Sensor network deployment and self-starting method for long-term performance monitoring of highway subgrade and pavement

Abstract

This invention discloses a sensor network deployment and self-startup method for monitoring the long-term service performance of highway subgrade and pavement. The method comprises the following steps: S1: analyzing the types and locations of defects based on historical highway inspection data to construct a subgrade and pavement condition assessment model; constructing a probability model for the highway monitoring sensor network based on the parameters of different sensor types; S2: optimizing network-level monitoring plan decisions; S3: combining the subgrade and pavement condition assessment model with the gridded sensor field location and granularity under different operating conditions, deducing monitoring probability thresholds, and establishing a genetic algorithm training data set; S4: optimizing the deployment of the highway monitoring sensor network using a multi-objective optimization genetic algorithm; S5: controlling the self-awakening of the highway monitoring sensors based on a game theory model; and S6: establishing a long-term highway service performance risk warning mechanism based on the traffic environment and monitoring data. This method can provide real-time early warning of long-term highway service performance risks.

Description

Technical Field

[0001] The present invention relates to the technical field of road engineering, and in particular to a sensor network deployment and self-starting implementation method for long-term service performance monitoring of highway roadbed and pavement. Background Art

[0002] With the rapid development of my country's expressway network and the continuous increase in road traffic volume, the long-term performance monitoring of highways has become increasingly important. Especially in complex traffic environments, the structural health of the roadbed and pavement directly affects the service life and driving safety of highways.

[0003] Current monitoring of roadbed and pavement health relies primarily on periodic manual inspections, making it difficult to achieve real-time monitoring and early warning of highway conditions. Furthermore, existing sensor network deployments are often guided by monitoring area coverage models, resulting in insufficient monitoring accuracy, high redundancy, high energy consumption, and high costs, hindering widespread application. Furthermore, existing sensor networks for highway monitoring are constantly transmitting data, consuming high energy and requiring frequent sensor replacement, resulting in high costs. Summary of the Invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a sensor network deployment and self-starting implementation method for long-term service performance monitoring of highway roadbed and pavement, to achieve low-energy consumption and high-efficiency monitoring under different working conditions, and to make up for the problems of insufficient accuracy, high energy consumption and high cost of traditional monitoring methods.

[0005] Technical Solution: A sensor network deployment and self-starting implementation method for long-term service performance monitoring of highway subgrade and pavement includes the following steps:

[0006] S1, based on historical highway inspection data, analyzes the types and locations of defects and builds a roadbed and pavement condition assessment model; based on the parameters of different types of sensors, builds a highway monitoring sensor network probability model;

[0007] S2, based on the improved weighted random forest algorithm, performs road network-level monitoring solution decision optimization;

[0008] S3, combined with the roadbed and pavement condition assessment model, uses simulation software to analyze the location and granularity of the sensor field grid under different working conditions, deduce the monitoring probability threshold, and establish a genetic algorithm training data set;

[0009] S4, using the genetic algorithm dataset, optimizes the layout of highway monitoring sensor networks through multi-objective optimization genetic algorithm;

[0010] S5, based on the game theory model, realizes the self-wake-up control of highway monitoring sensors;

[0011] S6. Establish a risk warning mechanism for long-term highway service performance based on traffic environment and monitoring data.

[0012] Furthermore, in step S1, the implementation steps of establishing the roadbed and pavement condition assessment model are as follows:

[0013] SA1: Divide the road network to be monitored into several sections. Regularly use a drop weight deflectometer and ground penetrating radar to detect and evaluate the subgrade and pavement conditions in each section, obtaining deflection basin change data and GPR data trends of the subgrade and pavement respectively.

[0014] SA2 uses the law of large numbers and Gaussian distribution to quantitatively evaluate the data in the roadbed and pavement condition assessment of each road section, and uses deep neural networks to mark and interpolate the data with abnormalities in the deflection basin data and GPR data. At the same time, based on DNN, feature extraction is performed on the deflection basin data and GPR data to inversely calculate the structural modulus E of each structural layer of the roadbed and pavement, and extract the dielectric constant ε r and reflection waveform characteristics;

[0015] SA3 uses the YOLO network to determine the weak locations of the detected road sections and obtain the specific types of diseases that often occur;

[0016] SA4, combined with historical inspection data, spatially aligns the processed GPR data and FWD data to verify whether the bearing capacity meets the design standards; evaluates the degradation trend of the bearing capacity through historical inspection data, and identifies potential voids, voids, or subsidence areas in the roadbed; verifies the severity of the disease by combining the abnormal areas of FWD deflection basins, evaluates the structural strength, bearing capacity, and disease status of the roadbed and pavement, and establishes a roadbed and pavement condition assessment model for this section.

[0017] Furthermore, in step S1, the implementation process of constructing the highway monitoring sensor network probability model is as follows:

[0018] The monitoring probability model of different types of sensors is calculated according to the following formula:

[0019] P(d)=P0·e -αd

[0020] Where P(d) is the detection probability, P0 is the detection probability of the sensor at zero distance, α is the attenuation factor, and d is the distance between the sensor and the target;

[0021] A type of sensor is selected, and the monitoring probability at different distances is measured experimentally. The target is placed at different distances from the sensor, the number of successful monitoring times of the sensor is recorded, the monitoring probability is calculated, and the data is saved in a computer. After obtaining a sufficient amount of experimental data, the attenuation factor α is fitted using the least squares method to obtain the monitoring probability model of this type of sensor. The monitoring probability model of all types of sensors is fitted using the same method as the probability model of the highway monitoring sensor network.

