Method for predicting life of catenary based on data enhancement and spatial feature extraction

Abstract

This invention relates to the field of high-speed rail data processing technology, providing a method for predicting the lifespan of overhead contact lines based on data augmentation and spatial feature extraction. The method includes: collecting dynamic and static data sets of the overhead contact lines and a data table of overhead contact line defect records to obtain a dataset to be processed; extracting the optimal feature dataset; extracting the dataset to be augmented from the optimal feature dataset and constructing a generative adversarial network (GAN) to form an augmented dataset; constructing a graph structure and generating several adjacency matrices to build a multi-scale graph convolutional layer; and predicting the augmented dataset based on the multi-scale graph convolutional layer to obtain the remaining lifespan prediction result. By employing the above method, the accuracy of predicting the remaining lifespan of the overhead contact lines can be improved.

Description

Technical Field

[0001] This invention relates to the field of high-speed rail data processing technology, and in particular to a method for predicting the lifespan of overhead contact lines based on data augmentation and spatial feature extraction. Background Technology

[0002] As a vital component of modern transportation, high-speed railways play a crucial role in safeguarding passenger lives and property and promoting socio-economic development. During high-speed operation, high-speed trains are powered by a pantograph-overhead contact system. The overhead contact system is key to providing a stable and reliable power supply to high-speed trains, and its stability and reliability directly affect the normal operation of the train.

[0003] To ensure the normal operation of trains, fault prediction and health management of the overhead contact system are crucial, with remaining service life prediction being a key technology in this process. Accurate prediction of the remaining service life of the overhead contact system can reduce unnecessary maintenance, enabling efficient maintenance while also preventing serious problems such as train downtime, electrical fires, and economic losses caused by overhead contact system failures. Existing methods for predicting the remaining service life of the overhead contact system mainly fall into two categories: physical model-based methods and data-driven model-based methods. Physical model methods predict remaining service life by establishing physical degradation models of the equipment, suitable for equipment systems with clear failure mechanisms and sufficient domain knowledge. However, high-speed rail overhead contact systems are highly complex, making it difficult to fully quantify physical models. Data-driven model methods, on the other hand, construct remaining service life prediction models based on collected condition monitoring data through mathematical statistics and artificial neural networks. Therefore, data-driven model methods are widely used in the field of remaining service life prediction.

[0004] However, the prediction accuracy of data-driven models depends on the quantity and quality of data and the quality of feature extraction. Since the maintenance of the high-speed rail catenary is relatively strict and frequent, the amount of fault data is relatively small. Existing models are unable to fully capture global structural information for objects like the catenary with complex structures and multi-scale features. When extracting features, they tend to ignore the relationships between samples, resulting in insufficient accuracy of prediction results. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a method for predicting the remaining service life of overhead contact lines based on data augmentation and spatial feature extraction. This invention achieves more accurate predictions of the remaining service life of overhead contact lines by performing data augmentation and spatial feature extraction. The present invention aims to solve the technical problem of insufficient accuracy in predicting the remaining service life of overhead contact lines in existing technologies.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0007] A method for predicting the life of overhead contact lines based on data augmentation and spatial feature extraction includes the following steps:

[0008] Collect dynamic and static data sets of the overhead contact system and a data table of overhead contact system defects. Based on the dynamic and static data sets of the overhead contact system and the data table of overhead contact system defects, obtain the dataset to be processed.

[0009] The dataset to be processed is preprocessed to form a basic dataset, and the optimal feature dataset is extracted from the basic dataset using a filtering method.

[0010] The dataset to be augmented is extracted from the optimal feature dataset. A generative adversarial network is constructed based on the penalized gradient and Wasserstein distance. The dataset to be augmented is then augmented using the generative adversarial network to form an expanded dataset.

[0011] Based on the augmented dataset, a graph structure is constructed, and several adjacency matrices are generated to construct multi-scale graph convolutional layers;

[0012] Based on the multi-scale graph convolutional layer, predictions are made on the augmented dataset to obtain the remaining lifetime prediction results.

[0013] Furthermore, the overhead contact line defect record data table includes several defect data points, and the step of obtaining the dataset to be processed based on the overhead contact line dynamic and static data group and the overhead contact line defect record data table includes:

[0014] Identify the contact wire support pillars corresponding to the defect data, and identify the kilometer markers corresponding to the contact wire support pillars;

[0015] Based on the kilometer markers, a catenary parameter dataset is extracted from the catenary dynamic and static data set. The data types of the catenary parameter dataset include speed, conductor height, pull-out value, hard point, grid voltage, spark time, contact force, and intra-span height difference. Several kilometer markers and the catenary parameter dataset constitute a dataset to be processed.

