle, axle weight of 100kN) is applied to the pavement surface. The moving speed is set to 60km / h. The vibration response (acceleration time history curve) of the pavement surface and key points inside (such as crack tips, void edges, and settlement centers) is recorded. Perform fast Fourier transform (FFT) on the recorded acceleration time history curve to obtain the amplitude spectrum. Calculate the characteristic parameters of the amplitude spectrum, such as peak frequency and frequency band energy ratio (0-25Hz, 25-75Hz, 75-250Hz). Associate the damage type, location, and size with the characteristic parameters of the amplitude spectrum. For example, a record is: "There is a transverse crack with a length of 100mm and a depth of 40mm in the surface layer, which causes the vibration energy ratio of the 25-75Hz frequency band to increase by 18%." The vibration response characteristic parameters under all damage conditions are organized into a database. The database is indexed by damage type, location, and size, and has amplitude spectrum characteristic parameters (peak frequency, energy ratio of each frequency band) as entries.

[0033] A multi-layer Bayesian network is constructed. The nodes of the network represent the status of each layer of the pavement structure (such as intact surface layer, slight cracks in the surface layer, severe cracks in the surface layer, intact base layer, base layer voiding, etc.). The directed edges between the nodes represent the causal relationship between the layers (such as cracks in the surface layer lead to stress concentration in the base layer, which accelerates base layer voiding). The conditional probability table (CPT) of the Bayesian network is determined in the following way: According to the damage vibration transfer function library, the distribution of vibration characteristic parameters under different damage types and severity is statistically analyzed to obtain the prior probability. The prior probability is corrected using the abnormal probability value in the pavement abnormal probability distribution diagram. Specifically, each time point in the abnormal probability distribution diagram is used as an evidence node. The status of the evidence node is discretized according to the abnormal probability value (low probability: 0-0.3, medium probability: 0.3-0.7, high probability: 0.7-1). The Bayesian network inference algorithm is used for inference. After inputting the status of the evidence node, the algorithm updates the posterior probability of each node in the network. The state probability distribution of each layer structure at each time point is output. For example, the output is: "At time t1, the probability of slight cracks in the surface layer is 0.7, the probability of severe cracks is 0.1, and the probability of voids in the base layer is 0.2". The probability distribution of each layer state at all time points is combined into a three-dimensional matrix (time-structure layer-state probability).

[0034] Analyze the probability map of the multi-layer state of the pavement and extract the association rules between the vibration characteristics of the surface layer of the pavement and the deep structural state. Focus on the combination of high-probability states. For example, when the probability of severe cracks in the surface layer is high (>0.8) and the probability of voids in the base layer is also high (>0.7), it is recorded as a potential association pattern. Combined with the information in the damage vibration transfer function library, analyze the vibration characteristics corresponding to the pattern. For example, the above pattern corresponds to a significant increase in the proportion of vibration energy in the 25-75Hz frequency band. Use the Apriori algorithm to mine frequent item sets from the probability map of the multi-layer state of the pavement to find the association rules between the abnormal surface vibration pattern and the deep structural state. For example, a rule is: "If the proportion of vibration energy in the 25-75Hz frequency band is continuously high and lasts for more than 3 time windows (i.e. 216 hours, because each time window is set to 72 hours), and the posterior probability of base layer debonding in the pavement multi-layer state probability map is greater than 0.7, and the posterior probability of severe cracks in the surface layer is greater than 0.6, then it is considered that there is a strong correlation between base layer debonding and surface layer cracks in this road section." Set the minimum support threshold (e.g., 0.05) and the minimum confidence threshold (e.g., 0.8). The minimum support indicates the frequency of the rule in the data set, and the minimum confidence indicates the reliability of the rule. The Apriori algorithm first finds all frequent 1-item sets (i.e., the item sets where the frequency of a single state exceeds the minimum support), and then generates frequent 2-item sets (i.e., the item sets where the frequency of two states appearing at the same time exceeds the minimum support) from the frequent 1-item sets, and so on, until no higher-order frequent item sets can be generated. For each frequent item set, calculate its confidence. The confidence calculation formula is: Confidence(A→B)=Support(A∪B) / Support(A), where A and B represent different pavement conditions (e.g., A represents "the proportion of vibration energy in the 25-75Hz frequency band is continuously high", B represents "the base layer is empty"), A∪B represents that A and B occur at the same time, and Support(X) represents the support of X (i.e., the frequency of X in the data set). Filter out the rules whose confidence exceeds the minimum confidence threshold as the final association rules. These rules represent the statistically significant association between the abnormal vibration mode of the pavement surface and the deep structural state. Store the association rules in a structured form. Store all mined association rules as a rule base. This rule base is the association data between the pavement structure layers.

[0035] Preferably, the traffic load assessment module includes the following functions: Collect vehicle images and weather data on key road sections; Carry out multi-dimensional monitoring and classification of traffic flow based on vehicle images to obtain vehicle type load classification data; Carry out vehicle model loss contribution analysis based on vehicle model load classification data to obtain vehicle model loss contribution; The environmental impact matrix is ​​constructed based on meteorological data to obtain the environmental impact coefficient matrix; A dynamic weight model is constructed according to the vehicle type loss contribution and environmental impact coefficient matrix to obtain a road surface loss dynamic weight model.

[0036] In an embodiment of the present invention, high-definition cameras and weather stations are installed on key sections (high-risk sections determined by the road condition monitoring module, such as the section from K100+000 to K101+000 of the G30 Expressway). The camera is installed on a roadside gantry at a height of 6.5 meters from the ground, and the lens is perpendicular to the road surface to ensure that a complete side image of all vehicles passing through the section can be clearly captured. The camera resolution is set to 2560×1440 pixels and the frame rate is set to 30 frames per second. The video stream is compressed using the H.265 encoding format and transmitted to the data processing center via optical fiber. The weather station is installed on the roadside, 2.5 meters from the edge of the road surface and 9.5 meters from the ground. The weather station is equipped with the following sensors: temperature sensor (measuring range -35℃ to 55℃, accuracy ±0.3℃), humidity sensor (measuring range 0-100%RH, accuracy ±2%RH), rainfall sensor (measuring range 0-40mm / h, accuracy ±0.05mm), wind speed sensor (measuring range 0-50m / s, accuracy ±0.3m / s), wind direction sensor (measuring range 0-360°, accuracy ±3°). The weather station collects data every 30 seconds and transmits it to the data processing center via a 4G wireless network. The time synchronization between the camera and the weather station is achieved through a GPS module with an error of less than 1 second. The vehicle images and meteorological data received by the data processing center are all accurately time-stamped.

[0037] The collected vehicle images are processed to extract vehicle feature information. A deep learning-based object detection algorithm (such as YOLOv7) is used to identify the vehicles in the image. For each detected vehicle, the area containing the complete side image of the vehicle is cropped. A deep learning-based vehicle classification algorithm (such as ResNet-101) is used to classify the cropped vehicle images. Vehicle classification system According to the "Terms and Definitions of Road Vehicle Models" (GB / T3730.1-2001), vehicles are divided into: mini buses (M1 category, less than or equal to 9 seats), small buses (M1 category, less than or equal to 9 seats), medium buses (M2 category, more than 9 seats, less than or equal to 19 seats), large buses (M3 category, more than 19 seats), mini trucks (N1 category, total mass less than or equal to 1.8 tons), light trucks (N1 category, total mass greater than 1.8 tons, less than or equal to 3.5 tons), medium trucks (N2 category, total mass greater than 3.5 tons, less than or equal to 12 tons), heavy trucks (N3 category, total mass greater than 12 tons), articulated trains (including tractors and trailers). For each identified vehicle, its model, passing time (accurate to milliseconds), and driving lane are recorded. The vehicle contour recognition algorithm is used to measure the number of axles and wheelbase of the vehicle. The number of axles is determined by detecting the number of wheels. The wheelbase is determined by measuring the horizontal distance between the centers of two adjacent axles, and the measurement error is controlled within 5cm. The vehicle model, passing time, driving lane, number of axles, and wheelbase information of all vehicles passing through the road section within each hour are summarized into a table.