[0022] Furthermore, the decision optimization of the road network-level monitoring solution includes the following steps:

[0023] S21, performing weighted processing on the extracted feature data based on the actual situation of the specific road section; if the road section has a specific disease risk, then increasing the weight of the feature data corresponding to the specific disease;

[0024] S22 uses Gini impurity minimization as the growth principle of the decision tree, accumulates the probability density functions of the monitoring schemes under the same indicator, and divides the subtrees into similar sizes. The growth condition is set as the Gini value growth rate less than the threshold, and a classifier model based on the training sample is obtained;

[0025] S23, performing out-of-band data detection on the generated decision tree, eliminating and reconstructing unqualified decision trees to ensure the accuracy of the final synthesized classifier; using the layer analysis method to calculate the sensitivity of the decision tree to different features, and determine the weights of the sub-classifiers based on the feature sensitivity, and form the overall classifier of the random forest based on the weights of the sub-classifiers;

[0026] S24, based on the characteristic data of the roadbed and pavement in the road section, manually annotate the data set according to the roadbed and pavement condition assessment model, and divide the road section into different monitoring level labels according to the monitoring level: no monitoring required - label 0, general monitoring - label 1, key monitoring - label 2;

[0027] S25, using the feature data as an input feature vector and the monitoring degree label as an output vector to create a data set; and using the data set to train the overall classifier.

[0028] Furthermore, analyzing the sensor field gridding and deducing the monitoring probability threshold under different working conditions includes the following steps:

[0029] S31, selecting sensor types and placement locations based on the roadbed and pavement condition assessment model; selecting sensor installation locations based on the types of damage and weak locations that may appear in the roadbed and pavement condition assessment model;

[0030] S32, based on the different sensor types, weak locations and possible defects, use MATLAB to simulate and analyze the sensor field grid of the designated road sections based on the optimized road network level monitoring scheme;

[0031] S33, combining the subgrade and pavement condition assessment model with the gridded sensor field for a specific road section, analyzes the most unfavorable locations of the subgrade and pavement structure and sets the monitoring probability threshold to 90%. For non-key monitoring areas, the monitoring probability threshold is set to 70%. A multi-sensor joint monitoring probability model is established in MATLAB. When multiple sensors jointly monitor a grid point, the monitoring probability is calculated as follows:

[0032] P miss =(1-P1)(1-P2)...(1-P n )

[0033] P detect =1-P miss

[0034] Where, P miss is the probability of monitoring failure, P n is the probability of the nth sensor of the same type monitoring a specific point, P detect is the joint monitoring probability;

[0035] Among them, the most unfavorable locations include locations where diseases may occur and locations where bearing capacity decays fastest.

[0036] Furthermore, in step S4, the optimization objectives of the genetic algorithm are determined to be: the number of sensors and the sensor monitoring coverage. By weighted combination of the two optimization objectives, a comprehensive objective function is obtained:

[0037] minF(n,P(d))=ω1n-ω2Q P

[0038] Where minF(n,P(d)) is the objective function, which is reduced during the iteration process; P(d) is the monitoring probability model of this type of sensor; n is the number of sensors; Q P is the sensor monitoring coverage, with a value between 0 and 1; ω1 and ω2 are weight parameters;

[0039] Select the initial population. Each individual is usually encoded in the form of a binary string or integer array, representing the deployment status of sensors at different locations. Based on the probabilistic coverage model of sensor monitoring capabilities, the requirements of different damaged areas of the structure are considered. While meeting the reliability requirements, the required monitoring capacity of each monitoring point under different environmental conditions is quantified as the optimization boundary to guide the selection of the initial population.

[0040] The objective function is used to comprehensively evaluate the reliability of coverage monitoring probability and the number of sensors. Through a multi-objective optimization iterative process, the reliability of sensor collaborative monitoring probability is integrated and redundant deployment is eliminated. The number of sensors and deployment locations of the monitoring system are gradually determined. After iteration, the Pareto optimal solution set is obtained, which is used to characterize the optimal deployment location of sensors on this road section.

[0041] Furthermore, in step S5, the steps of implementing the self-wake-up control of the monitoring sensor based on the game theory model are as follows:

[0042] S51, establish the sensor energy consumption model, the total energy consumption C of the i-th sensor i Calculate according to the following formula:

[0043] C i =E s (S i )+E p (S i )+E t (S i )

[0044] Among them, E p (S i ) represents the sensing energy of the i-th sensor, E s (S i ) represents the processing energy of the i-th sensor, E t (S i ) represents the communication energy of the i-th sensor, S i is the i-th sensor in the sensor network;

[0045] In step S52, the switching of the sensor between the awake state and the dormant state is considered as a game problem. The payoff function is defined based on the number of data successfully transmitted by the sensor in the awake state. The sensor node comprehensively considers the payoff function and the amount of energy consumed to make a trade-off. The game space is: {from awake to dormant, from awake to not dormant}. The game model GT of the game player is defined as follows:

[0046] GT={N,K,{u i}}

[0047] Where N represents the game party, which is all sensor nodes in the network; K is the strategy space adopted; u i is the utility function;

[0048] Iterative game model, in this process, the dynamic sleep threshold of the sensor node is determined based on the utility function. In one round of game, when the utility function value is greater than the sleep threshold, the sensor node remains awake; when the utility function value is less than the sleep threshold, the sensor enters the sleep state;

[0049] S53 introduces a sensor node penalty mechanism. When a sensor node chooses to enter a dormant state to save its own energy consumption, it is called selfish behavior of the sensor node. When a sensor node adopts selfish behavior, the sensor network will mark the node and force the sensor node to stay awake in the following M rounds. The M-th round game is defined as follows:

[0050] u i (s i ,s -i )=U i (s i ,s -i )-C i (s i ,s -i )

[0051] Among them, U i (s i ,s -i ) is the payoff function of the i-th sensor game in the sensor network; C i (s i ,s -i ) is the energy consumption function of the ith sensor game in the sensor network; s i The decision to wake up the i-th sensor in the sensor network; s -i Take the sleep decision for the i-th sensor in the sensor network.

