Airport pavement performance prediction method and system based on multi-source feature coding

Abstract

The invention provides an airport pavement performance prediction method and system based on multi-source feature coding, and belongs to the technical field of pavement monitoring, and the method comprises the steps: obtaining a plurality of historical scores and a plurality of pavement attributes of a preset unit, the historical score is a historical airport pavement performance score of a corresponding part of the preset unit; on the basis of all the historical scores, initial time sequence features are constructed, and on the basis of all the pavement attributes, initial attribute features are constructed; inputting the initial time sequence features and the initial attribute features into an airport pavement performance prediction model to obtain a prediction score; and determining the airport pavement performance based on the predicted score. According to the method, the historical scores and the pavement attributes are combined, the initial time sequence features and the attribute features are constructed, and the features are input into the performance prediction model, so that the airport pavement performance is accurately predicted based on the multi-source data.

Description

Technical Field

[0001] The present invention relates to the technical field of pavement monitoring, and in particular to a method and system for predicting airport pavement performance based on multi-source feature coding. Background Art

[0002] In the operation and maintenance of modern airports, the performance of airport pavements, as important infrastructure for aircraft takeoff, landing and taxiing, directly affects the airport's operational efficiency and flight safety. With the increase in the frequency of airport use and the increase in aircraft load, the service life and performance of pavements are facing increasing challenges.

[0003] Traditional airport pavement management mainly relies on manual inspections and periodic testing. Although this method can provide information on the current status of the pavement, it has significant deficiencies in predicting future pavement performance changes. Therefore, in order to improve this situation, pavement monitoring systems based on sensors and data analysis technologies have emerged in recent years. These systems collect physical data such as stress, deformation, and temperature of the pavement in real time through sensors deployed on the pavement, and use these data to evaluate the pavement status. However, this method still faces the problems of single data dimension and high data processing complexity. Especially in the face of the diverse use environments and structural characteristics of airport pavements, existing technologies find it difficult to accurately integrate multi-source data for the prediction and management of pavement performance.

[0004] Therefore, in airport pavement management, how to accurately predict airport pavement performance based on multi-source data has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] The present invention provides an airport pavement performance prediction method, system, electronic device and storage medium based on multi-source feature coding, which are used to solve the defects in the prior art and realize accurate prediction of airport pavement performance based on multi-source data.

[0006] The present invention provides an airport pavement performance prediction method based on multi-source feature coding, comprising the following steps: obtaining a plurality of historical scores and a plurality of pavement attributes of a preset unit, wherein the preset unit is obtained after the airport pavement is divided, and the historical score is a historical performance score of the airport pavement portion corresponding to the preset unit; Based on all of the historical scores, construct an initial time series feature, and based on all of the road surface attributes, construct an initial attribute feature; Inputting the initial time series features and the initial attribute features into an airport pavement performance prediction model to obtain a prediction score; Based on the predicted score, an airport pavement performance is determined.

[0007] According to a method for predicting airport pavement performance based on multi-source feature coding provided by the present invention, the airport pavement performance prediction model includes a multi-source feature encoder and a performance predictor; the initial time series features and the initial attribute features are input into the airport pavement performance prediction model to obtain a prediction score, specifically including: Inputting the initial time series features and the initial attribute features into the multi-source feature encoder to obtain a first reconstruction mean parameter and a second reconstruction mean parameter; The first reconstructed mean parameter and the second reconstructed mean parameter are input into the performance predictor to obtain the prediction score.

[0008] According to a method for predicting airport pavement performance based on multi-source feature coding provided by the present invention, the multi-source feature encoder includes a time series feature encoding module, an attribute feature encoding module, a fusion encoding module, a time series feature decoding module and an attribute feature decoding module; the initial time series feature and the initial attribute feature are input into the multi-source feature encoder to obtain a first reconstructed mean parameter and a second reconstructed mean parameter, specifically including: Inputting the initial time series feature into the time series feature encoding module to obtain a time series feature vector, and inputting the initial attribute feature into the attribute feature encoding module to obtain an attribute feature vector; Inputting the time series feature vector and the attribute feature vector into the fusion coding module for fusion, generating a first mean parameter, a first variance parameter, a second mean parameter and a second variance parameter; The first mean parameter and the first variance parameter are input into the time series feature decoding module to obtain the first reconstructed mean parameter, and the second mean parameter and the second variance parameter are input into the attribute feature decoding module to obtain the second reconstructed mean parameter.

[0009] According to a method for predicting airport pavement performance based on multi-source feature coding provided by the present invention, the multi-source feature encoder is trained by the following method: Based on the multiple historical score labels of each unit sample, construct a time series feature sample of each unit sample, and based on the multiple road surface attribute labels of each unit sample, construct an attribute feature sample of each unit sample; The initial encoder is trained based on all the time series feature samples and all the attribute feature samples to obtain the multi-source feature encoder.

[0010] According to a method for predicting airport pavement performance based on multi-source feature coding provided by the present invention, the initial encoder is trained based on all the time series feature samples and all the attribute feature samples to obtain the multi-source feature encoder, specifically comprising: Input each of the time series feature samples and each of the attribute feature samples into the initial encoder to obtain a first training reconstruction mean parameter, a second training reconstruction mean parameter, a reconstructed attribute feature, and a reconstructed time series feature of each of the unit samples; Inputting each of the first training reconstruction mean parameters and each of the second training reconstruction mean parameters into the performance predictor to obtain a training prediction score for each of the unit samples; Obtaining a feature reconstruction error value based on each of the reconstructed time series features, each of the reconstructed attribute features, each of the time series feature samples, and each of the attribute feature samples; Obtaining a prediction error value based on each of the training prediction scores and each of the historical score labels; Adding the feature reconstruction error value and the prediction error value to obtain a total error value; The initial encoder is trained with minimizing the total error value as a training target to obtain the multi-source feature encoder.

[0011] According to a method for predicting airport pavement performance based on multi-source feature coding provided by the present invention, the initial encoder includes a time series feature encoding module, an attribute feature encoding module, a fusion encoding module, a time series feature decoding module and an attribute feature decoding module; each of the time series feature samples and each of the attribute feature samples are input into the initial encoder to obtain a first training reconstruction mean parameter, a second training reconstruction mean parameter, a reconstructed attribute feature and a reconstructed time series feature of each of the unit samples, specifically including: Input each of the time series feature samples into the time series feature encoding module to obtain a time series feature vector sample of each of the unit samples; Input each of the attribute feature samples into the attribute feature encoding module to obtain an attribute feature vector sample of each of the unit samples; Inputting each of the time series feature vector samples and each of the attribute feature vector samples into the fusion coding module for fusion, and generating a first training mean parameter, a first training variance parameter, a second training mean parameter and a second training variance parameter for each of the unit samples; Inputting each of the first training mean parameters and each of the first training variance parameters into the time series feature decoding module to obtain a first training reconstruction mean parameter of each of the unit samples, and obtaining a reconstructed time series feature of each of the unit samples based on each of the first training reconstruction mean parameters; Each of the second training mean parameters and each of the second training variance parameters are input into the attribute feature decoding module to obtain the second training reconstruction mean parameter of each of the unit samples, and based on each of the second training reconstruction mean parameters, the reconstructed attribute feature of each of the unit samples is obtained.