[0038] Based on the vehicle load classification data, combined with mechanical analysis and empirical formulas, the relative loss of different vehicle models to the road surface is calculated. The standard axle load (BZZ-100, double-wheel group single axle, axle weight 100kN) is used as the benchmark. First, according to the number of axles and axle weight of the vehicle, the axle weight equivalent of each axle is calculated. For a single axle, the axle weight equivalent is calculated using the fourth power law: equivalent axle number = (actual axle weight / standard axle weight)^4. For multi-axle groups (double-axle, triple-axle), the axle group equivalent conversion coefficient provided in the "Highway Asphalt Pavement Design Code" (JTGD50-2017) is used for correction. For example, the equivalent conversion coefficient of the double-axle is 1.9, and the equivalent conversion coefficient of the triple-axle is 2.6. Then, according to the wheelbase of the vehicle, the wheelbase correction coefficient is calculated. Wheelbase correction coefficient = 1 + 0.05 × (3.0-actual wheelbase) (unit: meter), where the standard wheelbase is 3.0 meters. Multiply the equivalent axle times of each axle by the wheelbase correction factor, and then add them up to get the total equivalent axle times of the vehicle. The total equivalent axle times is used as the loss contribution of the vehicle model to the road surface. For all identified vehicle models, calculate their loss contributions respectively. Create a vehicle model-loss contribution comparison table. The table is indexed by vehicle model (minibus, small bus...heavy truck, articulated train) and has loss contribution (equivalent axle times) as an entry. For example: minibus 0.01, small bus 0.03, medium truck 3.5, heavy truck 8.2, articulated train 15.6. This comparison table is the vehicle model loss contribution.

[0039] Based on meteorological data, the impact of different environmental factors on pavement performance is analyzed. Meteorological data are divided into the following dimensions: temperature (high temperature, normal temperature, low temperature, freeze-thaw cycle), humidity (dry, moist, saturated), precipitation (no precipitation, light rain, moderate rain, heavy rain). Temperature classification standards: high temperature (≥32℃), normal temperature (4℃-32℃), low temperature (≤4℃). The number of freeze-thaw cycles is counted based on whether the daily minimum and maximum temperatures cross 0℃. Humidity classification standards: dry (<55%RH), moist (55%-75%RH), saturated (>75%RH). The precipitation classification standard is determined based on 24-hour precipitation: light rain (<10mm), moderate rain (10-25mm), heavy rain (>25mm). According to the "Highway Asphalt Pavement Design Code" (JTGD50-2017) and related research literature, the attenuation coefficient of pavement material performance under different environmental conditions is determined. Construct an environmental impact coefficient matrix. The rows of the matrix represent different combinations of environmental conditions (high temperature and dryness, low temperature and humidity, moderate rain, etc., a total of 434 = 48 combinations), and the columns represent different pavement performance indicators (modulus, tensile strength, shear strength). The elements in the matrix represent the attenuation coefficients (value range 0-1) of the pavement performance indicators relative to the normal state (normal temperature, dryness, no precipitation) under the corresponding environmental conditions. For example, under high temperature and dry conditions, the modulus attenuation coefficient of the asphalt surface layer is 0.75, the tensile strength attenuation coefficient is 0.85, and the shear strength attenuation coefficient is 0.8.

[0040] Combining the vehicle type loss contribution and the environmental impact coefficient matrix, a dynamic weight model of pavement loss considering vehicle type and environmental factors is constructed. According to the current environmental conditions, the corresponding pavement performance attenuation coefficient is found from the environmental impact coefficient matrix. Since the environmental impact coefficient matrix is ​​discrete, and the actual environmental conditions are continuously changing, the attenuation coefficient is calculated using a linear interpolation method. For example, if the current temperature is 28°C and the humidity is 65%, linear interpolation is performed between high temperature and normal temperature, and between wet and dry. Multiply the vehicle type loss contribution (equivalent axle order) by the corresponding pavement performance attenuation coefficient. For example, if the equivalent axle order of a certain vehicle type (such as a heavy truck) is 8.2, the pavement modulus attenuation coefficient under the current environmental conditions (such as high temperature and dryness) is 0.75, and the tensile strength attenuation coefficient is 0.85, then the weighted equivalent axle order of the vehicle type in the current environment is: weighted equivalent axle order = 8.2 × (0.75 × w1 + 0.85 × w2), where w1 and w2 are the weights of modulus and tensile strength in calculating the weighted equivalent axle order. For asphalt pavement, the modulus weight w1 is 0.65 and the tensile strength weight w2 is 0.35. For all vehicles passing through the road section per hour, the weighted equivalent axle times are calculated respectively. The weighted equivalent axle times of all vehicles are accumulated to obtain the total weighted equivalent axle times of the road section for that hour. The pavement loss dynamic weight model is stored and output in the form of function expressions and parameter tables.

[0041] Preferably, the dynamic weight model construction includes: The four-dimensional key factor data is standardized according to the vehicle model loss contribution and environmental impact coefficient matrix to obtain a standardized factor data table, wherein the standardized factor data table includes vehicle weight data, vehicle speed data, road stiffness data and temperature data; Use vehicle weight data to analyze the relationship between vehicle weight and road surface to obtain vehicle weight impact data; Use vehicle speed data to analyze the speed-road dynamic response relationship and obtain speed response data; Use the pavement stiffness data to analyze the pavement stiffness-bearing characteristics and obtain stiffness characteristic data; Using temperature data to analyze temperature modulation effect, obtaining temperature modulation data; A comprehensive dynamic weight calculation is performed based on vehicle weight impact data, speed response data, stiffness characteristic data and temperature modulation data to obtain a pavement loss dynamic weight model.

[0042] In an embodiment of the present invention, the gross vehicle weight data is extracted from the vehicle model loss contribution. The gross vehicle weight is divided into four levels: light (≤3 tons), medium (3-10 tons), heavy (10-25 tons), and extra heavy (>25 tons). The weight range of each level is mapped to the 0-1 interval: light (0-0.2), medium (0.2-0.4), heavy (0.4-0.7), and extra heavy (0.7-1). For example, a vehicle with a gross weight of 18 tons has a standardized vehicle weight data of 0.56. The vehicle speed data is extracted from the vehicle model load classification data. The vehicle speed is divided into three levels: low speed (≤40km / h), medium speed (40-80km / h), and high speed (>80km / h). The speed range of each level is mapped to the 0-1 interval: low speed (0-0.3), medium speed (0.3-0.6), and high speed (0.6-1). For example, a vehicle traveling at 70km / h has a standardized speed data of 0.525. Extract the road deflection value data from the road anomaly data (from the road condition monitoring module) as a representation of the road stiffness. Divide the deflection value into three levels: high stiffness (≤0.2mm), medium stiffness (0.2-0.5mm), and low stiffness (>0.5mm). Map the deflection value range of each level to the 0-1 interval: high stiffness (0-0.3), medium stiffness (0.3-0.6), and low stiffness (0.6-1). For example, the deflection value of a certain road section is 0.4mm, and its standardized road stiffness data is 0.45. Extract road surface temperature data from meteorological data. Divide the temperature into three intervals: low temperature (≤5℃), normal temperature (5-30℃), and high temperature (>30℃). Map each temperature interval to the 0-1 interval: low temperature (0-0.2), normal temperature (0.2-0.8), and high temperature (0.8-1). For example, the current road surface temperature is 35°C, and its standardized temperature data is 0.9. The standardized vehicle weight data, vehicle speed data, road surface stiffness data, and temperature data of each vehicle are combined into a four-dimensional vector. The four-dimensional vectors of all vehicles are arranged in chronological order to form a standardized factor data table.

