Road health prediction method and device

Abstract

The invention provides a road health prediction method and device, and the method comprises the steps: 1, collecting road surface data and non-road surface data, and carrying out the preprocessing of the road surface data and the non-road surface data; 2, based on a pavement structure performance evaluation model and a pavement safety performance evaluation model, cleaning the pavement data and the non-pavement data; and step 3, based on a dynamic model of a recurrent neural network, inputting the cleaned road surface data and non-road surface data, and predicting the change condition of the road surface performance index in the whole life cycle. According to the method, road abnormal values, missing values, non-road traffic volume data, micro-meteorological data and maintenance data are integrated, a machine learning method is used, firstly, multi-source data are preprocessed, and then based on a road structure performance evaluation model and a road safety performance evaluation model, a multi-source heterogeneous big data model is proposed, so that the road safety performance is evaluated. Therefore, more accurate and comprehensive road health prediction can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of road network maintenance, and in particular to a highway health prediction method and device. Background Art

[0002] Highway disease prediction is a key technology to ensure road safety and extend road service life. Commonly used highway disease prediction methods mainly include empirical and statistical models, digital technology (such as drones, lidar), and multi-source data fusion technology.

[0003] The empirical and statistical model summarizes and generalizes the occurrence patterns of diseases through historical data and field investigations, and establishes a prediction model suitable for different regions and road conditions. This method relies on a large amount of historical data and statistical analysis, and can more accurately predict the time and extent of disease occurrence.

[0004] Digital technologies, such as drones and LiDAR, can be used for comprehensive and efficient inspections of highways, improving the detection and prediction accuracy of road surface defects. Drones can quickly cover large areas, and LiDAR can provide high-precision three-dimensional data to help detect subtle signs of defects.

[0005] Multi-source data fusion technology uses multiple data sources such as satellite remote sensing, mobile sensors and traffic monitoring data to conduct comprehensive assessment and prediction of road conditions, thereby improving the accuracy and reliability of predictions.

[0006] However, although these methods have improved the accuracy of highway disease prediction to a certain extent, they still have shortcomings. Several current methods mainly focus on the state of the road surface itself, but pay less attention to road surface anomalies, missing values, and do not consider the impact of non-road surface factors on highway health, such as traffic volume data, micro-meteorological data, and maintenance data. In addition, existing prediction models often lack effective data preprocessing and fusion mechanisms when processing multi-source heterogeneous big data, resulting in limited accuracy and generalization of prediction results. Summary of the invention

[0007] The present invention provides a highway health prediction method and device, which comprehensively considers multi-source data of road surface and non-road surface, and performs data preprocessing and model construction through advanced technologies such as machine learning to achieve more accurate and comprehensive highway health prediction.

[0008] The present invention provides a highway health prediction method and device, comprising: Step 1, collecting road surface data and non-road surface data, and preprocessing the road surface data and non-road surface data; Step 2, based on the pavement structure performance evaluation model and the pavement safety performance evaluation model, cleaning the pavement data and the non-pavement data; Step 3, based on the dynamic model of the recurrent neural network, input the cleaned road surface data and non-road surface data to predict the changes of road surface performance indicators over the entire life cycle.

[0009] Wherein, the step 1 further comprises: Collecting road surface data including road surface damage conditions, road surface roughness index and / or road surface images through equipment including high-definition cameras, laser rangefinders and / or roughness sensors; Non-road data including traffic volume data, micro-meteorological data and / or maintenance data are collected through equipment including strain sensors, vibration sensors and / or temperature sensors.

[0010] Wherein, the step 1 further comprises: Deleting repeated occurrences of the road surface data; Using mean interpolation method, nearest neighbor interpolation method, regression interpolation method, linear interpolation method and / or curve interpolation method to fill in the missing road surface data; A statistical hypothesis testing method, a clustering algorithm, a classification method and / or an isolation forest algorithm are used to detect and exclude abnormal data in the road surface data or the non-road surface data.

[0011] Wherein, the step 2 further comprises: Constructing a linearly separable sample set of the road surface data and / or the non-road surface data on a hyperplane; Through the Lagrangian function, the sample set with the largest classification interval is obtained; Based on the support vectors corresponding to the samples in the sample set, the road surface data and / or non-road surface data of different categories are distinguished through a classification decision function.

