Low-carbon asphalt road periodic maintenance method based on edge computing gateway

Abstract

The application discloses a low-carbon asphalt road periodic maintenance method based on an edge computing gateway, and comprises the following steps: accessing multiple source sensing devices through the edge computing gateway, collecting road surface images, structural stress, environment and maintenance material carbon emission associated data and shunting conversion; calling a multi-modal feature fusion crack identification algorithm to extract fused crack features and synchronously acquiring historical maintenance carbon footprint data; transmitting the multi-dimensional data after correlation mapping to a carbon-efficiency double-target maintenance cycle prediction model; outputting a cycle prediction result through model feature weighting and dimension recombination, and dynamically correcting in combination with material consumption rate data; calling a time series data driven performance degradation early warning algorithm to analyze time series correlation, and generating early warning information; synchronously transmitting information to a terminal through an edge cloud collaborative carbon footprint monitoring platform, and generating a maintenance execution scheme in combination with different parameters. The method realizes dynamic adjustment of maintenance decision, guarantees road performance requirements, and realizes minimization of carbon emission.

Description

Technical Field

[0001] This invention relates to the field of periodic road maintenance technology, and in particular to a periodic maintenance method for low-carbon asphalt roads based on an edge computing gateway. Background Technology

[0002] With the deep penetration of low-carbon development concepts into the infrastructure sector, the need for carbon footprint management and performance assurance in asphalt road maintenance is becoming increasingly prominent. Current road maintenance relies heavily on manual inspections and experience-based periodic planning, which struggles to adapt to the differentiated damage characteristics and carbon emission constraints of various road sections. The rise of edge computing and multi-source sensing technologies makes real-time monitoring possible, but integrating multi-dimensional data such as pavement images, structural stress, and environmental impacts to achieve carbon-efficiency synergistic optimization of maintenance decisions remains a challenge. Furthermore, road performance degradation exhibits significant temporal characteristics, and traditional maintenance methods lack dynamic correlation analysis between historical data and real-time parameters, resulting in insufficiently targeted maintenance solutions. Therefore, there is an urgent need to construct an intelligent maintenance technology system that integrates edge and cloud computing.

[0003] Existing technologies suffer from two significant drawbacks: First, the accuracy of multi-source data fusion and crack identification is insufficient. They rely heavily on single-modal data for pavement damage assessment, failing to achieve a deep correlation mapping between image features, structural stress, and carbon emission data. This makes it difficult to comprehensively reflect the nature of road damage and the impact of maintenance carbon efficiency, resulting in significant biases in damage identification and an inability to provide accurate data support for carbon efficiency optimization. Second, there is a lack of synergy between maintenance cycle prediction and carbon footprint management. Existing models often consider either performance degradation or carbon emission as a single objective in isolation, without establishing a coupled optimization mechanism between the two. Furthermore, they lack dynamic interaction and correction between real-time edge data and cloud-based models, making it impossible to adjust cycle plans based on real-time parameters such as maintenance material consumption and environmental changes. Consequently, maintenance decisions fail to meet both performance assurance requirements and the goal of minimizing carbon emissions. Summary of the Invention

[0004] To overcome the shortcomings and deficiencies of existing technologies, this invention provides a method for periodic maintenance of low-carbon asphalt roads based on edge computing gateways.

[0005] A periodic maintenance method for low-carbon asphalt roads based on edge computing gateways includes:

[0006] Step S1: Access multi-source sensing devices deployed at different monitoring sections of the road through an edge computing gateway to collect pavement image data streams, structural stress response data, environmental impact factor data, and carbon emission correlation data of maintenance materials for low-carbon asphalt roads. Perform preliminary data splitting and protocol conversion on the collected data through edge nodes.

[0007] Step S2: Call the multimodal feature fusion crack recognition algorithm integrated at the edge to perform feature extraction and fusion processing on the image data stream, identify the geometric and distribution features of road cracks, and simultaneously obtain the historical maintenance carbon footprint data of the corresponding monitoring section through the edge-cloud collaborative carbon footprint monitoring platform;

[0008] Step S3: Correlate and map the crack identification results, structural stress response data, environmental impact factor data, and historical maintenance carbon footprint data, and transmit them to the input layer of the carbon-efficiency dual-objective maintenance cycle prediction model deployed in the cloud;

[0009] Step S4: The associated data is weighted and reorganized in terms of features using the carbon-efficiency dual-objective maintenance cycle prediction model. The cycle prediction results under different maintenance strategies are output and dynamically corrected by combining the maintenance material consumption rate data collected in real time at the edge.

[0010] Step S5: Call the time-series data-driven road performance degradation early warning algorithm to perform time-series correlation analysis on the corrected cycle prediction results, explore the potential mapping relationship between road performance parameters and maintenance cycle, and generate phased performance degradation early warning information;

[0011] Step S6: Synchronize the early warning information and maintenance cycle parameters to the terminal device through the edge cloud collaborative carbon footprint monitoring platform, and generate a periodic maintenance execution plan based on the carbon emission coefficient of maintenance materials, pavement damage rate and performance degradation rate parameters.

[0012] Furthermore, the multimodal feature fusion crack identification algorithm adopts the following expression:

[0013]

[0014] in, The fused crack feature vector These are the fusion weights for edge features, texture features, shape features, and contextual features, respectively. The coordinates of the pixels in the road surface image. For edge feature extraction operators, For texture feature extraction operators, For shape feature extraction operators, Context feature extraction operator.

[0015] Furthermore, the carbon-efficiency dual-objective maintenance cycle prediction model adopts the following expression:

[0016]

[0017] in, For optimal maintenance cycle, These are the weighting coefficients for carbon emission targets and performance targets, respectively. For carbon emission calculation functions, For performance benefit function, As a maintenance cycle variable, For maintenance process types, The carbon emission coefficient of maintenance materials, For road surface damage rate, These are the initial performance parameters. The comprehensive environmental impact index, This represents the performance degradation rate.

[0018] Furthermore, the time-series data-driven road performance degradation early warning algorithm adopts the following expression:

[0019]

[0020] in, This represents the performance degradation warning value at time t. The time window length, The weight of the k-th time series point. The sampling time interval, for Road performance parameters at any given time This is a timing error correction term. For early warning correlation functions, To predict maintenance cycles, This represents the cumulative carbon footprint value.

[0021] Furthermore, the edge-cloud collaborative carbon footprint monitoring platform adopts a carbon flow tracing model:

[0022]

[0023] in, For total carbon footprint, Carbon emissions from material production For transporting carbon emissions, For construction energy consumption and carbon emissions, To reduce carbon emissions, The amount of the i-th type of maintenance material is... The carbon emission coefficient for producing the i-th material. For transport volume, For unit carbon emissions transportation, For the j-th type of energy consumption, Let j be the carbon emission coefficient of the j-th energy source. To recover the amount of material, The unit is the carbon emission reduction factor for recycling.

[0024] Furthermore, the periodic maintenance parameters for the low-carbon asphalt road are optimized using a collaborative optimization model:

[0025]

[0026] in, To collaboratively optimize the target value, To optimize weights, This represents the total carbon emissions during the cycle. Let i be the amount of material used. Let be the carbon emission coefficient of the i-th material. Let t be the energy consumption at time t. Energy carbon emission coefficient, This represents the performance residual rate at the end of the cycle. For initial performance, Let t be the performance degradation rate at time t.

[0027] Furthermore, step S3 includes the following sub-steps:

[0028] S31: The local data processing module of the edge computing gateway performs spatiotemporal alignment of the crack length, width, and density parameters obtained from crack identification with the deflection and shear stress parameters in the structural stress response data, and establishes a multi-dimensional data association matrix based on the monitoring section coordinates.

[0029] S32: Retrieve carbon emission data from the historical database of the edge-cloud collaborative carbon footprint monitoring platform for the three most recent maintenance operations of the corresponding monitoring section, including detailed parameters of carbon emission from material production, transportation, and construction, and perform time-dimensional correlation matching with the currently collected data;

[0030] S33: The monitoring data of different dimensions are transformed into a unified feature space through the data mapping algorithm. The crack feature parameters are encoded using geometric features, the stress data are processed by physical quantity normalization, and the carbon footprint data are encoded using time series to form the model input dataset.