[0043] Calculate the vehicle weight influence factor based on the standardized vehicle weight data. Use the modified fourth power law. For single-axle vehicles, the vehicle weight influence factor calculation formula is: W_single=(w / 0.4)^4, where w is the standardized vehicle weight data and 0.4 is the standardized vehicle weight data corresponding to the standard axle load (10 tons). For multi-axle vehicles (double-axle, triple-axle, etc.), make corrections based on the axle group equivalent conversion factor. The "Highway Asphalt Pavement Design Code" (JTGD50-2017) stipulates that the equivalent conversion factor for double-axle is 1.9 and the equivalent conversion factor for triple-axle is 2.6. Assuming that the vehicle axle group type is known, determine the corresponding equivalent conversion factor C_axle according to the vehicle axle group type. The vehicle weight influence factor calculation formula for multi-axle vehicles is: W_multi=C_axle×(w / 0.4)^4. Calculate the vehicle weight influence factor W for each vehicle. For example, for a double-axle vehicle with a standardized vehicle weight of 0.6, its vehicle weight influence factor W_multi=1.9×(0.6 / 0.4)^4=9.639.

[0044] Calculate the speed influence factor based on the standardized vehicle speed data. Use the exponential function form: V=1+0.12×v^1.8, where v is the standardized vehicle speed data, 0.12 and 1.8 are model parameters determined based on empirical values ​​and test data. Calculate the speed influence factor V for each vehicle. For example, for a vehicle with a standardized vehicle speed data of 0.525, its speed influence factor V=1+0.12×0.525^1.8=1.034. Store the speed influence factors V of all vehicles as speed response data.

[0045] Calculate the stiffness influence factor based on the standardized pavement stiffness data. Use the exponential function form: K=exp(-1.5×(1-k)), where k is the standardized pavement stiffness data, and 1.5 is a model parameter that reflects the sensitivity of pavement stiffness to bearing capacity, determined based on empirical values ​​and test data. Calculate the stiffness influence factor K for each road section. For example, if the standardized pavement stiffness data for a certain road section is 0.45, its stiffness influence factor K=exp(-1.5×(1-0.45))=0.438. Store the stiffness influence factor K of all road sections as stiffness characteristic data.

[0046] Calculate the temperature influence factor based on the standardized temperature data. Use the exponential function form: T=exp(-2.2×|t-0.5|), where t is the standardized temperature data, 0.5 is the standardized data corresponding to the reference temperature (20°C), and 2.2 is the model parameter, which reflects the sensitivity of temperature to pavement material performance and is determined based on empirical values ​​and test data. Calculate the temperature influence factor T at each moment. For example, the standardized temperature data at the current moment is 0.9, and its temperature influence factor T=exp(-2.2×|0.9-0.5|)=0.415. Store the temperature influence factor T at all moments as temperature modulation data.

[0047] The dynamic weight is calculated by combining the four influencing factors. The following formula is used: DW=W×V×K×T, where DW is the dynamic weight, W is the vehicle weight influencing factor, V is the speed influencing factor, K is the stiffness influencing factor, and T is the temperature influencing factor. For each vehicle, the corresponding W, V, K, and T values ​​are obtained according to its vehicle model, speed, road section, and current time, and then substituted into the above formula to calculate the dynamic weight of the vehicle. For example, the vehicle weight influencing factor W=9.639, the speed influencing factor V=1.034, the stiffness influencing factor K=0.438, and the temperature influencing factor T=0.415 of a vehicle at a certain moment, then the dynamic weight of the vehicle at that moment DW=9.639×1.034×0.438×0.415=1.81. The pavement loss dynamic weight model is stored in the form of a function expression (DW=W×V×K×T) and a parameter table (the calculation formula and parameters of each influencing factor).

[0048] Preferably, the traffic flow redistribution module includes the following functions: Analyze the travel demand characteristics of vehicles and obtain a travel demand characteristic library; Carry out traffic priority assessment based on the travel demand feature database to obtain the vehicle demand priority matrix; Multi-path load impact prediction is performed based on the vehicle demand priority matrix and the road section load sensitivity coefficient to obtain path load impact prediction data; Demand-aware path generation is performed based on the path load impact prediction data and the vehicle demand priority matrix to obtain a demand matching path set.

[0049] In the embodiment of the present invention, the travel demand characteristics of the vehicle are analyzed through multi-source data such as vehicle registration information, historical driving trajectory data, and real-time traffic information. For registered vehicles, the vehicle type (passenger car / truck), usage nature (operational / non-operational), and affiliated unit (such as XX transportation company) information are obtained from the vehicle management department. For operating vehicles, their business scope (ordinary passenger transport, cargo transport, and dangerous goods transport) is further obtained. By analyzing the historical driving trajectory data of the vehicle (such as GPS data, with a time resolution of 10 seconds and a position accuracy of 5 meters), the common starting and ending points (OD, accurate to the street level), travel time rules (weekdays / weekends, peak / off-peak, with a time interval of 30 minutes), and driving path preferences (such as high-speed priority, shortest path priority) of the vehicle are extracted. For example, for a vehicle that often travels between residential area A and commercial area B during the morning and evening rush hours (7:00-9:00, 17:00-19:00) on weekdays, it is inferred that its travel purpose is commuting. Combined with real-time traffic information (such as road conditions and traffic control information), the real-time travel demand of the vehicle is analyzed. For example, for a vehicle heading to Hospital C, it is inferred that the purpose of travel is to seek medical treatment, and it is of high urgency. The above information is integrated to build a travel demand feature library. The library is indexed by vehicle ID and contains the following fields: vehicle type, usage nature, affiliated unit, business scope, common OD, travel time pattern, route preference, real-time travel purpose, and time sensitivity (high / medium / low). The library is stored in the form of a database table and updated every 5 minutes.

[0050] Based on the travel demand feature database, combined with traffic regulations and management objectives, formulate traffic priority assessment rules. Determine the basic priority of vehicles according to vehicle type and usage nature. Emergency rescue vehicles (ambulances, fire trucks, police cars) have the highest priority (level 1), public transportation vehicles (buses, shuttle buses) are second (level 2), ordinary passenger cars (level 3), ordinary trucks (level 4), and illegal vehicles (overloaded, over-limit, and uninspected vehicles) have the lowest priority (level 5). Adjust the priority of vehicles according to their real-time travel purpose and time sensitivity. For vehicles performing emergency tasks (ambulances transporting critically ill patients, engineering vehicles performing rescue tasks), their priority is raised to level 1. For travel with higher time sensitivity (catching a plane or a train), their priority is raised by one level (up to level 2). Dynamically adjust the vehicle priority by comprehensively considering the operation status of the road network. In traffic congested sections (congestion index greater than 0.8), priority is given to vehicles of levels 1 and 2. In sections with poor road conditions (road damage index greater than 0.7), the passage of vehicles of levels 4 and 5 is restricted. The above rules are quantified into scoring criteria. Level 1 is 100 points, level 2 is 80 points, level 3 is 60 points, level 4 is 40 points, and level 5 is 20 points. 20 points are added for high time sensitivity, 10 points for medium time sensitivity, and no points are added for low time sensitivity. The road network operation status is scored according to the congestion index and the degree of road damage, and then deducted from the total score of the vehicle. According to the scoring results, the vehicles are divided into different priority levels (1-5). Construct a vehicle demand priority matrix. The matrix has vehicle ID as rows and priority levels (1-5) as columns. The matrix elements represent the priority level of the vehicle at the current moment. The matrix is ​​stored in sparse matrix format and updated every 1 minute.