[0012] According to a method for predicting airport pavement performance based on multi-source feature coding provided by the present invention, each of the first training reconstruction mean parameters and each of the second training reconstruction mean parameters are input into the performance predictor to obtain a training prediction score for each of the unit samples, specifically comprising: The prediction score is calculated by the following formula: ; Where t is the use time of the airport pavement, is the predicted score corresponding to the i-th unit sample at the t-th usage time, is the first reconstructed mean parameter of the i-th unit sample, is the second reconstructed mean parameter of the i-th unit sample.

[0013] According to a method for predicting airport pavement performance based on multi-source feature coding provided by the present invention, the feature reconstruction error value is obtained based on each of the reconstructed time series features, each of the reconstructed attribute features, each of the time series feature samples and each of the attribute feature samples, specifically including: The feature reconstruction error value is calculated by the following formula: ; in, is the feature reconstruction error value, Q is the number of unit samples, is the reconstructed attribute feature of the i-th unit sample, is the attribute feature sample of the i-th unit sample, is the reconstructed time series feature of the i-th unit sample, is the timing feature sample of the i-th unit sample.

[0014] According to a method for predicting airport pavement performance based on multi-source feature coding provided by the present invention, obtaining a prediction error value based on each of the training prediction scores and each of the historical score labels specifically includes: The prediction error value is calculated by the following formula: ; in, is the prediction error value, Q is the number of unit samples, T is the time length required for the airport pavement to be predicted, t is the usage time of the airport pavement, is the predicted score corresponding to the i-th unit sample at the t-th usage time, is the historical scoring label corresponding to the i-th unit sample at the t-th usage time.

[0015] The present invention also provides an airport pavement performance prediction system based on multi-source feature coding, comprising the following modules: a first acquisition module, a first processing module and a performance evaluation module; The first acquisition module is used to acquire multiple historical scores and multiple pavement attributes of a preset unit, wherein the preset unit is obtained after the airport pavement is divided, and the historical score is the historical airport pavement performance score corresponding to the preset unit; The first processing module is used to construct an initial time series feature based on all the historical scores, and to construct an initial attribute feature based on all the road surface attributes; The performance evaluation module is used to input the initial time series features and the initial attribute features into the airport pavement performance prediction model to obtain a prediction score; The performance evaluation module is further used to determine the airport pavement performance based on the predicted score.

[0016] The present invention also provides 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 program, the airport pavement performance prediction method based on multi-source feature coding as described in any one of the above is implemented.

[0017] 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, the method for predicting airport pavement performance based on multi-source feature coding as described in any one of the above is implemented.

[0018] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method for predicting airport pavement performance based on multi-source feature coding as described above is implemented.

[0019] In summary, one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: By dividing the pavement into multiple preset units, the performance status of each preset unit can be evaluated independently, so that the areas that need to be focused and maintained can be accurately identified during the management process. By obtaining multiple historical scores and multiple pavement attributes of the preset units, comprehensive and multi-dimensional data support is provided for the subsequent pavement performance prediction. The historical scores reflect the changes in the status of each preset unit at different time points, while the pavement attributes cover the long-term impact of physical and environmental factors on pavement performance. By fully obtaining these data, the model can better understand and predict the future status of the pavement, thereby improving the accuracy of the prediction. By constructing initial time series features based on multiple historical scores and initial attribute features based on multiple pavement attributes, key features closely related to pavement performance can be effectively extracted. These features not only capture the trend of pavement performance over time, but also reflect the potential impact of physical properties and environmental conditions on pavement performance. Through the construction of these features, the model can better capture the changing laws of pavement performance in subsequent predictions, thereby improving the prediction effect of the model. By inputting the initial time series features and initial attribute features into the airport pavement performance prediction model, the prediction score is obtained, and the accurate prediction of the future performance of the pavement is achieved. Through comprehensive analysis of time series features and attribute features, the model can output the prediction score of each pavement preset unit. Finally, the airport pavement performance is determined based on the prediction score, thereby achieving accurate prediction of airport pavement performance based on multi-source data. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. 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.

[0021] Figure 1 It is one of the flow charts of the airport pavement performance prediction method based on multi-source feature coding provided by the present invention.

[0022] Figure 2 It is a schematic diagram of the network architecture of the airport pavement performance prediction model provided by the present invention.

[0023] Figure 3 This is the second flow chart of the airport pavement performance prediction method based on multi-source feature coding provided by the present invention.

[0024] Figure 4 This is the third flow chart of the airport pavement performance prediction method based on multi-source feature coding provided by the present invention.

[0025] Figure 5This is the fourth flow chart of the airport pavement performance prediction method based on multi-source feature coding provided by the present invention.

[0026] Figure 6 It is a structural schematic diagram of the airport pavement performance prediction system based on multi-source feature coding provided by the present invention.

[0027] Figure 7 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0028] 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.

[0029] It should be noted that, in the description of the present invention, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "include one..." do not exclude the existence of other identical elements in the process, method, article or device including the elements. The orientation or positional relationship indicated by the terms "upper", "lower", etc. is based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the system or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances.

[0030] The terms "first", "second", etc. in the present invention are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present invention can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are generally of the same type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" means at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0031] Combine the following Figure 1-Figure 7The invention describes an airport pavement performance prediction method, system, electronic device and storage medium based on multi-source feature coding.

[0032] Figure 1 This is one of the flow charts of the airport pavement performance prediction method based on multi-source feature coding provided by the present invention, such as Figure 1 As shown, including but not limited to the following steps: Step 11: Obtain multiple historical scores and multiple pavement attributes of preset units, wherein the preset unit is obtained after the airport pavement is divided, and the historical score is the historical airport pavement performance score of the corresponding part of the preset unit.

[0033] When implementing step 11, multiple historical scores and multiple pavement attributes of the preset unit are obtained in order to provide a comprehensive and accurate data basis for the subsequent airport pavement performance prediction. The historical score is the historical airport pavement performance score of the corresponding part of the preset unit, and the airport pavement performance score is the PCI value (the historical airport pavement performance score is uniformly referred to as the PCI value in subsequent embodiments). The historical PCI value and pavement attributes are key data reflecting the airport pavement conditions and influencing factors. By collecting these data, the changing trend of pavement performance can be better modeled and the accuracy of the prediction can be improved.

[0034] The purpose of dividing the airport pavement into multiple preset units is to achieve refined management of the airport pavement and accurate performance prediction. This division process follows the three-level rule of "part-area-preset unit" for airport pavement zoning. Specifically, the division is based on the physical structure differences of the runway and the impact of aircraft loads.

[0035] In the specific implementation, firstly, according to the overall layout of the airport pavement, the pavement is divided into different parts, such as runway, parallel taxiway, connecting road, apron and asphalt transition section. These parts are divided according to their functions and structural characteristics in the airport pavement to reflect the usage characteristics and maintenance requirements of different areas.

[0036] Next, the pavement is further divided into several zones based on the pavement structure and the load level it is subjected to. The criteria for this division include factors such as the type of pavement material, thickness, structural strength, and load distribution. For example, a runway may be divided into multiple zones based on different load distributions at its centerline and edge areas, and each zone may have different physical properties and wear characteristics.

[0037] Finally, on the basis of the region, each region is further divided into multiple preset units. Each preset unit is usually composed of multiple cement concrete slabs and is the smallest observation object. Such preset unit division can ensure that when monitoring and predicting performance, it can be refined to a specific small area for analysis, ensuring the accuracy of the prediction and the pertinence of subsequent management.