[0031] S34: Perform outlier detection on the input dataset, remove outlier data points that exceed the reasonable fluctuation range by comparing neighborhood data, and encapsulate the corrected dataset into a data format that meets the interface requirements of the carbon-efficiency dual-objective maintenance cycle prediction model.

[0032] Furthermore, step S4 includes the following sub-steps:

[0033] S41: Transmit the encapsulated input dataset to the feature processing layer of the cloud model, and dynamically weight the crack features, stress features, environmental features and carbon footprint features through a hierarchical attention mechanism to generate a multi-dimensional feature vector;

[0034] S42: Calls the core calculation module of the carbon-efficiency dual-objective maintenance cycle prediction model, performs multi-scenario maintenance cycle simulation calculations based on feature vectors, and outputs carbon emission values ​​and performance retention rate parameters corresponding to maintenance intervals of 6 months, 12 months and 24 months.

[0035] S43: Receive maintenance material inventory data and consumption rate data in real time at the edge, combine them with material replenishment cycle parameters to verify the feasibility of cycle prediction results for different scenarios, and mark prediction results that do not meet the material supply conditions.

[0036] S44: The weighted correction algorithm is used to adjust the prediction results that have passed the verification. The weight of material consumption rate is integrated with the weight of carbon emission and performance to obtain the preliminary value of the corrected optimal maintenance cycle.

[0037] Furthermore, step S5 includes the following sub-steps:

[0038] S51: Retrieve the time series data of road performance parameters for the corresponding monitoring section over the past five years from the time series database, including road surface smoothness, skid resistance, and structural strength parameters, and match them with the corrected preliminary value of the maintenance cycle on a time scale.

[0039] S52: Call the time series feature extraction module of the time series data-driven road performance degradation early warning algorithm to perform trend decomposition on the time series data of performance parameters and separate the long-term degradation trend, seasonal fluctuations and random disturbance components.

[0040] S53: Substitute the preliminary value of the maintenance cycle into the decay trend model to predict the theoretical value of the performance parameters at each time point within the cycle, perform deviation analysis with the real-time collected performance monitoring data, and calculate the deviation rate and fluctuation amplitude parameters.

[0041] S54: Adjust the warning threshold based on the deviation analysis results. When the probability of the predicted performance parameter being lower than the preset threshold exceeds the set proportion, generate a phased performance degradation warning information including the decay rate and the expected time to reach the target.

[0042] The present invention has the following beneficial effects:

[0043] This invention proposes a low-carbon asphalt road periodic maintenance method based on an edge computing gateway. This method achieves significant benefits through multi-dimensional technological innovation, effectively overcoming the shortcomings of existing technologies. By leveraging an edge computing gateway to connect to multi-source sensing devices, it integrates pavement images, structural stress, environmental impact, and carbon emission-related data. Combined with multi-modal feature fusion crack identification technology, it achieves deep correlation mapping of multi-modal data, replacing the single-modal judgment mode and significantly improving crack identification accuracy. This provides comprehensive data support for carbon efficiency optimization and solves the problems of insufficient accuracy in multi-source data fusion and defect identification. Through an edge-cloud collaborative carbon footprint monitoring platform linked to a cloud-based carbon-efficiency dual-objective maintenance cycle prediction model, it establishes a coupled optimization mechanism between performance degradation and carbon emissions. It dynamically corrects the cycle prediction results by combining real-time collected parameters such as material consumption and environmental changes from the edge. Simultaneously, it utilizes a time-series data-driven degradation early warning algorithm to link historical and real-time performance data, enabling dynamic adjustment of maintenance decisions. This ensures both road performance requirements and minimizes carbon emissions, compensating for the lack of synergy between maintenance cycle prediction and carbon footprint management, and constructing a precise, efficient, and intelligent maintenance system. Attached Figure Description

[0044] Figure 1 This is a flowchart of the method steps of the present invention;

[0045] Figure 2 This is a diagram showing the unit composition for implementing the method of the present invention. Detailed Implementation

[0046] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] like Figure 1 As shown, the periodic maintenance method for low-carbon asphalt roads based on edge computing gateways includes:

[0048] Step S1: Access multi-source sensing devices deployed at different monitoring sections of the road through an edge computing gateway to collect pavement image data streams, structural stress response data, environmental impact factor data, and carbon emission correlation data of maintenance materials for low-carbon asphalt roads. Perform preliminary data splitting and protocol conversion on the collected data through edge nodes.

[0049] Specifically, step S1 is the initial step in maintenance data collection and preprocessing, constructing a multi-dimensional, high-fidelity raw data pool to provide basic support for subsequent algorithm analysis and model calculations, avoiding decision-making biases caused by missing data or disordered formats. In the specific implementation process, the edge computing gateway adopts an industrial-grade hardware architecture, connecting to multi-source sensing devices linearly distributed along the road via Ethernet, 4G / 5G, and LoRa communication interfaces. Monitoring sections are set up according to a standard of three per kilometer for a two-way four-lane road, located at the lane wheel track, road centerline, and shoulder, with four sets of sensing devices deployed at each section. The high-definition image sensor uses a 2-megapixel CMOS chip, acquiring RGB image data streams with a resolution of 1920×1080 and a frame rate of 2 frames / second. The fiber optic stress sensor has a range of 0-50MPa, a sampling frequency of 10Hz, and captures the stress response of the road structure in real time. The temperature, humidity, and precipitation sensors have measurement accuracies of ±0.5℃, ±5%RH, and ±0.1mm, respectively, and acquire environmental impact factor data once per minute. The material weight sensor uses a strain gauge structure, with a measurement range of 0-1000kg, and acquires data on the remaining amount and consumption of maintenance materials once per hour. After acquisition, the multi-channel data classifier built into the edge node automatically sorts the data according to data type, importing image data into a dedicated image processing cache channel, stress data into a physical quantity analysis channel, and environmental and material data into a comprehensive parameter channel. At the same time, through ModbusRTU and TCP / IP protocol conversion modules, the different formats of data output from heterogeneous devices are uniformly converted into UTF-8 encoded JSON format. The data fields include acquisition timestamp, device ID, monitoring section coordinates, and parameter values, providing format compatibility for cross-module data calls.

[0050] Step S2: Call the multimodal feature fusion crack recognition algorithm integrated at the edge to perform feature extraction and fusion processing on the image data stream, identify the geometric and distribution features of road cracks, and simultaneously obtain the historical maintenance carbon footprint data of the corresponding monitoring section through the edge-cloud collaborative carbon footprint monitoring platform;

[0051] Specifically, step S2 is a crucial step in disease feature extraction and historical data association. It transforms the raw image data into quantifiable disease parameters and associates them with historical maintenance carbon emission information, bridging the gap between subsequent data fusion and cycle prediction, and solving the problem of disconnect between disease identification and carbon efficiency analysis. In the specific implementation, the edge device pre-loads a multimodal feature fusion crack identification algorithm through an embedded operating system. The algorithm's memory usage is controlled within 512MB, and the startup response time is less than 1 second. The algorithm first processes the image data stream transmitted in S1 into blocks, using a sliding window method to divide each frame into 16 512×512 pixel sub-blocks. Edge features are extracted using the Canny operator, texture features using the gray-level co-occurrence matrix, shape features using the Hu moment, and contextual features using a fully connected layer. These four types of features are fused using a dynamically updated weight matrix, with initial weight values ​​set to 0.3, 0.25, 0.25, and 0.2, respectively. The output includes geometric and distribution feature data such as crack length, width, density, orientation, and area percentage, with a feature extraction accuracy of no less than 92%. Meanwhile, the edge computing gateway establishes a point-to-point connection with the edge cloud collaborative carbon footprint monitoring platform through an IPSec VPN encrypted channel. Based on the GPS coordinates (error ±0.5m) of the monitoring section in the WGS84 coordinate system, it retrieves detailed carbon footprint data of all maintenance records of the section in the past 5 years, including carbon emissions from the production of materials such as asphalt and aggregates (accurate to the unit), carbon emissions from transportation (calculated by transportation distance × carbon emissions per unit mileage), carbon emissions from construction machinery energy consumption, and carbon emissions associated with human activities. Each record includes maintenance time, maintenance process type, maintenance area, material usage, and corresponding carbon emission value. Through timestamp matching, the spatiotemporal alignment of crack identification results with historical carbon footprint data is achieved, with the alignment deviation controlled within 1 minute.