[0016] Furthermore, the step of preprocessing the dataset to be processed to form the base dataset includes:

[0017] Several scatter plots are drawn based on the dataset to be processed to identify several outliers in the dataset, and the outliers are removed to form the first dataset.

[0018] Linear interpolation is used to fill in several missing values ​​in the first dataset to form the second dataset;

[0019] The second dataset is normalized to form the base dataset.

[0020] Furthermore, the step of extracting the optimal feature dataset from the basic dataset using a filtering method includes:

[0021] The RUL value is set according to the maintenance interval of the overhead contact line equipment. Several related features are established based on the data types of the basic dataset. The data types of the basic dataset are consistent with the data types of the overhead contact line parameter dataset. Several feature values ​​corresponding to the related features are extracted from the basic dataset to calculate the Pearson correlation coefficient between the related features and the RUL value.

[0022] The absolute values ​​of several Pearson correlation coefficients are sorted by size to select several optimal features from several related features, and the feature values ​​corresponding to several optimal features are combined to form an optimal feature dataset.

[0023] Furthermore, the steps for constructing a generative adversarial network based on penalized gradients and Wasserstein distance include:

[0024] Based on the GAN model, the Wasserstein distance is used as the loss function in the GAN model to form a basic adversarial network;

[0025] Lipschitz constraints are applied to the base adversarial network based on the penalized gradient to construct a generative adversarial network.

[0026] Furthermore, the generative adversarial network includes a generator and a discriminator, and the loss function of the generator is:

[0027]

[0028] in, Represents the loss function of the generator. Indicates random noise. express The sampling sample, express The expectation of the joint distribution, This indicates the output of the discriminator. This indicates the generator output.

[0029] The loss function of the discriminator is:

[0030]

[0031] in, This represents the loss function of the discriminator. This represents actual overhead contact line sample data. express The sampling sample.

[0032] Furthermore, the step of constructing a graph structure based on the augmented dataset and generating several adjacency matrices to construct a multi-scale graph convolutional layer includes:

[0033] Using the catenary support as a node and the connection between two nodes as an edge, a graph structure consisting of several nodes and several edges is constructed based on the extended dataset;

[0034] Based on several edges, the Euclidean distance formula is used for calculation and analysis to obtain several adjacency matrices;

[0035] A multi-scale graph convolutional layer is constructed based on several adjacency matrices and several nodes.

[0036] Furthermore, the step of setting up a multi-scale graph convolutional layer based on several adjacency matrices and several nodes includes:

[0037] Based on several adjacency matrices, several neighboring nodes corresponding to each node are obtained;

[0038] Based on the nodes and the neighboring nodes corresponding to the nodes, construct several subgraph convolutional layers;

[0039] Based on the number of nodes and the number of neighboring nodes corresponding to the nodes, the scale of the subgraph convolutional layer is set so as to form a multi-scale graph convolutional layer from the subgraph convolutional layers according to the scale.

[0040] Furthermore, the step of predicting the augmented dataset based on the multi-scale graph convolutional layer to obtain the remaining lifetime prediction result includes:

[0041] Based on the attention mechanism, weights are assigned to all nodes in the subgraph convolutional layer to form a weight matrix, and dynamic attention weights are calculated between all nodes in the subgraph convolutional layer and the neighboring nodes corresponding to the nodes.

[0042] Based on several weight matrices, several dynamic attention weights, and several adjacency matrices, the aggregated features of several subgraph convolutional layers are calculated;

[0043] The aggregated features are weighted and summed to obtain several final node outputs. The final node outputs are then mapped using a fully connected layer to obtain the RUL prediction value, thereby obtaining the remaining lifetime prediction result.