[0038] By dividing the airport pavement into multiple preset units, the precision of data collection and processing can be significantly improved, making the subsequent airport pavement performance prediction more accurate. This detailed division can also better capture the local characteristics and changes of the pavement in different areas and preset units, thereby providing more detailed guidance for pavement management.

[0039] To obtain multiple historical scores of preset units, first, the PCI value of each preset unit needs to be extracted from the historical data. These PCI values ​​are comprehensive scores of pavement conditions, usually obtained through regular inspections and assessments, reflecting the degree of damage, cracks, dents, etc. of the pavement. Obtaining these historical PCI values ​​can capture the performance changes of the pavement in different periods, forming a time series, and providing the necessary basis for constructing time series features.

[0040] At the same time, it is also necessary to obtain multiple pavement attributes related to each preset unit. These attributes include but are not limited to traffic volume, structural strength, surface thickness, base type, environmental conditions, and material type. These attribute data are important factors affecting pavement performance, and they can be obtained from design documents, construction records, sensor data, and environmental monitoring systems. The collection of these attribute data can not only help understand the factors affecting pavement performance, but also provide multi-dimensional input features for the model, thereby improving the performance of the prediction model.

[0041] By collecting and integrating these historical PCI values ​​and pavement attributes, we can provide rich data support for subsequent prediction models. These data will be used to construct initial time series features and initial attribute features as inputs to the prediction model. In the model, these data will be used to capture the relationship between the time series evolution law of the pavement and the attribute features, thereby helping the model to make more accurate future performance predictions.

[0042] Step 12: Based on all historical scores, construct initial time series features, and based on all road surface attributes, construct initial attribute features.

[0043] When implementing step 12, first construct the initial time series features for the historical PCI values ​​of each preset unit. These historical PCI values, as time series data, can reflect the performance changes of the pavement at different time points. In order to capture the trends and periodic changes in these time series data, the historical PCI values ​​need to be processed. For example, key features in the time series, such as trend terms, seasonal terms, etc., can be extracted through sliding windows, differential processing and other methods. These processed time series features can better reflect the aging process of the pavement, changes in usage intensity and other information, thereby providing more valuable input for the model.

[0044] At the same time, it is also necessary to construct initial attribute features based on the pavement attribute data. Pavement attributes include traffic volume, structural strength, surface thickness, base type, environmental conditions and material type, etc. These attributes play an important role in the change of pavement performance. In order to convert these attribute data into input features that can be used by the model, standardization, normalization or other forms of feature engineering are usually required. For example, the numerical features of different attributes can be normalized to the same scale, or the categorical variables can be converted into binary vectors through One-Hot Encoding. These processing steps ensure that the model can use each feature in a balanced way when processing data of different dimensions and types, thereby improving the overall prediction effect.

[0045] Through the above processing, the constructed initial time series features and initial attribute features can be used as the input of the model, so that the model can simultaneously consider the temporal dynamic changes of pavement performance and the influence of various attributes. The time series features help the model understand how pavement performance evolves over time, while the attribute features help the model capture the impact of specific physical conditions and usage environment on pavement performance.

[0046] Step 13: Input the initial time series features and initial attribute features into the airport pavement performance prediction model to obtain the prediction score.

[0047] Specifically, the airport pavement performance prediction model consists of two key components: a multi-source feature encoder and a performance predictor. That is, the initial time series features and initial attribute features are input into the multi-source feature encoder and the performance predictor to obtain the prediction score.

[0048] In the subsequent embodiments, the training process of the multi-source feature encoder will be described in detail. First, in order to facilitate the understanding of the training process of the multi-source feature encoder, refer to Figure 2 , Figure 2 It is a schematic diagram of the network architecture of the airport pavement performance prediction model provided by the present invention.

[0049] In one possible implementation, refer to Figure 3 , Figure 3This is the second flow chart of the airport pavement performance prediction method based on multi-source feature coding provided by the present invention. Before step 13, the method further includes steps 21-22: Step 21: Based on multiple historical score labels of each unit sample, construct a time series feature sample of each unit sample, and based on multiple road surface attribute labels of each unit sample, construct an attribute feature sample of each unit sample.

[0050] Specifically, the specific construction method in step 21 is similar to the specific construction method in step 12 of the aforementioned embodiment, so it will not be described in detail here.

[0051] Step 22: Train the initial encoder based on all time series feature samples and all attribute feature samples to obtain a multi-source feature encoder.

[0052] When implementing step 22, the initial encoder is trained based on all time series feature samples and all attribute feature samples to obtain a multi-source feature encoder, so that the model can effectively capture and fuse these features by learning the time series and attribute features in the historical data, thereby providing accurate feature representation for future pavement performance prediction. The core goal of this step is to enable the multi-source feature encoder to establish complex relationships between different input data through training, and to output accurate feature expressions, laying the foundation for performance prediction.

[0053] In subsequent embodiments, a specific training method for training the initial encoder based on the time series feature samples and the attribute feature samples will be described in detail.

[0054] In one possible implementation, refer to Figure 4 , Figure 4 This is the third flow chart of the airport pavement performance prediction method based on multi-source feature coding provided by the present invention, step 22 specifically includes steps 221-226: Step 221: Input each time series feature sample and each attribute feature sample into the initial encoder to obtain the first training reconstruction mean parameter, the second training reconstruction mean parameter, the reconstructed attribute feature and the reconstructed time series feature of each unit sample.

[0055] Specifically, the initial encoder includes the following key components: a time series feature encoding module, an attribute feature encoding module, a fusion encoding module, a time series feature decoding module, and an attribute feature decoding module. When implementing step 221, each time series feature sample and each attribute feature sample are input into the initial encoder to obtain the first training reconstruction mean parameter, the second training reconstruction mean parameter, the reconstructed attribute feature, and the reconstructed time series feature of each unit sample. This is to encode and reconstruct the training data through the initial encoder and learn the key features that can represent the performance of the airport pavement. The purpose of this step is to enable the model to extract high-dimensional feature representations through the feature encoding and decoding process, and to verify the validity of these features through reconstruction.

[0056] In one possible implementation, refer to Figure 5 , Figure 5 This is the fourth flow chart of the airport pavement performance prediction method based on multi-source feature coding provided by the present invention. Step 221 specifically includes steps 2211-2215: Step 2211: Input each time series feature sample into the time series feature encoding module to obtain a time series feature vector sample of each unit sample.

[0057] When implementing step 2211, each time series feature sample is input into the time series feature encoding module to obtain the time series feature vector sample of each unit sample, in order to deeply encode the historical PCI data through the graph neural network (GCN) and the recurrent neural network (LSTM), and extract high-dimensional features that can reflect the spatial correlation and time dependence between the preset units. The purpose of this step is to capture the performance change rules of each preset unit of the airport pavement in the time series data, and convert these rules into feature vectors suitable for model processing, so as to provide effective time series information support for subsequent predictions.

[0058] In the specific implementation, first, the historical PCI sequence of each unit sample is It is passed as input data to the temporal feature encoding module. For each observation time point, the temporal feature encoding module first constructs a correlation network Gm=(V,E) between preset units, where the node set V represents the preset units in the area and the edge set E represents the connection relationship between adjacent preset units. Through this network structure, the spatial relationship between the preset units of the airport pavement can be effectively captured, as well as how these relationships evolve over time.