[0012] Wherein, the step 2 further comprises: When the training set is linearly inseparable, the training samples are mapped to a high-dimensional feature space through a nonlinear mapping function based on the introduction of penalty factors and slack variables; Construct a hyperplane in the high-dimensional feature space, use the kernel function for linear classification, and obtain a linearly separable sample set.

[0013] Wherein, the step 3 further comprises: By principal component analysis, key parameters of the road surface data and non-road surface data are focused on; Carry out single-index and multi-index prediction to obtain the changes of pavement performance indicators throughout the life cycle.

[0014] The present invention also provides a highway health prediction device, comprising: A collection module, the collection module is used to collect road data and non-road data, and pre-process the road data and non-road data; A cleaning module, the cleaning module is used to clean the pavement data and non-pavement data based on a pavement structure performance evaluation model and a pavement safety performance evaluation model; A prediction module is used to predict the changes of pavement performance indicators over the entire life cycle based on a dynamic model of a recurrent neural network, input the cleaned pavement data and non-pavement data.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the computer program implements any of the above-mentioned highway health prediction methods.

[0016] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned highway health prediction methods.

[0017] The present invention provides a highway health prediction method and device, which uses machine learning to preprocess multi-source heterogeneous data, builds a road performance evaluation model, and realizes accurate and comprehensive highway health prediction. The method and device of the present invention are not limited to the state of the road surface itself, but also integrate road surface anomalies and missing values, and at the same time include multi-source heterogeneous data such as traffic volume data, micro-meteorological data and maintenance data.

[0018] Specifically, the method and device of the present invention construct a comprehensive multi-source heterogeneous big data model based on the pavement structure performance evaluation model and the pavement safety performance evaluation model. The model can comprehensively consider the physical state, traffic conditions, climate conditions and maintenance history of the pavement, thereby providing a comprehensive pavement performance evaluation, which can not only improve the efficiency and effect of pavement maintenance, but also provide more scientific and accurate decision support for road management departments. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 It is a flow chart of the highway health prediction method provided by the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] Highway health prediction refers to monitoring and analyzing the health status of highways, predicting their performance changes and potential damage risks, so as to ensure traffic safety and extend the service life of highways. The existing main methods for highway health prediction include empirical and statistical models, digital technology and multi-source data fusion technology.

[0023] Highway health prediction is crucial to ensuring traffic safety and extending the life of roads. It plays a significant role in preventing traffic accidents and reducing maintenance costs, and is an important part of improving the intrinsic safety of road traffic. However, current highway health prediction methods mostly focus on the state of the road surface itself, ignoring key variables such as environmental factors, road network information, and maintenance data. This limits the comprehensiveness and accuracy of the prediction results, and a more comprehensive evaluation method is urgently needed.

[0024] In general, the present invention can achieve accurate and comprehensive highway health prediction through a highway health prediction method that uses machine learning to preprocess multi-source heterogeneous data and construct a pavement performance evaluation model. The method of the present invention constructs a comprehensive multi-source heterogeneous big data model based on the pavement structure and safety performance evaluation model. The data source is not limited to the pavement itself, but also comprehensively considers the pavement physical state, traffic, meteorology and maintenance data, thereby greatly improving the comprehensiveness and accuracy of the prediction. In addition, during the information preprocessing process, key information is extracted, data interference is eliminated, and features that significantly affect pavement performance are retained to improve prediction accuracy.

[0025] Among them, in the method of the present invention, machine learning technology is used to pre-process these multi-source heterogeneous big data, and key information that has a significant impact on pavement performance evaluation is extracted from the massive data. Through this step, the interference caused by missing values ​​and outliers can be eliminated, while retaining data features that have a significant impact on pavement performance, thereby improving the accuracy of highway health prediction.

[0026] Figure 1 FIG. 1 is a flow chart of the highway health prediction method of the present invention. Figure 1As shown, the present invention provides a highway health prediction method, including: step 1, collecting road surface data and non-road surface data, and preprocessing the road surface data and non-road surface data; step 2, cleaning the road surface data and non-road surface data based on a road surface structure performance evaluation model and a road surface safety performance evaluation model; step 3, based on a dynamic model of a recurrent neural network, inputting the cleaned road surface data and non-road surface data, and predicting the changes of road surface performance indicators over the entire life cycle.