[0052] Step S3: Correlate and map the crack identification results, structural stress response data, environmental impact factor data, and historical maintenance carbon footprint data, and transmit them to the input layer of the carbon-efficiency dual-objective maintenance cycle prediction model deployed in the cloud;

[0053] Specifically, step S3 breaks down data silos, integrating scattered disease, stress, environmental, and carbon emission data into a standardized dataset. This improves the completeness and relevance of the input data for the prediction model, ensuring the accuracy of model calculations. In the implementation process, the local data processing module of the edge computing gateway is first activated. This module uses an ARM Cortex-A9 processor with a processing frequency of 1GHz. Using the GPS coordinates of the monitoring section as a reference index, it aligns the crack length (accurate to 0.1cm), width (accurate to 0.01cm), and density (strips / m²) parameters output by S2 with the deflection values ​​(0-10mm) and shear stress (0-30MPa) parameters in the structural stress response data collected by S1 in a spatiotemporal manner. Based on the timestamp, a multi-dimensional data association matrix with 12 dimensions is synchronously constructed, where the row index represents the acquisition time and the column index represents the parameter type. Subsequently, detailed data from the three most recent maintenance operations of this section were screened from the historical carbon footprint database. Sub-parameters such as material carbon emission coefficient (kgCO2 / kg), carbon emission values ​​(kgCO2) corresponding to transportation distance (km), construction energy consumption (kWh), and carbon emissions were extracted. Pearson correlation analysis was then performed with the currently collected environmental temperature, humidity, and precipitation data to identify historical data segments with environmental similarity exceeding 80%. Next, a data mapping algorithm based on principal component analysis was used to transform all data into a unified feature space within the [0,1] interval. Crack features were converted into 6-dimensional numerical vectors using geometric feature encoding rules, stress data were linearly transformed according to the sensor range ratio, carbon footprint data were encoded into 4-dimensional vectors in chronological order, and environmental data were directly normalized to measured values, forming an initial dataset containing 2000 data records. Finally, the 3σ criterion was used to detect outliers in the dataset. The mean and standard deviation of each parameter were calculated, and outliers exceeding the range of [mean - 3σ, mean + 3σ] were removed. The corrected dataset was then packaged into an HDF5 format file that includes feature names, data dimensions, sample size, and numerical matrix, conforming to the input interface specification of the carbon-efficiency dual-objective maintenance cycle prediction model.

[0054] Step S4: The associated data is weighted and reorganized in terms of features using the carbon-efficiency dual-objective maintenance cycle prediction model. The cycle prediction results under different maintenance strategies are output and dynamically corrected by combining the maintenance material consumption rate data collected in real time at the edge.

[0055] Specifically, step S4 combines the dual objectives of carbon efficiency and environmental efficiency to generate a scientific maintenance cycle plan, and improves the plan's feasibility through real-time data correction, solving the problems of strong subjectivity and disconnect from actual conditions in traditional cycle prediction. In the implementation process, the HDF5 format dataset encapsulated in S3 is transmitted to the cloud server via a 5G private network at a transmission rate of no less than 10Mbps and a latency controlled within 50ms, and then connected to the input layer of the carbon-efficiency dual-objective maintenance cycle prediction model deployed in the cloud. This model adopts an architecture combining deep learning and traditional optimization algorithms. The input layer includes 12 feature neurons. First, through the hierarchical attention mechanism of the feature processing layer, initial weights of 0.3, 0.25, 0.2, and 0.25 are assigned to crack features, stress features, environmental features, and carbon footprint features, respectively. A 32-dimensional multi-dimensional feature vector is then generated through a fully connected layer. The core computational module of the model simulates maintenance cycles across multiple scenarios based on feature vectors. The maintenance cycle variable is set to range from 6 to 36 months, with a step size of 6 months. A gradient descent algorithm is used to solve the dual-objective optimization problem of carbon emissions and performance, outputting carbon emission values ​​and performance retention rates for six different cycle plans. Specifically, for a 12-month cycle, carbon emissions are 120 tons with a performance retention rate of 85%; for an 18-month cycle, carbon emissions are 160 tons with a performance retention rate of 80%; and for a 24-month cycle, carbon emissions are 210 tons with a performance retention rate of 70%. Simultaneously, the edge device receives data in real-time from the maintenance material inventory management system via the MQTT protocol, obtaining the current inventory (accurate to kg), average daily consumption rate (kg / day), and replenishment cycle (days) of major materials such as asphalt and aggregates. If the material demand for a given cycle plan exceeds 70% of the current inventory and the replenishment cycle exceeds 15 days, it is marked as an infeasible plan. Finally, a weighted correction algorithm is adopted, which introduces a material consumption rate weight of 0.2, and integrates it with a carbon emission weight of 0.4 and a performance weight of 0.4 to perform calculations. The cycle value of the feasible scheme is fine-tuned, and the preliminary value of the corrected optimal maintenance cycle is output, with the error controlled within ±1 month.

[0056] Step S5: Call the time-series data-driven road performance degradation early warning algorithm to perform time-series correlation analysis on the corrected cycle prediction results, explore the potential mapping relationship between road performance parameters and maintenance cycle, and generate phased performance degradation early warning information;

[0057] Specifically, step S5 is a crucial step in performance degradation early warning analysis. It uses time-series data mining to predict road performance trends, identify critical points for maintenance needs in advance, and provide forward-looking validation for maintenance cycle plans, avoiding passive maintenance caused by sudden performance changes. In the implementation process, the cloud-based InfluxDB time-series database retrieves performance parameters from the past 5 years (once a month, 180 sets in total) by section ID, including smoothness (IRI, m / km), skid resistance (BPN), and structural strength (deflection basin parameters). A time-series data-driven early warning algorithm (developed in Python, using the TensorFlow framework) is activated, employing the db4 wavelet basis function to perform a three-level decomposition of the data, separating long-term degradation trends (univariate linear regression, R²≥0.85), seasonal fluctuations (quarterly cycle), and random disturbances (moving average smoothing). The preliminary cycle values ​​from S4 are substituted into the degradation model, and linear interpolation is used to predict the theoretical monthly average performance value within the cycle. This is compared with the latest three months of measured data to calculate the deviation rate (difference between theoretical and measured values / measured value) and the fluctuation amplitude. When the deviation rate is greater than 5% and the fluctuation range is greater than 3% for two consecutive months, the warning threshold (smoothness 6.0m / km, anti-skid 45BPN, deflection 2.0mm) is adjusted. The performance compliance time is calculated by combining the attenuation rate (monthly index change) and a warning message containing attenuation rate, compliance time and influencing factors is generated. The warning is divided into three levels according to the attenuation rate.

[0058] Step S6: Synchronize the early warning information and maintenance cycle parameters to the terminal device through the edge cloud collaborative carbon footprint monitoring platform, and generate a periodic maintenance execution plan based on the carbon emission coefficient of maintenance materials, pavement damage rate and performance degradation rate parameters.

[0059] Specifically, step S6 transforms the early warning information and periodic parameters into an operable maintenance execution plan, achieving a closed loop from data processing to on-site operations and improving the efficiency of maintenance decision implementation. During implementation, the edge-cloud collaborative carbon footprint monitoring platform (B / S architecture) transmits the early warning information and periodic parameters to the terminal (command center industrial computer: i7-10700K CPU, 16GB memory; on-site tablet: Android 11, 10.1-inch screen) via a RESTful API interface, with a latency of <3 seconds. The platform retrieves the carbon emission coefficients of 12 types of maintenance materials (e.g., base asphalt 3.2, modified asphalt 3.5), pavement damage rate (accuracy 0.1%), the average performance degradation rate over the past 6 months, and the personnel / equipment configuration and efficiency parameters for 8 types of processes. The plan generation engine uses a greedy algorithm for optimization, determining the process based on the damage rate and early warning level, calculating the material quantity based on the maintenance area, dividing the road into sections based on traffic flow (each section ≤500 meters), determining the operation window based on environmental forecasts (avoiding rain and high temperatures), and calculating the estimated total carbon emissions. The final solution consists of 12 modules (maintenance scope, material usage, process standards, etc.), which is stored in encrypted PDF format and pushed to the terminal, supporting online viewing and printing.