[0044] Compared with existing technologies, the beneficial effects of this invention are as follows: By preprocessing and the filtering method, the collected data is processed to remove outliers and reduce irrelevant features, thereby improving data quality and feature extraction quality; by constructing the generative adversarial network, the problem of insufficient contact network defect data is effectively solved; and the generative adversarial network constructed using penalized gradients and Wasserstein distance overcomes the problems of pattern collapse and gradient vanishing that easily occur in traditional adversarial networks, ensuring the quality of the generated data; by analyzing the spatial characteristics of the contact network using Euclidean distance and employing dynamic attention weights when extracting features, the prediction results are obtained through the multi-scale graph convolutional layer, effectively improving the accuracy of the prediction results. Attached Figure Description

[0045] Figure 1 This is a flowchart of the overhead contact line life prediction method based on data augmentation and spatial feature extraction in the first embodiment of the present invention.

[0046] Figure 2 This is a structural block diagram of the overhead contact line life prediction system based on data augmentation and spatial feature extraction in the second embodiment of the present invention;

[0047] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0048] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0049] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0050] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0051] Please see Figure 1The first embodiment of the present invention provides a method for predicting the life of overhead contact lines based on data augmentation and spatial feature extraction, comprising the following steps:

[0052] Step S10: Collect the dynamic and static data set of the overhead contact system and the data table of overhead contact system defects. Based on the dynamic and static data set of the overhead contact system and the data table of overhead contact system defects, obtain the dataset to be processed.

[0053] Preferably, the contact network dynamic and static data set can be collected by an 1C dynamic inspection vehicle, with a collection period of half a month, and the contact network defect record data table is obtained from the fault repair report.

[0054] Step S10 includes:

[0055] S110: Establish the contact wire support corresponding to the defect data, and establish the kilometer marker corresponding to the contact wire support;

[0056] Preferably, the high-speed railway overhead contact system consists of several overhead contact system supports, and the overhead contact system supports are connected by lines.

[0057] S120: Based on the kilometer markers, extract the catenary parameter dataset from the catenary dynamic and static data group. The data types of the catenary parameter dataset include speed, conductor height, pull-out value, hard point, grid voltage, spark time, contact force, and intra-span height difference. Several kilometer markers and the catenary parameter dataset constitute a dataset to be processed.

[0058] Understandably, the maintenance of high-speed rail overhead contact lines is quite rigorous and frequent, hence the limited amount of defect data.

[0059] Step S20: Preprocess the dataset to be processed to form a basic dataset, and use a filtering method to extract the optimal feature dataset from the basic dataset;

[0060] Understandably, preprocessing the dataset to be processed, removing outliers, and extracting features that are highly correlated with the remaining service life of the overhead contact line can effectively improve the quality of the data and the quality of feature extraction.

[0061] Step S20 includes:

[0062] S210: Draw several scatter plots based on the dataset to be processed to identify several outliers in the dataset to be processed, remove the outliers, and form a first dataset;

[0063] Preferably, a scatter plot is created according to each data type. For example, a scatter plot corresponding to the guide height is created, with the vertical axis representing the guide height value and the horizontal axis representing the data sequence number. Several guide height values ​​are arranged sequentially in the plot to form the scatter plot corresponding to the guide height. Outliers are removed based on the data point graph in the scatter plot.

[0064] S220: Use linear interpolation to fill in several missing values ​​in the first dataset to form a second dataset;

[0065] Understandably, supplementing the missing values ​​is beneficial to improving the data quality of the second dataset.

[0066] S230: Normalize the second dataset to form the base dataset;

[0067] The second dataset includes catenary data with different units and amplitudes. Data with smaller values ​​are easily lost. Therefore, a data normalization method is used to map all data in the second dataset to [-1,1], which is beneficial to improve the convergence speed and accuracy of model training.

[0068] S240: Set the RUL value according to the maintenance interval of the overhead contact line equipment, establish several related features based on the data types of the basic dataset, the data types of the basic dataset are consistent with the data types of the overhead contact line parameter dataset, extract several feature values ​​corresponding to the related features from the basic dataset, and calculate the Pearson correlation coefficient between the related features and the RUL value;

[0069] Preferably, the maintenance time interval of the overhead contact line equipment is the difference between the previous maintenance time and the next maintenance time, each data type corresponds to a related feature, and the several feature values ​​corresponding to a related feature are several data values ​​under that data type.

[0070] S250: Sort the absolute values ​​of the Pearson correlation coefficients by size, select several optimal features from the several correlation features, and form an optimal feature dataset by combining the feature values ​​corresponding to the several optimal features.

[0071] Preferably, six optimal features are selected, and the absolute values ​​of the Pearson correlation coefficients are arranged from largest to smallest. The correlation features corresponding to the absolute values ​​of the top six Pearson correlation coefficients are selected as the optimal features.