[0027] Wherein, the step 1 further comprises: Collecting road surface data including road surface damage conditions, road surface roughness index and / or road surface images through equipment including high-definition cameras, laser rangefinders and / or roughness sensors; Non-road data including traffic volume data, micro-meteorological data and / or maintenance data are collected through equipment including strain sensors, vibration sensors and / or temperature sensors.

[0028] Among them, the road data collection equipment, such as high-definition cameras, laser rangefinders and flatness sensors, are usually installed at key locations on highways, such as bridges, tunnels, intersections and representative sections. These devices can capture the damage condition, flatness index and road image of the road in real time, providing intuitive and accurate data support for highway health detection. High-definition cameras can capture road cracks, potholes and other damage conditions, laser rangefinders can accurately measure the flatness of the road, and flatness sensors can further refine the slight changes in the road surface.

[0029] Among them, the non-road data acquisition equipment, such as strain sensors, vibration sensors and temperature sensors, are more often installed on highway infrastructure or vehicles. Strain sensors and vibration sensors are often installed in key parts of structures such as bridges and tunnels to monitor their stress state and vibration. Temperature sensors may be installed under the road surface or in the surrounding environment to monitor micro-meteorological changes in the road, such as temperature and humidity. At the same time, some sensors can also be installed on maintenance vehicles to collect data in real time during the maintenance process.

[0030] Wherein, the step 1 further comprises: Deleting repeated occurrences of the road surface data; Using mean interpolation method, nearest neighbor interpolation method, regression interpolation method, linear interpolation method and / or curve interpolation method to fill in the missing road surface data; A statistical hypothesis testing method, a clustering algorithm, a classification method and / or an isolation forest algorithm are used to detect and exclude abnormal data in the road surface data or the non-road surface data.

[0031] The duplicate data refers to the data of damage condition of a road at a certain point in time being recorded twice or more due to sensor failure or data recording error when collecting road surface data. In this case, these data are duplicate data.

[0032] For example, two records show that the width of cracks on a road is 3 mm at the same time point, and other relevant fields (such as location, date, etc.) are also exactly the same.

[0033] The missing data refers to that when collecting road surface data, if the rutting depth data at a certain time point is not recorded, then the value of this field is missing.

[0034] For example, a record may show the crack width and pavement roughness index for a road at a point in time, but the rutting depth field may be empty or marked as "unknown".

[0035] The abnormal data refers to when collecting pavement data, if the crack width data at a certain time point suddenly becomes abnormally large or abnormally small, and there is a significant difference compared with the data at other time points, then these data may be abnormal data.

[0036] For example, a record shows that the crack width of a certain road suddenly increased to 100 mm at a certain point in time (while normally it is only a few millimeters), while other related fields (such as traffic flow, meteorological conditions, etc.) do not show corresponding abnormal changes.

[0037] Among them, the detection methods of abnormal data include statistical hypothesis testing, clustering algorithms, classification methods, and isolation forest algorithms.

[0038] Among them, statistical hypothesis testing is to put forward some hypotheses about the unknown or partially unknown population under study, and to test which hypothesis should be accepted in a probability sense based on the actual results of the sample after certain calculations. This is called statistical hypothesis testing.

[0039] For example, assuming that the road surface roughness data obeys a Gaussian distribution, the μ and σ of the data are calculated respectively, and then the μ±3 σ interval is obtained. The data points that are not in this interval can be considered as abnormal data.

[0040] Among them, clustering algorithm is to study the classification method based on the characteristics of a large amount of data or samples, and reasonably classify the data according to this classification method, and finally classify similar data into a group, that is, "same type, different type". For example, DBSCAN algorithm, K-Means clustering method, BP neural network method.

[0041] Among them, the classification method is to judge and divide according to the established program mode and standards. For example, the MOne ClassSV algorithm is an unsupervised classification algorithm that works better in solving the classification of unbalanced samples.

