Deformation monitoring and control method and system for road surface and storage medium

Abstract

The invention relates to the technical field of road surface monitoring and control, and discloses a deformation monitoring and control method and system for a road surface and a storage medium. The method comprises the following steps: carrying out multi-level sensing data acquisition on a road surface; preprocessing the data to obtain a time series data set; inputting an LSTM network to predict a deformation risk; determining a dominant factor generation grading strategy; converting the instructions into execution instructions and responding according to priorities; and constructing a knowledge graph based on feedback, modeling as a Markov process, and training a deep Q network to realize adaptive regulation and control. According to the method, accurate identification of the pavement deformation state, advanced prediction of the deformation trend, accurate analysis of the deformation reason and adaptive optimization of the control strategy are realized, so that the road maintenance efficiency is improved, the service life of the road is prolonged, and the driving safety is guaranteed.

Description

Technical Field

[0001] The present application relates to the technical field of road surface monitoring and control, and in particular to a deformation monitoring and control method, system and storage medium for road surface. Background Art

[0002] Road pavement deformation monitoring and control is a key link in the maintenance and management of transportation infrastructure, and is of great significance for ensuring the service life of roads and driving safety. Traditional road pavement deformation monitoring mainly relies on manual inspections and fixed sensor networks. Manual inspections usually use equipment such as levels and displacement meters for regular measurements. The data collection frequency is low and it is difficult to capture the dynamic process of pavement deformation. Fixed sensor networks monitor through strain sensors, displacement sensors and other equipment buried under the road surface. Although the collection frequency is high, the coverage is limited and the installation cost is high. In terms of data processing, traditional methods mainly use statistical analysis and threshold judgment, lacking the ability to integrate and process multi-source heterogeneous data; in terms of deformation control, it mainly relies on post-maintenance and regular maintenance, lacking active prevention and precise intervention mechanisms. Domestic and foreign scholars have conducted a lot of research on road pavement deformation monitoring and control, such as the pavement management system (PMS) developed by the Federal Highway Administration of the United States and the Intelligent Road Monitoring Network (IRMN) of the European Union, which have initially realized remote monitoring and evaluation of pavement conditions.

[0003] However, the existing technologies still have many deficiencies in road pavement deformation monitoring and control: first, the data collection link lacks the coordinated monitoring of multi-level sensor networks, making it difficult to fully capture the spatiotemporal characteristics of road pavement deformation; second, the data processing link has limited fusion analysis capabilities for multi-source heterogeneous data, and cannot fully tap the value of data; third, the deformation risk prediction link lacks the support of advanced deep learning algorithms, and the prediction accuracy and advance amount are insufficient; in addition, the control strategy generation link lacks accurate identification of deformation causes and a hierarchical response mechanism, and the control measures are not targeted; finally, the system generally lacks self-learning and continuous optimization capabilities, and cannot continuously improve the control effect based on historical experience. These deficiencies make it difficult for the existing road pavement deformation monitoring and control system to cope with complex and changeable road conditions, and cannot achieve accurate prediction and active control of road pavement deformation, which in turn affects road safety and maintenance efficiency. Summary of the invention

[0004] The present application provides a deformation monitoring and control method, system and storage medium for road pavement, which are used to accurately identify the deformation state of the road surface, predict the deformation trend in advance, accurately analyze the cause of deformation and adaptively optimize the control strategy, so as to improve the efficiency of road maintenance, extend the service life of the road and ensure driving safety.

[0005] In a first aspect, the present application provides a method for monitoring and controlling deformation of a road pavement, the method comprising: performing multi-level sensor data collection on the road pavement to obtain an original deformation data set including fixed monitoring point data, mobile monitoring unit data and road surface images; preprocessing the original deformation data set to obtain a road surface state time series data set; inputting the road surface state time series data set into a long short-term memory network for risk prediction to obtain a road surface deformation risk level assessment result; according to the road surface deformation risk level assessment result, determining the deformation dominant factor and generating a hierarchical control strategy; converting the hierarchical control strategy into a standardized execution instruction set, judging the priority of each instruction in the execution instruction set, and when the instruction priority is 1, starting the emergency response mechanism and sending an emergency intervention notification within 15 minutes; when the instruction priority is 2 to 4, sending an execution notification 24 hours in advance, collecting execution feedback and calculating the execution deviation rate; based on the execution feedback and execution deviation rate, constructing a road health knowledge graph, modeling the road surface deformation control process as a Markov decision process, and training a deep Q network to achieve adaptive traffic flow control.

[0006] In a second aspect, the present application provides a deformation monitoring and control system for a road surface, the deformation monitoring and control system for a road surface comprising: The acquisition module is used to collect multi-level sensor data of the road surface and obtain the original deformation data set including fixed monitoring point data, mobile monitoring unit data and road surface images; A processing module, used for preprocessing the original deformation data set to obtain a road surface state time series data set; A prediction module, used for inputting the road surface state time series data set into a long short-term memory network for risk prediction to obtain a road surface deformation risk level assessment result; A grading module, used to determine the dominant factors of deformation and generate a grading control strategy according to the pavement deformation risk level assessment result; A judgment module is used to convert the hierarchical control strategy into a standardized execution instruction set, judge the priority of each instruction in the execution instruction set, and when the instruction priority is 1, start the emergency response mechanism and send an emergency intervention notice within 15 minutes; when the instruction priority is 2 to 4, send an execution notice 24 hours in advance, collect execution feedback and calculate the execution deviation rate; The control module is used to construct a road health knowledge graph based on the execution feedback and execution deviation rate, model the road deformation control process as a Markov decision process, and train a deep Q network to achieve adaptive traffic flow control.

[0007] In a third aspect, a deformation monitoring and control device for a road pavement is provided, comprising: a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor calls the instructions in the memory so that the deformation monitoring and control device for a road pavement executes the above-mentioned deformation monitoring and control method for a road pavement.

[0008] In a fourth aspect, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium, and when the computer-readable storage medium is run on a computer, the computer executes the above-mentioned method for deformation monitoring and control of a road surface.

[0009] In the technical solution provided by the present application, by collecting multi-level sensor data on the road surface, the original deformation data set including fixed monitoring point data, mobile monitoring unit data and road surface images is obtained, thereby realizing comprehensive monitoring of the road surface state, avoiding the limitations of the traditional single data source, and greatly improving the monitoring coverage and data integrity; the original deformation data set is preprocessed to obtain a road surface state time series data set, effectively eliminating data noise and discontinuity, and improving the accuracy of subsequent analysis; the road surface state time series data set is input into the long short-term memory network for risk prediction to obtain the road surface deformation risk level assessment result. As a deep learning algorithm, the long short-term memory network can effectively capture the long-term dependencies in the time series data and accurately predict the road surface deformation development trend. Compared with the traditional threshold judgment method, the warning lead time is significantly improved, providing a sufficient time window for preventive maintenance; according to the road surface deformation risk level assessment result, the deformation dominant factor is determined and a hierarchical control strategy is generated, realizing an in-depth analysis from symptoms to causes. The control measures are more targeted, avoiding waste of resources and ineffective intervention; the hierarchical control strategy is converted into a standardized execution instruction set, and the priority of each instruction in the execution instruction set is judged, realizing differentiated response based on the degree of urgency. Emergency-level instructions can trigger emergency response within 15 minutes, greatly improving the system's ability to respond to emergencies; based on execution feedback and execution deviation rate, a road health knowledge graph is constructed, the road deformation control process is modeled as a Markov decision process, and a deep Q network is trained to achieve adaptive traffic flow control. This link fully reflects the unique value of artificial intelligence algorithms in specific application fields. Knowledge graph technology converts discrete monitoring data and control experience into structured knowledge. As a representative algorithm in the field of reinforcement learning, the deep Q network can gradually learn the optimal control strategy through continuous trial and error and optimization, realizing the transition from "manual control" to "intelligent control". As the system is used for a longer time, the control effect continues to improve, which significantly reduces road maintenance costs, extends road service life, and improves driving safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0011] Figure 1 A schematic diagram of an embodiment of a method for monitoring and controlling deformation of a road surface in an embodiment of the present application; Figure 2 A schematic diagram of an embodiment of a deformation monitoring and control system for a road surface in an embodiment of the present application; Figure 3 It is a schematic block diagram of the structure of a deformation monitoring and control device for a road surface in an embodiment of the present invention. DETAILED DESCRIPTION

[0012] Embodiments of the present application provide a method, system and storage medium for monitoring and controlling deformation of road pavement. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0013] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiment of the present application, an embodiment of the method for monitoring and controlling deformation of a road surface includes: Step S101, performing multi-level sensor data collection on the road surface to obtain an original deformation data set including fixed monitoring point data, mobile monitoring unit data and road surface images; Step S102, preprocessing the original deformation data set to obtain a road surface state time series data set; Step S103, inputting the road surface state time series data set into the long short-term memory network for risk prediction to obtain the road surface deformation risk level assessment result; Step S104: Determine the dominant factors of deformation and generate a hierarchical control strategy based on the road deformation risk level assessment result; Step S105: convert the hierarchical control strategy into a standardized execution instruction set, determine the priority of each instruction in the execution instruction set, and when the instruction priority is 1, start the emergency response mechanism and send an emergency intervention notice within 15 minutes; when the instruction priority is 2 to 4, send an execution notice 24 hours in advance, collect execution feedback and calculate the execution deviation rate; Step S106: Based on the execution feedback and the execution deviation rate, a road health knowledge graph is constructed, the road deformation control process is modeled as a Markov decision process, and a deep Q network is trained to realize adaptive traffic flow control.