[0072] Step S30: Extract the dataset to be augmented from the optimal feature dataset, construct a generative adversarial network based on the penalized gradient and Wasserstein distance, and augment the dataset to be augmented through the generative adversarial network to form an expanded dataset;

[0073] Preferably, 80% of the data in the optimal feature dataset constitutes the dataset to be augmented, and the remaining data is used to test the model's performance. Understandably, by constructing the generative adversarial network (GAN), the original data is augmented to form the expanded dataset, effectively solving the problem of insufficient contact network defect data. Furthermore, the GAN constructed using penalized gradients and Wasserstein distance overcomes the problems of mode collapse and gradient vanishing that easily occur in traditional adversarial networks, ensuring the quality of the generated data.

[0074] Step S30 includes:

[0075] S310: Based on the GAN model, the Wasserstein distance is used as the loss function in the GAN model to form a basic adversarial network;

[0076] In traditional GAN ​​models, if two data distributions differ significantly, the JS divergence may be zero, causing the gradient update of the discriminator to the generator in the GAN model to stagnate, resulting in the gradient vanishing problem. By using Wasserstein distance instead of the loss function in traditional GAN ​​models, even if the two data distributions do not overlap, it is still possible to maintain the measurement of the difference between the two data distributions.

[0077] S320: Apply Lipschitz constraints to the base adversarial network based on the penalized gradient to construct a generative adversarial network.

[0078] By replacing weight pruning with a gradient penalty mechanism, gradient explosion and non-convergence can be effectively prevented.

[0079] In step S30, the generative adversarial network includes a generator and a discriminator, and the loss function of the generator is:

[0080]

[0081] in, Represents the loss function of the generator. Indicates random noise. express The sampling sample, express The expectation of the joint distribution, This indicates the output of the discriminator. This indicates the generator output.

[0082] The loss function of the discriminator is:

[0083]

[0084] in, This represents the loss function of the discriminator. This represents actual overhead contact line sample data. express The sampling sample.

[0085] Preferably, the generator consists of an input layer, multiple deconvolutional layers, and activation layers. It upsamples random noise from a one-dimensional space to a high-dimensional space of the target sample. The deconvolutional layers expand the low-dimensional input into a high-dimensional output. Several deconvolutional layers are followed by activation layers, which use ReLU as the activation function to enhance the nonlinear expressive power of the network. The last deconvolutional layer in the generator uses the Tanh activation function. The discriminator is used to distinguish between generated data and real data. It consists of convolutional layers and activation layers. Each convolutional layer is followed by an activation layer and a batch normalization layer. The activation function used is LeakyReLU, which is beneficial for introducing nonlinearity and preventing gradient vanishing.

[0086] Step S40: Based on the augmented dataset, construct a graph structure and generate several adjacency matrices to construct a multi-scale graph convolutional layer;

[0087] As an object with complex structure and multi-scale features, the catenary is difficult to extract spatial features from traditional LSTM and CNN networks, making it hard to capture global structural information. The multi-scale graph convolutional layer described above can extract the multi-scale spatial features of the catenary, which is beneficial to improving the accuracy of predicting the remaining service life of the catenary.

[0088] Step S40 includes:

[0089] S410: Using the catenary support as a node and the connection between two nodes as an edge, construct a graph structure based on the extended dataset, consisting of several nodes and several edges;

[0090] S420: Based on several edges, perform calculation and analysis using the Euclidean distance formula to obtain several adjacency matrices;

[0091] S430: Construct a multi-scale graph convolutional layer based on several adjacency matrices and several nodes.

[0092] In S410~S430, for the catenary system, the degree of mutual influence between the nodes is different, and the influence range of the neighboring nodes of each node is also different. Traditional graph convolutional neural networks determine fixed weights for all neighboring nodes according to the topology, and the influence between the nodes gradually weakens, which limits the expressive power of the graph convolutional neural network. Understandably, by analyzing the spatial characteristics of the catenary through Euclidean distance and analyzing the interrelationship between the catenary supports, the quality of spatial feature extraction can be improved.

[0093] The S430 also includes:

[0094] S4310: Based on several adjacency matrices, obtain several neighboring nodes corresponding to each node;

[0095] S4320: Construct several subgraph convolutional layers based on several nodes and several neighboring nodes corresponding to several nodes;

[0096] S4330: Based on the number of nodes and the number of neighboring nodes corresponding to the nodes, set the scale of the subgraph convolutional layer to form a multi-scale graph convolutional layer from the subgraph convolutional layers according to the scale.