[0059] Next, the correlation network is processed using a graph neural network (GCN). GCN can learn the spatial correlation characteristics between each preset unit by aggregating the information of neighboring nodes layer by layer based on the network graph. Specifically, at each time point, the current PCI value of the preset unit is used as the input feature, and a high-dimensional feature vector containing 64 dimensions is generated through the hierarchical propagation of GCN. ,Right now This feature vector not only contains the performance information of the preset unit at the current time point, but also combines the influence of its neighbor preset units to form a comprehensive spatial correlation representation. Assuming there are L observation time points, L correlation features can be obtained, namely , for the i-th preset unit, the L correlation features can be expressed as: , express The i-th row in express The i-th row in express The i-th row in , where T represents the transpose operation of the vector.

[0060] In order to further capture the dependencies between these time series features, especially the long-term time dependencies, the time series feature sequences are input into the recurrent neural network (LSTM). LSTM is good at processing long sequence data. It can capture the long-term dependencies in the time series through its internal memory preset units, thereby generating more time-dependent time series feature vector samples. .

[0061] Through the above process, the time series feature encoding module can extract a high-dimensional, comprehensive time series feature vector from the original PCI time series data. This feature vector not only retains the key time information in the original data, but also combines spatial correlation and time dependence, so that the model can more accurately capture the changing trend of pavement performance in subsequent training and prediction.

[0062] Step 2212: Input each attribute feature sample into the attribute feature encoding module to obtain the attribute feature vector sample of each unit sample.

[0063] When implementing step 2212, each attribute feature sample is input into the attribute feature encoding module to obtain the attribute feature vector sample of each unit sample, in order to extract the feature information that has the greatest impact on the pavement performance from the physical and environmental attributes of the pavement, and convert this information into a high-dimensional vector representation suitable for model processing. The purpose of this step is to deeply encode the original attribute data through a multi-layer perceptron (MLP) to capture the complex relationship between different attributes, thereby providing reliable input for subsequent feature fusion and performance prediction.

[0064] In the specific implementation, the attribute feature data of each unit sample is first input into the attribute feature encoding module. Pavement attributes include traffic volume, structural strength, surface thickness, base type, environmental conditions, and material type. These feature data may have different dimensions and types (such as continuous and discrete data). In order to process these diverse data, the original data must first be preprocessed to ensure the normalization of the input features. Continuous data is usually standardized or normalized to eliminate the impact of different dimensions on model training. For discrete data, such as material type or base type, it is converted into a binary vector suitable for model input through One-Hot Encoding. .

[0065] After preprocessing, the normalized and encoded attribute feature data is input into a multi-layer perceptron (MLP). MLP is a feed-forward neural network that can extract high-order attribute features from input features through nonlinear transformations of multiple hidden layers. Specifically, MLP maps the original attribute features to a higher-dimensional space through a series of linear transformations and nonlinear activation functions, and extracts the feature information that has the greatest impact on road surface performance layer by layer. Each hidden layer gradually extracts more abstract and expressive feature vectors through the action of weight matrices and activation functions.

[0066] Finally, the attribute feature encoding module outputs an attribute feature vector sample E containing 64 dimensions. K , This high-dimensional vector condenses the key information of various physical properties and environmental factors of the road surface, and can effectively represent the attribute characteristics of the preset unit in the current state. This vector not only retains the important information in the original data, but also enhances the relationship between the features through nonlinear transformation, so that the model can better understand the impact of these attribute data on road surface performance when processing them.

[0067] Step 2213: Input each time series feature vector sample and each attribute feature vector sample into the fusion coding module for fusion, and generate the first training mean parameter, the first training variance parameter, the second training mean parameter and the second training variance parameter of each unit sample.

[0068] In the specific implementation, firstly, the time series feature vector sample E obtained by the time series feature encoding module is G and the attribute feature vector sample E obtained by the attribute feature encoding module KAs input, it is passed to the fusion encoding module. In this module, the fusion of temporal features and attribute features is completed by concatenating the two vectors. The concatenation operation concatenates the two vectors into a higher-dimensional vector, which contains both the temporal dynamic information of the road surface and the physical attribute information of the road surface.

[0069] Once the time series feature and attribute feature vectors are connected together, the fusion coding module processes this high-dimensional vector through a multi-layer perceptron (MLP). MLP extracts the complex interactions hidden in the time series and attribute data through multiple layers of nonlinear transformations, and converts these interactions into statistical parameters that are crucial for pavement performance prediction. Specifically, the output of MLP is two mean parameters and two variance parameters, namely the first training mean parameter , the first training variance parameter , the second training mean parameter and the second training variance parameter These parameters are generated by fusing the temporal features and attribute features of the training data, and they can summarize and characterize the key feature information in the entire training data set, that is, , , , .

[0070] The generation process of these parameters is the key to the model capturing the complex relationship between time series and attribute features. The mean parameter represents the central tendency of the fused features, while the variance parameter reflects the distribution and uncertainty of the feature data. These parameters not only provide a statistical description of the pavement performance prediction, but also help the model consider the uncertainty of the data during the prediction process, making the model output more robust and reliable.

[0071] Step 2214: Input each first training mean parameter and each first training variance parameter into the time series feature decoding module to obtain the first training reconstruction mean parameter of each unit sample, and based on each first training reconstruction mean parameter, obtain the reconstructed time series feature of each unit sample.

[0072] When implementing step 2214, each first training mean parameter and each first training variance parameter are input into the time series feature decoding module to obtain the first training reconstruction mean parameter of each unit sample, and based on each first training reconstruction mean parameter, the reconstructed time series feature of each unit sample is obtained, in order to verify and optimize the encoding effect of the time series feature, and ensure that the model can accurately capture and reproduce the core information of the road surface time series feature. The purpose of this step is to detect whether the encoder has effectively extracted the time series feature through the decoding process, and adjust the model parameters through the feedback of the reconstruction error, so as to improve the quality of feature representation.

[0073] In the specific implementation, the first training mean parameter and the first training variance parameter generated in step 2213 are first input into the time series feature decoding module. The main task of the time series feature decoding module is to use these statistical parameters to generate the corresponding reconstructed time series features. This process simulates how to restore the original time series features from the encoded feature representation and verify the effectiveness of the encoding.

[0074] In the decoding process, a sampling operation of the normal distribution is first performed to generate random samples using the first training mean parameter and the first training variance parameter. These samples represent the possible distribution in the time series feature space. Specifically, this sampling operation uses and As the mean and standard deviation of the normal distribution, a set of random time series features with similar statistical properties are generated, namely This group of samples is called the first training reconstruction mean parameters, which represent the possible reconstruction results of the time series features under given conditions.

[0075] Next, the first training reconstruction mean parameter is input into the multi-layer perceptron (MLP) to further generate the reconstructed time series features, i.e. The task of MLP is to convert these reconstruction parameters into a complete time series feature vector, which is actually a reproduction of the original time series features. The reconstructed time series features generated by the decoding module will be compared with the original time series feature samples to calculate the reconstruction error. This error value reflects the accuracy of the encoder and decoder in capturing and expressing time series features.

[0076] By comparing the difference between the reconstructed time series features and the original time series features, the model can obtain feedback to adjust the parameters of the encoder and decoder. As training progresses, this feedback mechanism will gradually reduce the reconstruction error and ensure that the model can accurately reconstruct the original time series features. This not only verifies the encoder's ability to extract features, but also provides a guarantee for the model to better utilize time series features in actual predictions.