[0014] It is understandable that the execution subject of the present application may be a deformation monitoring and control system for a road surface, or a terminal or a server, which is not specifically limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0015] Specifically, multi-level sensor data collection is performed on the road surface to obtain the original deformation data set including fixed monitoring point data, mobile monitoring unit data and road surface images. This step forms fixed monitoring points by burying strain sensors, displacement sensors, pressure sensors and temperature and humidity sensors every 50 meters on the road, and collects data at a frequency of once an hour. At the same time, an acceleration sensor, a gyroscope and a GPS positioning module are installed on a specific vehicle to form a mobile monitoring unit, and the change value of the vehicle's vertical acceleration is recorded at a sampling rate of 100Hz. The road surface image acquisition device performs a full-scale scan of the road once a day to obtain high-definition road surface images. During the data collection process, when a single sensor data deviates from the historical average by more than 30%, the data point will be marked and a repeated collection instruction will be triggered to ensure the accuracy of the data. When severe weather conditions are detected, such as rainfall exceeding 50mm / day or temperature below minus 10℃, the data collection frequency will be automatically adjusted to four times per hour. The original deformation data set is preprocessed to obtain a road surface state time series data set. The wavelet transform method is applied to the fixed monitoring point data to remove high-frequency noise, and the Kalman filter algorithm is applied to the mobile monitoring unit data to eliminate the vehicle's own vibration interference. The missing values ​​in the data are supplemented by linear interpolation of time series neighboring points. When there are too many consecutive missing data, the mean of historical data in the same period is used to replace them. The processed data are standardized, and data from different sources and dimensions are uniformly converted to standard normal distribution. The standardized data are processed for spatiotemporal alignment, and the mobile monitoring unit data is mapped to the road mileage pile number according to the GPS coordinates, and located to the road grid unit. Finally, multi-source data fusion is performed through the evidence theory method, weights are assigned to data from different sources, and the feature layer fusion algorithm is used to integrate and form a road state time series data set.

[0016] The pavement status time series data set is input into the long short-term memory network for risk prediction, and the pavement deformation risk level assessment result is obtained. First, the pavement status time series data set is analyzed in the time domain to extract the basic characteristic parameters such as the mean and standard deviation of the pavement displacement. The frequency domain is converted by fast Fourier transform to extract the frequency domain features, and the wavelet packet decomposition method is applied to obtain the time-frequency features. After the main feature representation is extracted by principal component analysis, the dimension reduction feature set is input into the long short-term memory network. The network includes an input layer, two memory unit layers and an output layer. The first memory unit layer processes the input data through the forget gate, input gate and output gate to generate a hidden state vector; the second memory unit layer generates high-level time series features through cell state update and output control gate; finally, it is mapped to the future deformation prediction value through the fully connected output layer. Based on the prediction results, the pavement deformation risk is divided into safety level, attention level, warning level, danger level and emergency level. According to the pavement deformation risk level assessment results, the deformation dominant factors are determined and the hierarchical control strategy is generated. A multi-factor pavement deformation cause analysis model is constructed, which includes environmental factor sub-model, material factor sub-model and load factor sub-model to analyze the contribution of each factor to the deformation. Through Bayesian network integration analysis, the contribution percentage of the dominant deformation cause is output. According to the dominant deformation factors and risk levels, matching items are retrieved from the predefined control solution knowledge base to generate an initial control solution set. The life cycle cost analysis of the control solution is carried out to screen out the solution with the best cost-effectiveness ratio. Global optimization is performed through the integer programming algorithm to generate a hierarchical control strategy containing an implementation schedule, a resource requirement list, expected effects and emergency plans corresponding to different risk levels. The hierarchical control strategy is converted into a standardized execution instruction set, the priority of each instruction in the execution instruction set is determined, and corresponding measures are taken according to the priority. The standardized execution instruction set contains six elements: execution subject, execution time, execution location, execution content, execution standard and acceptance requirements. Instructions are assigned priorities according to risk levels. The priority of emergency-level instructions is 1, the priority of dangerous-level instructions is 2, and so on. When the instruction priority is 1, an emergency intervention notice is sent to the relevant responsible units and the emergency resource scheduling system is activated; when the instruction priority is 2 to 4, an execution notice is sent 24 hours in advance. Execution status data is collected through multiple channels, the execution deviation rate is calculated, and when the deviation rate exceeds the preset threshold, an abnormal alarm and adjustment suggestions are sent.

[0017] Based on execution feedback and execution deviation rate, a pavement health knowledge graph is constructed, the pavement deformation control process is modeled as a Markov decision process, and a deep Q network is trained to achieve adaptive traffic flow control. The pavement health knowledge graph contains five types of nodes: road section information, deformation characteristics, environmental factors, intervention measures, and effect evaluation. The nodes are connected by causal and temporal relationships. The state space of the Markov decision process is the pavement deformation feature vector, the action space is the set of optional control strategies, the transition probability is the change relationship between states, and the reward function is the comprehensive score of deformation control effect and resource consumption. The deep Q network contains three fully connected layers, with 128, 64, and 32 neurons in each layer. The input layer receives the pavement state feature vector, and the output layer generates the Q value of each control strategy. After training, the deep Q network can analyze the current pavement deformation state and predicted trend, calculate the long-term benefits of different traffic flow allocation schemes, obtain the optimal traffic control strategy, and adjust the travel time, speed limit value, and vehicle traffic ratio through traffic signal control and variable information signs to achieve adaptive traffic flow control on key deformed sections.

[0018] In the embodiment of the present application, by performing multi-level sensor data collection on the road surface, the original deformation data set including fixed monitoring point data, mobile monitoring unit data and road surface images is obtained, thereby realizing comprehensive monitoring of the road surface state, avoiding the limitations of the traditional single data source, and greatly improving the monitoring coverage and data integrity; the original deformation data set is preprocessed to obtain a road surface state time series data set, effectively eliminating data noise and discontinuity, and improving the accuracy of subsequent analysis; the road surface state time series data set is input into the long short-term memory network for risk prediction, and the road surface deformation risk level assessment result is obtained. As a deep learning algorithm, the long short-term memory network can effectively capture the long-term dependencies in the time series data and accurately predict the road surface deformation development trend. Compared with the traditional threshold judgment method, the warning lead time is significantly improved, providing a sufficient time window for preventive maintenance; according to the road surface deformation risk level assessment result, the deformation dominant factors are determined and a hierarchical control strategy is generated, realizing an in-depth analysis from symptoms to causes, and controlling The control measures are more targeted, avoiding waste of resources and ineffective intervention; the hierarchical control strategy is converted into a standardized execution instruction set, and the priority of each instruction in the execution instruction set is judged, realizing differentiated response based on the degree of urgency. Emergency-level instructions can trigger emergency response within 15 minutes, greatly improving the system's ability to respond to emergencies; based on execution feedback and execution deviation rate, a road health knowledge graph is constructed, the road deformation control process is modeled as a Markov decision process, and a deep Q network is trained to achieve adaptive traffic flow control. This link fully reflects the unique value of artificial intelligence algorithms in specific application fields. Knowledge graph technology converts discrete monitoring data and control experience into structured knowledge. As a representative algorithm in the field of reinforcement learning, the deep Q network can gradually learn the optimal control strategy through continuous trial and error and optimization, realizing the transformation from "manual control" to "intelligent control". As the system is used for a longer time, the control effect continues to improve, which significantly reduces the road maintenance cost, extends the road service life, and improves driving safety.

[0019] In a specific embodiment, the process of executing step S101 may specifically include the following steps: Strain sensors, displacement sensors, pressure sensors, and temperature and humidity sensors are buried every 50 meters on the road to form fixed monitoring points. Sensor data is collected once an hour to obtain fixed monitoring point data. An acceleration sensor, a gyroscope and a GPS positioning module are installed on a specific vehicle to form a mobile monitoring unit, and the acceleration change value of the vehicle in the vertical direction is recorded at a sampling rate of 100 Hz to obtain the mobile monitoring unit data; Use road surface image acquisition devices to conduct a full-scale scan of the road once a day to obtain high-definition images with a resolution of no less than 4K to form road surface images; Upload fixed monitoring point data, mobile monitoring unit data and road surface images to the central data processing server in real time through the 4G / 5G wireless transmission module; The deviation of single sensor data in the original deformation data set is detected. When the data deviates from the historical average value by more than 30%, the deviated data point is marked and a repeated collection instruction is triggered to ensure data validity. An automatic weather station is used to record environmental parameters in real time. When the rainfall exceeds 50 mm / day or the temperature is below minus 10°C, the data collection frequency is adjusted to four times per hour to obtain intensive monitoring data under harsh conditions.