[0097] Step S50: Based on the multi-scale graph convolutional layer, make predictions on the augmented dataset to obtain the remaining lifetime prediction results.

[0098] Preferably, the subgraph convolutional layer at each scale is a GCN graph convolutional layer. The connection structures formed by the nodes and their neighboring nodes in different subgraph convolutional layers are different. Through multi-scale aggregation, it is beneficial to extract the spatial features of the catenary system and effectively improve the accuracy of the prediction results.

[0099] Step S50 includes:

[0100] S510: Based on the attention mechanism, assign weights to all nodes in the subgraph convolutional layer to form a weight matrix, and calculate the dynamic attention weights between all nodes in the subgraph convolutional layer and the neighboring nodes corresponding to the nodes.

[0101] S520: Based on several weight matrices, several dynamic attention weights, and several adjacency matrices, calculate the aggregate features of several subgraph convolutional layers;

[0102] S530: The aggregated features are weighted and summed to obtain several final node outputs. The final node outputs are then mapped using a fully connected layer to obtain the RUL prediction value, thereby obtaining the remaining lifetime prediction result.

[0103] Preferably, real overhead contact line data is obtained. In step S30, 80% of the data in the optimal feature dataset is used to train the prediction model formed by the overhead contact line life prediction method based on data augmentation and spatial feature extraction described in this embodiment. The remaining 20% ​​of the data is used to test the prediction model, and the predicted RUL value is calculated. The actual RUL value is obtained by combining the collected fault inspection reports. The prediction accuracy of the prediction model can be evaluated by calculating the mean absolute error, root mean square error, and mean absolute percentage error between the actual RUL value and the predicted RUL value. The mean absolute error of the prediction model is 2.291%, the root mean square error is 3.249%, and the mean absolute percentage error is 4.64%. The same data is used to train and test CNN and LSTM models respectively. The proposed methods for predicting the lifespan of overhead contact lines using data augmentation and spatial feature extraction effectively improve the accuracy of such predictions. Specifically, the CNN model has a mean absolute error of 8.284%, a root mean square error of 11.806%, and a mean absolute percentage error of 12.94%; the LSTM model has a mean absolute error of 10.902%, a root mean square error of 14.951%, and a mean absolute percentage error of 15.01%; the DCNN model has a mean absolute error of 4.296%, a root mean square error of 6.045%, and a mean absolute percentage error of 8.88%; and the GCN model has a mean absolute error of 4.649%, a root mean square error of 6.219%, and a mean absolute percentage error of 8.75%.

[0104] Please see Figure 2 The second embodiment of the present invention provides a catenary life prediction system based on data augmentation and spatial feature extraction, which applies the catenary life prediction method based on data augmentation and spatial feature extraction described in the first embodiment above. The system includes:

[0105] The acquisition module 10 is used to acquire the dynamic and static data set of the overhead contact system and the data table of overhead contact system defects, and to obtain the dataset to be processed based on the dynamic and static data set of the overhead contact system and the data table of overhead contact system defects.

[0106] The acquisition module 10 includes:

[0107] The first unit is used to identify the contact wire support corresponding to the defect data and to identify the kilometer marker corresponding to the contact wire support.

[0108] The second unit is used to extract the catenary parameter dataset from the catenary dynamic and static data group according to the kilometer markers. The data types of the catenary parameter dataset include speed, conductor height, pull-out value, hard point, grid voltage, spark time, contact force and intra-span height difference. Several kilometer markers and the catenary parameter dataset constitute a dataset to be processed.

[0109] Processing module 20 is used to preprocess the dataset to be processed to form a basic dataset, and to extract the optimal feature dataset from the basic dataset using a filtering method;

[0110] The processing module 20 includes:

[0111] The third unit is used to draw several scatter plots based on the dataset to be processed, in order to identify several outliers in the dataset to be processed, remove the outliers, and form the first dataset.

[0112] The fourth unit is used to fill in several missing values ​​in the first dataset using linear interpolation to form the second dataset;

[0113] The fifth unit is used to normalize the second dataset to form the base dataset;

[0114] The sixth unit is used to set the RUL value according to the maintenance interval of the overhead contact line equipment, establish several relevant features based on the data types of the basic dataset, the data types of the basic dataset are consistent with the data types of the overhead contact line parameter dataset, and extract several feature values ​​corresponding to the relevant features from the basic dataset to calculate the Pearson correlation coefficient between the relevant features and the RUL value.

