Method and system for identifying key factors of road surface behavior based on feature perturbation

Abstract

The present disclosure provides a feature disturbance-based pavement performance key factor identification method and system, relating to the technical field of performance factor identification, comprising obtaining the characteristic performance and geometric parameters of the road pavement, and constructing a scaled model of the road pavement; obtaining a plurality of performance influencing factors that affect the road pavement, inputting the performance influencing factors into a neural network model for extreme feature disturbance, adjusting the values within the range of [0, 1] by setting the maximum and minimum values, observing the changes in the output results of the neural network model, and preliminarily determining the key factors that have a great influence on the output results of the neural network model; repeatedly inputting the obtained key factors into the neural network model and verifying the accuracy of the selected key factors in a probabilistic manner. The present disclosure can accurately obtain the performance key factors generated by the change of the pavement.

Description

Technical Field

[0001] This disclosure relates to the field of performance factor identification technology, specifically to a method and system for identifying key road performance factors based on feature perturbation. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] In road construction and maintenance, accurately identifying key performance factors of the pavement is crucial for assessing pavement condition and developing maintenance strategies. Key performance factors refer to sensitive indicators that best reflect the service performance of infrastructure, such as stress and top tension. "Factor" is a general term encompassing all similar stress and top tension indicators.

[0004] Currently, traditional methods for identifying road surface behavior influencing factors mostly rely on data collection from actual roads. However, this approach is limited by data collection conditions, making it difficult to conduct relevant experiments on real road surfaces. This makes it difficult to trace back certain behaviors in real-world scenarios and further select more critical influencing factors. At the same time, the high cost, long cycle, and difficulty in data acquisition hinder its promotion and application in practice. On the other hand, the calculation methods of indices such as PCL are mainly based on expert experience, and the importance of various factor weights lacks theoretical explanation. Summary of the Invention

[0005] To address the aforementioned issues, this disclosure proposes a method and system for identifying key factors of pavement performance based on feature perturbation. By establishing a scaled model and corresponding pavement influencing factors, introducing a neural network model, and perturbing the influencing factors with features, the weight of each performance influencing factor is determined, thereby accurately identifying key factors of pavement performance.

[0006] According to some embodiments, the present disclosure adopts the following technical solutions:

[0007] Methods for identifying key factors of pavement behavior based on feature perturbations include:

[0008] Obtain the characteristic behavior and geometric parameters of the road surface;

[0009] A scaled-down model of the road surface is constructed based on the aforementioned characteristic properties and geometric parameters.

[0010] Based on the scaled-down model, multiple behavior factors that affect the road surface are obtained and the data is normalized.

[0011] Train the neural network model, input the behavior influencing factors into the neural network model to perform extreme feature perturbation, and adjust the values ​​within the range of [0,1] by setting the maximum and minimum values, observe the changes in the output of the neural network model, and preliminarily determine the factors that have a great impact on the output of the neural network model as key factors;

[0012] The acquired key factors are repeatedly input into the neural network model, and the accuracy of the selected key factors is verified using a probabilistic approach.

[0013] According to some embodiments, the present disclosure adopts the following technical solutions:

[0014] A key factor identification system for road surface behavior based on feature perturbation includes:

[0015] The data acquisition module is used to acquire the characteristic properties and geometric parameters of the road surface;

[0016] The preprocessing module is used to construct a scaled-down model of the road surface based on the characteristic properties and geometric parameters; based on the scaled-down model, it obtains multiple property influencing factors that affect the road surface and performs data normalization processing.

[0017] The key factor identification module is used to train the neural network model. It inputs the behavior influencing factors into the neural network model to perform extreme feature perturbation. By setting the maximum and minimum values, it makes numerical adjustments within the range of [0,1] and observes the changes in the output of the neural network model. The factors that have a large impact on the output of the neural network model are initially identified as key factors.

[0018] The acquired key factors are repeatedly input into the neural network model, and the accuracy of the selected key factors is verified using a probabilistic approach.

[0019] Compared with the prior art, the beneficial effects of this disclosure are as follows:

[0020] The proposed method for identifying key factors of pavement behavior based on feature disturbances utilizes 3D printing to construct scaled-down models of the same material for testing. Compared to testing on actual pavement, this method has lower cost and time requirements, and allows for systematic research on different feature disturbances.

[0021] This disclosure employs a neural network method, which can establish a mapping relationship between the training response of the model and the influencing factors of road surface features. It can comprehensively and accurately identify key factors, solving the limitations and inaccuracies caused by traditional manual feature extraction methods, and providing an important basis for road maintenance.