[0020] Specifically, strain sensors, displacement sensors, pressure sensors and temperature and humidity sensors are buried every 50 meters on the road to form fixed monitoring points. The strain sensor measures the deformation of the pavement material under the action of external force, the displacement sensor directly measures the vertical displacement change of the pavement, the pressure sensor detects the pressure distribution of the pavement, and the temperature and humidity sensor records the changes in ambient temperature and humidity. These sensors collect data at a frequency of once per hour to form a time series. The data generated by each monitoring point contains a timestamp, a monitoring point ID, a measurement value and a quality mark. The data of the fixed monitoring point directly reflects the deformation state of the pavement under static and dynamic loads and is the core data source for deformation monitoring. The mobile monitoring unit is composed of an acceleration sensor, a gyroscope and a GPS positioning module installed on a specific vehicle. The acceleration sensor is installed on the vehicle chassis, perpendicular to the road surface direction, and records the vertical acceleration changes of the vehicle during driving at a sampling rate of 100Hz. This high-frequency sampling can capture subtle road surface unevenness, and the acceleration fluctuations generated when the vehicle passes through an uneven road surface directly reflect the road surface conditions. The gyroscope monitors the posture changes of the vehicle in three-dimensional space and supplements the acceleration data. The GPS positioning module records the precise location information of the vehicle in real time and associates the acceleration data with the geographic location. The advantage of the mobile monitoring unit data is that it has a wide coverage and can dynamically collect data from the entire road network, making up for the spatial limitations of fixed monitoring points.

[0021] The road image acquisition device is a high-resolution camera group installed on a special vehicle. It performs a full-scale scan of the road once a day to obtain high-definition images with a resolution of no less than 4K. During the image acquisition process, the camera works synchronously with the GPS to add accurate geographic location tags to each image. After processing, these high-definition images can identify surface deformation features such as cracks, potholes, and ruts to form road image data. The road image data provides an intuitive deformation representation, which complements the sensor data. The fixed monitoring point data, mobile monitoring unit data and road image are uploaded to the central data processing server in real time through the 4G / 5G wireless transmission module. Data compression and block transmission technology are used during the transmission process to ensure the efficient transmission of large amounts of data. For mobile monitoring units and image acquisition devices, the data is first stored in the local cache and then uploaded centrally when the vehicle enters the network coverage area to avoid data loss. After receiving the data, the server immediately stores and preliminarily verifies it in preparation for subsequent processing.

[0022] Deviation detection of single sensor data in the original deformation data set is an important part of ensuring data quality. Deviation detection evaluates data credibility by calculating the difference between the current data and the mean of historical data at the same time and location. The specific calculation method is to take the average value of the data at the same time in the last 30 days as the benchmark value, and calculate the percentage deviation between the current data and the benchmark value. When the data deviates from the historical average by more than 30%, the system will automatically mark the data point as an abnormal point and trigger a repeated collection instruction. Repeated collection will perform short-term intensive sampling of abnormal sensors at a speed three times the normal frequency to verify whether the data anomaly is caused by changes in road conditions or sensor failure.

[0023] Automatic weather stations are special meteorological monitoring equipment installed along roads, which record environmental parameters such as rainfall, temperature, humidity, and wind speed in real time. When extreme weather conditions are detected, such as rainfall exceeding 50 mm / day or temperature below minus 10°C, the system will automatically adjust the data collection frequency of all sensors from once per hour to four times per hour. This is because the risk of road deformation increases significantly under extreme weather conditions, and more intensive monitoring is required to capture changing trends. This mechanism of dynamically adjusting the sampling frequency ensures data density during critical periods while avoiding excessive redundant data under normal circumstances.

[0024] In a specific embodiment, the process of executing step S102 may specifically include the following steps: Applying wavelet transform method to fixed monitoring point data, high frequency noise is removed at a specific frequency to obtain filtered fixed monitoring point signals; Apply Kalman filter algorithm to the mobile monitoring unit data to eliminate the interference caused by the vehicle's own vibration and obtain a pure road response signal; The missing values ​​in the filtered fixed monitoring point signals and the pure road surface response signals are supplemented by the time series neighboring point linear interpolation method. When the continuous missing data exceeds the preset threshold, the mean value of the historical data in the same period is used to replace it to obtain a complete data sequence. The complete data sequence is processed by a standardization method, and the data from different sources are uniformly converted to a standard normal distribution to obtain a standardized data set; Perform spatiotemporal alignment on the standardized data set, establish a mapping relationship between the mobile monitoring unit data and the road mileage pile number according to the GPS coordinates, locate it to the road grid unit of a specific size, and obtain an aligned data set; The evidence theory method is applied to the aligned data set to perform multi-source data fusion. Different weights are assigned to the fixed monitoring point data, mobile monitoring unit data and road surface images, and the road surface status time series data set is formed through the feature layer fusion algorithm.

[0025] Specifically, the preprocessing of the original deformation data set is a key link in the road pavement deformation monitoring and control method. First, the wavelet transform method is applied to the fixed monitoring point data for processing. Wavelet transform is a time-frequency analysis tool that can remove high-frequency noise while retaining the main characteristics of the signal. In this method, the Daubechies wavelet function is used to decompose the fixed monitoring point data with a cutoff frequency of 40Hz. The specific operation is to decompose the original signal into approximate coefficients and detailed coefficients. The approximate coefficients represent the low-frequency part of the signal, and the detailed coefficients represent the high-frequency part. By setting the threshold, retaining the approximate coefficients and some detailed coefficients, filtering out the noise components above the cutoff frequency, and then reconstructing the signal to obtain the filtered fixed monitoring point signal. This processing process effectively removes high-frequency noise caused by non-road deformation such as environmental vibration and electrical interference, so that the sensor data more accurately reflects the actual deformation of the road surface.

[0026] The Kalman filter algorithm is applied to the mobile monitoring unit data to eliminate the interference caused by the vehicle's own vibration. Kalman filtering is a recursive state estimation algorithm that is particularly suitable for processing dynamic systems containing random noise. In the processing process, the state space model of the vehicle-road system is first established, and the state variables include road displacement, velocity and acceleration. Through the prediction step, the current state is estimated based on the state of the previous moment; and then through the update step, the predicted value is corrected in combination with the actual measurement value. The Kalman gain is dynamically adjusted according to the ratio of the prediction error and the measurement error. When the vehicle is driving on a smooth road, the model can learn and filter out the vibration characteristics of the vehicle itself, retaining only the response caused by the uneven road surface, thereby obtaining a pure road surface response signal.

[0027] The missing values ​​in the filtered fixed monitoring point signals and the pure road surface response signals are supplemented by the linear interpolation method of time series neighboring points. This method uses the valid data of the time points before and after the missing point to calculate the interpolation according to the time weight. The specific operation is that for a single missing point, one valid point before and after it is taken, and the weighted average is performed according to the inverse of the time distance; for multiple consecutive missing points, the valid points at both ends of the missing interval are used for linear interpolation. However, when the continuous missing data exceeds the preset threshold (set to 5 data points), simple linear interpolation may lead to large errors. At this time, the mean of the historical data in the same period is used instead. That is, the data of the same monitoring point and the same time period (such as the same hour of a day) are extracted from the historical database, and the mean is calculated as the filling value of the missing interval. This method makes full use of the periodic characteristics of road deformation and obtains a more reasonable and complete data sequence.

[0028] The purpose of processing the complete data sequence through the standardization method is to solve the problem of inconsistent dimensions of different types of data. The standardization process uses the Z-score method to convert various types of data into a standard normal distribution with a mean of 0 and a standard deviation of 1. The mean and standard deviation are calculated separately for each data type, and then converted according to the formula of (original value-mean) / standard deviation. For example, the displacement sensor data unit is millimeter, and the acceleration sensor data unit is m / s². After standardization, both data become dimensionless standard scores, which is convenient for subsequent comprehensive analysis. Standardization not only unifies the scale of the data, but also highlights outliers, making them easier to identify in subsequent analysis.

[0029] The standardized data set is processed for spatiotemporal alignment, focusing on solving the spatial mapping problem of fixed monitoring point data and mobile monitoring unit data. First, the road is divided into 1m×1m grid units as the basic spatial reference unit. The fixed monitoring point data is directly mapped to the grid unit where it is located; while the mobile monitoring unit data needs to establish a mapping relationship with the road mileage pile number through GPS coordinates. In the specific process, the longitude and latitude coordinates recorded by the vehicle GPS are combined with the road GIS database to locate each measurement point to the nearest road grid unit. When multiple mobile measurement points are mapped to the same grid unit, the average value is taken as the representative value of the unit. In this way, both the fixed point data and the mobile data are aligned with a unified spatial reference system, forming an aligned data set that is continuous in time and consistent in space.

[0030] Applying evidence theory methods to fusion of multi-source data on aligned data sets is a key step in integrating information from different sources. Dempster-Shafer evidence theory is suitable for processing uncertain and incomplete data. It combines evidence from different sensors by defining basic probability distribution functions, confidence functions, and likelihood functions. In this method, according to the reliability and accuracy of each data source, basic weights of 0.6, 0.3, and 0.1 are assigned to fixed monitoring point data, mobile monitoring unit data, and road surface images, respectively. Then, the three types of data are integrated in the feature space through the feature layer fusion algorithm to generate a feature vector containing 20 dimensions, covering key parameters such as road surface vertical displacement, displacement change rate, and vibration frequency characteristics. Finally, the fused feature vectors are reorganized in time series and processed using the sliding window method with a window size of 24 hours and a sliding step of 1 hour to form a continuous road surface state time series data set.