[0077] Step 2215: Input each second training mean parameter and each second training variance parameter into the attribute feature decoding module to obtain the second training reconstruction mean parameter of each unit sample, and based on each second training reconstruction mean parameter, obtain the reconstructed attribute feature of each unit sample.

[0078] When implementing step 2215, each second training mean parameter and each second training variance parameter are input into the attribute feature decoding module to obtain the second training reconstruction mean parameter of each unit sample, and based on each second training reconstruction mean parameter, the reconstructed attribute feature of each unit sample is obtained, in order to verify and optimize the encoding effect of the attribute feature, and ensure that the model can accurately capture and reproduce the core information of the pavement attribute feature. The purpose of this step is to detect whether the encoder has effectively extracted the attribute feature through the decoding process, and adjust the model parameters through the feedback of the reconstruction error, so as to improve the quality of feature representation and the overall performance of the model.

[0079] In specific implementation, first, the second training mean parameter and the second training variance parameter generated in step 2213 are input into the attribute feature decoding module. The main task of the attribute feature decoding module is to generate corresponding reconstructed attribute features using these statistical parameters. This process is similar to the decoding of time series features. Through the decoding process, the original attribute features are restored from the encoded feature representation to verify the effectiveness of the encoder.

[0080] In the decoding process, firstly, a sampling operation of the normal distribution is performed to generate random samples using the second training mean parameter and the second training variance parameter. These samples represent the possible distribution in the attribute feature space. Specifically, this sampling operation uses and As the mean and standard deviation of the normal distribution, a set of random time series features with similar statistical properties are generated, namely This group of samples is called the second training reconstruction mean parameters, which represent the possible reconstruction results of the time series features under given conditions.

[0081] Subsequently, these second training reconstruction mean parameters are input into a multi-layer perceptron (MLP) to further generate a complete reconstruction attribute feature vector through nonlinear transformation, i.e. The role of MLP is to convert these reconstruction parameters into complete attribute feature representations to reproduce the original attribute feature data. The generated reconstructed attribute features will be compared with the original attribute feature samples to calculate the reconstruction error. This error value will be fed back to the model to optimize the parameter settings of the encoder and decoder.

[0082] By calculating and minimizing the reconstruction error, the model can obtain feedback on the effectiveness of the encoding and decoding process, thereby gradually optimizing the feature representation. This process ensures that the model can not only effectively encode and reconstruct time series features, but also achieve high accuracy in the representation and reproduction of attribute features. This comprehensive optimization improves the performance of the model in practical applications and ensures that it has stronger expressive and predictive capabilities when processing complex multi-source data.

[0083] By implementing step 2215, the attribute feature decoding module successfully generates a reconstructed feature that is as consistent as possible with the original attribute feature. This process verifies the effectiveness of attribute feature encoding and optimizes the model's ability to capture and express attribute data through a feedback mechanism. Through this two-way optimization cycle, the model gradually improves its ability to represent pavement attribute features, thereby providing more accurate and reliable feature input for subsequent pavement performance prediction.

[0084] Step 222: Input each first training reconstruction mean parameter and each second training reconstruction mean parameter into the performance predictor to obtain a training prediction score for each unit sample.

[0085] In a possible implementation, step 222 specifically includes the following steps: The predicted score is calculated using the following formula: ; Where t is the use time of the airport pavement, is the predicted score corresponding to the i-th unit sample at the t-th usage time, is the first reconstruction mean parameter of the i-th unit sample, is the second reconstruction mean parameter of the i-th unit sample.

[0086] In the specific implementation, the first training reconstruction mean parameter and the second training reconstruction mean parameter obtained in steps 2214 and 2215 are first input into the performance predictor. The performance predictor is designed based on a mathematical model, which uses these reconstruction mean parameters to calculate the PCI value of the unit sample at a specific time point in the future. Specifically, the performance predictor uses the following formula to calculate the prediction score: ; Where t is the use time of the airport pavement, in years. is the predicted score corresponding to the i-th unit sample at the t-th usage time, is the first reconstruction mean parameter of the i-th unit sample, is the second reconstruction mean parameter of the i-th unit sample.

[0087] This formula converts the reconstruction parameters into specific performance scores through nonlinear transformation. The exponential term in the formula reflects the decay trend of pavement performance over time, while the mean parameter and variance parameter capture the combined impact of timing characteristics and attribute characteristics on pavement performance. In this way, the performance predictor can generate future PCI values ​​based on the input feature parameters.

[0088] Step 223: Based on each reconstructed time series feature, each reconstructed attribute feature, each time series feature sample and each attribute feature sample, a feature reconstruction error value is obtained.

[0089] In a possible implementation, step 223 specifically includes the following steps: The feature reconstruction error value is calculated by the following formula: ; in, is the feature reconstruction error value, Q is the number of unit samples, is the reconstructed attribute feature of the i-th unit sample, is the attribute feature sample of the i-th unit sample, is the reconstructed time series feature of the i-th unit sample, is the time series feature sample of the i-th unit sample.

[0090] In the specific implementation, it is first necessary to compare the reconstructed time series features and attribute features with the original time series feature samples and attribute feature samples. The reconstructed time series features are generated by the time series feature decoding module in step 2214, and the reconstructed attribute features are generated by the attribute feature decoding module in step 2215. They respectively represent the reconstruction results of the model on the original features.

[0091] In order to quantify the error of the model in the reconstruction process, the feature reconstruction error value is calculated This is the key step. This error value is calculated by the following formula: ; in, is the feature reconstruction error value, Q is the number of unit samples, is the reconstructed attribute feature of the i-th unit sample, is the attribute feature sample of the i-th unit sample, is the reconstructed time series feature of the i-th unit sample, is the time series feature sample of the i-th unit sample.

[0092] In this formula, the calculation of reconstruction error includes two parts: the reconstruction error of attribute features and the reconstruction error of time series features. By calculating the square error between the reconstructed features and the original features of each unit sample, and then averaging all unit samples and taking the square root, the overall error value of the entire model in the feature reconstruction process is obtained. This error value directly reflects the accuracy of the model in capturing and reproducing the original features.

[0093] By implementing step 223, the calculated feature reconstruction error value can provide important feedback information for the model. When the error value is large, it indicates that the model may be insufficient in the feature extraction and reconstruction process, and the parameters and structure of the model need to be further optimized. On the contrary, when the error value is small, it means that the model can capture and reconstruct the time series features and attribute features well.

[0094] Step 224: Based on each training prediction score and each historical score label, a prediction error value is obtained.

[0095] In a possible implementation, step 224 specifically includes the following steps: The prediction error value is calculated by the following formula: ; in, is the prediction error value, Q is the number of unit samples, T is the time length required to predict the use of the airport pavement, t is the use time of the airport pavement, is the predicted score corresponding to the i-th unit sample at the t-th usage time, is the historical rating label corresponding to the i-th unit sample at the t-th usage time.

[0096] In the specific implementation, firstly, the training prediction score generated by step 222 is With the corresponding historical rating label The training prediction score is the PCI value at a specific point in the future predicted by the model based on the reconstructed feature parameters, while the historical score label is the PCI value obtained through actual observation, which represents the actual performance status of the pavement.