[0052] Furthermore, in step S6, the implementation process of establishing a highway long-term service performance risk early warning mechanism is as follows:

[0053] After determining the layout position and self-starting conditions, according to the monitoring data: vertical stress σ z , stress σ along the vehicle travel direction x , stress perpendicular to the vehicle's direction of travel σ y , vertical strain ε z , strain ε along the vehicle's travel direction x , strain perpendicular to the vehicle's direction of travel ε y , the temperature T and humidity H of each structural layer, and use DNN to inversely calculate the pavement structure modulus E as a characterization indicator of the health of the roadbed and pavement structure. The inverse calculation process is as follows:

[0054] Where, is the structural modulus output prediction vector, E i is the structural modulus of the i-th layer of the pavement structure; A=[σ z ,σ x ,σ y ,ε z ,ε x ,εy ,T,H]; θ is the parameter in the prediction model, including weights and biases;

[0055] The structural strength and bearing capacity of each structural layer of the roadbed and pavement are evaluated by the changes in the structural modulus E of each layer;

[0056] Use automatic data collection technology to monitor and record the traffic and environmental parameters of a designated road section in real time, and evaluate the traffic and environmental conditions of the road section;

[0057] Combining monitoring data with traffic environment models, we can predict the downward trend in the bearing capacity of the roadbed and pavement structure, the location and development trend of cracks, voids, interlayer delamination, and subsidence, and ultimately achieve real-time early warning of the long-term service performance risks of the highway.

[0058] Compared with the prior art, the present invention has the following significant effects:

[0059] 1. Based on historical detection data and an improved optimization algorithm, this invention constructs a roadbed and pavement condition assessment model based on historical monitoring data, and a probabilistic coverage model based on sensor parameters and actual highway conditions. Through a multi-objective optimization iterative process, it integrates the probability reliability of sensor collaborative monitoring, eliminates redundant deployment, and gradually determines the number and deployment locations of sensors in the monitoring system. This establishes a dynamically adjustable monitoring foundation to ensure that the monitoring needs of key areas are met and further improve monitoring accuracy. This overcomes the problems of traditional monitoring methods such as insufficient accuracy, excessive energy consumption, and excessive cost, and enables real-time monitoring of the roadbed and pavement structure status.

[0060] 2. Based on the game theory model, the present invention optimizes the system collaboration strategy while ensuring monitoring reliability, gradually determines the self-starting strategy of each node sensor in the network, ensures the monitoring accuracy of key areas while meeting the reliability while reducing system power consumption, thereby realizing efficient and low-redundancy monitoring of the structural health status, and providing technical support and theoretical basis for improving the long-term service performance and intelligent management of highways. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 A flowchart of the invention;

[0062] Figure 2 The present invention is a flow chart of the self-awakening control of the highway monitoring sensor based on the game theory model. DETAILED DESCRIPTION

[0063] The present invention will be described in further detail below with reference to the accompanying drawings and specific implementations.

[0064] like Figure 1As shown, the present invention proposes a sensor network deployment and self-starting strategic optimization method for long-term service performance monitoring of highway subgrade and pavement. This method uses a strategic optimization method to build a subgrade and pavement condition assessment model, extract structural state characteristic parameters, iterate genetic algorithms, and iterate game theory to analyze the optimal sensor deployment plan and self-starting strategy under different working condition combinations, achieve real-time early warning of long-term service performance risks of highways, and provide suggestions for subsequent subgrade and pavement inspection plans. The method specifically includes the following steps:

[0065] Step 1: Based on historical highway inspection data, analyze the types and locations of defects and build a roadbed and pavement condition assessment model; based on the parameters of different types of sensors, build a highway monitoring sensor network probability model;

[0066] (1) Based on historical data from FWD (Falling Weight Deflectometer) and ground penetrating radar inspections of road sections, a DNN (Deep Neural Network) was used for feature extraction, and a YOLO network was used to analyze the type and location of damage. This information was then integrated to establish a roadbed and pavement condition assessment model.

[0067] The implementation steps for establishing a roadbed and pavement condition assessment model are as follows:

[0068] Step A1: Divide the road network to be monitored into several sections. Monthly, use a drop weight deflectometer and ground penetrating radar to inspect and evaluate the subgrade and pavement conditions within each section. Measure the deflection of each pavement point using the drop weight deflectometer, and compare the deflection basin data of the subgrade and pavement at different times to obtain deflection basin change data of the subgrade and pavement. Use ground penetrating radar to inspect the subgrade and pavement, obtain raw GPR (Ground Penetrating Radar) data at a specific moment, and compare the GPR detection results at different times to analyze the changing trends of the subgrade and pavement GPR data.