[0031] Taking the road surface monitoring of a certain highway as an example, the original data of the displacement sensor at a fixed monitoring point contains about 15% high-frequency noise components. After wavelet transformation, the smoothness of the signal is significantly improved, which can more clearly reflect the long-term settlement trend of the road surface. During the driving process of the mobile monitoring vehicle, about 60% of the vertical vibration data recorded by its acceleration sensor comes from the vibration of the vehicle's own suspension system. After Kalman filtering, this part of the interference is effectively eliminated, highlighting the characteristic vibration caused by potholes on the road. During the data collection process, due to communication interruption, four consecutive points of data were missing for a period of time. Through the linear interpolation method of time-series neighboring points, according to the value before the missing value of 2.3mm and the value after the missing value of 2.5mm, the filling values ​​of the missing points were calculated to be 2.35mm, 2.4mm, 2.45mm and 2.475mm, maintaining the continuity and trend of the data. For data from different sources, such as the displacement data range of fixed points in the range of 0-10mm and the acceleration data range of mobile units in the range of 0-3g, after standardization, all data are mapped to a similar distribution range, which is convenient for comprehensive comparison. During the data fusion stage, at the same location on a certain road section, the fixed monitoring point showed a displacement of 3.2 mm (1.5 after standardization), the mobile monitoring showed abnormal vibration acceleration (1.8 after standardization), and the road surface image showed slight cracks (1.0 after standardization). After fusion using the evidence theory method, the comprehensive score was 1.5, indicating that there was a moderate deformation risk at this location and that close monitoring was required.

[0032] In a specific embodiment, the process of executing step S103 may specifically include the following steps: Perform time domain analysis on the road surface state time series data set, extract basic characteristic parameters including road surface displacement mean, standard deviation, peak value, valley value, peak-to-valley difference and its change rate, and obtain the time domain feature set; The time domain feature set is converted into the frequency domain through fast Fourier transform, and the main frequency component, power spectrum density and frequency bandwidth are extracted to obtain the frequency domain feature set; Apply wavelet packet decomposition method to the frequency domain feature set, set the decomposition layer number to five, obtain the energy distribution of each frequency band and its proportional relationship, and obtain the time-frequency feature set; The main features of the time-frequency feature set are extracted by principal component analysis, and the first five principal components are selected as the morphological feature description of the road surface deformation to obtain the reduced dimension feature set. The reduced dimension feature set is input into the input layer of the long short-term memory network in time series order, and the hidden state vector is obtained after being processed by the first memory unit layer including the forget gate, input gate and output gate. The high-level time series features are then generated through the cell state update and output control gate of the second memory unit layer, and finally mapped into the deformation prediction value of the next 24 hours through the fully connected output layer to obtain the deformation prediction result; The pavement deformation risk is graded based on the deformation prediction results. According to the comprehensive scores of the three dimensions of displacement, deformation rate and prediction trend, the risk levels are divided into safety level, attention level, warning level, danger level and emergency level, forming the pavement deformation risk level assessment results.

[0033] Specifically, the time domain analysis of the pavement state time series data set is performed to extract basic characteristic parameters. The time domain analysis directly calculates statistical features from the time series data, including the mean, standard deviation, peak, valley, peak-to-valley difference and its rate of change of the pavement displacement. The mean pavement displacement reflects the overall deformation level of the pavement, which is obtained by the arithmetic average of the continuous 24-hour data; the standard deviation measures the degree of deformation fluctuation and characterizes the stability of the pavement; the peak and valley are the maximum and minimum displacement values ​​in the time window, respectively, reflecting the extreme deformation; the peak-to-valley difference is the difference between the peak and valley values, indicating the deformation amplitude; the rate of change is the difference between the means of adjacent time windows divided by the time interval, reflecting the speed of deformation development. These basic characteristic parameters together constitute the time domain feature set, which intuitively describes the static and dynamic characteristics of pavement deformation. The frequency domain conversion of the time domain feature set through fast Fourier transform is to reveal the periodic characteristics implicit in the deformation data. Fast Fourier transform decomposes the time domain signal into the superposition of sine waves of different frequencies, which can identify deformation components of different frequencies. During the processing, the fast Fourier transform is applied to the 24-hour time series data of each monitoring point to obtain a spectrum, and then three types of key frequency domain features are extracted from it: the main frequency component, that is, the frequency point with the most concentrated energy, which is usually related to the main deformation factor; the power spectrum density, which represents the energy distribution of each frequency component and reflects the intensity of the deformation; the frequency bandwidth, which represents the range of effective frequency components and reflects the complexity of the deformation. These frequency domain features constitute the frequency domain feature set, which can distinguish different types of deformation modes, such as high-frequency vibration caused by traffic load and low-frequency expansion and contraction caused by temperature changes.

[0034] The wavelet packet decomposition method is applied to the frequency domain feature set in order to obtain more refined time-frequency features. Wavelet packet decomposition can analyze signals at different scales while taking into account both time domain and frequency domain information. Setting the number of decomposition layers to five means dividing the spectrum into 32 equal-width bands. The energy is calculated for each band, that is, the sum of the squares of all coefficients in the band, and then the proportion of the energy of each band to the total energy is calculated. This decomposition method can capture deformation characteristics of different scales and frequency bands, and is particularly suitable for analyzing non-stationary signals, such as the response characteristics of the road surface under different traffic flow and temperature conditions. The energy distribution and proportional relationship of each band obtained by wavelet packet decomposition constitute the time-frequency feature set, which provides a richer description of deformation characteristics than simple time domain or frequency domain analysis.

[0035] The principal component analysis method is used to extract the main features from the time-frequency feature set. It is a dimensionality reduction technique that aims to reduce the number of features while retaining the most critical information. Principal component analysis converts the original features into a set of linearly independent new features (principal components) through orthogonal transformation. These principal components are sorted according to the size of the variance. The principal components with larger variance carry more information. In this method, principal component analysis is performed on a high-dimensional data set containing multiple time-frequency features, and the eigenvalues ​​and eigenvectors of the feature covariance matrix are calculated. Then, the eigenvectors corresponding to the five largest eigenvalues ​​are selected as principal components. These five principal components cumulatively explain more than 85% of the variance of the original data. In this way, the high-dimensional time-frequency feature set is compressed into a five-dimensional reduced dimensionality feature set. Each principal component is a linear combination of the original features and represents a key aspect of pavement deformation.

[0036] Inputting the reduced-dimensionality feature set into the long-short-term memory network for risk prediction is to use deep learning technology to capture the temporal law of deformation. The long-short-term memory network is a special recurrent neural network that is good at processing sequence data with long-term dependencies. The core of the network is the memory unit, which includes three control mechanisms: the forget gate, the input gate, and the output gate. The reduced-dimensionality feature set is input into the input layer of the network in time series order, and then processed by the first memory unit layer: the forget gate determines which historical information needs to be discarded, the input gate determines which new information needs to be added, and the output gate controls the information output by the current state. These three gates together generate a hidden state vector. The vector is further processed by the second memory unit layer. The cell state update mechanism integrates long-term memory and short-term input, and the output control gate filters useful information to generate high-level temporal features. Finally, these features are converted into specific prediction values ​​through the fully connected output layer, that is, the road deformation trend in the next 24 hours. The entire network is trained by the back-propagation algorithm to minimize the error between the predicted value and the actual observed value.

[0037] Grading the pavement deformation risk based on the deformation prediction results is a key step in transforming quantitative analysis into a basis for decision-making. Risk classification considers three dimensions: displacement, which directly indicates the current degree of deformation; deformation rate, which indicates the speed of deformation development; and predicted trend, which indicates the direction of future deformation development. The three dimensions are scored separately, and then the weighted sum is used to obtain a comprehensive score. According to the comprehensive score, the risk level is divided into five levels: safety level (green), indicating that the deformation is within the normal range and no intervention is required; concern level (blue), indicating that the deformation is slightly abnormal and the monitoring frequency needs to be increased; warning level (yellow), indicating that the deformation is obvious and preventive maintenance needs to be arranged; danger level (orange), indicating that the deformation is serious and needs to be repaired as soon as possible; emergency level (red), indicating that the deformation is extreme and emergency measures need to be taken immediately. This multi-level risk assessment method can intuitively reflect the pavement condition and provide clear guidance for the formulation of subsequent control strategies.