[0060] In a further implementation, the multimodal feature fusion crack identification algorithm adopts the following expression:

[0061]

[0062] in, The fused crack feature vector; The fusion weights are for edge features, texture features, shape features, and context features, respectively. The initial fusion weights for the four types of features are set to 0.3, 0.25, 0.25, and 0.2, respectively. The fusion weights of the four types of features are dynamically adjusted according to different road surface types. For example, the weight of shape features is appropriately increased in areas with high asphalt pavement smoothness, and the weight of texture features is increased in areas with slight wear on the pavement. The coordinates of the pixels in the road surface image. For edge feature extraction operators, For texture feature extraction operators, For shape feature extraction operators, Context feature extraction operator.

[0063] Specifically, the multimodal feature fusion crack identification algorithm improves crack identification accuracy through weighted fusion of multi-dimensional features, solving the problems of susceptibility to noise interference and incomplete feature extraction in single feature identification, and providing reliable damage data support for subsequent maintenance decisions. In its implementation, the algorithm is deployed in the embedded processing unit of an edge computing gateway, with memory usage controlled within 512MB and a startup response time of no more than 1 second, adapting to the limited computing resources at the edge. During algorithm execution, the input road surface image is first divided into blocks, each with a size of 512×512 pixels. Then, four types of key features are extracted: edge features are extracted using the Canny operator to capture the gray-level difference boundary between the crack and the road surface; texture features are extracted using the gray-level co-occurrence matrix to identify the texture distribution pattern in the crack area; shape features are extracted using Hu moments to quantify the geometric contour of the crack; and contextual features are extracted using a fully connected layer to supplement crack location information in conjunction with the surrounding road environment. The fusion weights of the four types of features are dynamically adjusted according to different road surface types. The initial weights are set to 0.3, 0.25, 0.25 and 0.2 respectively. The weights of shape features are appropriately increased in areas with high asphalt pavement smoothness, and the weights of texture features are increased in areas with slight wear on the pavement. The final output fusion feature vector can comprehensively reflect the geometric and distribution characteristics of cracks. According to actual tests, the feature extraction accuracy is stable at over 92%, which meets the requirements for real-time edge recognition.

[0064] In a further embodiment, the carbon-efficiency dual-objective maintenance cycle prediction model adopts the following expression:

[0065]

[0066] Where argmin() represents the value of the variable that minimizes the objective function, i.e., the minimum value over the maintenance period T; To determine the optimal maintenance cycle, the gradient descent algorithm is used to solve for the optimal maintenance cycle, and the corresponding carbon emission values ​​and performance retention rates are calculated for different cycle schemes. These are the weighting coefficients for carbon emission targets and performance targets, respectively. This is a carbon emission calculation function that integrates parameters such as the carbon emission coefficient of maintenance materials, transportation distance, and construction energy consumption. This is the performance benefit function, which is associated with indicators such as initial performance parameters, comprehensive environmental impact index, and performance degradation rate. For the maintenance cycle variable, the cycle variable range is set to 6-36 months, with a step size of 6 months; For maintenance process types, The carbon emission coefficient of maintenance materials, For road surface damage rate, These are the initial performance parameters. The comprehensive environmental impact index, This represents the performance degradation rate.

[0067] Specifically, the carbon-efficiency dual-objective maintenance cycle prediction model establishes a quantitative correlation between maintenance cycle and carbon emissions and performance, achieving synergistic optimization between the two. This breaks the carbon-efficiency imbalance problem caused by the single-objective decision-making of traditional models, providing a quantitative basis for the scientific formulation of maintenance cycles. In its implementation, the model is deployed on a cloud server, employing an architecture combining deep learning and traditional optimization algorithms. The input layer includes 12 feature neurons, adapting to the dimensional requirements of multi-source fusion data. During model execution, a hierarchical attention mechanism is first used to assign weights to the input crack features, stress features, environmental features, and carbon footprint features. The initial weight values ​​are 0.3, 0.25, 0.2, and 0.25, respectively, and can be dynamically adjusted according to maintenance priorities. The core of the model achieves dual-objective optimization by constructing a carbon emission calculation function and a performance benefit function. The carbon emission calculation function integrates parameters such as the carbon emission coefficient of maintenance materials, transportation distance, and construction energy consumption, while the performance benefit function correlates initial performance parameters, comprehensive environmental impact index, performance degradation rate, and other indicators. The optimal maintenance cycle is solved using the gradient descent algorithm, with the cycle variable set to a range of 6-36 months and a step size of 6 months. The corresponding carbon emission values ​​and performance retention rates are calculated for different cycle schemes. For example, a 12-month cycle corresponds to 120 tons of carbon emissions and a performance retention rate of 85%, while a 24-month cycle corresponds to 210 tons of carbon emissions and a performance retention rate of 70%. The final output is the optimal cycle value that balances minimizing carbon emissions and maximizing performance.

[0068] In a further implementation, the time-series data-driven road performance degradation early warning algorithm adopts the following expression:

[0069]

[0070] in, This represents the performance degradation warning value at time t. The timing window length is set to 12 sampling points. The weight of the kth time series point is assigned according to the time decay law, with recent data having a higher weight than older data. The sampling time interval is fixed at 30 days. for Road performance parameters at any given time; This is a timing error correction term; The early warning correlation function is constructed by combining the predicted maintenance cycle with the cumulative carbon footprint value; Let be the road performance parameters at time t; To predict maintenance cycles, The cumulative carbon footprint value is calculated; the performance degradation warning value at time t is calculated by adding the two parts.

[0071] Specifically, the time-series data-driven road performance degradation early warning algorithm achieves accurate prediction of performance degradation trends through in-depth mining of time-series data, identifies maintenance critical points in advance, and solves the problems of lack of time-series correlation analysis and delayed early warning in traditional early warning systems, providing forward-looking support for the dynamic adjustment of maintenance cycles. In its implementation, the algorithm is developed based on the TensorFlow framework, written in Python, and deployed in a cloud-based time-series data processing module. It supports the analysis of performance parameter data from the past 5 years, with data collection occurring monthly, totaling 180 valid data points. During algorithm execution, the time-series window length is first set to 12 sampling points, with the weight of each time-series point allocated according to the time decay law, with recent data having a higher weight than older data. The sampling interval is fixed at 30 days. Wavelet decomposition is used to perform a three-level decomposition of the time-series data, separating long-term degradation trends, seasonal fluctuations, and random disturbance components. The long-term degradation trend is fitted using a univariate linear regression model, with a goodness of fit R² not less than 0.85. Subsequently, a time-series error correction term is introduced, and an early warning correlation function is constructed by combining the predicted maintenance cycle with the cumulative carbon footprint value. By adding the two parts, the performance degradation warning value at different times is calculated, realizing a comprehensive warning calculation based on historical trends and current multiple factors. When the warning value exceeds the set threshold and the duration reaches 2 months, the corresponding level of warning is triggered. The warning information includes the degradation rate, the expected time to reach the target, and the main influencing factors.

[0072] In a further implementation, the edge-cloud collaborative carbon footprint monitoring platform employs a carbon flow tracing model:

[0073]

[0074] in, To determine the total carbon footprint, the entire maintenance process is divided into four key stages: material production, transportation, construction energy consumption, and recycling, with carbon emissions calculated for each stage separately. Carbon emissions from material production are calculated by statistically analyzing the usage of each maintenance material and its corresponding carbon emission coefficient, including 12 commonly used materials such as base asphalt, modified asphalt, and limestone aggregate. For transport carbon emissions, transport carbon emissions are calculated as the product of transport volume and unit transport carbon emissions. The transport volume is determined based on a combination of material weight and transport distance. Construction energy consumption carbon emissions integrate the consumption of different energy types and their corresponding carbon emission coefficients, including electricity, diesel, etc. To achieve carbon reduction through recycling, carbon reduction is calculated based on the amount of recycled materials and the unit carbon reduction coefficient. The amount of the i-th type of maintenance material is... The carbon emission coefficient for producing the i-th material. For transport volume, For unit carbon emissions transportation, For the j-th type of energy consumption, Let j be the carbon emission coefficient of the j-th energy source. To recover the amount of material, The unit is the carbon emission reduction factor for recycling.