[0069] In step A2, the law of large numbers and Gaussian distribution are used to quantitatively evaluate the data in the roadbed and pavement condition assessment of each road section. A deep neural network is used to mark and interpolate the data with abnormalities in the deflection basin data and GPR data. Simultaneously, feature extraction is performed on the deflection basin data and GPR data based on the DNN to inversely calculate the structural modulus E of each structural layer of the roadbed and pavement, reflecting the changing trend of the bearing capacity of the roadbed and pavement structure. The dielectric constant ε is extracted. r , reflected waveform characteristics and other characteristic data are used to characterize the development trend of cracks, voids, interlayer delamination, settlement and other defects in the roadbed and pavement structure.

[0070] Step A3: Use the YOLO network to determine the weak locations of the detected road section and obtain the specific types of common diseases;

[0071] In step A4, the processed GPR data and FWD data are spatially aligned, combined with the aforementioned historical inspection data, to verify whether the bearing capacity meets the design standards. The historical inspection data is then used to assess the degradation trend of the bearing capacity and identify potential cavities, voids, or subsidence areas in the roadbed. The severity of the damage is verified by combining the abnormal FWD deflection basin areas, assessing the structural strength, bearing capacity, and damage status of the roadbed and pavement, and establishing a roadbed and pavement condition assessment model for the section.

[0072] (2) By analyzing the parameters of different types of sensors and conducting experiments, we established a probabilistic model for the highway monitoring sensor network using experimental data and an exponential decay model. The implementation process is as follows:

[0073] In the laboratory, install one sensor of each type required for road surface monitoring and connect the sensor to the computer wirelessly. Turn on the computer to ensure that the connection between the computer and the sensor is normal. Calibrate the different sensors and prepare to collect data. Select the exponential decay model to describe the effect of distance on the monitoring probability. The model probability is calculated according to formula (1):

[0074] P(d)=P0·e -αd (1)

[0075] Where P(d) is the detection probability, P0 is the detection probability of the sensor at zero distance (contact monitoring), α is the attenuation factor, and d is the distance between the sensor and the target.

[0076] A type of sensor is selected, and the monitoring probability at different distances is measured experimentally. The target is placed at different distances from the sensor, the number of successful monitoring times of the sensor is recorded, the monitoring probability is calculated, and the data is saved in a computer. After obtaining a sufficient amount of experimental data, the attenuation factor α in the exponential decay model is fitted using the least squares method to obtain the monitoring probability model of this type of sensor. The monitoring probability model of all types of sensors is fitted using the same method as the probability model of the highway monitoring sensor network.

[0077] Step 2: Optimize the road network-level monitoring solution decision based on the improved weighted random forest algorithm;

[0078] The decision optimization of the road network-level monitoring solution includes the following steps:

[0079] Step 21: Introduce a weighted mechanism to generate a decision tree: Combined with the actual situation of a specific road section, extract the dielectric constant ε from the extracted feature data structure modulus E r , reflection waveform characteristics A, signal attenuation rate α r, two-way travel time t, and signal phase difference φ are weighted. If the road section has no disease risk in historical detection, the weight of the structural modulus E is increased; if the road section has a specific disease risk, the weight of the characteristic data corresponding to the disease is increased, thereby increasing the influence weight of the characteristic indicators representing the disease and highlighting the important disease characteristic indicators;

[0080] Step 22: Using the minimization of Gini impurity as the growth principle of the decision tree, the probability density functions of the monitoring schemes under the same indicator are accumulated and divided into subtrees of similar size; the growth rate of the Gini value is less than the threshold as the growth stop condition, and the classifier model based on the training sample is obtained. The process is shown in formulas (2) and (3).

[0081]

[0082]

[0083] Where D represents the divided training sample set; is the number of decision subtrees; β is the partition feature of the training sample set; GiniIndex(D,β) is the sum of the Gini values ​​of the current partitioned sample training set; N is the number of samples in set D; N v is the vth set D v The number of samples; d is the number of different samples in set D, p i is the proportion of the i-th sample in the set D to all samples; p(β) is the probability density coincidence distribution function of the feature β, which is the probability set of the partition points of the feature β in random sampling; is the probability density coincidence distribution function of the partition point set

[0084] Step 23: Generate a high-precision classifier based on the random forest algorithm: perform out-of-band data detection on the generated decision tree, remove and reconstruct unqualified decision trees, and ensure the accuracy of the final synthesized classifier; use the analytic hierarchy process to calculate the sensitivity of the decision tree to different features according to formula (4), and determine the weight of the sub-classifier based on the feature sensitivity, and form the overall classifier of the random forest based on the weight combination of each sub-classifier;

[0085]

[0086] Where, The roadbed and pavement condition evaluation feature d (the feature index extracted in step 1, i.e., the distance between the sensor and the target) is used to evaluate the decision tree J i Sensitivity; T OOB is the classification accuracy of the decision tree classifier; T OOB_d is the decision tree J after deleting feature d i Out-Of-Bag Accuracy

[0087] Step 24: Set the road section monitoring intensity label: Extract the dielectric constant ε based on the structural modulus E of the roadbed and pavement in the road section r , reflection waveform characteristics A, signal attenuation rate α r , two-way travel time t, signal phase difference φ feature data, refer to the road section roadbed and pavement condition assessment model for manual annotation to create a data set, and divide it into different levels of road section monitoring: no monitoring required (label 0), ordinary monitoring (label 1), and key monitoring (label 2);

[0088] Step 25: Train the model to select the monitoring scheme: extract the dielectric constant ε from the structural modulus E r , reflection waveform characteristics A, signal attenuation rate α r , two-way travel time t, signal phase difference φ and other feature data are used as input feature vectors, and the monitoring degree label is used as the output vector to make a data set. The data set is used to train the overall classifier, and finally a data-driven monitoring solution is obtained. The process is shown in formula (5).