[0038] Taking the monitoring of a national road section as an example, the data processing process of a bridgehead jumping section is as follows: First, the time domain features are extracted from the displacement sensor data for 72 consecutive hours, and the displacement mean is calculated to be 2.8mm, the standard deviation is 0.5mm, the peak is 3.9mm (occurring during the afternoon traffic rush hour), the valley is 1.8mm (occurring during the low traffic period in the early morning), the peak-to-valley difference is 2.1mm, and the displacement change rate is 0.2mm / day, indicating that the deformation is slowly increasing. These time domain data are converted to the frequency domain through fast Fourier transform, and it is found that the main frequency component is located at 0.5Hz, which coincides with the frequency of heavy vehicles passing. The power spectrum density reaches the maximum value at this frequency point, and the frequency bandwidth is concentrated in the range of 0.3-0.8Hz, indicating that the deformation is mainly caused by vehicle load. Through the five-layer decomposition of wavelet packets, it is found that the energy is mainly concentrated in the low-frequency and medium-frequency bands. The energy of the low-frequency band (0-0.2Hz) accounts for 35%, reflecting the characteristics of foundation settlement; the energy of the medium-frequency band (0.2-0.6Hz) accounts for 45%, reflecting the impact of traffic load; the energy of the high-frequency band is relatively low, accounting for only 20%. The principal component analysis of these time-frequency characteristics shows that the first five principal components reflect the five aspects of load response, temperature effect, material fatigue, foundation settlement and seasonal changes. The time series data formed by these five principal components are input into the long-term and short-term memory network. After the network is trained with historical data, it is predicted that the displacement will increase to 3.5mm and the deformation rate will accelerate to 0.3mm / day in the next 24 hours. Based on the three-dimensional scores of current displacement, deformation rate and predicted trend, the risk level of this section is rated as dangerous (orange), which triggers the subsequent control strategy generation process and arranges the pavement structural repair plan.

[0039] In a specific embodiment, the process of executing step S104 may specifically include the following steps: A multi-factor pavement deformation cause analysis model is constructed, which includes an environmental factor sub-model, a material factor sub-model, and a load factor sub-model. The relationship between environmental parameters and deformation is analyzed through multiple linear regression to obtain the contribution of environmental factors. The contribution of environmental factors, material aging degree and traffic load impact are evaluated by fuzzy reasoning system, and the functional relationship between the cumulative equivalent axle number and the deformation development rate is calculated to obtain the intensity of each factor. The intensity of each factor is input into the Bayesian network for integrated analysis, and the contribution percentage of the dominant deformation cause is output to obtain the dominant deformation factor; According to the deformation dominant factors and the pavement deformation risk level assessment results, matching items are retrieved from the knowledge base containing predefined control schemes, corresponding control types are generated for sections with different risk levels, and an initial control scheme set is obtained; Conduct life cycle cost analysis on the initial control scheme set, calculate the long-term cost-effectiveness ratio of each scheme, screen out the scheme with the best cost-effectiveness ratio, and obtain the candidate control strategy; The candidate control strategies are globally optimized through an integer programming algorithm, and the resource allocation of the entire road network is balanced while meeting safety requirements. A hierarchical control strategy is generated that includes an implementation schedule, resource requirement list, expected effects, and emergency plans corresponding to different risk levels.

[0040] Specifically, a multi-factor pavement deformation cause analysis model is constructed, which contains three sub-models: environmental factor sub-model, material factor sub-model and load factor sub-model. The environmental factor sub-model considers parameters such as temperature cycle changes, precipitation, and freeze-thaw cycle times, and establishes a quantitative relationship between environmental parameters and deformation through the multivariate linear regression method. Multivariate linear regression analysis estimates the regression coefficients through the least squares method, establishes a linear relationship between multiple environmental variables and deformation, and calculates the degree of influence of each environmental factor on deformation. For example, the deformation increment corresponding to a temperature increase of 10 degrees, the deformation increment corresponding to a precipitation increase of 10 mm, etc. These regression coefficients directly reflect the influence intensity of each environmental factor, thereby obtaining the contribution of environmental factors. The fuzzy reasoning system is used to evaluate the contribution of environmental factors, material aging degree and traffic load impact in order to deal with the uncertainty and ambiguity in these factors. The fuzzy reasoning system first converts the precise values ​​of each factor into fuzzy sets, such as dividing the material aging degree into three fuzzy levels of "mild", "medium" and "severe". Then reasoning is performed through a predefined fuzzy rule base, which contains rules in the form of "if...then...", such as "if the temperature changes greatly and the material ages severely, then the combined impact of the environment and the material is strong." The impact of traffic load is quantified by calculating the functional relationship between the cumulative equivalent axle times and the deformation development rate. The cumulative equivalent axle times is the cumulative number of times different types of vehicles are converted to standard axle loads, calculated based on vehicle type, axle weight and traffic frequency. The cumulative equivalent axle times is functionally related to the observed deformation development rate to obtain the sensitivity coefficient of load intensity and deformation. The fuzzy reasoning system ultimately outputs the intensity of each factor, expressed as a clear quantitative value.

[0041] Inputting the intensity of each factor into the Bayesian network for integrated analysis is a probabilistic graphical model method that can express the conditional dependency relationship between variables. The Bayesian network consists of nodes and directed edges. The nodes represent random variables (various influencing factors) and the directed edges represent causal relationships. The network contains two parts of information: the network structure (graph) and the conditional probability table. The network structure is predefined based on domain knowledge, such as environmental factors affecting material properties, material properties and loads jointly affecting deformation, etc. The conditional probability table defines the probability distribution of child nodes under a given parent node state and is obtained through historical data training. After inputting the observed intensity of each factor, the Bayesian network calculates the posterior probability of each factor on deformation through probabilistic reasoning, and finally outputs the contribution percentage of the dominant deformation cause, clearly indicating which factor is the dominant cause of deformation. According to the deformation dominant factor and the pavement deformation risk level assessment results, matching items are retrieved from the predefined control scheme knowledge base to generate corresponding control types for sections with different risk levels. The control scheme knowledge base contains 250 predefined schemes, covering combinations of different deformation causes, different deformation degrees, and different urgency levels. The retrieval process uses a multi-condition matching algorithm to calculate the similarity between the deformation dominant factors and risk levels of the current road section and the cases in the knowledge base, and select several solutions with the highest similarity. For example, for minor cracks caused by temperature stress (attention level), the retrieval results include sealing treatment solutions; for moderate deformation caused by foundation settlement (warning level), the retrieval results include base reinforcement solutions; for severe cracking caused by material aging (danger level), the retrieval results include milling and resurfacing solutions. The retrieved solutions constitute the initial control solution set, providing candidates for subsequent optimization.

[0042] A life cycle cost analysis is performed on the initial set of control schemes to calculate the long-term cost-benefit ratio of each scheme. The life cycle cost analysis considers the full life cycle cost of the scheme, including the initial construction cost, maintenance cost, user delay cost and risk cost. The initial construction cost is the direct cost of the implementation plan; the maintenance cost is the cumulative cost of subsequent regular maintenance; the user delay cost is the economic loss caused by traffic delays due to construction; and the risk cost is the expected value of the loss that may be caused by the failure of the scheme. Future costs are converted into present values ​​through the discount rate to calculate the total life cycle cost. At the same time, the benefits of the scheme are evaluated, including extending the service life and improving the service quality, and quantified as economic benefits. The cost-benefit ratio is obtained by dividing the benefit by the cost, and the scheme with the best cost-benefit ratio is selected as the candidate control strategy, taking into account both technical feasibility and economic rationality.

[0043] The purpose of globally optimizing the candidate control strategies through the integer programming algorithm is to reasonably allocate limited maintenance resources while meeting the safety requirements of the entire road network. Integer programming is a mathematical optimization method in which decision variables can only take integer values, which is suitable for dealing with resource allocation problems. In this scheme, an integer programming model is established, and the objective function is to minimize the total risk or total cost. The constraints include budget constraints, human resource constraints, equipment resource constraints, time constraints, etc. By solving the model, the optimal control strategy combination for each section of the entire road network is obtained, including implementation sequence, resource allocation, etc. The final generated hierarchical control strategy contains four parts: implementation schedule (when to implement what measures), resource requirement list (required manpower, materials, equipment), expected results (performance indicators after repair) and emergency plan (alternative plan when the main plan fails).

[0044] In a specific embodiment, the process of executing step S105 may specifically include the following steps: The hierarchical control strategy is converted into a standardized execution instruction set containing six elements: execution subject, execution time, execution location, execution content, execution standard and acceptance requirements. The execution subject is divided into two categories: automatic execution system and manual intervention. The priority of the standardized execution instruction set is assigned according to the risk level, and the priority of the emergency-level instruction is set to 1, the priority of the danger-level instruction is set to 2, the priority of the warning-level instruction is set to 3, and the priority of the attention-level instruction is set to 4, and the priority ranking result is obtained; The instruction type is determined based on the priority sorting results. When the instruction priority is 1, an emergency intervention notice containing the intervention location, intervention content, and intervention criteria is sent to the relevant responsible units, and the emergency resource dispatch system is activated to obtain an emergency response record; Determine the instruction type based on the priority sorting results. When the instruction priority is 2 to 4, send an execution notice containing the execution plan, resource requirements and acceptance criteria to the relevant responsible units according to the scheduled time plan to obtain a regular execution record. Collect execution status data of emergency response records and routine execution records through multiple channels, including automatic execution status data and manual intervention progress data, compare and analyze the execution progress with the planned progress, and obtain execution feedback; Based on the execution feedback, the ratio of the actual execution time to the planned execution time is calculated to obtain the execution deviation rate. When the execution deviation rate exceeds the preset threshold, abnormal alarms and adjustment suggestions are sent to managers.