[0051] Combined with the vehicle demand priority matrix and the load sensitivity coefficient of the road section (from the traffic load assessment module), the impact of different path selection schemes on the road network load is predicted. A road network topology model is constructed. The highway network is abstracted into a directed graph consisting of nodes (intersections) and road sections (roads connecting intersections). The attributes of each road section include: length (unit: meter), number of lanes, design speed (unit: km / h), pavement type (asphalt pavement / cement pavement), pavement condition (expressed by the pavement damage index, between 0-1, the larger the value, the worse the pavement condition), and load sensitivity coefficient (from the traffic load assessment module). According to the vehicle's origin and destination (OD) and current location, the Dijkstra algorithm is used to calculate multiple (for example, 3) feasible paths. For each feasible path, its path load is calculated. The path load consists of two parts: one is the inherent load of each road section on the path (related to the length of the road section, the number of lanes, and the pavement condition), and the other is the additional load generated by vehicle traffic. The additional load is calculated based on the vehicle model, load, driving speed (from the pavement loss dynamic weight model) and the load sensitivity coefficient of the road section. Specific calculation formula: Path load = ∑(road section inherent load i + ∑(vehicle additional load ij)), where i represents the road section number and j represents the vehicle number. The inherent load of the road section is obtained based on historical traffic data and road condition assessment. Vehicle additional load = vehicle dynamic weight (from the road loss dynamic weight model) × road section load sensitivity coefficient. For all feasible paths, their path loads are calculated separately and weighted according to the priority of the vehicle. Vehicles with high priority (level 1 and 2) have lower path load weights (for example, 0.8); vehicles with low priority (level 4 and 5) have higher path load weights (for example, 1.2). The weighted load of each path is used as the path load impact prediction data.

[0052] Based on the path load impact prediction data and the vehicle demand priority matrix, differentiated recommended paths are generated for vehicles of different priorities. A multi-objective optimization method is used to comprehensively consider factors such as path length, travel time, path load, and user preference. Optimization objective function: Minimize: α×travel time+β×path load+γ×path length, where α, β, and γ are weight coefficients that are adjusted according to vehicle priority. For level 1 and 2 vehicles, α=0.6, β=0.2, and γ=0.2; for level 3 vehicles, α=0.4, β=0.3, and γ=0.3; for level 4 and 5 vehicles, α=0.2, β=0.5, and γ=0.3. The travel time is calculated based on the length of the road section and the design speed. The path length is directly obtained from the road network topology model. For each vehicle, the objective function value of each path is calculated from the candidate path set according to its priority and current location, and the path with the smallest objective function value is selected as the recommended path. Construct a demand matching path set. The collection is indexed by vehicle ID and contains the following fields: recommended route (including road segment number sequence, estimated travel time, estimated route load), recommendation level ("recommended"). The collection is stored in JSON format and pushed to the driver through navigation and APP.

[0053] Preferably, the demand-aware path generation includes: Extract vehicle load characteristics based on route load impact prediction data and a predefined vehicle classification system to obtain a vehicle load characteristic library; According to the vehicle model load characteristic library, the vehicle demand priority matrix is ​​analyzed to obtain the priority management strategy set; Based on the priority management strategy set and the path load impact prediction data, a detailed analysis of path load sensitivity is performed to obtain a road network load sensitivity map. According to the road network load sensitivity map and the priority management strategy set, the vehicle type characteristics and priority collaborative path planning are performed to obtain the collaborative path planning strategy; Optimize the road network load balancing according to the collaborative path planning strategy to obtain the road network load balancing solution; Differentiated path matching is performed according to the road network load balancing solution and vehicle type load feature library to obtain a demand matching path set.

[0054] In the embodiment of the present invention, the estimated number, average load and average speed of each vehicle type on each candidate path are extracted from the path load impact prediction data. The vehicle classification system is divided into: mini passenger car (M1 category, less than or equal to 9 seats), small passenger car (M1 category, less than or equal to 9 seats), medium-sized passenger car (M2 category, greater than 9 seats, less than or equal to 19 seats), large passenger car (M3 category, greater than 19 seats), mini truck (N1 category, total mass less than or equal to 1.8 tons), light truck (N1 category, total mass greater than 1.8 tons, less than or equal to 3.5 tons), medium truck (N2 category, total mass greater than 3.5 tons, less than or equal to 12 tons), heavy truck (N3 category, total mass greater than 12 tons), articulated train (including tractor and trailer) for each candidate path. For each candidate path, the load index of each vehicle type is calculated. The load index calculation formula is: vehicle type load index = vehicle type number × average load × average speed × vehicle type loss contribution. For example, for path P1, it is expected that there will be 100 minibuses, with an average load of 0.8 tons, an average speed of 80 km / h, and a loss contribution of 0.01. Then, the load index of minibuses on this path = 100 × 0.8 × 80 × 0.01 = 64. Similarly, calculate the load index of all other vehicle models on path P1. Summarize the load index, number of models, average load, and average speed of all vehicle models on all candidate paths into a table, which is indexed by path ID and vehicle model. Then, obtain the type, usage nature (operational / non-operational), and affiliated unit of the registered vehicle from the vehicle management department. For operating vehicles, further obtain their business scope (ordinary passenger transport, cargo transport, and dangerous goods transport). Add these static vehicle attribute information to the above table. This expanded table is the vehicle model load feature library, which is stored in CSV file format.

[0055] Read the vehicle load feature library and the vehicle demand priority matrix. The vehicle demand priority matrix has vehicle ID as the row and priority level (1-5) as the column. The matrix elements represent the priority level of the vehicle at the current moment (1 is the highest and 5 is the lowest). For each record in the vehicle load feature library, find the priority level of the corresponding vehicle ID in the priority matrix. According to the priority level, the records in the vehicle load feature library are divided into five subsets, corresponding to five priority levels. For each priority level subset, count the proportion of vehicles of different models, different usage properties, and different units. For example, for the subset with priority level 1, count the proportion of emergency rescue vehicles such as ambulances, fire trucks, and police cars, as well as the proportion of ordinary vehicles performing emergency tasks. According to the statistical results, formulate a priority management strategy. The strategy is expressed in the form of rules. For example: Rule 1: If the vehicle type is an ambulance, fire truck, or police car, set its priority to level 1. Rule 2: If the vehicle is performing an emergency task (based on real-time traffic information and vehicle driving trajectory, such as transporting critically ill patients to the hospital), raise its priority to level 1. Rule 3: If the congestion index of the current road section is greater than 0.8 (the congestion index is calculated based on real-time traffic flow data) and the vehicle priority is level 1 or 2, its right of way will be prioritized. Rule 4: If the road damage index of the current road section is greater than 0.7 (the road damage index comes from the road condition monitoring module) and the vehicle priority is level 4 or 5, its passage will be restricted or it will be guided to detour. Rule 5: For vehicles carrying dangerous goods, if their driving path passes through densely populated areas or water source protection areas, their priority will be reduced by one level. All formulated priority management strategies are stored as a rule base, which is the priority management strategy set, stored in XML file format.

[0056] Read the priority management strategy set and path load impact prediction data. The path load impact prediction data includes the expected traffic flow, vehicle type composition, average speed, and weighted path load (from the traffic load assessment module) of each candidate path after considering vehicle type and environmental factors. For each road segment in the road network, calculate its load change under different priority management strategies according to the candidate path set to which it belongs. For example, for road segment L1, assume that it belongs to candidate paths P1 and P2. In the baseline case (no priority management strategy is implemented), the expected traffic flow of P1 and P2 is F1 and F2 respectively. According to the rules in the priority management strategy set, adjust the traffic flow distribution on P1 and P2. For example, according to rule 3, if the congestion index of L1 is greater than 0.8, vehicles with priority levels 4 and 5 on P1 and P2 are guided to other paths, thereby reducing the load of L1. Calculate the load change rate of L1 under different priority management strategies. Load change rate = (adjusted load - baseline load) / baseline load. The larger the absolute value of the load change rate, the more sensitive the road segment is to the priority management strategy. Visualize the load change rate data of all road sections in the road network. Using geographic information system (GIS) technology, the load change rate of the road section is represented on the map with color depth. The darker the color, the greater the load change rate, that is, the more sensitive the road section is to the priority management strategy. The generated map is the road network load sensitivity map, which is stored in GeoTIFF format.