[0089]

[0090] Where, For classification labels The predicted probability of the final classification label is the largest probability category, where c = 0, 1, 2; It is the feature data extracted from the original detection data. T is the temperature of each structural layer; H is the humidity of each structural layer; is the predicted probability of category c by the kth tree.

[0091] Step 3: Combined with the roadbed and pavement condition assessment model, simulation software is used to analyze the location and granularity of the sensor field grid under different working conditions, deduce the monitoring probability threshold, and establish a genetic algorithm training data set;

[0092] Sensor field gridding and monitoring probability threshold analysis include the following steps:

[0093] Step 31: Select the sensor type and placement location based on the roadbed and pavement condition assessment model. In this embodiment, the sensor types selected are stress sensor, strain sensor, temperature sensor, and humidity sensor, which are used to respectively detect the three-dimensional stress (vertical stress σ z , stress σ along the vehicle travel direction x , stress perpendicular to the vehicle's direction of travel σ y ), three-dimensional strain (vertical strain ε z , strain ε along the vehicle's travel direction x , strain perpendicular to the vehicle's direction of travel εy ), the temperature T and humidity H of each structural layer are monitored; according to the possible disease types and weak locations in the road section roadbed and pavement condition assessment model, the sensor installation locations are selected. For locations where fatigue cracking of the asphalt layer may occur, the focus is on monitoring the tensile strain at the bottom of the asphalt layer; for locations where fatigue cracking of the inorganic binder stabilized layer may occur, the focus is on monitoring the horizontal tensile stress at the bottom of the base layer; for locations where subsidence may occur, the focus is on monitoring the compressive strain on the top surface of the roadbed; temperature sensors and humidity sensors are mainly installed at the bottom of each structural layer.

[0094] Step 32, divide the sensor field granularity based on the monitoring plan and the roadbed and pavement condition assessment model: according to the different sensor types, weak locations and possible diseases, use MATLAB to perform grid simulation analysis of the sensor field of the specified road section based on the optimized road network-level monitoring plan. The specific method is as follows: according to the type and characteristics of the roadbed and pavement structure and the main installation location of the monitoring sensor, the geometric model of the grid division of the specific area is selected as three-dimensional, and each grid point uses (x, y, z) coordinates to represent its position; the granularity of the grid division is determined according to the importance of the monitoring plan, the load-bearing capacity decline trend, the possible disease type and the monitoring accuracy requirement; the geometric boundary of the monitoring area is defined, and the three-dimensional grid points are generated through the meshgrid function, and the probabilistic monitoring model of the sensor is defined in MATLAB; considering the actual situation, in the specified road section, a Boolean matrix is ​​created to simulate the existence of obstacles, and each grid point is marked whether it is blocked by the obstacle, so as to obtain the appropriate grid sensor field for the section and provide data support for the establishment of the probability threshold.

[0095] Step 33, simulation and deduction of sensor coverage probability threshold: combining the roadbed and pavement condition assessment model with the gridded sensor field of a specific road section, analyzing the most unfavorable position of the roadbed and pavement structure (including the position where disease may occur and the position where the bearing capacity decays fastest), setting the monitoring probability threshold to 90%, and setting the monitoring probability threshold to 70% for non-key monitoring areas; establishing a multi-sensor joint monitoring probability model in MATLAB, and calculating the monitoring probability of grid points according to formulas (6) and (7) when multiple sensors jointly monitor the grid points; for non-key areas, a higher overall coverage rate can be achieved by appropriately lowering the monitoring probability threshold of a single point; determining a suitable threshold by continuously adjusting the sensor distribution position and the monitoring probability threshold to ensure that the sensors have sufficient coverage while the monitoring redundancy is as small as possible; organizing the data obtained in the adjustment process into a genetic algorithm data set for further optimization.

[0096] P miss =(1-P1)(1-P2)...(1-P n ) (6)

[0097] P detect=1-P miss (7)

[0098] Where, P miss is the probability of monitoring failure, P n is the probability of the nth sensor of the same type monitoring a specific point, P detect is the joint monitoring probability.

[0099] Step 4: Using the established genetic algorithm dataset, the layout of the highway monitoring sensor network is optimized through a multi-objective optimization genetic algorithm;

[0100] A non-dominated sorting genetic algorithm (NSG-II) with an elite strategy based on multi-objective optimization is used. First, the optimization objectives and boundary conditions of the genetic algorithm are determined. There are two optimization objectives: the number of sensors and the sensor monitoring coverage rate. By weighted combination of the two optimization objectives, the comprehensive objective function is obtained as shown in Equation (8). Taking the key monitoring area as an example, the boundary conditions are shown in Equation (9). The initial population is selected. Each individual is usually encoded in the form of a binary string or integer array, representing the deployment status of sensors at different locations. Based on the sensor monitoring capability probabilistic coverage model, the differences in requirements of different structural damage areas are considered. Under the condition of meeting the reliability, the monitoring capability required by each monitoring point under different environmental conditions is quantified as the optimization boundary and guides the selection of the initial population. The objective function is used to comprehensively evaluate the coverage monitoring probability reliability and the number of sensors. Through the multi-objective optimization iterative process, the sensor collaborative monitoring probability reliability is integrated and redundant deployment is eliminated. The number of sensors and deployment locations of the monitoring system are gradually determined, and a monitoring foundation that can be dynamically adjusted is established. After iteration, the Pareto optimal solution set is obtained, which is used to characterize the optimal deployment location of sensors in this section.