[0022] This disclosure introduces an extreme feature perturbation method, and for the first time proposes an input data processing method that transforms the maximum and minimum feature values ​​based on the numerical values ​​of the input feature influence factors, to explore the model output results under extreme feature perturbation and preliminarily screen out key road surface influence factors.

[0023] This disclosure employs a probabilistic method, which involves inputting the same data into the model multiple times, observing the output of the neural network model each time, and further determining the accuracy of the proposed key factors by comparing the posterior probability values. Attached Figure Description

[0024] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0025] Figure 1 This is a schematic diagram of the overall process of an embodiment of this disclosure;

[0026] Figure 2 This is a schematic diagram of the pavement performance influencing factor database and coding according to an embodiment of the present disclosure;

[0027] Figure 3 This is a schematic diagram illustrating the training and input / output of a neural network model according to an embodiment of this disclosure. Detailed Implementation

[0028] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0029] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0030] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0031] Terminology Explanation

[0032] Scaled-down model: refers to a model created in research or experimentation by scaling down the actual object or system while retaining its key features and behaviors.

[0033] 3D printing is an additive manufacturing technology that first selects relevant materials and then creates three-dimensional objects by stacking the materials layer by layer. It typically uses model files created using computer-aided design (CAD).

[0034] Example 1

[0035] One embodiment of this disclosure provides a method for identifying key factors of road surface behavior based on feature perturbation, including:

[0036] Step 1: Obtain the characteristic behavior and geometric parameters of the road surface;

[0037] Step 2: Construct a scaled-down model of the road surface based on the described characteristic properties and geometric parameters;

[0038] Step 3: Based on the scaled-down model, obtain multiple behavior factors that affect the road surface and perform data normalization processing;

[0039] Step 4: Train the neural network model. Input the behavior influencing factors into the neural network model to perform extreme feature perturbation. By setting the maximum and minimum values, adjust the values ​​within the range of [0,1] and observe the changes in the output of the neural network model. Preliminarily identify the factors that have a significant impact on the output of the neural network model as key factors.

[0040] Step 5: Repeatedly input the obtained key factors into the neural network model, and verify the accuracy of the selected key factors using a probabilistic approach.

[0041] As one embodiment, the overall idea of ​​the pavement behavior key factor identification method based on feature perturbation disclosed herein is as follows:

[0042] This disclosure utilizes 3D printing technology to construct scaled-down models. Based on the material mix ratio of road surfaces (asphalt and cement), scaled-down models of the same type are constructed at equal scale, effectively achieving experimental results with small-scale models. Deep learning-related technologies are employed to process, classify, train, and analyze feature data collected by sensors, ultimately establishing a mapping relationship between the scaled-down model under extreme feature perturbations and key performance factors, and building a complete CNN convolutional neural network model. Extreme feature perturbations are introduced into the method for identifying key road surface performance factors in the scaled-down model. Road surface performance influencing factors are collected based on the scaled-down model, and different feature perturbations are applied to the collected performance factors. Changes in relevant model data indicators are observed, achieving small-scale, high-efficiency, and high-precision indoor monitoring of key road surface factor data, initially screening out key road surface influencing factors. Further determination of whether a factor is a key factor involves repeatedly inputting data and observing the output results of the network model each time. The total likelihood of the output level is calculated, and the posterior probability of each output level is estimated using Bayesian network principles. Finally, by comparing the posterior probability values ​​of the levels, it is further determined whether the performance influencing factor is a key influencing factor.

[0043] As one embodiment, the specific implementation process of the pavement behavior key factor identification method based on feature perturbation disclosed herein is as follows:

[0044] Step 1: Obtain the characteristic properties and geometric parameters of the road surface, and construct a scaled-down model of the road surface based on the characteristic properties and geometric parameters. This includes: obtaining the characteristic properties and geometric parameters of the actual road surface by consulting relevant data and conducting on-site investigations of the corresponding road surface; using 3D printing technology to print scaled-down models with different material ratios and 3D printing parameter combinations to ensure that the scaled-down model has similar geometric shape, texture and material properties to the actual road surface.

[0045] Preferably, the details of the method for constructing the scaled model of the road surface in step 1 are as follows:

[0046] S1-1: In the scaled-down model construction step, by consulting relevant data and conducting on-site investigations of the corresponding road surfaces, suitable materials and dimensions are found and confirmed.