[0057] Read the road network load sensitivity map and priority management strategy set. For each vehicle type, formulate differentiated path planning strategies based on its load index, average load, average speed and other characteristics in the vehicle type load feature library, as well as its priority level in the priority management strategy set. For example, for heavy trucks (N3 category), their load index is high and their priority is usually low (level 4 or 5). When planning the path, it is preferred to guide them to sections with lower network load sensitivity (i.e. sections with lighter colors in the road network load sensitivity map), even if this means that the path length may increase. For ambulances (priority level 1), the shortest path or the fastest path is preferred, even if the path passes through sections with higher network load sensitivity. Integrate the path planning strategies of different vehicle types and priorities. For example: Strategy 1: For vehicles with priority level 1 (such as ambulances and fire trucks), choose the shortest path or the fastest path without considering the load sensitivity of the section. Strategy 2: For vehicles with priority level 2 (such as buses and shuttles), on the basis of the shortest path and the fastest path, the load sensitivity of the road section is appropriately considered to avoid selecting sections with excessive load sensitivity. Strategy 3: For vehicles with priority level 3 (such as ordinary passenger cars), a balanced path is selected by comprehensively considering the path length, travel time and section load sensitivity. Strategy 4: For vehicles with priority levels 4 and 5 (such as ordinary trucks and heavy trucks), sections with lower load sensitivity of the road network are given priority, even if this means an increase in path length or travel time. The integrated path planning strategy is stored as a rule base, which is the collaborative path planning strategy, stored in JSON format.

[0058] Read the collaborative path planning strategy. Use an iterative optimization algorithm to redistribute traffic flow in the road network to achieve a balanced road network load. Initially, all vehicles follow the shortest or fastest path. In each iteration, perform the following operations: 1. According to the collaborative path planning strategy, a candidate path is selected for each vehicle.

[0059] 2. Calculate the load of each road section in the road network. Road section load = ∑ (weighted equivalent axle times of all vehicles passing through the road section), where the weighted equivalent axle times comes from the pavement loss dynamic weight model (traffic load assessment module).

[0060] 3. Calculate the balance of the road network load. Use standard deviation or coefficient of variation to measure the balance of the road network load. The smaller the standard deviation, the more balanced the road network load.

[0061] 4. If the balance of the road network load does not meet the preset threshold (for example, the standard deviation is greater than a certain set value), adjust the paths of some vehicles. The adjustment strategy is: prioritize low-priority vehicles (level 4 and 5) on high-load sections to low-load sections, while considering the path selection preferences for different vehicle types in the collaborative path planning strategy. During the adjustment process, avoid assigning vehicles to paths that do not match their priorities (for example, avoid assigning ambulances to paths with too long detours).

[0062] 5. Repeat steps 1-4 until the balance of the road network load meets the preset threshold or the maximum number of iterations is reached.

[0063] The optimization algorithm can be genetic algorithm, simulated annealing algorithm, particle swarm algorithm, etc. Taking genetic algorithm as an example, the specific implementation is as follows: 1. Encoding: Encode each vehicle’s path selection into a gene. For example, if a vehicle has 3 candidate paths, use a gene of length 1 to represent its path selection (1, 2, or 3). The path selections of all vehicles in the entire road network form a chromosome.

[0064] 2. Initial population: Randomly generate a set of chromosomes (for example, 100) as the initial population.

[0065] 3. Fitness function: The fitness function value is proportional to the balance of the road network load (the inverse of the standard deviation). The smaller the standard deviation, the larger the fitness function value.

[0066] 4. Selection: Use the roulette wheel selection method to select chromosomes to enter the next generation based on the fitness function value.

[0067] 5. Crossover: Perform a crossover operation on the selected chromosomes. Randomly select two chromosomes and exchange some of their gene fragments.

[0068] 6. Mutation: Perform mutation operations on chromosomes. Randomly select a gene and change its value with a certain probability (for example, change path 1 to path 2).

[0069] 7. Iteration: Repeat steps 3-6 until the termination condition is met (the maximum number of iterations is reached or the fitness function value no longer increases significantly).

[0070] After multiple iterations of optimization, a set of optimal chromosomes is obtained, each chromosome represents a road network load distribution scheme. The scheme corresponding to the chromosome with the highest fitness function value is selected as the final road network load balancing scheme. The scheme is stored in a table, and each row of the table represents a vehicle, which contains the following fields: vehicle ID, recommended route (section number sequence), estimated travel time, and estimated route load.

[0071] Read the road network load balancing scheme and vehicle model load feature library. For each vehicle in the road network load balancing scheme, perform the following operations: Find the vehicle model, usage nature, affiliated unit, common OD, travel time pattern, route preference and other information from the vehicle model load feature library. Generate personalized route recommendation information based on the characteristics of the vehicle and the route assigned to it in the road network load balancing scheme. For example, for a commuter vehicle that often travels between residential area A and commercial area B during the morning and evening rush hours on weekdays, the recommendation information may include: "It is recommended that you choose route X, with an estimated travel time of Y minutes, which can avoid congested section Z." For a heavy truck, the recommendation information may include: "In order to reduce the damage to the road surface, it is recommended that you choose route W, which passes through the road section with better road conditions." Encapsulate the recommendation information and route information (section number sequence, estimated travel time, estimated route load) into JSON format data.

[0072] The personalized route recommendation information of all vehicles is aggregated into a set, which is the demand matching route set. The set is stored in JSON format and pushed to the driver through the navigation system, mobile phone APP, etc. Each element in the set corresponds to a vehicle and contains the following fields: vehicle ID, recommended route (section number sequence), estimated travel time, estimated route load, personalized recommendation information, and recommendation level ("recommendation").

[0073] Preferably, the maintenance opportunity determination module includes the following functions: According to the differentiated diversion scheme, the traffic flow time series pattern analysis is carried out to obtain the traffic flow time series characteristic map; According to the flow time series characteristic map, the maintenance environment suitability is evaluated to obtain the environmental suitability time series; Perform maintenance resource availability analysis based on the environmental suitability time series to obtain a resource availability configuration table; Determine maintenance priorities based on pavement loss warning indicators and resource availability configuration tables to obtain segmented maintenance plans; The maintenance time window is optimized according to the traffic time series characteristic map, environmental suitability time series, resource availability configuration table and segmented maintenance plan to obtain the maintenance activity time window.

[0074] In an embodiment of the present invention, a differentiated diversion scheme is read (from a traffic flow redistribution module). The differentiated diversion scheme includes a recommended route (section number sequence) for each vehicle in a specific time period. For each section in the road network, the number of vehicles and vehicle type composition (minivans, small buses, medium buses, large buses, mini trucks, light trucks, medium trucks, heavy trucks, articulated trains) passing through the section in different time periods (with a time interval of 15 minutes) are counted. The traffic flow data of each section is analyzed using a time series analysis method. The traffic flow time series is fitted and predicted using an autoregressive moving average model (ARMA) or an exponential smoothing model. For example, using an ARMA (p, q) model, where p and q represent the autoregressive order and the moving average order, respectively, the optimal p and q values ​​are determined using the Akaike Information Criterion (AIC) or the Bayesian Information Criterion (BIC). Using the trained model, the traffic flow of each section in the future (such as the next 24 hours) is predicted. Identify the periodic patterns of traffic flow, such as morning and evening rush hours on weekdays, changes in traffic flow on weekends, etc. Visualize the traffic flow time series data, prediction results, and periodic patterns of each road section. Draw a curve chart of traffic flow changes over time, and mark the peak hours and trough hours in the chart. Summarize the traffic flow time series analysis results of all road sections into a map, which is the traffic time series feature map. The traffic time series feature map is indexed by the road section ID and contains the following fields: road section ID, historical traffic flow time series data (one data point every 15 minutes), predicted traffic flow time series data (one data point every 15 minutes), periodic patterns (weekday morning rush hour, weekday evening rush hour, weekend daytime, etc.), peak period start and end time, trough period start and end time. The map is stored in HDF5 file format.