[0101] minF(n,P(d))=ω1n-ω2Q P (8)

[0102] n>0,Q P >0.9 (9)

[0103] Where minF(n,P(d)) is the objective function, which is reduced during the iteration process; P(d) is the monitoring probability model of this type of sensor; n is the number of sensors; Q P is the sensor monitoring coverage, which ranges from 0 to 1; ω1 and ω2 are weight parameters.

[0104] Step 5: Establish a game model in PyCharm based on the determined sensor layout plan, and implement sensor self-wake-up control while ensuring the reliability of roadbed and pavement structure monitoring;

[0105] The steps to implement the self-wake-up control of the monitoring sensor based on the game theory model are as follows:

[0106] Step 51: First, establish a sensor energy consumption model: the energy consumed by the sensor during the detection process is related to multiple factors, including the sensing energy E p (S i ), processing energy E s (S i ), communication energy E t (S i ), the above three energy consumptions are determined by the sensor itself, so the total energy consumption of a single sensor is defined according to formula (10);

[0107] Step 52, then establish a sensor network game model: the rational switching of the sensor between the awake state and the dormant state can be regarded as a game problem, and the game model is expressed as formula (10); the profit function is defined according to the number of data successfully transmitted by the sensor in the awake state, and the sensor node comprehensively considers the profit function and the amount of energy consumption to make a trade-off, and its game space is: {from awake to dormant, from awake to not dormant}, and the game model GT of the game party is defined as shown in formula (11); the game model is iterated, and in this process, the dynamic dormant threshold of the sensor node is determined based on the utility function. In a round of game, when the utility function value is greater than the dormant threshold, the sensor node remains awake, and when the utility function value is less than the dormant threshold, the sensor enters the dormant state;

[0108] Step 53, finally introduces the sensor node penalty mechanism: when the sensor node always chooses to enter the dormant state to save its own energy consumption, it is called the selfish behavior of the sensor node; when the sensor node takes selfish behavior, the sensor network will mark the node and force the sensor node to stay awake in the subsequent M rounds, where the M-th round game is defined according to formula (12), so as to ensure that the sensor node can wake up in time in the subsequent game process, and finally the entire sensor network enters the Nash equilibrium, realizing the self-starting control of the sensor network monitoring under the condition that the deployment plan is determined and the reliability is met. The process is as follows Figure 2 shown.

[0109] C i =E s (S i )+E p (S i )+E t (S i ) (10)

[0110] Where S i is the i-th sensor in the sensor network; C i is the total energy consumed by the sensor.

[0111] GT={N,K,{u i}} (11)

[0112] u i (s i ,s -i )=U i (s i ,s -i )-C i (s i ,s -i ) (12)

[0113] Where N represents the game party. In this step, every node in the sensor network participates in sending sensor information, so the game party is all sensor nodes in the network; K is the strategy space adopted; u i is the utility function; U i (s i ,s -i ) is the payoff function of the i-th sensor game in the sensor network; C i (s i ,s -i ) is the energy consumption function of the ith sensor game in the sensor network; s i The decision to wake up the i-th sensor in the sensor network; s -i Take the sleep decision for the i-th sensor in the sensor network.

[0114] Step 6: Establish a long-term highway service performance risk warning mechanism based on traffic environment and monitoring data. Perform modulus back calculation based on real-time monitoring data, and combine the subgrade and pavement condition assessment model with traffic environment information to provide guidance for subsequent subgrade and pavement disease monitoring and maintenance.

[0115] The implementation process of establishing a long-term highway service performance risk warning mechanism based on traffic environment and monitoring data is as follows:

[0116] After determining the layout location and self-starting conditions, use DNN to calculate the vertical stress σ based on the monitoring data. z , stress σ along the vehicle travel direction x , stress perpendicular to the vehicle's direction of travel σ y , vertical strain ε z , strain ε along the vehicle's travel direction x , strain perpendicular to the vehicle's direction of travel ε yThe pavement structural modulus E is calculated from the temperature T and humidity H of each structural layer, which serves as a characterization of the health of the subgrade and pavement structure. The calculation process is shown in Equation (13). The structural strength and bearing capacity of each structural layer of the subgrade and pavement are evaluated by the changes in the structural modulus E of each layer. Ring coils, microwave sensors, video monitoring systems, thermometers and hygrometers are installed on designated road sections. Traffic parameters (traffic volume, traffic composition, vehicle axle load information) and environmental parameters (temperature and humidity) of the designated road section are monitored and recorded in real time using automatic data collection technology to evaluate the traffic and environmental conditions of the section. Combining monitoring data with traffic environment models, the downward trend of the subgrade and pavement structural bearing capacity, the location and development trend of diseases such as cracks, voids, interlayer delamination, and subsidence are predicted, guiding subsequent inspection and maintenance, and ultimately achieving real-time early warning of long-term service performance risks of the highway.

[0117]

[0118] Where, is the structural modulus output prediction vector, Among them E i is the structural modulus of the i-th layer of the pavement structure; A=[σ z ,σ x ,σ y ,ε z ,ε x ,ε y ,T,H]; θ is the parameter in the prediction model, including weights and biases.

[0119] The examples described in the present invention are merely descriptions of the preferred embodiments of the present invention and are not intended to limit the concept and scope of the present invention. Without departing from the design concept of the present invention, various modifications and improvements made to the technical solutions of the present invention by engineers and technicians in this field should fall within the scope of protection of the present invention.