[0097] In order to quantify the error of the model in the prediction process, the prediction error value is calculated This is the key step. This error value is calculated by the following formula: ; in, is the prediction error value, Q is the number of unit samples, T is the length of time required to predict the use of the airport pavement, in years, and t is the use time of the airport pavement, in years. is the predicted score corresponding to the i-th unit sample at the t-th usage time, is the historical rating label corresponding to the i-th unit sample at the t-th usage time.

[0098] In this formula, the calculation of the prediction error includes the squared error of all unit samples at all prediction time points. By calculating the difference between the predicted value and the true value of each unit sample, summing the squares of these differences, averaging all unit samples and time points and taking the square root, the overall prediction error value is finally obtained. This error value directly reflects the accuracy of the model in predicting airport pavement performance.

[0099] By implementing step 224, the calculated prediction error value can provide important feedback information for the model. When the prediction error value is large, it indicates that the model may be insufficient in the performance prediction process and needs to further optimize the parameters and structure of the model. On the contrary, when the prediction error value is small, it indicates that the model can better predict the future airport pavement performance.

[0100] Step 225: Add the feature reconstruction error value and the prediction error value to obtain a total error value.

[0101] Specifically, this total error value is used as a global indicator to directly guide the training and optimization of the model. By minimizing the total error value, the model can simultaneously optimize the parameter settings of the feature encoder / decoder and the performance predictor during the training process. As the training progresses, the model will gradually adjust the parameters to reduce the feature reconstruction error and prediction error, thereby improving the overall performance of the model.

[0102] Step 226: Taking minimizing the total error value as the training objective, the initial encoder is trained to obtain a multi-source feature encoder.

[0103] In specific implementation, it is first necessary to define a loss function, which uses the total error value calculated in step 225 as an optimization target.

[0104] Next, an appropriate optimization algorithm is used to minimize the loss function. In the present invention, it is preferred to use the Adam optimizer for training. The Adam optimizer is an optimization algorithm based on gradient descent, which has the advantages of adaptive learning rate and momentum adjustment and can perform well in high-dimensional non-convex problems. Specifically, by iteratively updating the model parameters, the Adam optimizer can gradually reduce the value of the loss function, thereby reducing the total error value.

[0105] During the training process, the model repeatedly uses the training data for forward propagation and back propagation. In each iteration, the model first calculates the feature reconstruction error and prediction error , and then add them together to get the total error value. Then, the gradient of the loss function relative to the model parameters is calculated through the back-propagation algorithm, and the model parameters are updated using the Adam optimizer. This process continues in each round of iteration until the total error value converges to a minimum value or the number of iterations is reached (e.g., the maximum number of iterations is 500).

[0106] As the training progresses, the parameters of the model will gradually adjust to the optimal state, so that the multi-source feature encoder can efficiently extract and fuse time series feature samples and attribute feature samples, and the performance predictor can more accurately predict the future performance score of the airport pavement. Ultimately, the fully trained multi-source feature encoder will have strong feature representation and prediction capabilities, and can provide high-quality feature input for accurate prediction of airport pavement performance.

[0107] By implementing step 226, the model can gradually optimize and adjust the parameters of the multi-source feature encoder and the performance predictor by minimizing the total error value, and finally obtain a fully trained multi-source feature encoder.

[0108] In a possible implementation manner, step 13 specifically includes the following steps: Inputting the initial time series features and the initial attribute features into a multi-source feature encoder to obtain a first reconstruction mean parameter and a second reconstruction mean parameter; The first reconstructed mean parameter and the second reconstructed mean parameter are input into a performance predictor to obtain a prediction score.

[0109] In a possible implementation, the initial time series features and the initial attribute features are input into a multi-source feature encoder to obtain a first reconstruction mean parameter and a second reconstruction mean parameter, which specifically includes the following steps: Inputting the initial time series feature into the time series feature encoding module to obtain a time series feature vector, and inputting the initial attribute feature into the attribute feature encoding module to obtain an attribute feature vector; Inputting the time series feature vector and the attribute feature vector into a fusion coding module for fusion, generating a first mean parameter, a first variance parameter, a second mean parameter and a second variance parameter; The first mean parameter and the first variance parameter are input into the time series feature decoding module to obtain a first reconstructed mean parameter, and the second mean parameter and the second variance parameter are input into the attribute feature decoding module to obtain a second reconstructed mean parameter.

[0110] In the specific implementation, the initial time series features and initial attribute features obtained from the actual data are first passed as input to the multi-source feature encoder, which is the same as the steps for training the multi-source feature encoder. The encoder has been optimized through the previous training steps and can efficiently process and fuse multi-source data. Inside the multi-source feature encoder, the time series features are first processed by the time series feature encoding module, using the graph neural network (GCN) and the recurrent neural network (LSTM) to capture the spatial correlation and time dependency between the preset units of the road surface. At the same time, the attribute features are passed through the attribute feature encoding module, using the multi-layer perceptron (MLP) to extract key physical and environmental attributes.

[0111] Next, the encoded time series features and attribute features are input into the fusion coding module for fusion processing. Through the concatenation operation, these features are integrated into a high-dimensional feature vector, which is then further processed by the multi-layer perceptron to generate two key statistical parameters: the first reconstruction mean parameter and the second reconstruction mean parameter.

[0112] After obtaining the first reconstruction mean parameter and the second reconstruction mean parameter, substitute the first reconstruction mean parameter and the second reconstruction mean parameter into the formula , where Y is the prediction score, A is the first reconstruction mean parameter, B is the second reconstruction mean parameter, and t is the airport pavement usage time in years.

[0113] Step 14: Based on the predicted scores, determine the airport pavement performance.

[0114] Specifically, the PCI value of each pavement preset unit reflects its expected condition at a specific point in the future, and the high or low PCI value directly corresponds to the quality of the pavement. Based on the model prediction, managers can determine which preset units may need special attention or maintenance in the future by comparing the PCI values ​​of different preset units at different points in time.

[0115] For example, when the predicted PCI value is below a certain set threshold, this usually indicates that the performance of the preset units of the pavement is declining, and there may be problems such as cracks, wear, and insufficient structural strength. At this time, the system can automatically mark these preset units as high-risk areas and recommend priority inspection and repair. Through the systematic analysis of these PCI values, a specific maintenance plan can be formulated, including when to conduct regular inspections, when to arrange repairs, and even when to consider the overall renewal of the pavement.

[0116] In addition, by analyzing the changing trends of PCI values ​​of different preset units at different time periods, managers can identify the long-term changing patterns of pavement performance. For example, if the PCI values ​​of certain preset units show a continuous downward trend, this may indicate that the current maintenance strategy is not sufficient to maintain the long-term performance of the pavement and that the maintenance frequency or method needs to be adjusted. On the contrary, if the PCI value is relatively stable, it indicates that the existing maintenance measures are effective and the existing maintenance strategy can continue to be maintained.

[0117] Reference Figure 6 , Figure 6 It is a structural schematic diagram of an airport pavement performance prediction system based on multi-source feature coding provided by the present invention, the system comprising: a first acquisition module, a first processing module and a performance evaluation module; A first acquisition module is used to acquire multiple historical scores and multiple pavement attributes of preset units, wherein the preset unit is obtained after the airport pavement is divided, and the historical score is the historical airport pavement performance score of the corresponding part of the preset unit; A first processing module is used to construct an initial time series feature based on all historical scores and to construct an initial attribute feature based on all road surface attributes; The performance evaluation module is used to input the initial time series features and the initial attribute features into the airport pavement performance prediction model to obtain the prediction score; The performance evaluation module is also used to determine the airport pavement performance based on the predicted score.