[0075] Specifically, the carbon flow tracking model of the edge-cloud collaborative carbon footprint monitoring platform enables accurate calculation and dynamic tracking of the carbon footprint throughout the entire maintenance process. This breaks through the problems of fragmented processes and data lag in traditional carbon emission statistics, providing accurate carbon emission data support for optimizing the dual objectives of carbon and efficiency. In the implementation process, the model is integrated into the carbon accounting module of the edge-cloud collaborative carbon footprint monitoring platform. The platform adopts a B / S architecture, supporting real-time data synchronization between the edge and the cloud, with a synchronization delay of no more than 3 seconds. During model operation, the entire maintenance process is divided into four key stages: material production, transportation, construction energy consumption, and recycling, with carbon emissions calculated for each stage. Material production carbon emissions are calculated by statistically analyzing the usage of each maintenance material and its corresponding production carbon emission coefficient, including 12 commonly used materials such as base asphalt, modified asphalt, and limestone aggregate, with material usage accuracy controlled at the kilogram level. Transportation carbon emissions are calculated by multiplying the transportation volume by the carbon emission per unit of transportation, with the transportation volume determined comprehensively based on material weight and transportation distance. Construction energy consumption carbon emissions integrate the consumption of different energy types and their corresponding carbon emission coefficients, including electricity and diesel. Carbon emission reduction through recycling is calculated based on the amount of recycled materials and the carbon emission reduction coefficient per unit of recycled materials. The final total carbon footprint of the entire maintenance process is obtained by summarizing the data. The data update frequency is synchronized with material consumption and construction progress to ensure the real-time and accuracy of carbon emission data.

[0076] In a further embodiment, the periodic maintenance parameters for the low-carbon asphalt road adopt a collaborative optimization model:

[0077]

[0078] in, To collaboratively optimize the target value, The initial weights for carbon emission optimization and performance optimization are set to 0.4 and 0.6 respectively, and can be dynamically adjusted according to low-carbon development requirements and road performance conditions. The total carbon emissions within the period are calculated by integrating carbon emissions from material production, transportation, construction energy consumption, and other aspects, and processing energy consumption changes over different time periods through integral calculations. Let i be the amount of material used. Let be the carbon emission coefficient of the i-th material. Let t be the energy consumption at time t. Energy carbon emission coefficient, The performance residual rate at the end of the cycle is calculated based on the integral difference between the initial performance parameters and the performance decay rate within the cycle, reflecting the road performance status at the end of the maintenance cycle. For initial performance, Let t be the performance degradation rate at time t.

[0079] Specifically, the collaborative optimization model for periodic maintenance parameters of low-carbon asphalt roads establishes a quantitative optimization relationship between maintenance parameters and carbon emissions and performance, achieving efficient allocation of maintenance resources and solving the problems of empirical parameter setting and poor carbon-efficiency synergy in traditional maintenance. This provides a basis for parameter optimization in the precise formulation of maintenance plans. In implementation, the model is deployed in a cloud-based maintenance plan optimization module, running in conjunction with a carbon-efficiency dual-objective maintenance cycle prediction model. Input parameters include the usage and carbon emission coefficients of 12 types of maintenance materials, energy consumption data at each stage of construction, initial road performance parameters, and performance degradation rates. During model execution, a collaborative optimization objective function is first constructed, balancing the priority of carbon emission optimization and performance optimization through weight coefficients. The initial weight values ​​are set to 0.4 and 0.6, respectively, and can be dynamically adjusted according to low-carbon development requirements and road performance conditions. The carbon emission calculation integrates carbon emissions from material production, transportation, and construction energy consumption, processing energy consumption changes over different time periods through integral calculations. The performance residual rate calculation is based on the integral difference between the initial performance parameters and the performance degradation rate within the cycle, reflecting the road performance status at the end of the maintenance cycle. The greedy algorithm optimizes and adjusts parameters such as the amount of maintenance materials, the type of maintenance process, and the maintenance time nodes, and finally outputs the parameter combination that maximizes the collaborative optimization target value, ensuring that the maintenance plan minimizes carbon emissions while meeting performance requirements.

[0080] In a further implementation, step S3 includes the following sub-steps: S31: The local data processing module of the edge computing gateway performs spatiotemporal alignment of the crack length, width, and density parameters obtained from crack identification with the deflection and shear stress parameters in the structural stress response data, and establishes a multi-dimensional data association matrix based on the monitoring section coordinates; S32: The carbon emission data of the corresponding monitoring section for the last three maintenance operations are retrieved from the historical database of the edge-cloud collaborative carbon footprint monitoring platform, including detailed parameters of carbon emission from material production, transportation, and construction, and is matched with the currently collected data in terms of temporal correlation; S33: The monitoring data of different dimensions are converted into a unified feature space through a data mapping algorithm, wherein the crack feature parameters are encoded using geometric features, the stress data is processed using physical quantity normalization, and the carbon footprint data is encoded using time series, forming a model input dataset; S34: Outlier detection is performed on the input dataset, and abnormal data points exceeding the reasonable fluctuation range are removed by comparing neighborhood data, and the corrected dataset is encapsulated into a data format that meets the interface requirements of the carbon-efficiency dual-objective maintenance cycle prediction model.

[0081] Specifically, step S3 is implemented in stages to achieve efficient association and standardized conversion of multi-source data through a systematic data processing workflow. This solves the problem of insufficient model input quality caused by spatiotemporal misalignment and heterogeneous formats of different types of data, providing a highly adaptable dataset for the carbon-efficiency dual-objective prediction model. In the specific implementation process, S31 utilizes the local data processing module of the edge computing gateway (using an ARM Cortex-A9 processor with a computing frequency of 1GHz) to align the length, width, and density parameters obtained from crack identification with the deflection and shear stress parameters in the structural stress response, using the GPS coordinates of the monitoring section as a reference. A multi-dimensional data association matrix with 12 dimensions is established at 1-second timestamp intervals to ensure data time synchronization accuracy within 100 milliseconds. S32 retrieves detailed carbon emission data from the historical database of the edge-cloud collaborative carbon footprint monitoring platform for the corresponding section from the past three maintenance operations via an encrypted channel. This data includes values ​​from each stage of material production, transportation, and construction. It is then matched with the currently collected data based on a time similarity algorithm, with a matching threshold set at 80%. S33 employs a data mapping algorithm to transform multi-dimensional data into a unified feature space. Crack features are converted into numerical vectors according to geometric feature encoding rules, stress data is normalized according to the sensor range ratio, and carbon footprint data is encoded according to time series. The feature space dimension is unified to 32 dimensions. S34 uses a neighborhood data comparison method for outlier detection, setting a reasonable fluctuation range of the mean plus or minus three standard deviations. After removing outliers, the dataset is packaged into a format compatible with the model interface, ensuring data integrity of over 99%.

[0082] In a further implementation, step S4 includes the following sub-steps: S41: The encapsulated input dataset is transmitted to the feature processing layer of the cloud model. The crack features, stress features, environmental features, and carbon footprint features are dynamically weighted through a hierarchical attention mechanism to generate a multi-dimensional feature vector; S42: The core calculation module of the carbon-efficiency dual-objective maintenance cycle prediction model is called to perform multi-scenario maintenance cycle simulation calculations based on the feature vectors, and outputs carbon emission values ​​and performance retention rate parameters corresponding to maintenance intervals of 6 months, 12 months, and 24 months; S43: Maintenance material inventory data and consumption rate data are received in real time at the edge terminal. The feasibility of the cycle prediction results for different scenarios is verified by combining the material replenishment cycle parameters, and prediction results that do not meet the material supply conditions are marked; S44: The weighted correction algorithm is used to adjust the prediction results that have passed the verification. The material consumption rate weight, carbon emission weight, and performance weight are fused and calculated to obtain the preliminary value of the optimal maintenance cycle after correction.