Claims

1. A sensor network deployment and self-starting implementation method for long-term service performance monitoring of highway roadbed and pavement, characterized in that: The steps are as follows: S1, based on historical highway inspection data, analyzes the types and locations of defects and builds a roadbed and pavement condition assessment model; based on the parameters of different types of sensors, builds a highway monitoring sensor network probability model; S2, based on the improved weighted random forest algorithm, performs road network-level monitoring solution decision optimization; S3, combined with the roadbed and pavement condition assessment model, uses simulation software to analyze the location and granularity of the sensor field grid under different working conditions, deduce the monitoring probability threshold, and establish a genetic algorithm training data set; S4, using the genetic algorithm dataset, optimizes the layout of highway monitoring sensor networks through multi-objective optimization genetic algorithm; S5, based on the game theory model, realizes the self-wake-up control of highway monitoring sensors; S6. Establish a risk warning mechanism for long-term highway service performance based on traffic environment and monitoring data.

2. The sensor network deployment and self-starting implementation method for long-term service performance monitoring of highway subgrade and pavement according to claim 1 is characterized in that: In step S1, the implementation steps for establishing the roadbed and pavement condition assessment model are as follows: SA1: Divide the road network to be monitored into several sections. Regularly use a drop weight deflectometer and ground penetrating radar to detect and evaluate the subgrade and pavement conditions in each section, obtaining deflection basin change data and GPR data trends of the subgrade and pavement respectively. SA2 uses the law of large numbers and Gaussian distribution to quantitatively evaluate the data in the roadbed and pavement condition assessment of each road section, and uses deep neural networks to mark and interpolate the data with abnormalities in the deflection basin data and GPR data. At the same time, based on DNN, feature extraction is performed on the deflection basin data and GPR data to inversely calculate the structural modulus E of each structural layer of the roadbed and pavement, and extract the dielectric constant ε r and reflection waveform characteristics; SA3 uses the YOLO network to determine the weak locations of the detected road sections and obtain the specific types of diseases that often occur; SA4, combined with historical inspection data, spatially aligns the processed GPR data and FWD data to verify whether the bearing capacity meets the design standards; evaluates the degradation trend of the bearing capacity through historical inspection data, and identifies potential voids, voids, or subsidence areas in the roadbed; verifies the severity of the disease by combining the abnormal areas of FWD deflection basins, evaluates the structural strength, bearing capacity, and disease status of the roadbed and pavement, and establishes a roadbed and pavement condition assessment model for this section.

3. The sensor network deployment and self-starting implementation method for long-term service performance monitoring of highway subgrade and pavement according to claim 1 is characterized in that: In step S1, the implementation process of constructing the highway monitoring sensor network probability model is as follows: The monitoring probability model of different types of sensors is calculated according to the following formula: P(d)=P0·e -αd Where P(d) is the detection probability, P0 is the detection probability of the sensor at zero distance, α is the attenuation factor, and d is the distance between the sensor and the target; A type of sensor is selected, and the monitoring probability at different distances is measured experimentally. The target is placed at different distances from the sensor, the number of successful monitoring times of the sensor is recorded, the monitoring probability is calculated, and the data is saved in a computer. After obtaining a sufficient amount of experimental data, the attenuation factor α is fitted using the least squares method to obtain the monitoring probability model of this type of sensor. The monitoring probability model of all types of sensors is fitted using the same method as the probability model of the highway monitoring sensor network.

4. The sensor network deployment and self-starting implementation method for long-term service performance monitoring of highway subgrade and pavement according to claim 1 is characterized in that: The decision optimization of the road network-level monitoring solution includes the following steps: S21, performing weighted processing on the extracted feature data based on the actual situation of the specific road section; if the road section has a specific disease risk, then increasing the weight of the feature data corresponding to the specific disease; S22 uses Gini impurity minimization as the growth principle of the decision tree, accumulates the probability density functions of the monitoring schemes under the same indicator, and divides the subtrees into similar sizes. The growth condition is set as the Gini value growth rate less than the threshold, and a classifier model based on the training sample is obtained; S23, performing out-of-band data detection on the generated decision tree, eliminating and reconstructing unqualified decision trees to ensure the accuracy of the final synthesized classifier; using the layer analysis method to calculate the sensitivity of the decision tree to different features, and determine the weights of the sub-classifiers based on the feature sensitivity, and form the overall classifier of the random forest based on the weights of the sub-classifiers; S24, based on the characteristic data of the roadbed and pavement in the road section, manually annotate the data set according to the roadbed and pavement condition assessment model, and divide the road section into different monitoring level labels according to the monitoring level: no monitoring required - label 0, general monitoring - label 1, key monitoring - label 2; S25, using the feature data as an input feature vector and the monitoring degree label as an output vector to create a data set; and using the data set to train the overall classifier.

5. The sensor network deployment and self-starting implementation method for long-term service performance monitoring of highway subgrade and pavement according to claim 4 is characterized in that: Analyzing sensor field gridding and deducing monitoring probability thresholds under different operating conditions includes the following steps: S31, selecting sensor types and placement locations based on the roadbed and pavement condition assessment model; selecting sensor installation locations based on the types of damage and weak locations that may appear in the roadbed and pavement condition assessment model; S32, based on the different sensor types, weak locations and possible defects, use MATLAB to simulate and analyze the sensor field grid of the designated road section based on the optimized road network level monitoring scheme; S33, combining the subgrade and pavement condition assessment model with the gridded sensor field for a specific road section, analyzes the most unfavorable locations of the subgrade and pavement structure and sets the monitoring probability threshold to 90%. For non-key monitoring areas, the monitoring probability threshold is set to 70%. A multi-sensor joint monitoring probability model is established in MATLAB. When multiple sensors jointly monitor a grid point, the monitoring probability is calculated as follows: P miss =(1-P1)(1-P2)...(1-P n ) P detect =1-P miss Where, P miss is the probability of monitoring failure, P n is the monitoring probability of the nth sensor of the same type of sensor to the grid point, P detect is the joint monitoring probability; Among them, the most unfavorable locations include locations where diseases may occur and locations where bearing capacity decays fastest.