[0045] Specifically, converting the hierarchical control strategy into a standardized execution instruction set is the execution link in the road pavement deformation monitoring and control method. The standardized execution instruction set contains six core elements: execution subject, execution time, execution location, execution content, execution standard and acceptance requirements. The execution subject is divided into two categories: automatic execution system and manual intervention. The automatic execution system includes non-physical intervention equipment such as traffic flow control equipment, information release platform, and environmental parameter monitoring device; manual intervention includes maintenance construction teams, professional inspection personnel, road administration personnel, etc., who are responsible for specific physical maintenance work. The execution time defines the start time and completion deadline of the instruction, accurate to the hour level. The execution location is accurately located by highway number, mileage pile number and GPS coordinates, and is marked with an electronic map. The execution content describes in detail the specific measures to be taken, such as "milling 3cm asphalt top layer and re-paving modified asphalt mixture". The execution standard stipulates the operating specifications and quality requirements, such as "milling flatness deviation does not exceed 3mm / 3m, and the compaction degree of the new pavement layer is not less than 98%". The acceptance requirements specify the inspection methods and qualification standards after completion, such as "passing the deflection meter test, the deflection value does not exceed 0.4mm". The conversion process maps the maintenance plan elements in the hierarchical control strategy to the standardized instruction format through a structured template to form a complete set of execution instructions. The priority assignment adopts a direct mapping method to convert the five-level risk level into a four-level priority. The priority of the emergency level instruction (corresponding to the red risk) is set to 1, indicating the highest priority and needs to be processed immediately; the priority of the danger level instruction (corresponding to the orange risk) is set to 2, indicating a high priority and needs to be processed as soon as possible; the priority of the warning level instruction (corresponding to the yellow risk) is set to 3, indicating a medium priority and is processed as planned; the priority of the attention level instruction (corresponding to the blue risk) is set to 4, indicating a low priority and can be processed later. No execution instructions are generated for the safety level section. The priority assignment process also takes into account the importance coefficient of the section, and appropriately increases the priority of instructions at key locations such as important arterial roads and transportation hubs. After the assignment is completed, the instructions are sorted from high to low according to priority, and the priority sorting result is obtained, which provides a basis for subsequent differentiated processing.

[0046] The type of instruction is determined based on the priority sorting results, and an emergency response mechanism is adopted for instructions with a priority of 1. When an instruction with a priority of 1 is identified, the emergency intervention process is immediately initiated and an emergency intervention notification is sent to the relevant responsible units. Emergency intervention notifications are sent simultaneously through multiple methods such as SMS, phone calls, and work platform push to ensure that the information is delivered quickly. The notification content contains three core elements: intervention location (accurate description of the location of the problem section), intervention content (detailed description of the emergency measures to be taken), and intervention standards (specifying the quality requirements for emergency disposal). At the same time, the emergency resource scheduling process is activated to mobilize personnel, equipment and materials from the nearest maintenance base to give priority to the resource requirements of emergency tasks. The entire notification process is completed within 15 minutes after the emergency problem is detected to ensure a rapid response. The entire emergency response process is recorded, including the time of notification sending, receipt confirmation time, resource scheduling status, etc., to form an emergency response record for subsequent tracking and evaluation.

[0047] For instructions with priorities 2 to 4, the conventional execution process is used for processing. According to the priority sorting results, an execution notice is sent to the relevant responsible units 24 hours in advance according to the pre-arranged time plan. The content of the execution notice is more comprehensive, including the execution plan (detailed work schedule and step arrangement), resource requirements (specific quantity and specifications of required personnel, equipment, and materials) and acceptance criteria (indicators and methods for completion acceptance). Conventional execution notices are sent through work platforms, emails, etc. to ensure complete information transmission. After receiving the notice, the responsible unit prepares resources and arranges work as required and starts execution at the planned time. Each notification and the corresponding execution process form an execution record, which contains key information such as notification time, planned execution time, and actual start time, forming a conventional execution record and providing a data basis for execution monitoring. Collecting execution status data through multiple channels is an important link to ensure the effective implementation of control measures. Data collection channels are divided into two categories: automatic execution status data is automatically uploaded through IoT devices, including the working status of traffic control equipment, information update status of electronic display screens, etc.; manual intervention progress data is uploaded in real time by on-site personnel through mobile terminal applications, including text descriptions, on-site photos, key parameter measurements, etc. On-site personnel need to update progress information at each key node (such as start, half completion, completion, etc.), and increase the reporting frequency in special circumstances. All collected data is summarized in the execution monitoring platform and automatically compared and analyzed with the planned progress. The comparison process calculates the completion time difference and quality indicator deviation of each node, forming structured execution feedback data to provide a basis for subsequent evaluation and adjustment.

[0048] Calculating the execution deviation rate based on execution feedback is the core step to quantify the execution effect. The execution deviation rate is mainly calculated by the ratio of the actual execution time to the planned execution time, indicating the degree of delay or advance of the work progress. The specific calculation method is that for each execution node, the actual completion time is subtracted from the planned completion time to obtain the time difference, which is then divided by the total planned execution time and multiplied by 100% to obtain the execution deviation rate of the node. The deviation rates of all nodes are combined and the weighted average is used to obtain the overall execution deviation rate. When the execution deviation rate exceeds the preset threshold (usually set to 20%), the abnormal alarm mechanism is triggered, and an alarm message containing abnormal cause analysis and adjustment suggestions is sent to the pavement management personnel. The adjustment suggestions include resource supplementation plans, construction period adjustment plans, technical method optimization plans, etc., to help managers make decisions quickly. Abnormal alarms are sent through instant messaging methods such as text messages and phone calls to ensure that problems are handled in a timely manner.

[0049] In a specific embodiment, the process of executing step S106 may specifically include the following steps: Based on the execution feedback and execution deviation rate, a pavement health knowledge graph is constructed, which contains five types of nodes: road section information, deformation characteristics, environmental factors, intervention measures, and effect evaluation. The nodes are connected by edges of causal and temporal relationships to obtain a deformation control knowledge base. The road deformation control process is modeled as a Markov decision process, where the state space is the road deformation feature vector, the action space is the set of optional control strategies, the transition probability is the relationship between state changes, and the reward function is the comprehensive score of deformation control effect and resource consumption, thus obtaining a decision process model. A deep Q network consisting of three fully connected layers is constructed, with the number of neurons in each layer being 128, 64, and 32 respectively. The input layer receives the road state feature vector, and the output layer generates the Q value of each control strategy, thus obtaining the reinforcement learning model structure. The reinforcement learning model structure is trained using the experience replay mechanism. The replay buffer size is set to 10,000 records. The target network parameters are updated every 500 interactions. The Q value estimation is optimized using the gradient descent algorithm to obtain a trained deep Q network. Based on the trained deep Q network, the traffic flow is dynamically analyzed. According to the current road deformation state and predicted trend, the long-term benefits of different traffic flow allocation schemes are calculated to obtain the optimal traffic control strategy. The optimal traffic control strategy is sent to the traffic signal control system and variable information signs to dynamically adjust the travel time, speed limit and vehicle traffic ratio to achieve adaptive control of traffic flow on key deformed sections.

[0050] Specifically, the pavement health knowledge graph consists of five core nodes: road section information nodes, deformation feature nodes, environmental factor nodes, intervention measure nodes, and effect evaluation nodes. The road section information node records the basic attributes of the road, including road grade, paving materials, service life, and historical maintenance records; the deformation feature node stores technical indicators such as the displacement, deformation rate, and vibration frequency characteristics of the pavement; the environmental factor node contains environmental data such as temperature cycle changes, precipitation, and freeze-thaw cycle times; the intervention measure node records the pavement maintenance operations that have been performed, such as crack sealing, pavement milling and resurfacing, and other specific technical means; the effect evaluation node saves the status comparison data before and after the intervention, such as improvement rate and stability index. In the knowledge graph, nodes are connected by two types of edges: causal relationship edges and temporal relationship edges. The causal relationship edge describes the influence mechanism between different nodes, such as the "cause" relationship edge between the environmental factor node of precipitation exceeding 50mm / day and the deformation feature node of the road displacement increasing by 0.8mm / day; the temporal relationship edge records the order of events, such as the "predecessor" relationship edge between the intervention measure node of road milling and resurfacing and the effect evaluation node of reducing displacement by 85%. Through this structured representation, the knowledge graph transforms discrete monitoring data and control experience into a structured deformation control knowledge base.

[0051] Modeling the pavement deformation control process as a Markov decision process is a key step in achieving adaptive control. In this model, the state space is composed of pavement deformation feature vectors, including multi-dimensional parameters such as displacement, deformation rate, and vibration characteristics; the action space corresponds to a set of optional control strategies, covering all technical means from light intervention (such as crack sealing) to heavy intervention (such as pavement reconstruction); the transition probability represents the change law of the pavement state after adopting a specific control strategy; the reward function is based on the trade-off between deformation control effect and resource consumption, and comprehensively evaluates the effectiveness of the control strategy through the weighted summation method, with the weight ratio of effect accounting for 70% and resource consumption accounting for 30%.

[0052] The deep Q network is the core algorithm in reinforcement learning and is used to learn the optimal control strategy. The network structure consists of three fully connected layers. The first layer has 128 neurons, which receive the road state feature vector as input; the second layer has 64 neurons, which are responsible for feature extraction and conversion; the third layer has 32 neurons, which generate the Q value of each control strategy. Each neuron uses the ReLU activation function to improve the nonlinear expression ability of the network. The road state feature vector received by the input layer has a dimension of 20, including key indicators such as displacement, deformation rate, and vibration frequency; the output layer generates the Q value of each control strategy, which represents the long-term expected benefit of selecting a specific control strategy under the current state.