[0075] Read the traffic time series characteristic map and meteorological data (from the traffic load assessment module). Meteorological data includes temperature, humidity, precipitation, wind speed, and wind direction. According to the highway maintenance technical specifications, determine the suitability of pavement maintenance operations under different meteorological conditions. For example, asphalt pavement paving operations are required to be carried out under weather conditions with a temperature above 10°C, no precipitation, and a wind speed of less than level 4. Cement concrete pavement maintenance requires temperatures above 5°C and relative humidity greater than 80%. According to meteorological data and maintenance technical specifications, calculate the environmental suitability of each time period (with 1 hour as a time interval). Environmental suitability is represented by a value between 0 and 1. The larger the value, the more suitable the environmental conditions are for maintenance operations. For example, if the current temperature is 15°C, there is no precipitation, and the wind speed is level 3, the environmental suitability of asphalt pavement paving operations is 0.9. If the current temperature is 8°C and there is light rain, the environmental suitability of asphalt pavement paving operations is 0.1. Arrange the environmental suitability of each time period in chronological order to form a time series, which is the environmental suitability time series. The environmental suitability time series is indexed by time and contains the following fields: time (accurate to the hour), environmental suitability (a value between 0 and 1), and the type of maintenance operation that is suitable (asphalt pavement paving, cement concrete pavement maintenance, road marking, etc.). The time series is stored in CSV file format.

[0076] Read the environmental suitability time series. Obtain maintenance resource information from the highway maintenance management department, including: the number of maintenance personnel, the type and quantity of maintenance equipment (asphalt pavers, rollers, milling machines, caulking machines, road marking machines, etc.), and the inventory of maintenance materials (asphalt mixture, cement, gravel, road marking paint, etc.). Determine the maintenance resources required for each time period based on the type of maintenance operation suitable for each time period in the environmental suitability time series. For example, if a certain time period is suitable for asphalt pavement paving operations, resources such as asphalt pavers, rollers, and asphalt mixtures are required. Determine the number of maintenance personnel available in each time period based on the work schedule of maintenance personnel (for example, 8-hour work system, three shifts, etc.). Determine the number of maintenance equipment available in each time period based on the number and working status of maintenance equipment (normal, repair, maintenance). Determine the number of maintenance materials available in each time period based on the inventory and consumption of maintenance materials. Summarize the available maintenance resource information for each time period into a table, which is the resource availability configuration table. The resource availability configuration table is indexed by time and contains the following fields: time (accurate to the hour), number of available maintenance personnel, type and quantity of available maintenance equipment, type and quantity of available maintenance materials, and type of maintenance operations that can be performed. This table is stored in the form of a database table.

[0077] Read the pavement loss warning indicators (from the road condition monitoring module) and the resource availability configuration table. The pavement loss warning indicators include the pavement damage type (cracks, potholes, subsidence, etc.), damage degree, and damage location (pile number) of each road section. Determine the maintenance priority of each road section based on the severity of the pavement damage and the impact on driving safety. For example, damage that seriously affects driving safety (such as potholes with a depth greater than 5cm and longitudinal cracks with a length greater than 10m) has the highest priority (level 1), and minor damage (such as cracks with a width less than 3mm) has a lower priority (level 3). According to the resource availability configuration table, determine the maintenance resources available for each time period. According to the maintenance priority and available resources, formulate a segmented maintenance plan. The maintenance plan needs to clearly define the maintenance operation type, maintenance time period, and required maintenance resources for each road section. For example, for section A, there are multiple potholes with a depth greater than 5cm on its pavement, and the maintenance priority is level 1. According to the resource availability configuration table, the pothole repair work is arranged for section A in time period T1 (for example, from 22:00 at night to 6:00 the next morning, when the traffic volume is small and the environmental suitability is high). The required resources are: 1 asphalt paver, 2 rollers, 6 maintenance personnel, and 20 tons of asphalt mixture. The section maintenance plan is stored in a table. Each row of the table represents a section and contains the following fields: section ID, maintenance priority (level 1-3), maintenance operation type, planned maintenance time period, and required maintenance resources.

[0078] Read the traffic time series characteristic map, environmental suitability time series, resource availability configuration table and segmented maintenance plan. For each section in the segmented maintenance plan, optimize the maintenance time window according to its planned maintenance time period, traffic time series characteristic map, environmental suitability time series and resource availability configuration table. The optimization goal is to minimize the impact on traffic while meeting the resources and environmental conditions required for the maintenance operation. The specific operations are as follows: 1. Identify traffic flow changes during the planned maintenance period based on the traffic time series characteristic map.

[0079] 2. Based on the environmental suitability time series, identify changes in environmental suitability during the planned maintenance period.

[0080] 3. According to the resource availability configuration table, confirm the resource availability during the planned maintenance period.

[0081] 4. If the traffic volume is high, the environmental suitability is low, or resources are unavailable during the planned maintenance period, adjust the maintenance time window. The adjustment strategy is: give priority to time periods with low traffic volume, high environmental suitability, and available resources. For example, if the maintenance work was originally planned to be carried out during the day, but the traffic volume is high during the day, the maintenance time window can be adjusted to the night. If the maintenance work was originally planned to be carried out in a certain time period, but the environmental suitability during that time period is low (such as in case of rainfall), the maintenance time window can be postponed until the weather improves.

[0082] 5. Consider the connection of maintenance time windows of adjacent sections. Avoid performing maintenance work on adjacent sections in the same time period to avoid causing large-scale traffic congestion.

[0083] 6. Repeat steps 1-5 to finally determine the maintenance time window for each road section.

[0084] The maintenance activity time window is represented in the form of a Gantt chart, with the horizontal axis representing time and the vertical axis representing the road section. Each rectangular bar represents the maintenance time window of a road section, and the length of the rectangular bar represents the duration of the maintenance operation.

[0085] Preferably, the maintenance effect verification module includes the following functions: According to the maintenance activity time window and the road loss warning index, the road surface physical parameters before and after maintenance are compared and collected to obtain the road surface physical parameter recovery index; The vehicle vibration response characteristics are collected according to the road surface physical parameter recovery index to obtain the vehicle vibration response data; Maintenance feedback optimization processing is performed based on vehicle vibration response data to achieve highway maintenance safety management tasks.

[0086] In an embodiment of the present invention, before the specific maintenance start time determined by the maintenance activity time window (from the maintenance timing determination module), the pavement physical parameter collection equipment is deployed immediately. A multifunctional pavement inspection vehicle is used, which integrates a laser deflection meter, a ground penetrating radar, a pavement flatness meter and a pavement image acquisition system. The laser deflection meter is used to measure the deflection value of the pavement under the standard axle load, with a measurement accuracy of 0.01mm and a measurement interval of 5m. The ground penetrating radar is used to detect the internal structure of the pavement, including the thickness of each structural layer and whether there are defects such as voids and cavities. The detection depth is greater than 1m and the horizontal resolution is less than 10cm. The pavement flatness meter uses a laser profiler to measure the international flatness index (IRI) according to the method specified in the "Highway Roadbed and Pavement Field Test Regulations" (JTG 3450-2019). The measurement speed is 60km / h, and an IRI value is output every 20m. The pavement image acquisition system uses a high-speed linear array camera to capture the pavement surface image, with an image resolution of 1mm / pixel, and can identify cracks with a width greater than 0.5mm.