[0115] The seventh unit is used to sort the absolute values ​​of several Pearson correlation coefficients by size, so as to select several optimal features from several related features, and to form an optimal feature dataset by combining several feature values ​​corresponding to several optimal features.

[0116] The expansion module 30 is used to extract the dataset to be enhanced from the optimal feature dataset, construct a generative adversarial network based on the penalized gradient and Wasserstein distance, and perform data enhancement on the dataset to be enhanced through the generative adversarial network to form an expanded dataset.

[0117] The expansion module 30 includes:

[0118] The eighth unit is used to form a basic adversarial network based on the GAN model, using the Wasserstein distance as the loss function in the GAN model.

[0119] The ninth unit is used to apply Lipschitz constraints to the basic adversarial network based on the penalized gradient in order to construct a generative adversarial network.

[0120] In the expansion module 30, the generative adversarial network includes a generator and a discriminator, and the loss function of the generator is:

[0121]

[0122] in, Represents the loss function of the generator. Indicates random noise. express The sampling sample, express The expectation of the joint distribution, This indicates the output of the discriminator. This indicates the generator output.

[0123] The loss function of the discriminator is:

[0124]

[0125] in, This represents the loss function of the discriminator. This represents actual overhead contact line sample data. express The sampling sample.

[0126] The construction module 40 is used to construct a graph structure based on the augmented dataset and generate several adjacency matrices to construct a multi-scale graph convolutional layer;

[0127] The construction module 40 includes:

[0128] The tenth unit is used to construct a graph structure consisting of several nodes and several edges based on the extended dataset, using the catenary support as a node and the connection relationship between two nodes as an edge.

[0129] The eleventh unit is used to perform calculations and analyses based on several edges using the Euclidean distance formula to obtain several adjacency matrices;

[0130] The twelfth unit is used to construct a multi-scale graph convolutional layer based on several adjacency matrices and several nodes;

[0131] The twelfth unit is specifically used to obtain several neighboring nodes corresponding to each node based on several adjacency matrices;

[0132] Based on the nodes and the neighboring nodes corresponding to the nodes, construct several subgraph convolutional layers;

[0133] Based on the number of nodes and the number of neighboring nodes corresponding to the nodes, the scale of the subgraph convolutional layer is set so as to form a multi-scale graph convolutional layer from the subgraph convolutional layers according to the scale.

[0134] The prediction module 50 is used to make predictions on the augmented dataset based on the multi-scale graph convolutional layer to obtain the remaining lifetime prediction result.

[0135] The prediction module 50 includes:

[0136] The thirteenth unit is used to assign weights to all nodes in the subgraph convolutional layer based on an attention mechanism to form a weight matrix, and to calculate the dynamic attention weights between all nodes in the subgraph convolutional layer and the neighboring nodes corresponding to the nodes.

[0137] The fourteenth unit is used to calculate the aggregate features of several subgraph convolutional layers based on several weight matrices, several dynamic attention weights, and several adjacency matrices.

[0138] The fifteenth unit is used to perform a weighted summation of several aggregated features to obtain several final node outputs, and to perform a fully connected layer mapping on several final node outputs to obtain RUL prediction values, so as to obtain the remaining lifetime prediction results.