[0118] In a possible implementation, the performance evaluation module is further used to input the initial time series features and the initial attribute features into the multi-source feature encoder to obtain a first reconstruction mean parameter and a second reconstruction mean parameter; The performance evaluation module is further used to input the first reconstruction mean parameter and the second reconstruction mean parameter into the performance predictor to obtain a prediction score.

[0119] In a possible implementation, the performance evaluation module is further configured to input the initial time series feature into the time series feature encoding module to obtain a time series feature vector, and input the initial attribute feature into the attribute feature encoding module to obtain an attribute feature vector; The performance evaluation module is further used to input the time series feature vector and the attribute feature vector into the fusion coding module for fusion, and generate a first mean parameter, a first variance parameter, a second mean parameter and a second variance parameter; The performance evaluation module is also used to input the first mean parameter and the first variance parameter into the time series feature decoding module to obtain the first reconstructed mean parameter, and to input the second mean parameter and the second variance parameter into the attribute feature decoding module to obtain the second reconstructed mean parameter.

[0120] In a possible implementation, the system further includes a second processing module; the second processing module is used to construct a time series feature sample of each unit sample based on multiple historical score labels of each unit sample, and to construct an attribute feature sample of each unit sample based on multiple pavement attribute labels of each unit sample; The second processing module is also used to train the initial encoder based on all the time series feature samples and all the attribute feature samples to obtain a multi-source feature encoder.

[0121] In a possible implementation manner, the second processing module is further used to input each time series feature sample and each attribute feature sample into an initial encoder to obtain a first training reconstruction mean parameter, a second training reconstruction mean parameter, a reconstructed attribute feature, and a reconstructed time series feature of each unit sample; The second processing module is further used to input each first training reconstruction mean parameter and each second training reconstruction mean parameter into the performance predictor to obtain a training prediction score of each unit sample; The second processing module is further used to obtain a feature reconstruction error value based on each reconstructed time series feature, each reconstructed attribute feature, each time series feature sample and each attribute feature sample; The second processing module is further used to obtain a prediction error value based on each training prediction score and each historical score label; The second processing module is further used to add the feature reconstruction error value and the prediction error value to obtain a total error value; The second processing module is also used to train the initial encoder with minimizing the total error value as the training target to obtain the multi-source feature encoder.

[0122] In a possible implementation manner, the second processing module is further used to input each time series feature sample into the time series feature encoding module to obtain a time series feature vector sample of each unit sample; The second processing module is further used to input each attribute feature sample into the attribute feature encoding module to obtain an attribute feature vector sample of each unit sample; The second processing module is further used to input each time series feature vector sample and each attribute feature vector sample into the fusion coding module for fusion, so as to generate a first training mean parameter, a first training variance parameter, a second training mean parameter and a second training variance parameter of each unit sample; The second processing module is further used to input each first training mean parameter and each first training variance parameter into the time series feature decoding module to obtain the first training reconstruction mean parameter of each unit sample, and obtain the reconstructed time series feature of each unit sample based on each first training reconstruction mean parameter; The second processing module is also used to input each second training mean parameter and each second training variance parameter into the attribute feature decoding module to obtain the second training reconstruction mean parameter of each unit sample, and obtain the reconstructed attribute feature of each unit sample based on each second training reconstruction mean parameter.

[0123] In a possible implementation, the second processing module is further configured to calculate the prediction score using the following formula: ; Where t is the use time of the airport pavement, is the predicted score corresponding to the i-th unit sample at the t-th usage time, is the first reconstruction mean parameter of the i-th unit sample, is the second reconstruction mean parameter of the i-th unit sample.

[0124] In a possible implementation manner, the second processing module is further configured to calculate a feature reconstruction error value using the following formula: ; in, is the feature reconstruction error value, Q is the number of unit samples, is the reconstructed attribute feature of the i-th unit sample, is the attribute feature sample of the i-th unit sample, is the reconstructed time series feature of the i-th unit sample, is the time series feature sample of the i-th unit sample.

[0125] In a possible implementation manner, the second processing module is further configured to calculate the prediction error value using the following formula: ; in, is the prediction error value, Q is the number of unit samples, T is the time length required to predict the use of the airport pavement, t is the use time of the airport pavement, is the predicted score corresponding to the i-th unit sample at the t-th usage time, is the historical rating label corresponding to the i-th unit sample at the t-th usage time.

[0126] It should be noted that the airport pavement performance prediction system based on multi-source feature coding provided by the present invention can execute the airport pavement performance prediction method based on multi-source feature coding of any of the above-mentioned embodiments during specific operation, which will not be elaborated in this embodiment.

[0127] Figure 7 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 7As shown, the electronic device may include: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus. The processor may call the logic instructions in the memory to execute an airport pavement performance prediction method based on multi-source feature coding, the method comprising: obtaining multiple historical scores and multiple pavement attributes of a preset unit, wherein the preset unit is obtained after the airport pavement is divided, and the historical score is the historical airport pavement performance score of the corresponding part of the preset unit; constructing an initial time series feature based on all historical scores, and constructing an initial attribute feature based on all pavement attributes; inputting the initial time series feature and the initial attribute feature into the airport pavement performance prediction model to obtain a prediction score; determining the airport pavement performance based on the prediction score.

[0128] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of a software function preset unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0129] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the airport pavement performance prediction method based on multi-source feature coding provided in the above-mentioned embodiments, and the method includes: obtaining multiple historical scores and multiple pavement attributes of preset units, wherein the preset unit is obtained after the airport pavement is divided, and the historical score is the historical airport pavement performance score of the corresponding part of the preset unit; based on all historical scores, constructing initial time series features, and based on all pavement attributes, constructing initial attribute features; inputting the initial time series features and initial attribute features into the airport pavement performance prediction model to obtain a predicted score; based on the predicted score, determining the airport pavement performance.

[0130] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the airport pavement performance prediction method based on multi-source feature coding provided in the above-mentioned embodiments, the method comprising: obtaining multiple historical scores and multiple pavement attributes of preset units, wherein the preset units are obtained after the airport pavement is divided, and the historical scores are the historical airport pavement performance scores of the corresponding parts of the preset units; constructing initial time series features based on all historical scores, and constructing initial attribute features based on all pavement attributes; inputting the initial time series features and the initial attribute features into an airport pavement performance prediction model to obtain a predicted score; and determining the airport pavement performance based on the predicted score.

[0131] The system embodiment described above is merely illustrative, wherein the preset units described as separate components may or may not be physically separated, and the components displayed as preset units may or may not be physical preset units, that is, they may be located in one place, or they may be distributed on multiple network preset 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.

[0132] Through the description of the above implementation modes, those skilled in the art can clearly understand that each implementation mode can be implemented by means of software plus a necessary general hardware platform, or of course by hardware. Based on such an understanding, the above technical solution can essentially or in other words be embodied in the form of a software product that contributes to the prior art. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes 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 of each embodiment or some parts of the embodiment.

[0133] 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 method for predicting airport pavement performance based on multi-source feature coding, characterized in that: include: Acquire multiple historical scores and multiple pavement attributes of preset units, wherein the preset units are obtained after the airport pavement is divided, and the historical scores are historical airport pavement performance scores corresponding to the preset units; Based on all of the historical scores, construct an initial time series feature, and based on all of the road surface attributes, construct an initial attribute feature; Inputting the initial time series features and the initial attribute features into an airport pavement performance prediction model to obtain a prediction score; Based on the predicted score, an airport pavement performance is determined.