[0042] Among them, the Isolation Forest algorithm is based on Ensemble and can efficiently detect anomalies with linear time complexity. The data space is continuously cut by random hyperplanes until only one data point remains in each subspace. High-density clusters need to be cut multiple times, while low-density points are quickly isolated to subspaces. This algorithm processes large data quickly and accurately.

[0043] Wherein, the step 2 further comprises: Constructing a linearly separable sample set of the road surface data and / or the non-road surface data on a hyperplane; Through the Lagrangian function, the sample set with the largest classification interval is obtained; Based on the support vectors corresponding to the samples in the sample set, the road surface data and / or non-road surface data of different categories are distinguished through a classification decision function.

[0044] For the road data and / or non-road data sample set, a hyperplane is constructed so that all samples belonging to a certain category are on one side of the hyperplane, and all samples belonging to another category are on the other side of the hyperplane. And an optimization algorithm is used to find the sample set that maximizes the classification interval (i.e., the sum of the shortest distances from the two categories of sample points to the classification hyperplane).

[0045] In which, it is assumed that the multidimensional space is , the general form of the hyperplane is as follows: Among them, variable x refers to the sample data value, b refers to the spatial offset correction value, which is generated by the nonparametric test results, and w refers to the linear reconstruction coefficient (weight) value, which is generated by the average intrinsic correlation matrix of pavement performance indicators.

[0046] Normalize the above formula to make the sample set linearly separable satisfy , at this time, the classification interval is , if the classification interval is the largest, the following optimization problem must be satisfied: The Lagrangian function is used to transform the above optimization problem into a dual problem, namely: Among them, variables x and y refer to the spatial coordinates of sample data, n refers to the spatial dimension, i refers to the cumulative index, and t refers to the multidimensional space number. For each sample corresponding to the Lagrange multiplier, solve the above quadratic optimization problem, the optimal solution is obtained , select A positive component of To calculate , positive component The corresponding sample is the support vector, and the classification decision function is: Wherein, the step 2 further comprises: When the training set is linearly inseparable, the training samples are mapped to a high-dimensional feature space through a nonlinear mapping function based on the introduction of penalty factors and slack variables; Construct a hyperplane in the high-dimensional feature space, use the kernel function for linear classification, and obtain a linearly separable sample set.

[0047] Among them, the penalty factor plays a role in controlling the complexity of the model and balancing overfitting and underfitting in the recurrent neural network. It indicates the degree of emphasis on the loss caused by outliers (i.e., samples that do not meet the hard interval constraints). When the penalty factor takes a larger value, the model will pay more attention to outliers and try to classify them correctly. This may lead to better performance of the model on the training set, but the generalization performance may decrease because the model may be too complex and sensitive to noise points. When the penalty factor takes a smaller value, the model pays less attention to outliers and allows some samples not to meet the hard interval constraints. This helps to reduce the complexity of the model and improve the generalization performance, but may result in slightly worse performance on the training set.

[0048] Wherein, the slack variable is used to construct a soft margin classifier in the recurrent neural network, allowing the classifier to make mistakes on some samples, that is, allowing some samples not to meet the hard margin constraint. The slack variable indicates the degree of sample outlier. When the slack variable is zero, it means that the sample is not outlier and meets the hard margin constraint. When the slack variable increases, it means that the sample is farther outlier, that is, the distance between the sample and the classification hyperplane is smaller, and it may even be misclassified. By introducing slack variables, the recurrent neural network algorithm can handle linear inseparable problems and avoid overly complex and overfitting models. At the same time, the introduction of slack variables also makes the classification boundary smoother and improves the generalization performance of the model.

[0049] Among them, when the training set is linearly inseparable, a penalty factor and a slack variable are added, and a nonlinear mapping function is introduced. : , where variable n refers to the feature space, mapping the training samples of the input space to the high-order feature space ,exist The hyperplane is constructed in the training algorithm using the inner product in the feature space , where variable x refers to sample data, i, j refers to spatial coordinates; kernel function is introduced , so that the kernel function satisfies the Mercer condition and realizes linear classification of nonlinear transformation. Commonly used kernel functions include linear kernel functions , polynomial kernel function and radial basis kernel function , where the variable σ is the mean exponent, calculated from the correlation matrix.