[0139] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0140] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for predicting the lifespan of overhead contact lines based on data augmentation and spatial feature extraction, characterized in that, Includes the following steps: Collect dynamic and static data sets of the overhead contact system and a data table of overhead contact system defects. Based on the dynamic and static data sets of the overhead contact system and the data table of overhead contact system defects, obtain the dataset to be processed. The overhead contact line defect record data table includes several defect data. The step of obtaining the dataset to be processed based on the overhead contact line dynamic and static data group and the overhead contact line defect record data table includes: Identify the contact wire support pillars corresponding to the defect data, and identify the kilometer markers corresponding to the contact wire support pillars; Based on the kilometer markers, a catenary parameter dataset is extracted from the catenary dynamic and static data set. The data types of the catenary parameter dataset include speed, conductor height, pull-out value, hard point, grid voltage, spark time, contact force, and intra-span height difference. Several kilometer markers and the catenary parameter dataset constitute a dataset to be processed. The dataset to be processed is preprocessed to form a basic dataset, and the optimal feature dataset is extracted from the basic dataset using a filtering method. The dataset to be augmented is extracted from the optimal feature dataset. A generative adversarial network is constructed based on the penalized gradient and Wasserstein distance. The dataset to be augmented is then augmented using the generative adversarial network to form an expanded dataset. Based on the augmented dataset, a graph structure is constructed, and several adjacency matrices are generated to construct multi-scale graph convolutional layers; The steps of constructing a graph structure based on the augmented dataset and generating several adjacency matrices to construct a multi-scale graph convolutional layer include: Using the catenary support as a node and the connection between two nodes as an edge, a graph structure consisting of several nodes and several edges is constructed based on the extended dataset; Based on several edges, the Euclidean distance formula is used for calculation and analysis to obtain several adjacency matrices; A multi-scale graph convolutional layer is constructed based on several adjacency matrices and several nodes; The step of setting up a multi-scale graph convolutional layer based on several adjacency matrices and several nodes includes: Based on several adjacency matrices, several neighboring nodes corresponding to each node are obtained; Based on the nodes and the neighboring nodes corresponding to the nodes, construct several subgraph convolutional layers; Based on the number of nodes and the number of neighboring nodes corresponding to the nodes, the scale of the subgraph convolutional layer is set so as to form a multi-scale graph convolutional layer from the subgraph convolutional layers according to the scale. Based on the multi-scale graph convolutional layer, predictions are made on the augmented dataset to obtain the remaining lifetime prediction results.

2. The catenary life prediction method based on data augmentation and spatial feature extraction according to claim 1, characterized in that, The step of preprocessing the dataset to be processed to form the base dataset includes: Several scatter plots are drawn based on the dataset to be processed to identify several outliers in the dataset, and the outliers are removed to form the first dataset. Linear interpolation is used to fill in several missing values ​​in the first dataset to form the second dataset; The second dataset is normalized to form the base dataset.

3. The catenary life prediction method based on data augmentation and spatial feature extraction according to claim 1, characterized in that, The step of extracting the optimal feature dataset from the base dataset using a filtering method includes: The RUL value is set according to the maintenance interval of the overhead contact line equipment. Several related features are established based on the data types of the basic dataset. The data types of the basic dataset are consistent with the data types of the overhead contact line parameter dataset. Several feature values ​​corresponding to the related features are extracted from the basic dataset to calculate the Pearson correlation coefficient between the related features and the RUL value. The absolute values ​​of several Pearson correlation coefficients are sorted by size to select several optimal features from several related features, and the feature values ​​corresponding to several optimal features are combined to form an optimal feature dataset.

4. The catenary life prediction method based on data augmentation and spatial feature extraction according to claim 1, characterized in that, The steps for constructing a generative adversarial network based on penalized gradients and Wasserstein distance include: Based on the GAN model, the Wasserstein distance is used as the loss function in the GAN model to form a basic adversarial network; Lipschitz constraints are applied to the base adversarial network based on the penalized gradient to construct a generative adversarial network.

5. The catenary life prediction method based on data augmentation and spatial feature extraction according to claim 1, characterized in that, The generative adversarial network includes a generator and a discriminator, and the loss function of the generator is: in, Represents the loss function of the generator. Indicates random noise. express The sampling sample, express The expectation of the joint distribution, This indicates the output of the discriminator. This indicates the generator output. The loss function of the discriminator is: in, This represents the loss function of the discriminator. This represents actual overhead contact line sample data. express The sampling sample.

6. The catenary life prediction method based on data augmentation and spatial feature extraction according to claim 1, characterized in that, The step of predicting the augmented dataset based on the multi-scale graph convolutional layer to obtain the remaining lifetime prediction result includes: Based on the attention mechanism, weights are assigned to all nodes in the subgraph convolutional layer to form a weight matrix, and dynamic attention weights are calculated between all nodes in the subgraph convolutional layer and the neighboring nodes corresponding to the nodes. Based on several weight matrices, several dynamic attention weights, and several adjacency matrices, the aggregated features of several subgraph convolutional layers are calculated; The aggregated features are weighted and summed to obtain several final node outputs. The final node outputs are then mapped using a fully connected layer to obtain the RUL prediction value, thereby obtaining the remaining lifetime prediction result.

Images