[0053] The experience replay mechanism is the core technology for training deep Q networks. It improves learning efficiency by storing and reusing historical interaction data. The replay buffer size is set to 10,000 records, each of which contains four elements: state, action, reward, and next state. During training, the system updates the target network parameters every 500 interactions to ensure the stability of the learning process. The Q value estimation is optimized by the gradient descent algorithm, with the learning rate set to 0.001, the batch size to 32, and the training rounds to 10,000. Finally, a deep Q network that can select the optimal control strategy under various road conditions is obtained.

[0054] Based on the trained deep Q network, the system can dynamically analyze and control traffic flow. When the risk of road deformation is detected, the system will calculate the long-term benefits of each plan based on the current road deformation state and predicted trend, while considering the impact of different traffic flow distribution plans on the road surface. For plans with a high proportion of heavy-loaded vehicles and large damage to the road surface, the system will give a lower Q value; for plans that can reduce the road load and delay the development of deformation, the system will give a higher Q value. By comparing the Q values ​​of different plans, the system selects the optimal traffic control strategy.

[0055] Sending the optimal traffic control strategy to the traffic control equipment is the final step in achieving adaptive control. The system issues control instructions through the interface with the traffic signal control system and variable information signs, and dynamically adjusts the travel time, speed limit value and vehicle traffic ratio. For sections with higher deformation risk, the system will reduce the frequency of heavy-loaded vehicles and reduce the road load; for sections where deformation has reached the warning level, the system will implement speed limit measures to limit the speed to a range with less impact on the road surface; for sections close to the danger level, the system will guide part of the traffic flow to detour and reduce the road load.

[0056] In actual applications, the road displacement sensor of a certain main road K1 section showed a deformation rate of 0.7mm / day after continuous rainfall. The system retrieved historical cases under similar road conditions through the knowledge graph and found that the deformation was highly correlated with the frequency of heavy-loaded vehicles. After deep Q network analysis, it was calculated that reducing the traffic volume of heavy-loaded vehicles by 30% can control the deformation rate below 0.3mm / day, while minimizing the impact on overall traffic efficiency. Based on this analysis result, the system sent a command to the traffic signal control system to shorten the green light time for heavy-loaded vehicles on this section by 15 seconds, and displayed a detour suggestion for heavy-loaded vehicles on the variable information sign, realizing precise traffic flow control on the deformed section, effectively delaying the development of road deformation, avoiding serious damage, and extending the service life of the road.

[0057] The above describes the deformation monitoring and control method for road pavement in the embodiment of the present application. The following describes the deformation monitoring and control system for road pavement in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of a deformation monitoring and control system for a road surface includes: The acquisition module is used to collect multi-level sensor data of the road surface and obtain the original deformation data set including fixed monitoring point data, mobile monitoring unit data and road surface images; A processing module, used for preprocessing the original deformation data set to obtain a road surface state time series data set; A prediction module, used for inputting the road surface state time series data set into a long short-term memory network for risk prediction to obtain a road surface deformation risk level assessment result; A grading module, used to determine the dominant factors of deformation and generate a grading control strategy according to the pavement deformation risk level assessment result; A judgment module is used to convert the hierarchical control strategy into a standardized execution instruction set, judge the priority of each instruction in the execution instruction set, and when the instruction priority is 1, start the emergency response mechanism and send an emergency intervention notice within 15 minutes; when the instruction priority is 2 to 4, send an execution notice 24 hours in advance, collect execution feedback and calculate the execution deviation rate; The control module is used to construct a road health knowledge graph based on the execution feedback and execution deviation rate, model the road deformation control process as a Markov decision process, and train a deep Q network to achieve adaptive traffic flow control.

[0058] Through the coordinated cooperation of the above-mentioned components, multi-level sensor data collection on the road surface is carried out to obtain the original deformation data set including fixed monitoring point data, mobile monitoring unit data and road surface images, so as to realize comprehensive monitoring of the road surface status, avoid the limitations of the traditional single data source, and greatly improve the monitoring coverage and data integrity; the original deformation data set is preprocessed to obtain the road surface status time series data set, which effectively eliminates data noise and discontinuity and improves the accuracy of subsequent analysis; the road surface status time series data set is input into the long short-term memory network for risk prediction to obtain the road surface deformation risk level assessment result. As a kind of deep learning algorithm, the long short-term memory network can effectively capture the long-term dependency in the time series data and accurately predict the development trend of road surface deformation. Compared with the traditional threshold judgment method, the early warning lead time is significantly improved, providing a sufficient time window for preventive maintenance; according to the road surface deformation risk level assessment result, the deformation dominant factor is determined and a hierarchical control strategy is generated, realizing in-depth analysis from symptoms to causes. The analysis makes the control measures more targeted, avoiding waste of resources and ineffective intervention; the hierarchical control strategy is converted into a standardized execution instruction set, and the priority of each instruction in the execution instruction set is judged, realizing differentiated response based on the degree of urgency. Emergency-level instructions can trigger emergency response within 15 minutes, greatly improving the system's ability to respond to emergencies; based on execution feedback and execution deviation rate, a road health knowledge graph is constructed, the road deformation control process is modeled as a Markov decision process, and a deep Q network is trained to achieve adaptive traffic flow control. This link fully reflects the unique value of artificial intelligence algorithms in specific application fields. Knowledge graph technology converts discrete monitoring data and control experience into structured knowledge. As a representative algorithm in the field of reinforcement learning, the deep Q network can gradually learn the optimal control strategy through continuous trial and error and optimization, realizing the transformation from "manual control" to "intelligent control". As the system is used for a longer time, the control effect continues to improve, which significantly reduces the road maintenance cost, extends the road service life, and improves driving safety.

[0059] above Figure 2 The deformation monitoring and control system for road pavement in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The deformation monitoring and control device for road pavement in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0060] Figure 31 is a schematic diagram of a deformation monitoring and control device for a road surface provided by an embodiment of the present invention. The deformation monitoring and control device 300 for a road surface may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage device terminals) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 may be temporary storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the deformation monitoring and control device 300 for a road surface. Furthermore, the processor 310 may be configured to communicate with the storage medium 330, and execute a series of instruction operations in the storage medium 330 on the deformation monitoring and control device 300 for a road surface to implement the steps of the above-mentioned deformation monitoring and control method for a road surface.

[0061] The road pavement deformation monitoring and control device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Figure 3 The structure of the deformation monitoring and control device for road pavement shown does not constitute a limitation on the deformation monitoring and control device for road pavement provided by the present invention, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0062] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the method for deformation monitoring and control of a road pavement.

[0063] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0064] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for enabling a deformation monitoring and control device for road pavement (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0065] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for monitoring and controlling deformation of a road surface, characterized in that: The method comprises: Perform multi-level sensor data collection on the road surface to obtain the original deformation data set including fixed monitoring point data, mobile monitoring unit data and road surface images; Preprocessing the original deformation data set to obtain a road surface state time series data set; Inputting the road surface state time series data set into the long short-term memory network for risk prediction to obtain a road surface deformation risk level assessment result; According to the pavement deformation risk level assessment result, determining the deformation dominant factor and generating a graded control strategy; Convert the hierarchical control strategy into a standardized execution instruction set, determine the priority of each instruction in the execution instruction set, and when the instruction priority is 1, activate the emergency response mechanism and send an emergency intervention notice within 15 minutes; when the instruction priority is 2 to 4, send an execution notice 24 hours in advance, collect execution feedback and calculate the execution deviation rate; Based on the execution feedback and execution deviation rate, a pavement health knowledge graph is constructed, the pavement deformation control process is modeled as a Markov decision process, and a deep Q network is trained to achieve adaptive traffic flow control.

2. The method for monitoring and controlling deformation of a road surface according to claim 1, characterized in that: The multi-level sensor data collection on the road surface is performed to obtain the original deformation data set including fixed monitoring point data, mobile monitoring unit data and road surface images, including: Strain sensors, displacement sensors, pressure sensors and temperature and humidity sensors are buried every 50 meters on the road to form fixed monitoring points, and sensor data is collected once an hour to obtain the fixed monitoring point data; An acceleration sensor, a gyroscope and a GPS positioning module are installed on a specific vehicle to form a mobile monitoring unit, and the acceleration change value of the vehicle in the vertical direction is recorded at a sampling rate of 100 Hz to obtain the mobile monitoring unit data; Using a road surface image acquisition device to perform an all-round scan of the road once a day to obtain a high-definition image with a resolution of not less than 4K to form the road surface image; Upload the fixed monitoring point data, the mobile monitoring unit data and the road surface image to the central data processing server in real time through the 4G / 5G wireless transmission module; Perform deviation detection on the single sensor data in the original deformation data set. When the data deviates from the historical average value by more than 30%, mark the deviated data point and trigger a repeated collection instruction to ensure data validity; An automatic weather station is used to record environmental parameters in real time. When the rainfall exceeds 50 mm / day or the temperature is below minus 10°C, the data collection frequency is adjusted to four times per hour to obtain intensive monitoring data under harsh conditions.