[0050] Wherein, the step 3 further comprises: By principal component analysis, key parameters of the road surface data and non-road surface data are focused on; Single and multiple indicators are predicted, and the changes of pavement performance indicators over the entire life cycle are obtained based on recurrent neural networks.

[0051] In highway health prediction, the recurrent neural network can learn the changing trends of pavement performance indicators over time and make predictions based on these trends. The dynamic model can handle the dynamic changes of input data, including fluctuations in traffic flow, changes in meteorological conditions, etc. Through the application of the dynamic model, the prediction module can more accurately predict the changes of pavement performance indicators in the future.

[0052] Among them, the principal component analysis (PCA) is a widely used dimensionality reduction method. Based on the idea of ​​maximizing the variance of data, the original high-dimensional data is mapped to a low-dimensional space through linear transformation, while retaining the main change patterns in the data as much as possible. These change patterns are the principal components of the data, which are linear combinations of the original variables, sorted by variance, and can represent most of the information in the data.

[0053] Among them, in highway health prediction, the key parameters of pavement data and non-pavement data can be focused on through principal component analysis. Specifically, PCA compares the accuracy of actual accuracy after training with historical data samples to identify which variables (i.e. key parameters) have a significant impact on the prediction results. In this way, more attention is paid to these variables in subsequent analysis and prediction. This helps to reduce the dimension of the data and improve the efficiency of analysis, while retaining the main information in the data, providing a more accurate and reliable basis for single-indicator and multi-indicator predictions of highway health.

[0054] Among them, the pavement performance indicators are collected from a number of maintenance experience data, and the pavement performance evaluation model is used for data cleaning to form multiple training data sets and train the prediction model. After the training is completed, the optimal parameters of the model are selected according to the accuracy of the test samples to optimize the model performance and ensure the accuracy of the model prediction.

[0055] The present invention also provides a highway health prediction device, comprising: an acquisition module, the acquisition module is used to collect road surface data and non-road surface data, and pre-process the road surface data and non-road surface data; a cleaning module, the cleaning module is used to clean the road surface data and non-road surface data based on a road surface structure performance evaluation model and a road surface safety performance evaluation model; a prediction module, the prediction module is used to input the cleaned road surface data and non-road surface data based on a dynamic model of a recurrent neural network, and predict the changes in road surface performance indicators over the entire life cycle.

[0056] Among them, in the acquisition module, the road surface data includes road damage conditions (such as cracks, potholes, ruts, etc.), road surface flatness index, road surface material characteristics, etc. These data are usually obtained through high-definition cameras, laser rangefinders, displacement sensors, road surface material analyzers and other equipment. Non-road surface data includes traffic flow, vehicle type, driving speed, meteorological conditions (such as temperature, humidity, rainfall), maintenance history, etc. These data can be collected through traffic flow monitors, weather stations, vehicle identification systems, maintenance record systems and other equipment.

[0057] Among them, the acquisition module is also used to: pre-process the original data, including data cleaning (removing noise and outliers), data filling (processing missing values), data standardization (unifying data format and units) and other steps to ensure the accuracy and consistency of the data and provide high-quality data input for subsequent data cleaning and prediction modules.

[0058] Among them, in the cleaning module, the pavement structure performance evaluation model can evaluate the structural performance of the pavement by analyzing factors such as the mechanical properties of the pavement material, the thickness of the structural layer, and the drainage performance. The pavement safety performance evaluation model can evaluate the safety performance of the pavement by factors such as the pavement flatness, anti-skid performance, and visibility.

[0059] Among them, in the cleaning module, the output of the cleaning module is a high-quality data set that has been deeply cleaned and screened.

[0060] Among them, in the prediction module, the recurrent neural network is a special neural network structure that can capture the time series characteristics in the data. In highway health prediction, the recurrent neural network can learn the changing trends of pavement performance indicators over time and make predictions based on these trends. The dynamic model can handle the dynamic changes of input data, including fluctuations in traffic flow, changes in meteorological conditions, etc. Through the application of the dynamic model, the prediction module can more accurately predict the changes in pavement performance indicators in the future.