[0047] S1-2: Use 3D printing technology or other suitable methods to create a scaled-down model. This model should have similar geometry, texture and material properties to the actual road surface identified in S1-1, ensuring that the model can simulate the response and behavior of the actual road surface, thereby determining the road surface behavior influencing factors.

[0048] Step 2: Based on the scaled model, obtain multiple behavior factors that affect the road surface and perform data normalization.

[0049] Based on the scaled-down road surface model constructed in step 1, sensors are used to collect data on road surface performance factors, including deflection, top tension, stress, displacement, vertical load, ambient humidity, wind speed, road surface smoothness, road surface friction coefficient, road surface texture, road surface structure, and degree of road surface damage. This ensures the dataset contains sufficient samples and diversity to cover different traffic scenarios and influencing factors. The quality and diversity of the dataset are crucial for accuracy and generalization performance.

[0050] Preferably, step 2 specifically includes:

[0051] S2-1: Collect data on factors influencing pavement properties To ensure that the dataset covers different traffic scenarios and conditions, n This indicates the types of input feature factors.

[0052] S2-2: Factors affecting pavement performance Data preprocessing includes data cleaning, outlier removal, and standardization to ensure data quality and consistency.

[0053] Step 3: Design and train the neural network model. Based on the characteristics of road surface condition influencing factors, design the structure of a convolutional neural network (CNN), including convolutional layers, pooling layers, and fully connected layers. This includes, but is not limited to, the structures of other neural network models such as CNN and Transformer, which also include convolutional layers, pooling layers, and fully connected layers. The model input values ​​are normalized 0-1 quantized values, and the model output consists of m classification results. Appropriate activation and loss functions are selected, and network parameters are set. The neural network model is trained using a road surface influencing factor database, and the network parameters are updated using the backpropagation algorithm, enabling the network to learn the correlation between road surface influencing factors and road surface conditions.

[0054] Preferably, in step 3, a neural network is used to train the dataset of road surface behavior influencing factors determined in step 2, as detailed below:

[0055] S3-1: The collected and labeled data are trained using deep learning algorithms. An appropriate deep learning model architecture is selected. Taking the CNN neural network as an example, the neural network automatically extracts features and learns the spatial relationships between features, automatically learning the key parts in the data, and improving the model's accuracy in identifying key factors. The activation function of the last layer of the neural network model is the Softmax function, as shown in Equation (1):

[0056] (1)

[0057] in, For the first The output value of each node, k This represents the number of output nodes, which indicates the number of categories. Denotes the base of the natural logarithm. This indicates the number of output influencing factor categories; finally, the Softmax function can convert the multi-class output values ​​into a range between [0,1], satisfying the output results of the neural network model in this disclosure.

[0058] The loss function used in this disclosure is the cross-entropy loss function, as shown in equation (2):

[0059] (2)

[0060] in, m This represents the total number of all impact factor categories. This represents the probability value predicted by the model. It is the probability of category node i in the actual target probability distribution.

[0061] S3-2: Employ appropriate parameter initialization methods, tuning and optimizing the model based on data characteristics and task requirements, such as random initialization, Xavier initialization, or He initialization, to avoid gradient vanishing and other problems. Deep learning models can learn the mapping relationship between model response and road surface features.

[0062] S3-3: Adjust the hyperparameters of the model. By adjusting parameters such as the learning rate and the number of iterations, the training process of the model can be optimized.

[0063] Step 4: Based on the neural network model trained in Step 3, apply extreme feature perturbation to the input behavior influencing factors after data normalization. By setting maximum and minimum values, adjust the values ​​within the range [0,1] and observe the changes in the output of the neural network model. If the maximum and minimum values ​​do not have a significant impact on the model output, it initially indicates that the behavior influencing factor is not a key factor; otherwise, it is a key factor.

[0064] Furthermore, extreme feature perturbation refers to making extreme or extreme changes or disturbances to features (usually input features) in the data, such as changing the maximum and minimum values ​​of the input values, to test the robustness and performance of the model under extreme conditions. This perturbation is often used to evaluate the model's behavior when faced with abnormal or unusual inputs.

[0065] Key pavement performance factors refer to critical factors that have a significant impact on the assessment of road pavement conditions and performance. These include, for example, pavement smoothness, pavement friction coefficient, pavement texture, pavement structure, and degree of pavement damage. In the fields of road engineering and transportation, the accurate identification and assessment of key pavement performance factors are of great significance for ensuring road safety, improving driving comfort, and extending pavement service life.