[0083] Specifically, step S4, through a hierarchical feature processing, scenario simulation, feasibility verification, and correction process, improves the accuracy and practicality of the maintenance cycle prediction results, solves the problem of the disconnect between model calculation and actual maintenance conditions, and provides rigorous support for determining the optimal cycle. In the specific implementation process, S41 transmits the encapsulated input dataset to the cloud model feature processing layer via a 5G private network at a transmission rate of no less than 10Mbps and a latency controlled within 50 milliseconds. A hierarchical attention mechanism dynamically assigns weights to crack, stress, environmental, and carbon footprint features, with a weight adjustment step size of 0.05, generating a 32-dimensional multi-dimensional feature vector. S42 calls the model's core calculation module to perform multi-scenario simulations, setting maintenance intervals to three gradients: 6 months, 12 months, and 24 months. Carbon emission values ​​and performance retention rate parameters are calculated simultaneously under each scenario, with at least 100 iterations to ensure result convergence. S43 receives maintenance material inventory and consumption rate data in real time via the MQTT protocol, setting an inventory warning threshold of 70%. When the predicted demand exceeds the threshold and the replenishment cycle exceeds 15 days, the solution is marked as infeasible. S44 uses a weighted correction algorithm to integrate the material consumption rate weight (0.2), carbon emission weight (0.4), and performance weight (0.4) to fine-tune the feasible scheme. After correction, the cycle error is controlled within ±1 month.

[0084] In a further implementation, step S5 includes the following sub-steps: S51: Retrieve the time-series data of road performance parameters for the corresponding monitoring section over the past five years from the time-series database, including pavement smoothness, skid resistance, and structural strength parameters, and match them with the corrected preliminary maintenance cycle values ​​on a time scale; S52: Call the time-series feature extraction module of the time-series data-driven road performance degradation early warning algorithm to perform trend decomposition on the time-series data of performance parameters, separating out long-term degradation trends, seasonal fluctuations, and random disturbance components; S53: Substitute the preliminary maintenance cycle values ​​into the degradation trend model to predict the theoretical values ​​of performance parameters at each time node within the cycle, perform deviation analysis with the real-time collected performance monitoring data, and calculate the deviation rate and fluctuation amplitude parameters; S54: Adjust the early warning threshold based on the deviation analysis results. When the probability that the predicted performance parameter is lower than the preset threshold exceeds a set proportion, generate phased performance degradation early warning information including the degradation rate and the expected time to reach the target.

[0085] Specifically, step S5 involves in-depth mining and trend analysis of time-series data to achieve accurate early warning of performance degradation, solving the problems of traditional early warning systems lacking historical data correlation and lagging deviation judgment, and providing a forward-looking basis for dynamic adjustment of the maintenance cycle. In the specific implementation process, S51 retrieves performance parameter data for the corresponding section over the past five years from the time-series database, collected monthly, totaling 180 sets of valid data points, including three core indicators: pavement smoothness, skid resistance, and structural strength. These data are aligned with the preliminary values ​​of the maintenance cycle using a time-scale matching algorithm, with a matching time window set to 3 months. S52 activates the time-series feature extraction module, using wavelet decomposition to perform a 3-level decomposition of the data, selecting the db4 wavelet basis function to separate long-term degradation trends, seasonal fluctuations, and random disturbance components, with a goodness of fit of no less than 0.85 for the long-term trend. S53 substitutes the preliminary cycle values ​​into the degradation trend model, using linear interpolation to predict the theoretical performance values ​​for each month within the cycle, and performs deviation analysis with the latest 3-month data collected in real time, calculating the deviation rate and fluctuation amplitude, with the deviation rate calculated to two decimal places. S54 dynamically adjusts the warning threshold based on the deviation results. When the probability of the predicted performance being lower than the threshold exceeds 60%, it generates warning information including the decay rate and the expected time to reach the target. The warning level is divided into three levels according to the decay rate, and the level switching threshold interval is set to 0.2 units / month.

[0086] The multimodal feature fusion crack recognition algorithm in this invention is an intelligent technology deployed at the edge that achieves accurate road crack recognition by integrating multi-dimensional image features. It breaks through the limitations of traditional single-feature recognition and constructs a more comprehensive disease feature description system. Specifically, the algorithm is integrated into the embedded processing unit of the edge computing gateway, adapting to the limited computing resources at the edge (memory usage ≤ 512MB, startup response time < 1 second). During runtime, it first receives a 1920×1080 resolution road image data stream, and uses a sliding window method to divide each frame into 16 512×512 pixel sub-blocks. Then, it extracts four core features through a combination of multiple operators: the Canny operator captures the gray-level difference edges between the crack and the road surface; the gray-level co-occurrence matrix analyzes the texture distribution pattern of the crack area; the Hu moment quantifies the geometric shape contour of the crack; and the fully connected layer mines the contextual association information between the crack and the surrounding road surface. Subsequently, the fusion weights are dynamically adjusted based on the actual road surface conditions (initial values ​​are 0.3, 0.25, 0.25, and 0.2 respectively), and the four types of features are weighted and integrated. The final output includes quantifiable parameters such as crack length, width, density, direction, and area percentage. Actual measurements show that the feature extraction accuracy is consistently above 92%. This algorithm transforms disordered raw image data into structured defect feature data, enabling real-time and accurate crack identification at the edge. This provides high-quality defect baseline data for subsequent multi-source data association mapping and maintenance cycle prediction. It solves the problems of traditional single-feature recognition being susceptible to noise interference from road stains, lighting changes, etc., and the identification bias caused by incomplete feature extraction. By reducing data transmission pressure through localized edge processing, it provides reliable defect judgment criteria for maintenance decisions, making it a primary technical support for achieving precise maintenance.

[0087] The carbon-efficiency dual-objective maintenance cycle prediction model in this invention is a cloud-based quantitative tool for maintenance decision-making that balances carbon emission control and road performance assurance. It aims to overcome the limitations of traditional single-objective decision-making and achieve scientific optimization of the maintenance cycle. The implementation process is as follows: The model adopts an architecture combining deep learning and traditional optimization algorithms. The input layer is adapted to 32-dimensional multi-source fusion data processed by S3 (including crack features, structural stress, environmental factors, and carbon footprint data). First, a hierarchical attention mechanism dynamically assigns weights to the input features (initial weights are 0.3, 0.25, 0.2, and 0.25 respectively) to highlight the impact of key factors on the cycle. Then, a dual-objective calculation function is constructed. The carbon emission calculation function integrates parameters such as the carbon emission coefficients of 12 types of maintenance materials, transportation distance, and construction energy consumption. The performance-efficiency function correlates indicators such as initial road performance parameters, comprehensive environmental impact index, and performance degradation rate. The maintenance cycle variable is set to range from 6 to 36 months (in 6-month increments). The gradient descent algorithm is used for at least 100 iterative calculations, outputting the carbon emission values ​​and performance retention rates for each cycle (e.g., a 12-month cycle corresponds to 120 tons of carbon emissions and an 85% performance retention rate, while a 24-month cycle corresponds to 210 tons of carbon emissions and a 70% performance retention rate). Simultaneously, real-time data on maintenance material inventory (accurate to the kg level), daily consumption rate, and replenishment cycle are received from the edge device. Infeasible solutions with material demand exceeding inventory by 70% and a replenishment cycle greater than 15 days are marked. Finally, a weighted correction algorithm (material consumption weight 0.2, carbon emission weight 0.4, performance weight 0.4) is used to output the optimal cycle, with an error controlled within ±1 month. The model's function is to generate maintenance cycle solutions that balance carbon emissions and performance based on multi-source data and dynamically correct them according to real-time operating conditions, ensuring the scientific validity and feasibility of the solutions. It changes the traditional decision-making model of maintenance cycle relying on experience judgment or a single goal (such as considering only performance), solves the industry pain point of imbalance between carbon emission control and performance guarantee, provides a quantitative basis for the formulation of maintenance cycle, promotes the transformation of road maintenance from "experience-driven" to "data-driven", and helps to achieve the goal of low-carbon and efficient maintenance.