[0087] According to the road section pile number range and damage type recorded in the road loss warning indicator (from the road condition monitoring module) that need maintenance, determine the specific scope of road surface physical parameter collection and the parameters to focus on. For example, if the road loss warning indicator shows that the main damage type of a road section is cracks, focus on the crack rate and road surface image; if the main damage type is rutting, focus on the deflection value and road surface flatness.

[0088] The multifunctional pavement inspection vehicle inspects the road sections that need maintenance according to the pre-set route and speed, and collects the pavement physical parameter data before maintenance. During the data collection process, the location information (latitude and longitude coordinates and stake number) of each measuring point is recorded through the GPS positioning system. Immediately after the maintenance end time determined by the maintenance activity time window, the same multifunctional pavement inspection vehicle is used to collect the pavement physical parameters of the same road section using exactly the same equipment parameters, inspection route and driving speed to obtain the pavement physical parameter data after maintenance. The pavement physical parameter data before and after maintenance are compared. For each pavement physical parameter, calculate its recovery index. The recovery index calculation formula is: Recovery index = [(parameter value after maintenance - parameter value before maintenance) / parameter value before maintenance] × 100%; for example: The average deflection value of a road section before maintenance was 0.45 mm, and the average deflection value after maintenance was 0.30 mm. The deflection recovery index = [(0.30-0.45) / 0.45]×100%=-33.3% (negative value indicates improvement).

[0089] The average IRI of a road section before maintenance was 3.8 m / km, and the average IRI after maintenance was 2.5 m / km. The IRI recovery index = [(2.5-3.8) / 3.8]×100%=-34.2% (negative values ​​indicate improvement).

[0090] The average structural depth of the layer before maintenance of a certain road section was 0.65mm, and it was 0.91mm after maintenance. The structural depth recovery index = [(0.91-0.65) / 0.65]×100%=40%.

[0091] The recovery index of all pavement physical parameters is summarized into a table, which is indexed by section ID, pile number range, pavement physical parameter type (deflection value, IRI, structural depth, crack rate, etc.) and has recovery index as an entry. The table is stored in the form of a database table.

[0092] The vibration sensor network is redeployed on the road section after maintenance. The type, quantity and layout of the sensors are exactly the same as those in the first layout in the road condition monitoring module. Piezoresistive pressure sensors are used to measure the surface pressure distribution of the road surface. The sensors are arranged along the wheel track line at a spacing of 1.5 meters to form a longitudinal pressure monitoring belt. Fiber Bragg grating vibration sensors are used to measure the internal vibration response of the road surface. The sensors are buried in the road base layer (25 cm deep from the surface) at a spacing of 4.5 meters to form a deep vibration monitoring network. All sensors are networked through the Zigbee wireless communication protocol, and the data are collected to the edge computing unit on the road side. The time synchronization of all sensors is completed through the GPS module, and the synchronization error is less than 1 millisecond. Start sensor data collection. The sampling frequency of the piezoresistive pressure sensor is set to once every 3 minutes, and the sampling frequency of the fiber Bragg grating vibration sensor is set to once every 20 minutes. The data collected by the pressure sensor is the surface pressure value of the road surface (unit: kPa), which reflects the stress distribution of the road surface under the action of the vehicle load. The data collected by the vibration sensor is the acceleration time history curve of the road surface in three orthogonal directions (X, Y, and Z axes) (unit: g, gravity acceleration). The received raw data was verified to remove invalid data that exceeded the sensor range or was generated by communication interruption. The sliding average filter algorithm was used to reduce the noise of the pressure data and filter out high-frequency noise interference (the filter window size was set to 7 sampling points). For the vibration data, a bandpass filter (cutoff frequency was 0.5Hz and 250Hz) was used to remove low-frequency drift and high-frequency noise. To ensure data synchronization, the pressure data was linearly interpolated based on the sampling time of the vibration sensor to generate a data set that was completely aligned with the vibration data timestamp. Fast Fourier transform (FFT) was performed on the synchronized vibration data (three-axis acceleration time history curve). A section of vibration data collected every 20 minutes was divided into multiple overlapping windows (the window length was 4096 sampling points and the overlap rate was 25%). The Hanning window function was applied to each window for windowing to reduce spectral leakage. FFT calculation was performed on each windowed data segment to obtain the amplitude spectrum of the window. The amplitude spectra of all windows were averaged to obtain the average amplitude spectrum of the 20-minute period. The frequency range is divided into three frequency bands: low frequency band (0.5-25Hz, reflecting the overall structural response of the road surface), medium frequency band (25-75Hz, reflecting the state of the road base and surface layer), and high frequency band (75-250Hz, reflecting the micro cracks in the road surface). Calculate the sum of the energy in each frequency band and normalize it (divide it by the total energy) to obtain the energy proportion of the three frequency bands. Calculate the peak frequency (the frequency corresponding to the maximum energy) and the spectral center of gravity (the weighted average of the frequency, with the weight being the amplitude) of the amplitude spectrum. Combine the average amplitude spectrum of each 20-minute period, the energy proportion of the three frequency bands, the peak frequency, and the spectral center of gravity into a feature vector. Arrange the feature vectors of all time periods in chronological order to form vehicle vibration response data.The data is stored in matrix form, with each row representing a time point and each column representing a vibration characteristic parameter.

[0093] Read the vehicle vibration response data collected after maintenance and the vehicle vibration response data collected before maintenance (from the road condition monitoring module). For each vibration characteristic parameter (average amplitude spectrum, energy proportion of each frequency band, peak frequency, spectral center of gravity), calculate the change rate before and after maintenance. The change rate calculation formula is: Change rate = [(parameter value after maintenance - parameter value before maintenance) / parameter value before maintenance] × 100%; for example: If the energy proportion of the mid-frequency band (25-75Hz) of the vibration response of a road section before maintenance is 35% and after maintenance is 28%, then the rate of change of the mid-frequency band energy proportion = [(28-35) / 35]×100%=-20% (negative value indicates a decrease).

[0094] If the peak frequency before maintenance is 62Hz and after maintenance is 55Hz, the rate of change of peak frequency = [(55-62) / 62]×100%=-11.3%.

[0095] Comprehensively evaluate the maintenance effect based on the change rate of vibration characteristic parameters and the recovery index of pavement physical parameters. Establish evaluation rules. For example: Rule 1: If the IRI recovery index is less than -30%, and the change rate of the energy proportion of the mid-frequency band and the high-frequency band is less than -15%, the maintenance effect is evaluated as "excellent". Rule 2: If the IRI recovery index is between -20% and -30%, and the change rate of the energy proportion of the mid-frequency band and the high-frequency band is between -10% and -15%, the maintenance effect is evaluated as "good". Rule 3: If the IRI recovery index is between -10% and -20%, and the change rate of the energy proportion of the mid-frequency band and the high-frequency band is between -5% and -10%, the maintenance effect is evaluated as "medium". Rule 4: If the IRI recovery index is greater than -10%, or the change rate of the energy proportion of the mid-frequency band and the high-frequency band is greater than -5%, the maintenance effect is evaluated as "poor". Rule 5: If the deflection recovery index is negative (indicating an increase in the road surface bearing capacity) and the rate of change of the energy share in the low frequency band (0.5-25Hz) is negative (indicating a weakening of the overall structural response of the road surface), the maintenance effect level is further improved. Rule 6: If the structural depth recovery index is positive (indicating an improvement in anti-skid performance) and the vibration energy in a specific frequency band (e.g., around 1000Hz) generated by the contact between the tire and the road surface when the vehicle passes by increases, the maintenance effect level is further improved (because good anti-skid performance is usually accompanied by a certain degree of vibration increase).