2. The airport pavement performance prediction method based on multi-source feature coding according to claim 1 is characterized in that: The airport pavement performance prediction model includes a multi-source feature encoder and a performance predictor; the initial time series features and the initial attribute features are input into the airport pavement performance prediction model to obtain a prediction score, specifically including: Inputting the initial time series features and the initial attribute features into the multi-source feature encoder to obtain a first reconstruction mean parameter and a second reconstruction mean parameter; The first reconstructed mean parameter and the second reconstructed mean parameter are input into the performance predictor to obtain the prediction score.

3. The airport pavement performance prediction method based on multi-source feature coding according to claim 2 is characterized in that: The multi-source feature encoder includes a temporal feature encoding module, an attribute feature encoding module, a fusion encoding module, a temporal feature decoding module and an attribute feature decoding module; The inputting the initial time series features and the initial attribute features into the multi-source feature encoder to obtain a first reconstruction mean parameter and a second reconstruction mean parameter specifically includes: Inputting the initial time series feature into the time series feature encoding module to obtain a time series feature vector, and inputting the initial attribute feature into the attribute feature encoding module to obtain an attribute feature vector; Inputting the time series feature vector and the attribute feature vector into the fusion coding module for fusion, generating a first mean parameter, a first variance parameter, a second mean parameter and a second variance parameter; The first mean parameter and the first variance parameter are input into the time series feature decoding module to obtain the first reconstructed mean parameter, and the second mean parameter and the second variance parameter are input into the attribute feature decoding module to obtain the second reconstructed mean parameter.

4. The airport pavement performance prediction method based on multi-source feature coding according to claim 2 is characterized in that: The multi-source feature encoder is trained by the following method: Based on the multiple historical score labels of each unit sample, construct a time series feature sample of each unit sample, and based on the multiple road surface attribute labels of each unit sample, construct an attribute feature sample of each unit sample; The initial encoder is trained based on all the time series feature samples and all the attribute feature samples to obtain the multi-source feature encoder.

5. The airport pavement performance prediction method based on multi-source feature coding according to claim 4 is characterized in that: The training of the initial encoder based on all the time series feature samples and all the attribute feature samples to obtain the multi-source feature encoder specifically includes: Input each of the time series feature samples and each of the attribute feature samples into the initial encoder to obtain a first training reconstruction mean parameter, a second training reconstruction mean parameter, a reconstructed attribute feature, and a reconstructed time series feature of each of the unit samples; Inputting each of the first training reconstruction mean parameters and each of the second training reconstruction mean parameters into the performance predictor to obtain a training prediction score for each of the unit samples; Obtaining a feature reconstruction error value based on each of the reconstructed time series features, each of the reconstructed attribute features, each of the time series feature samples, and each of the attribute feature samples; Obtaining a prediction error value based on each of the training prediction scores and each of the historical score labels; Adding the feature reconstruction error value and the prediction error value to obtain a total error value; The initial encoder is trained with minimizing the total error value as a training target to obtain the multi-source feature encoder.

6. The method for predicting airport pavement performance based on multi-source feature coding according to claim 5 is characterized in that: The initial encoder includes a temporal feature encoding module, an attribute feature encoding module, a fusion encoding module, a temporal feature decoding module and an attribute feature decoding module; The step of inputting each of the time series feature samples and each of the attribute feature samples into the initial encoder to obtain a first training reconstruction mean parameter, a second training reconstruction mean parameter, a reconstructed attribute feature, and a reconstructed time series feature of each of the unit samples specifically includes: Input each of the time series feature samples into the time series feature encoding module to obtain a time series feature vector sample of each of the unit samples; Input each of the attribute feature samples into the attribute feature encoding module to obtain an attribute feature vector sample of each of the unit samples; Inputting each of the time series feature vector samples and each of the attribute feature vector samples into the fusion coding module for fusion, and generating a first training mean parameter, a first training variance parameter, a second training mean parameter and a second training variance parameter for each of the unit samples; Inputting each of the first training mean parameters and each of the first training variance parameters into the time series feature decoding module to obtain a first training reconstruction mean parameter of each of the unit samples, and obtaining a reconstructed time series feature of each of the unit samples based on each of the first training reconstruction mean parameters; Each of the second training mean parameters and each of the second training variance parameters are input into the attribute feature decoding module to obtain the second training reconstruction mean parameter of each of the unit samples, and based on each of the second training reconstruction mean parameters, the reconstructed attribute feature of each of the unit samples is obtained.

7. The method for predicting airport pavement performance based on multi-source feature coding according to claim 5, characterized in that: The step of inputting each of the first training reconstruction mean parameters and each of the second training reconstruction mean parameters into the performance predictor to obtain a training prediction score for each of the unit samples specifically includes: The prediction score is calculated by the following formula: ； Where t is the use time of the airport pavement, is the predicted score corresponding to the i-th unit sample at the t-th usage time, is the first reconstructed mean parameter of the i-th unit sample, is the second reconstructed mean parameter of the i-th unit sample.

8. The method for predicting airport pavement performance based on multi-source feature coding according to claim 5, characterized in that: The obtaining of a feature reconstruction error value based on each of the reconstructed time series features, each of the reconstructed attribute features, each of the time series feature samples, and each of the attribute feature samples specifically includes: The feature reconstruction error value is calculated by the following formula: ； in, is the feature reconstruction error value, Q is the number of unit samples, is the reconstructed attribute feature of the i-th unit sample, is the attribute feature sample of the i-th unit sample, is the reconstructed time series feature of the i-th unit sample, is the timing feature sample of the i-th unit sample.

9. The method for predicting airport pavement performance based on multi-source feature coding according to claim 5, characterized in that: The obtaining of a prediction error value based on each of the training prediction scores and each of the historical score labels specifically includes: The prediction error value is calculated by the following formula: ； in, is the prediction error value, Q is the number of unit samples, T is the time length required for the airport pavement to be predicted, t is the usage time of the airport pavement, is the predicted score corresponding to the i-th unit sample at the t-th usage time, is the historical scoring label corresponding to the i-th unit sample at the t-th usage time.

10. An airport pavement performance prediction system based on multi-source feature coding, characterized in that: include: A first acquisition module, a first processing module and a performance evaluation module; The first acquisition module is used to acquire multiple historical scores and multiple pavement attributes of a preset unit, wherein the preset unit is obtained after the airport pavement is divided, and the historical score is the historical airport pavement performance score corresponding to the preset unit; The first processing module is used to construct an initial time series feature based on all the historical scores, and to construct an initial attribute feature based on all the road surface attributes; The performance evaluation module is used to input the initial time series features and the initial attribute features into the airport pavement performance prediction model to obtain a prediction score; The performance evaluation module is further used to determine the airport pavement performance based on the predicted score.

11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method for predicting airport pavement performance based on multi-source feature coding as described in any one of claims 1 to 9 is implemented.

12. 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, the method for predicting airport pavement performance based on multi-source feature coding as described in any one of claims 1 to 9 is implemented.

13. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for predicting airport pavement performance based on multi-source feature coding as described in any one of claims 1 to 9 is implemented.

Images