[0088] The time-series data-driven road performance degradation early warning algorithm in this invention is a technology that accurately predicts road performance degradation trends based on the correlation analysis of historical and real-time performance data. Its core lies in mining the temporal patterns of performance parameters to identify critical points for maintenance needs in advance. The implementation process is as follows: The algorithm is developed based on the TensorFlow framework and deployed in a cloud-based time-series data processing module. It supports the analysis of time-series data of road performance parameters over the past 5 years (collected monthly, totaling 180 valid data points), including three core indicators: pavement smoothness (IRI value), skid resistance (BPN value), and structural strength (deflection basin parameters). During runtime, an analysis window of 12 time-series points is first set, and weights are assigned according to the time decay pattern (recent data has higher weight than older data), with a fixed sampling interval of 30 days. Then, the time-series data is decomposed into three levels using the db4 wavelet basis function to separate the long-term degradation trend, seasonal fluctuations, and random disturbance components. The long-term degradation trend is fitted using a univariate linear regression model, with a goodness of fit R² of no less than 0.85. The preliminary maintenance cycle values ​​output from the carbon-efficiency dual-objective model are substituted into the degradation trend model. The theoretical performance parameters for each month within the cycle are predicted using linear interpolation. These are then compared point-by-point with the latest three-month performance monitoring data collected in real-time at the edge, calculating the deviation rate (difference between theoretical and measured values / measured value) and the fluctuation amplitude (difference between maximum and minimum deviation). When the probability that the predicted performance parameters are lower than the preset thresholds (smoothness 6.0 m / km, skid resistance 45 BPN, deflection 2.0 mm) exceeds 60%, a three-level early warning message is generated based on the degradation rate (interval of 0.2 units / month). This message includes the degradation rate value, the expected performance attainment time, and the main influencing factors (such as traffic volume and environmental corrosion). The algorithm aims to deeply explore the temporal evolution of road performance, achieve proactive early warning of performance degradation, and provide data support for the dynamic adjustment of the maintenance cycle. It solves the problems of delayed warnings and high misjudgment rates caused by the lack of historical data correlation and reliance on real-time data in traditional maintenance early warning systems. By predicting the risk of performance degradation in advance, it provides a buffer time for maintenance decisions, which can effectively reduce maintenance costs and road traffic impacts caused by sudden defects, and further improve the decision-making system for intelligent maintenance.

[0089] The edge-cloud collaborative carbon footprint monitoring platform of this invention is an integrated system that enables real-time tracking, accurate calculation, and data correlation of carbon emissions throughout the entire road maintenance process. It aims to overcome the fragmented processes and data lag issues inherent in traditional carbon emission statistics, providing high-quality carbon data support for optimizing both carbon and energy efficiency objectives. The implementation process is as follows: The platform adopts a B / S architecture, possessing real-time data collaboration capabilities between the edge and cloud. Data synchronization is achieved through an IPSec VPN encrypted channel, with a synchronization delay of no more than 3 seconds. The platform integrates a carbon flow tracking model, dividing the entire maintenance process into four key stages: material production, transportation, construction energy consumption, and recycling, and performing carbon emission calculation and aggregation for each stage. Carbon emissions from material production are calculated by statistically analyzing the usage (accurate to the kilogram level) of 12 commonly used maintenance materials, including base asphalt, modified asphalt, and limestone aggregate, and their corresponding carbon emission coefficients. Transportation carbon emissions are calculated by multiplying the material transport volume by the carbon emission per unit mileage; the transport volume is determined by combining material weight and transport distance. Construction energy consumption carbon emissions integrate the consumption of different energy types, such as electricity and diesel, with their corresponding carbon emission coefficients. Recycling carbon reduction is calculated based on the amount of waste material recycled and the carbon reduction coefficient per unit of recycling. The final result is the total carbon footprint of the entire maintenance process. The platform's data update frequency is synchronized with material consumption and construction progress, and it also supports retrieving historical carbon footprint data from the past 5 years. Timestamp matching achieves spatiotemporal alignment with real-time crack identification results and structural stress data (alignment deviation < 1 minute), forming a complete carbon efficiency data chain. The platform's role is to achieve accurate calculation, real-time tracking, and multi-source data correlation of carbon emissions throughout the entire maintenance process, providing highly timely and accurate carbon emission parameters for the carbon-efficiency dual-objective model. It fills the technological gap in the accurate monitoring of carbon footprint in the field of road maintenance, solves the problems of independent links, data lag and insufficient accuracy in traditional carbon emission statistics, and enables carbon emission control to shift from "post-event statistics" to "pre-event planning and in-event management". It provides core data support for low-carbon maintenance decision-making and helps the "dual carbon" goal be implemented in the infrastructure field.

[0090] like Figure 2As shown, a low-carbon asphalt road periodic maintenance method based on an edge computing gateway is implemented through different units, including: a multi-source heterogeneous data acquisition and protocol conversion unit, which is connected to image sensors, stress sensors, environmental sensors, and material monitoring sensors deployed at the road monitoring section. It uses the protocol parsing module built into the edge computing gateway to access and convert multiple types of data, outputting a standardized data stream; a multi-modal crack feature intelligent extraction and fusion unit, which is connected to the multi-source heterogeneous data acquisition and protocol conversion unit. This unit integrates a multi-modal feature fusion crack recognition algorithm, performing real-time edge feature extraction and multi-dimensional fusion processing on the image data stream, outputting a crack feature dataset; and an edge-cloud collaborative data transmission and association mapping unit, which is connected to both the multi-modal crack feature intelligent extraction and fusion unit and the edge-cloud collaborative carbon footprint unit. The system connects to a monitoring platform to perform spatiotemporal correlation and cloud transmission of crack characteristic data, stress data, and historical carbon footprint data; a dual-objective maintenance cycle intelligent prediction unit connects to an edge-cloud data collaborative transmission and correlation mapping unit to deploy a carbon-efficiency dual-objective maintenance cycle prediction model, dynamically calculate and correct the correlated data, and output maintenance cycle prediction results; a time-series driven road performance degradation early warning unit connects to the dual-objective maintenance cycle intelligent prediction unit, performs time-series analysis through a time-series data-driven road performance degradation early warning algorithm, and generates performance degradation early warning information; and a maintenance plan intelligent generation and terminal push unit connects to both the time-series driven road performance degradation early warning unit and the edge-cloud collaborative carbon footprint monitoring platform, integrates early warning information, carbon emission parameters, and maintenance cycle parameters, generates maintenance execution plans, and pushes them to terminal devices.

[0091] The low-carbon asphalt road periodic maintenance method based on edge computing gateways has significant advantages in multi-source data fusion and defect identification, accurately overcoming the shortcomings of existing technologies. It leverages edge computing gateways to achieve comprehensive access to image, stress, environmental, and carbon emission related data, overcoming the limitations of traditional single data sources. Simultaneously, it uses multi-modal feature fusion crack identification technology to deeply correlate and map various data types, rather than analyzing isolated single-modal information. This approach not only significantly improves the accuracy of crack geometry and distribution feature identification but also establishes a correlation link between defect features and carbon emission data, providing comprehensive support for subsequent decision-making. It effectively solves the problems of insufficient multi-source data fusion, large defect identification bias, and weak data support in existing technologies.

[0092] In terms of maintenance cycle prediction and carbon efficiency synergy, this method achieves a breakthrough optimization, overcoming the core deficiencies of traditional technologies. It integrates historical and real-time carbon emission data through an edge-cloud collaborative carbon footprint monitoring platform, linking a carbon-efficiency dual-objective maintenance cycle prediction model to establish a coupling relationship between performance and carbon emissions, thus changing the decision-making mode of existing technologies that considers a single objective in isolation. Simultaneously, it dynamically corrects the prediction results based on parameters such as material consumption and environmental changes collected in real time at the edge, ensuring that the cycle plan not only conforms to the actual road conditions but also balances performance assurance and carbon emission control requirements, completely solving the problem of the lack of synergy between maintenance cycle prediction and carbon footprint management.