3. The method for monitoring and controlling deformation of a road surface according to claim 1, characterized in that: The preprocessing of the original deformation data set to obtain a road surface state time series data set includes: Applying a wavelet transform method to the fixed monitoring point data to remove high-frequency noise at a specific frequency to obtain a filtered fixed monitoring point signal; Applying a Kalman filter algorithm to the mobile monitoring unit data to eliminate interference caused by the vehicle's own vibration and obtain a pure road surface response signal; The missing values ​​in the filtered fixed monitoring point signal and the pure road surface response signal are supplemented by the time series neighboring point linear interpolation method. When the continuous missing data exceeds the preset threshold, the mean value of the historical data in the same period is used to replace it to obtain a complete data sequence; The complete data sequence is processed by a standardization method, and data from different sources are uniformly converted to a standard normal distribution to obtain a standardized data set; Performing spatiotemporal alignment processing on the standardized data set, establishing a mapping relationship between the mobile monitoring unit data and the road mileage pile number according to the GPS coordinates, locating the mobile monitoring unit data to a road surface grid unit of a specific size, and obtaining an aligned data set; The evidence theory method is applied to the aligned data set to perform multi-source data fusion, different weights are assigned to the fixed monitoring point data, mobile monitoring unit data and road surface images, and the road surface state time series data set is formed by integrating them through the feature layer fusion algorithm.

4. The method for monitoring and controlling deformation of a road surface according to claim 1, characterized in that: The step of inputting the road surface state time series data set into the long short-term memory network for risk prediction to obtain a road surface deformation risk level assessment result includes: Performing time domain analysis on the road surface state time series data set, extracting basic characteristic parameters including road surface displacement mean, standard deviation, peak value, valley value, peak-to-valley difference and its change rate, and obtaining a time domain feature set; The time domain feature set is converted into frequency domain by fast Fourier transform, and the main frequency component, power spectrum density and frequency bandwidth are extracted to obtain the frequency domain feature set; Applying the wavelet packet decomposition method to the frequency domain feature set, setting the number of decomposition layers to five, obtaining the energy distribution of each frequency band and its proportional relationship, and obtaining the time-frequency feature set; The time-frequency feature set is extracted by principal component analysis method to represent the main features, and the first five principal components are selected as morphological feature descriptions of road surface deformation to obtain a reduced dimension feature set; The dimension reduction feature set is input into the input layer of the long short-term memory network in time series order, and a hidden state vector is obtained after being processed by the first memory unit layer including a forget gate, an input gate and an output gate, and then a high-level time series feature is generated through the cell state update and output control gate of the second memory unit layer, and finally mapped into a deformation prediction value for the next twenty-four hours through a fully connected output layer to obtain a deformation prediction result; The pavement deformation risk is graded based on the deformation prediction result. According to the comprehensive scores of the three dimensions of displacement, deformation rate and prediction trend, the risk levels are divided into safety level, concern level, warning level, danger level and emergency level to form the pavement deformation risk level assessment result.

5. The method for monitoring and controlling deformation of a road surface according to claim 1, characterized in that: Determining the dominant factors of deformation and generating a hierarchical control strategy based on the road deformation risk level assessment result includes: A multi-factor pavement deformation cause analysis model is constructed, which includes an environmental factor sub-model, a material factor sub-model, and a load factor sub-model. The relationship between environmental parameters and deformation is analyzed through multiple linear regression to obtain the contribution of environmental factors. The contribution of the environmental factors, the degree of material aging and the impact of traffic loads are evaluated by a fuzzy reasoning system, and the functional relationship between the cumulative equivalent axle times and the deformation development rate is calculated to obtain the intensity of each factor; The strength of each factor is input into the Bayesian network for integrated analysis, and the contribution percentage of the dominant deformation cause is output to obtain the dominant deformation factor; According to the deformation dominant factor and the road deformation risk level assessment result, matching items are retrieved from a knowledge base containing predefined control schemes, corresponding control types are generated for road sections with different risk levels, and an initial control scheme set is obtained; Performing a life cycle cost analysis on the initial control scheme set, calculating the long-term cost-effectiveness ratio of each scheme, screening out the scheme with the best cost-effectiveness ratio, and obtaining a candidate control strategy; The candidate control strategies are globally optimized through an integer programming algorithm, and the resource allocation of the entire road network is balanced under the premise of meeting safety requirements, generating a hierarchical control strategy that includes an implementation schedule, a resource requirement list, expected effects, and emergency plans corresponding to different risk levels.

6. The method for monitoring and controlling deformation of a road surface according to claim 1, characterized in that: The hierarchical control strategy is converted into a standardized execution instruction set, and the priority of each instruction in the execution instruction set is determined. When the instruction priority is 1, the emergency response mechanism is activated and an emergency intervention notice is sent within 15 minutes; when the instruction priority is 2 to 4, an execution notice is sent 24 hours in advance, execution feedback is collected, and the execution deviation rate is calculated, including: Convert the hierarchical control strategy into a standardized execution instruction set including six elements: execution subject, execution time, execution location, execution content, execution standard and acceptance requirements, wherein the execution subject is divided into two categories: automatic execution system and manual intervention; Assigning priorities to the standardized execution instruction set according to risk levels, setting the priority of emergency-level instructions to 1, the priority of danger-level instructions to 2, the priority of warning-level instructions to 3, and the priority of attention-level instructions to 4, to obtain a priority ranking result; Determine the instruction type according to the priority sorting result; when the instruction priority is 1, send an emergency intervention notice including the intervention location, intervention content and intervention criteria to the relevant responsible unit, activate the emergency resource dispatching system, and obtain an emergency response record; Determine the instruction type according to the priority sorting result. When the instruction priority is 2 to 4, send an execution notice including an execution plan, resource requirements and acceptance criteria to the relevant responsible unit according to the scheduled time plan to obtain a regular execution record; Collecting the execution status data of the emergency response record and the routine execution record through multiple channels, including automatic execution status data and manual intervention progress data, comparing and analyzing the execution progress with the planned progress, and obtaining the execution feedback; The ratio of the actual execution time to the planned execution time is calculated based on the execution feedback to obtain the execution deviation rate, and when the execution deviation rate exceeds a preset threshold, an abnormal alarm and adjustment suggestions are sent to the management personnel.

7. The method for monitoring and controlling deformation of a road surface according to claim 1, characterized in that: Based on the execution feedback and the execution deviation rate, a road health knowledge graph is constructed, the road deformation control process is modeled as a Markov decision process, and a deep Q network is trained to realize traffic flow adaptive control, including: Based on the execution feedback and the execution deviation rate, a pavement health knowledge graph including five types of nodes, namely, section information, deformation characteristics, environmental factors, intervention measures and effect evaluation, is constructed, and the nodes are connected by edges of causal relationships and temporal relationships to obtain a deformation control knowledge base; The road deformation control process is modeled as a Markov decision process, where the state space is the road deformation feature vector, the action space is the set of optional control strategies, the transition probability is the relationship between state changes, and the reward function is the comprehensive score of deformation control effect and resource consumption, thereby obtaining a decision process model; A deep Q network consisting of three fully connected layers is constructed, with the number of neurons in each layer being 128, 64, and 32 respectively. The input layer receives the road state feature vector, and the output layer generates the Q value of each control strategy, thus obtaining the reinforcement learning model structure. The reinforcement learning model structure is trained using an experience replay mechanism, the replay buffer size is set to 10,000 records, the target network parameters are updated every 500 interactions, and the Q value estimation is optimized by the gradient descent algorithm to obtain a trained deep Q network; Based on the trained deep Q network, the traffic flow is dynamically analyzed, and the long-term benefits of different traffic flow allocation schemes are calculated according to the current road deformation state and predicted trend to obtain the optimal traffic control strategy; The optimal traffic control strategy is sent to the traffic signal control system and variable information signs to dynamically adjust the travel time, speed limit value and vehicle traffic ratio to achieve adaptive control of traffic flow on key deformed sections.

8. A deformation monitoring and control system for a road surface, characterized in that: Used to implement the deformation monitoring and control method for a road pavement according to any one of claims 1 to 7, the deformation monitoring and control system for a road pavement comprises: The acquisition module is used to collect multi-level sensor data of the road surface and obtain the original deformation data set including fixed monitoring point data, mobile monitoring unit data and road surface images; A processing module, used for preprocessing the original deformation data set to obtain a road surface state time series data set; A prediction module, used for inputting the road surface state time series data set into a long short-term memory network for risk prediction to obtain a road surface deformation risk level assessment result; A grading module, used to determine the dominant factors of deformation and generate a grading control strategy according to the pavement deformation risk level assessment result; A judgment module is used to convert the hierarchical control strategy into a standardized execution instruction set, judge the priority of each instruction in the execution instruction set, and when the instruction priority is 1, start the emergency response mechanism and send an emergency intervention notice within 15 minutes; when the instruction priority is 2 to 4, send an execution notice 24 hours in advance, collect execution feedback and calculate the execution deviation rate; The control module is used to construct a road health knowledge graph based on the execution feedback and execution deviation rate, model the road deformation control process as a Markov decision process, and train a deep Q network to achieve adaptive traffic flow control.

9. A deformation monitoring and control device for road pavement, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and the processor implements the deformation monitoring and control method for a road surface as claimed in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the processor executes the deformation monitoring and control method for a road surface as claimed in any one of claims 1 to 7.

Images