[0066] Preferably, step 4 mainly involves perturbing the influencing factors with extreme features after the scaled model is constructed and the neural network model is trained. The specific details are as follows:

[0067] S4-1: Normalize the input performance influencing factor data to obtain values ​​in the range of [0,1]. Based on the trained CNN neural network model, perform extreme feature perturbation on the input feature influencing factors. By adjusting the value of the feature influencing factors to the maximum and minimum values, such as adjusting the value to be infinitely close to 0 or 1, observe the output of the neural network model to initially screen out the key factors of road performance.

[0068] First, the dataset of behavior influence factors established based on S2-1. Establish the number of inputs for multi-feature influencing factors ,in This indicates the number of types of input behavior influencing factors.

[0069] S4-2: First, the input value of the original behavior influence factor is... ,exist Based on this, the range of its values ​​is adjusted by taking maximum and minimum values, which can be expressed as the following formula:

[0070] Adjustment of the maximum value:

[0071] (3)

[0072] Adjustment for the minimum value:

[0073] (4)

[0074] in, These represent the maximum and minimum values ​​after extreme perturbation, respectively.

[0075] Step 5: To further verify the accuracy of the initial selection of key influencing factors in Step 4, the feature influencing factors in Step 4 are repeatedly input, and the output results of the model are observed after each input. Probabilistic methods, including but not limited to voting count and Bayesian principle, are used to further determine the key factors.

[0076] Specifically, it can be expressed as:

[0077] First, the behavior influence factor of the input is defined as follows: , in The input count can be represented as 1-10; the output is m from step 3, for example: m=3, which means... ; among which settings .

[0078] Defined as observed in the single-input behavior influence factor. "、" "、" The probabilities of " are respectively expressed as , , These three probabilities can be estimated based on the actual output of the network. For example, it was observed that at the "excellent" level, The probability of " is 0.6, which can be expressed as =0.6.

[0079] Then, for each output result, the likelihood of the output result with a given probability can be calculated. Assume the ten observations are as follows: Each of them Indicates a single observation result (" "、" "、" ").for Three levels are used. The likelihood of each output result observed at each level is calculated, and then the likelihoods of multiple results are multiplied to obtain the total likelihood of this series of results. For example, a level can be represented as:

[0080] (5)

[0081] in, express Total likelihood at each level.

[0082] Calculate the total likelihood at three levels, respectively Finally, Bayes' principle is used to estimate the posterior probability of each level, which can be expressed as:

[0083] (6)

[0084] (7)

[0085] (8)

[0086] Finally, the three posterior probabilities are compared, and the rank with the highest posterior probability is selected as the final result. This method allows for considering the probabilities of different ranks across multiple observations, and then estimating the most likely rank based on the observations.

[0087] Through the implementation of the above steps, based on the construction of the scaled model and the training and monitoring of the neural network model, and by performing extreme feature perturbation processing on the input feature influencing factors, the key performance factors of the road surface can be effectively identified, and strong support can be provided for road surface maintenance and management.

[0088] Example 2

[0089] One embodiment of this disclosure provides a road surface behavior key factor identification system based on feature perturbation, including:

[0090] The data acquisition module is used to acquire the characteristic properties and geometric parameters of the road surface;

[0091] The preprocessing module is used to construct a scaled-down model of the road surface based on the characteristic properties and geometric parameters; based on the scaled-down model, it obtains multiple property influencing factors that affect the road surface and performs data normalization processing.

[0092] The key factor identification module is used to train the neural network model. It inputs the behavior influencing factors into the neural network model to perform extreme feature perturbation. By setting the maximum and minimum values, it makes numerical adjustments within the range of [0,1] and observes the changes in the output of the neural network model. The factors that have a large impact on the output of the neural network model are initially identified as key factors.

[0093] The acquired key factors are repeatedly input into the neural network model, and the accuracy of the selected key factors is verified using a probabilistic approach.