[0093] In terms of performance early warning response, this method demonstrates high-efficiency linkage advantages, further improving the maintenance decision-making system. It utilizes a time-series data-driven road performance degradation early warning algorithm, deeply linking historical and real-time performance data. Through trend decomposition and deviation analysis, it generates accurate early warning information, breaking the limitations of traditional maintenance that relies on experience-based judgment and lacks early warning. Simultaneously, the end-to-end seamless design from data collection, feature extraction, and periodic prediction to early warning push achieves efficient collaboration between real-time processing at the edge and deep computing in the cloud. This ensures that early warning information and maintenance plans can be quickly synchronized to the terminal, providing support for timely intervention and effectively overcoming the problems of delayed response and insufficient decision-making specificity in existing technologies.

[0094] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0095] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for periodic maintenance of low-carbon asphalt roads based on edge computing gateways, characterized in that, include: Step S1: Access multi-source sensing devices deployed at different monitoring sections of the road through an edge computing gateway to collect pavement image data streams, structural stress response data, environmental impact factor data, and carbon emission correlation data of maintenance materials for low-carbon asphalt roads. Perform preliminary data splitting and protocol conversion on the collected data through edge nodes. Step S2: Call the multimodal feature fusion crack recognition algorithm integrated at the edge to perform feature extraction and fusion processing on the image data stream, identify the geometric and distribution features of road cracks, and simultaneously obtain the historical maintenance carbon footprint data of the corresponding monitoring section through the edge-cloud collaborative carbon footprint monitoring platform; Step S3: Correlate and map the crack identification results, structural stress response data, environmental impact factor data, and historical maintenance carbon footprint data, and transmit them to the input layer of the carbon-efficiency dual-objective maintenance cycle prediction model deployed in the cloud; Step S4: The associated data is weighted and reorganized in terms of features using the carbon-efficiency dual-objective maintenance cycle prediction model. The cycle prediction results under different maintenance strategies are output and dynamically corrected by combining the maintenance material consumption rate data collected in real time at the edge. Step S5: Call the time-series data-driven road performance degradation early warning algorithm to perform time-series correlation analysis on the corrected cycle prediction results, explore the potential mapping relationship between road performance parameters and maintenance cycle, and generate phased performance degradation early warning information; Step S6: Synchronize the early warning information and maintenance cycle parameters to the terminal device through the edge cloud collaborative carbon footprint monitoring platform, and generate a periodic maintenance execution plan based on the carbon emission coefficient of maintenance materials, pavement damage rate and performance degradation rate parameters; The carbon-efficiency dual-objective maintenance cycle prediction model uses the following expression: . in, For optimal maintenance cycle, These are the weighting coefficients for carbon emission targets and performance targets, respectively. For carbon emission calculation functions, For performance benefit function, As a maintenance cycle variable, For maintenance process types, The carbon emission coefficient of maintenance materials, For road surface damage rate, These are the initial performance parameters. The comprehensive environmental impact index, This refers to the performance degradation rate. The maintenance cycle parameters are optimized using a collaborative optimization model. . in, To collaboratively optimize the target value, To optimize weights, This represents the total carbon emissions during the cycle. Let i be the amount of material used. Let be the carbon emission coefficient of the i-th material. Let t be the energy consumption at time t. Energy carbon emission coefficient, This represents the performance residual rate at the end of the cycle. For initial performance, Let t be the performance degradation rate at time t; Step S4 includes the following sub-steps: S41: Transmit the encapsulated input dataset to the feature processing layer of the cloud model, and dynamically weight the crack features, stress features, environmental features and carbon footprint features through a hierarchical attention mechanism to generate a multi-dimensional feature vector; S42: Calls the core calculation module of the carbon-efficiency dual-objective maintenance cycle prediction model, performs multi-scenario maintenance cycle simulation calculations based on feature vectors, and outputs carbon emission values ​​and performance retention rate parameters corresponding to maintenance intervals of 6 months, 12 months and 24 months. S43: Receive maintenance material inventory data and consumption rate data in real time at the edge, combine them with material replenishment cycle parameters to verify the feasibility of cycle prediction results for different scenarios, and mark prediction results that do not meet the material supply conditions. S44: The weighted correction algorithm is used to adjust the prediction results that have passed the verification. The weight of material consumption rate is integrated with the weight of carbon emission and performance to obtain the preliminary value of the corrected optimal maintenance cycle.

2. The method for periodic maintenance of low-carbon asphalt roads based on edge computing gateways according to claim 1, characterized in that, The multimodal feature fusion crack identification algorithm uses the following expression: . in, The fused crack feature vector These are the fusion weights for edge features, texture features, shape features, and contextual features, respectively. The coordinates of the pixels in the road surface image. For edge feature extraction operators, For texture feature extraction operators, For shape feature extraction operators, Context feature extraction operator.

3. The method for periodic maintenance of low-carbon asphalt roads based on edge computing gateways according to claim 1, characterized in that, The time-series data-driven road performance degradation early warning algorithm uses the following expression: . in, This represents the performance degradation warning value at time t. The time window length, The weight of the k-th time series point. The sampling time interval, for Road performance parameters at any given time This is a timing error correction term. For early warning correlation functions, To predict maintenance cycles, This represents the cumulative carbon footprint value.

4. The method for periodic maintenance of low-carbon asphalt roads based on edge computing gateways according to claim 1, characterized in that, The edge-cloud collaborative carbon footprint monitoring platform adopts a carbon flow tracing model: . in, For total carbon footprint, Carbon emissions from material production For transporting carbon emissions, For construction energy consumption and carbon emissions, To reduce carbon emissions, The amount of the i-th type of maintenance material is... The carbon emission coefficient for producing the i-th material. For transport volume, For unit carbon emissions transportation, For the j-th type of energy consumption, Let j be the carbon emission coefficient of the j-th energy source. To recover the amount of material, The unit is the carbon emission reduction factor for recycling.

5. The method for periodic maintenance of low-carbon asphalt roads based on edge computing gateways according to claim 1, characterized in that, Step S3 includes the following sub-steps: S31: The local data processing module of the edge computing gateway performs spatiotemporal alignment of the crack length, width, and density parameters obtained from crack identification with the deflection and shear stress parameters in the structural stress response data, and establishes a multi-dimensional data association matrix based on the monitoring section coordinates. S32: Retrieve carbon emission data from the historical database of the edge-cloud collaborative carbon footprint monitoring platform for the three most recent maintenance operations of the corresponding monitoring section, including detailed parameters of carbon emission from material production, transportation, and construction, and perform time-dimensional correlation matching with the currently collected data; S33: The monitoring data of different dimensions are transformed into a unified feature space through the data mapping algorithm. The crack feature parameters are encoded using geometric features, the stress data are processed by physical quantity normalization, and the carbon footprint data are encoded using time series to form the model input dataset. S34: Perform outlier detection on the input dataset, remove outlier data points that exceed the reasonable fluctuation range by comparing neighborhood data, and encapsulate the corrected dataset into a data format that meets the interface requirements of the carbon-efficiency dual-objective maintenance cycle prediction model.

6. The method for periodic maintenance of low-carbon asphalt roads based on edge computing gateways according to claim 1, characterized in that, Step S5 It includes the following steps: S51: Retrieve the time series data of road performance parameters for the corresponding monitoring section over the past five years from the time series database, including road surface smoothness, skid resistance, and structural strength parameters, and match them with the corrected preliminary value of the maintenance cycle on a time scale. S52: Call the time series feature extraction module of the time series data-driven road performance degradation early warning algorithm to perform trend decomposition on the time series data of performance parameters and separate the long-term degradation trend, seasonal fluctuations and random disturbance components. S53: Substitute the preliminary value of the maintenance cycle into the decay trend model to predict the theoretical value of the performance parameters at each time point within the cycle, perform deviation analysis with the real-time collected performance monitoring data, and calculate the deviation rate and fluctuation amplitude parameters. S54: Adjust the warning threshold based on the deviation analysis results. When the probability of the predicted performance parameter being lower than the preset threshold exceeds the set proportion, generate a phased performance degradation warning information including the decay rate and the expected time to reach the target.

Images