[0061] It can be seen that the prediction method proposed in this invention, which uses machine learning to preprocess multi-source heterogeneous data and construct a pavement performance evaluation model, is not only a comprehensive transcendence of the traditional strategy of "focusing on the state of the pavement itself", but also a significant improvement in technology utilization and a profound enhancement of prediction accuracy. Through intelligent integration and flexible deployment of various technologies and data resources, this method has brought unprecedented comprehensiveness and accuracy to the field of highway health prediction.

[0062] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute a highway health prediction method provided by the above methods, and the method includes: step 1, collecting pavement data and non-pavement data, and preprocessing the pavement data and non-pavement data; step 2, based on a pavement structure performance evaluation model and a pavement safety performance evaluation model, cleaning the pavement data and non-pavement data; step 3, based on a dynamic model of a recurrent neural network, inputting the cleaned pavement data and non-pavement data, and predicting the changes in pavement performance indicators over the entire life cycle.

[0063] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it is implemented to execute a highway health prediction method provided by the above-mentioned methods. The method includes: step 1, collecting pavement data and non-pavement data, and preprocessing the pavement data and non-pavement data; step 2, cleaning the pavement data and non-pavement data based on a pavement structure performance evaluation model and a pavement safety performance evaluation model; step 3, based on a dynamic model of a recurrent neural network, inputting the cleaned pavement data and non-pavement data, and predicting changes in pavement performance indicators over the entire life cycle.

[0064] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0065] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0066] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. 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 highway health prediction method, characterized in that: include: Step 1, collecting road surface data and non-road surface data, and pre-processing the road surface data and non-road surface data; Step 2, based on the pavement structure performance evaluation model and the pavement safety performance evaluation model, cleaning the pavement data and the non-pavement data; Step 3, based on the dynamic model of the recurrent neural network, the cleaned road surface data and non-road surface data are input to predict the changes in road surface performance indicators over the entire life cycle.

2. The method according to claim 1, characterized in that The step 1 further comprises: Collecting road surface data including road surface damage conditions, road surface roughness index and / or road surface images through equipment including high-definition cameras, laser rangefinders and / or roughness sensors; Non-road data including traffic volume data, micro-meteorological data and / or maintenance data are collected through equipment including strain sensors, vibration sensors and / or temperature sensors.

3. The method according to claim 1, characterized in that The step 1 further comprises: Deleting repeated occurrences of the road surface data; Using mean interpolation method, nearest neighbor interpolation method, regression interpolation method, linear interpolation method and / or curve interpolation method to fill in the missing road surface data; A statistical hypothesis testing method, a clustering algorithm, a classification method and / or an isolation forest algorithm are used to detect and exclude abnormal data in the road surface data or the non-road surface data.

4. The method according to claim 1, characterized in that: The step 2 further comprises: Constructing a linearly separable sample set of the road surface data and / or the non-road surface data on a hyperplane; Through the Lagrangian function, the sample set with the largest classification interval is obtained; Based on the support vectors corresponding to the samples in the sample set, the road surface data and / or non-road surface data of different categories are distinguished through a classification decision function.

5. The method according to claim 4, characterized in that The step 2 further comprises: When the training set is linearly inseparable, the training samples are mapped to a high-dimensional feature space through a nonlinear mapping function based on the introduction of penalty factors and slack variables; Construct a hyperplane in the high-dimensional feature space, use the kernel function for linear classification, and obtain a linearly separable sample set.

6. The method according to claim 1, characterized in that The step 3 further comprises: By principal component analysis, key parameters of the road surface data and non-road surface data are focused on; Carry out single-index and multi-index prediction to obtain the changes of pavement performance indicators throughout the life cycle.

7. A highway health prediction device, characterized in that: include: A collection module, the collection module is used to collect road data and non-road data, and pre-process the road data and non-road data; A cleaning module, the cleaning module is used to clean the pavement data and non-pavement data based on a pavement structure performance evaluation model and a pavement safety performance evaluation model; A prediction module is used to predict the changes of pavement performance indicators over the entire life cycle based on a dynamic model of a recurrent neural network, input the cleaned pavement data and non-pavement data.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, a highway health prediction method as described in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, a highway health prediction method as described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, a highway health prediction method as described in any one of claims 1 to 6 is implemented.

Images