6. The sensor network deployment and self-starting implementation method for long-term service performance monitoring of highway subgrade and pavement according to claim 5 is characterized in that: In step S4, the optimization objectives of the genetic algorithm are determined to be: the number of sensors and the sensor monitoring coverage. By weighted combination of the two optimization objectives, a comprehensive objective function is obtained: minF(n,P(d))=ω1n-ω2Q P Where minF(n,P(d)) is the objective function, which decreases during the iteration process; P(d) is the monitoring probability model of this type of sensor; n is the number of sensors; Q P is the sensor monitoring coverage, with a value between 0 and 1; ω1 and ω2 are weight parameters; Select the initial population. Each individual is usually encoded in the form of a binary string or integer array, representing the deployment status of sensors at different locations. Based on the probabilistic coverage model of sensor monitoring capabilities, the requirements of different damaged areas of the structure are considered. While meeting the reliability requirements, the required monitoring capacity of each monitoring point under different environmental conditions is quantified as the optimization boundary to guide the selection of the initial population. The objective function is used to comprehensively evaluate the coverage monitoring probability reliability and the number of sensors. The multi-objective optimization iterative process integrates the sensor collaborative monitoring probability reliability and eliminates redundant deployment. The number of sensors and deployment locations of the monitoring system are gradually determined. After iteration, the Pareto optimal solution set is obtained, which is used to characterize the optimal deployment location of sensors on a specified road section.

7. The sensor network deployment and self-starting implementation method for long-term service performance monitoring of highway subgrade and pavement according to claim 1 is characterized in that: In step S5, the steps of implementing the self-wake-up control of the monitoring sensor based on the game theory model are as follows: S51, establish the sensor energy consumption model, the total energy consumption C of the i-th sensor i Calculate according to the following formula: C i =E s (S i )+E p (S i )+E t (S i ) Among them, E p (S i ) represents the sensing energy of the i-th sensor, E s (S i ) represents the processing energy of the i-th sensor, E t (S i ) represents the communication energy of the i-th sensor, S i is the i-th sensor in the sensor network; In step S52, the switching of the sensor between the awake state and the dormant state is considered as a game problem. The payoff function is defined based on the number of data successfully transmitted by the sensor in the awake state. The sensor node comprehensively considers the payoff function and the amount of energy consumed to make a trade-off. The game space is: {from awake to dormant, from awake to not dormant}. The game model GT of the game player is defined as follows: GT={N,K,{u i }} Where N represents the game party, which is all sensor nodes in the network; K is the strategy space adopted; u i is the utility function; Iterative game model, in this process, the dynamic sleep threshold of the sensor node is determined based on the utility function. In one round of game, when the utility function value is greater than the sleep threshold, the sensor node remains awake; when the utility function value is less than the sleep threshold, the sensor enters the sleep state; S53 introduces a sensor node penalty mechanism. When a sensor node chooses to enter a dormant state to save its own energy consumption, it is called selfish behavior of the sensor node. When a sensor node adopts selfish behavior, the sensor network will mark the node and force the sensor node to stay awake in the following M rounds. The M-th round game is defined as follows: u i (s i ,s -i )=U i (s i ,s -i )-C i (s i ,s -i ) Among them, U i (s i ,s -i ) is the payoff function of the i-th sensor game in the sensor network; C i (s i ,s -i ) is the energy consumption function of the ith sensor game in the sensor network; s i The decision to wake up the i-th sensor in the sensor network; s -i Take the sleep decision for the i-th sensor in the sensor network.

8. The sensor network deployment and self-starting implementation method for long-term service performance monitoring of highway subgrade and pavement according to claim 1 is characterized in that: In step S6, the implementation process of establishing a highway long-term service performance risk early warning mechanism is as follows: After determining the layout position and self-starting conditions, according to the monitoring data: vertical stress σ z , stress σ along the vehicle travel direction x , stress perpendicular to the vehicle's direction of travel σ y , vertical strain ε z , strain ε along the vehicle's travel direction x , strain perpendicular to the vehicle's direction of travel ε y , the temperature T and humidity H of each structural layer, and use DNN to inversely calculate the pavement structure modulus E as a characterization indicator of the health of the roadbed and pavement structure. The inverse calculation process is as follows: Where, is the structural modulus output prediction vector, E i is the structural modulus of the i-th layer of the pavement structure; A=[σ z ,σ x ,σ y ,ε z ,ε x ,ε y ,T,H]; θ is the parameter in the prediction model, including weights and biases; The structural strength and bearing capacity of each structural layer of the roadbed and pavement are evaluated by the changes in the structural modulus E of each layer; Use automatic data collection technology to monitor and record the traffic and environmental parameters of a designated road section in real time, and evaluate the traffic and environmental conditions of the road section; Combining monitoring data with traffic environment models, we can predict the downward trend in the bearing capacity of the roadbed and pavement structure, the location and development trend of cracks, voids, interlayer delamination, and subsidence, and ultimately achieve real-time early warning of the long-term service performance risks of the highway.

Images