[0096] If the maintenance effect is rated as "poor", it is necessary to analyze the reasons and adjust the maintenance plan. Possible reasons include: unqualified maintenance materials, improper construction technology, road damage beyond expectations, and drainage system failure. According to different reasons, take corresponding improvement measures: replace qualified maintenance materials, improve construction technology, conduct more comprehensive inspection and evaluation of the road surface, repair or rebuild the drainage system. Record the maintenance effect evaluation results, cause analysis, improvement measures and other information, and update the road condition data in the road loss warning indicator. This information will serve as an important basis for subsequent highway maintenance decisions. By establishing a maintenance effect verification and feedback optimization mechanism, a closed-loop management of "detection-evaluation-maintenance-verification-feedback" is formed to continuously improve the quality and efficiency of highway maintenance, and ultimately achieve the highway maintenance safety management task.

[0097] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0098] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A system for highway maintenance safety management, characterized in that: Includes the following modules: The road condition monitoring module is used to monitor the multi-layer road condition data of the highway to obtain the road surface vibration characteristic spectrum; perform surface-deep correlation detection on the road surface vibration characteristic spectrum to obtain the road surface abnormal point data; quantify the road surface loss risk of the road surface abnormal point data to obtain the road surface loss warning index; Traffic load assessment module, used to collect vehicle images and meteorological data; analyze vehicle model loss contribution based on vehicle images; construct an environmental impact coefficient matrix based on meteorological data; construct a dynamic weight model based on vehicle model loss contribution and environmental impact coefficient matrix to obtain a pavement loss dynamic weight model; perform section load sensitivity analysis based on the pavement loss dynamic weight model and pavement loss early warning indicators to obtain a section load sensitivity coefficient; The traffic flow redistribution module is used to predict the multi-path load impact according to the load sensitivity coefficient of the road section to obtain the path load impact data; generate demand-aware paths according to the path load impact data to obtain a demand matching path set; Perform differentiated induction strategy processing according to the demand matching path set to obtain differentiated diversion solutions; A maintenance timing determination module is used to determine the timing of road maintenance according to the differentiated diversion scheme and obtain a maintenance activity time window; The maintenance effect verification module is used to implement and optimize maintenance according to the maintenance activity time window to achieve highway maintenance management tasks.

2. The highway maintenance safety management system according to claim 1, characterized in that: The road condition monitoring module includes the following functions: Identify key sections of the highway and arrange sensors to obtain the topology of the sensor network; Perform multi-level data collection based on the road section sensor network topology map to obtain multi-layer road condition raw data; Extract vibration spectrum features from multi-layer road condition raw data to obtain road vibration characteristic spectrum; The road surface vibration characteristic spectrum is used to detect road surface stratification and obtain road surface abnormal point data.

3. The highway maintenance safety management system according to claim 2, characterized in that: The road surface delamination detection comprises: Perform multi-scale vibration feature extraction on the road vibration characteristic spectrum to obtain a multi-scale vibration feature tensor; Self-supervised abnormal pattern recognition is performed on multi-scale vibration feature tensors to obtain a road surface abnormality probability distribution map; According to the road surface anomaly probability distribution diagram, surface-deep association rules are mined to obtain the association data between the pavement structure layers; The damage evolution is predicted based on the inter-layer correlation data of the pavement structure and the pavement anomaly probability distribution map to obtain the pavement anomaly point data.

4. The highway maintenance safety management system according to claim 3, characterized in that: The table-deep association rule mining includes: Obtain pavement engineering archive data; construct a digital twin of the pavement multi-layer structure based on the pavement engineering archive data to obtain a digital twin model of the pavement structure; Design the physical constraint architecture based on the digital twin model of the pavement structure to obtain the physical constraint data; The damage-vibration transfer function is derived based on the physical constraint data and the digital twin model of the pavement structure to obtain a damage-vibration transfer function library; Multi-layer Bayesian network reasoning is performed based on the damage vibration transfer function library and the road surface abnormality probability distribution map to obtain the multi-layer state probability map of the road surface; The surface-deep symptom association knowledge is extracted from the multi-layer state probability map of the pavement to obtain the correlation data between the pavement structure layers.

5. The highway maintenance safety management system according to claim 1, characterized in that: The traffic load assessment module includes the following functions: Collect vehicle images and weather data on key road sections; Carry out multi-dimensional monitoring and classification of traffic flow based on vehicle images to obtain vehicle type load classification data; Carry out vehicle model loss contribution analysis based on vehicle model load classification data to obtain vehicle model loss contribution; The environmental impact matrix is ​​constructed based on meteorological data to obtain the environmental impact coefficient matrix; A dynamic weight model is constructed according to the vehicle type loss contribution and environmental impact coefficient matrix to obtain a road surface loss dynamic weight model.

6. The highway maintenance safety management system according to claim 5, characterized in that: The dynamic weight model construction includes: The four-dimensional key factor data is standardized according to the vehicle model loss contribution and environmental impact coefficient matrix to obtain a standardized factor data table, wherein the standardized factor data table includes vehicle weight data, vehicle speed data, road stiffness data and temperature data; Use vehicle weight data to analyze the relationship between vehicle weight and road surface to obtain vehicle weight impact data; Use vehicle speed data to analyze the speed-road dynamic response relationship and obtain speed response data; Use the pavement stiffness data to analyze the pavement stiffness-bearing characteristics and obtain stiffness characteristic data; Using temperature data to analyze temperature modulation effect, obtaining temperature modulation data; A comprehensive dynamic weight calculation is performed based on vehicle weight impact data, speed response data, stiffness characteristic data and temperature modulation data to obtain a pavement loss dynamic weight model.

7. The highway maintenance safety management system according to claim 1, characterized in that: The traffic flow redistribution module includes the following functions: Analyze the travel demand characteristics of vehicles and obtain a travel demand characteristic library; Carry out traffic priority assessment based on the travel demand feature database to obtain the vehicle demand priority matrix; Multi-path load impact prediction is performed based on the vehicle demand priority matrix and the road section load sensitivity coefficient to obtain path load impact prediction data; Demand-aware path generation is performed based on the path load impact prediction data and the vehicle demand priority matrix to obtain a demand matching path set.

8. The highway maintenance safety management system according to claim 7, characterized in that: The demand-aware path generation includes: Extract vehicle load characteristics based on route load impact prediction data and a predefined vehicle classification system to obtain a vehicle load characteristic library; According to the vehicle model load characteristic library, the vehicle demand priority matrix is ​​analyzed to obtain the priority management strategy set; Based on the priority management strategy set and the path load impact prediction data, a detailed analysis of path load sensitivity is performed to obtain a road network load sensitivity map. According to the road network load sensitivity map and the priority management strategy set, the vehicle type characteristics and priority collaborative path planning are performed to obtain the collaborative path planning strategy; Optimize the road network load balancing according to the collaborative path planning strategy to obtain the road network load balancing solution; Differentiated path matching is performed according to the road network load balancing solution and vehicle type load feature library to obtain a demand matching path set.

9. The highway maintenance safety management system according to claim 1, characterized in that: The maintenance opportunity determination module includes the following functions: According to the differentiated diversion scheme, the traffic flow time series pattern analysis is carried out to obtain the traffic flow time series characteristic map; According to the flow time series characteristic map, the maintenance environment suitability is evaluated to obtain the environmental suitability time series; Perform maintenance resource availability analysis based on the environmental suitability time series to obtain a resource availability configuration table; Determine maintenance priorities based on pavement loss warning indicators and resource availability configuration tables to obtain segmented maintenance plans; The maintenance time window is optimized according to the traffic time series characteristic map, environmental suitability time series, resource availability configuration table and segmented maintenance plan to obtain the maintenance activity time window.

10. The highway maintenance safety management system according to claim 1, characterized in that: The maintenance effect verification module includes the following functions: According to the maintenance activity time window and the road loss warning index, the road surface physical parameters before and after maintenance are compared and collected to obtain the road surface physical parameter recovery index; The vehicle vibration response characteristics are collected according to the road surface physical parameter recovery index to obtain the vehicle vibration response data; Maintenance feedback optimization processing is performed based on vehicle vibration response data to achieve highway maintenance safety management tasks.

Images