[0094] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0095] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0096] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A method for identifying key factors of pavement performance based on feature perturbation, characterized in that, The method comprises the following steps: Obtain the characteristic state and geometric parameters of the road pavement; Construct a scaled model of the road pavement in proportion according to the characteristic state and geometric parameters; Obtain a plurality of state influence factors affecting the road pavement based on the scaled model and perform data normalization processing; Train a neural network model, input the state influence factors into the neural network model to perform extreme feature disturbance, adjust the numerical value in the range of [0, 1] by setting the maximum value and the minimum value, and observe the change of the output result of the neural network model to preliminarily determine the key factors that have a great influence on the output result of the neural network model; Repeat inputting the obtained key factors into the neural network model and verify the accuracy of the selected key factors in a probabilistic manner; The method for preliminarily screening the road pavement state key factors by performing extreme feature disturbance on the input characteristic influence factors is as follows: Construct a state influence factor data set according to the obtained state influence factors, determine the input number of the multi-feature influence factors based on the state influence factor data set, input the state influence factors into the neural network model, and adjust the maximum and minimum values of the value range; Normalize the input state influence factor data to obtain a numerical value in the range of [0, 1], perform extreme feature disturbance on the input characteristic influence factors based on the CNN neural network model; According to the established behavior influence factor data set The input number of the multi-characteristic influence factor is established , wherein, is the number of input behavior influence factors; the input value of the original behavior influence factor is , and the maximum and minimum values of the value range of the original behavior influence factor are adjusted on the basis of . Adjustment of the maximum value: wherein, is the input value of the original state influence factor, denotes the maximum value after the extreme disturbance. Adjustment of the minimum value: wherein represents the minimum value after the extreme perturbation; Extreme feature disturbance refers to extreme or extreme degree of change or disturbance of the feature in the data, and the maximum value and the minimum value of the input numerical value are changed to test the robustness and performance of the model under extreme conditions. This disturbance is to evaluate the behavior of the model when facing abnormal or unusual inputs; Determine the key factors by using voting counting and Bayesian principle probability method through repeated input of the characteristic influence factors: The input state influence factor is , The input frequency is represented, and the output result is m, where ; in the single input state influence factor , , The probability of each of the three levels is represented as , , ; for three levels, the likelihood of observing each output result at each level is calculated, the likelihood of the result is multiplied, and the total likelihood of the result is obtained; the total likelihood of the three levels is calculated, which is ; the posterior probability of each level is estimated using the Bayes principle: Compare the posterior probabilities and estimate the grades according to the observation results.

2. The feature perturbation-based pavement condition key factor identification method of claim 1, wherein, Constructing a scaled model of the road pavement in proportion according to the characteristic state and geometric parameters comprises the following steps: obtaining the characteristic state and geometric parameters of the actual road pavement by referring to relevant data and corresponding field investigation of the road pavement, printing the scaled model under different material ratios and 3D printing parameter combinations by using 3D printing technology, and ensuring that the scaled model has similar geometric shape, texture and material properties to the actual road pavement.

3. The feature perturbation-based pavement condition key factor identification method of claim 1, wherein, Obtain a plurality of state influence factors affecting the road pavement and perform data normalization processing, comprising: Obtain state influence factor data so that the state influence factor data set covers different traffic scenarios and conditions, and perform data cleaning, removal of abnormal values and standardization of the road pavement state influence factor data.

4. The feature perturbation-based pavement condition key factor identification method of claim 1, wherein, The neural network model is trained on the road pavement state influence factors, automatically extracts features and learns the spatial relationship between the features through the neural network, automatically learns the key part of the data, and the activation function of the last layer of the neural network model adopts a Softmax function. The Softmax function converts the output value of multi-classification into a range of [0, 1].

5. The feature perturbation-based pavement condition key factor identification method of claim 4, wherein, A suitable parameter initialization method is adopted, and the neural network model is adjusted and optimized according to the characteristics of the data and the task requirements, so as to learn the mapping relationship between the model response and the characteristic factors of the pavement state.

6. The feature perturbation-based pavement condition key factor identification method of claim 4, wherein, The hyperparameters of the neural network model are adjusted, and the neural network model is optimized by adjusting the learning rate and the number of iterations.

7. A system for identifying key factors of pavement performance based on feature perturbation, which performs the method for identifying key factors of pavement performance based on feature perturbation according to any one of claims 1-6, characterized in that, It comprises: A data acquisition module for acquiring the characteristic state and geometric parameters of the road pavement; A preprocessing module for constructing a scaled model of the road pavement according to the characteristic state and geometric parameters; based on the scaled model, a plurality of state influence factors affecting the road pavement are obtained, and data normalization processing is performed; A key factor identification module for training a neural network model, inputting the state influence factors into the neural network model for extreme characteristic disturbance, adjusting the values within the range of [0, 1] by setting the maximum and minimum values, observing the changes in the output results of the neural network model, and preliminarily determining the key factors that have a great influence on the output results of the neural network model; The obtained key factors are repeatedly input into the neural network model, and the accuracy of the selected key factors is verified in a probabilistic manner.

Images