Multi-modal fusion intelligent detection system for railway track contact network suspension monitoring

Through multimodal data acquisition and processing units, combined with nonlinear feature extraction and multi-scale morphological analysis, the abnormal state index is calculated, which solves the problems of low monitoring efficiency and single data in the contact network suspension system in the existing technology, and achieves more accurate The fault detection and early warning improves the safety and reliability of the system.

CN118797532BActive Publication Date: 2024-11-15CHENGDU NUOBIKAN TECH CO LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411276756.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-12
Publication Date
2024-11-15
Estimated Expiration
2044-09-12

AI Technical Summary

Technical Problem

The existing contact network suspension system monitoring method is inefficient and cannot realize real-time monitoring. The data dimension of a single sensor is single, making it difficult to fully reflect the system status, resulting in false alarms or missed reports. The abnormal detection performance and robustness of multimodal data fusion systems are insufficient. .

Method used

Using multimodal data acquisition and processing units, including lidar, multispectral camera and electromagnetic induction sensor, the abnormal state index is calculated through nonlinear feature extraction and dimensionality reduction technology, combined with multi-scale morphological analysis, and the abnormal state index is calculated to realize the contact network suspension system. comprehensive and fine inspection.

Benefits of technology

It significantly improves the monitoring accuracy and fault warning accuracy of the contact network suspension system, avoids false alarms or missed reports caused by a single data source, and improves the real-time and security of the system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118797532B_ABST
    Figure CN118797532B_ABST
Patent Text Reader

Abstract

The present invention belongs to the field of contact network detection technology, and specifically relates to a multimodal fusion intelligent detection system for monitoring the contact network suspension of railway tracks, characterized in that the system comprises: a multimodal data acquisition and processing unit, a data processing and analysis unit, and an abnormal state determination unit; the multimodal data acquisition and processing unit is used to uniformly select multiple sampling points on the contact network suspension to construct an initialization composite data matrix of the contact network suspension; the data processing and analysis unit is used to perform nonlinear feature extraction and dimensionality reduction on the initialization composite data matrix to obtain a reduced dimensionality feature matrix; the abnormal state determination unit is used to calculate the abnormal state index of the contact network suspension according to the tension estimation result and the displacement function, and define the abnormal determination interval. The present invention not only overcomes the limitations of traditional methods, but also makes breakthrough progress in data processing, fault detection, and early warning accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of contact network detection, and in particular relates to a multi-modal fusion intelligent detection system for railway track contact network suspension monitoring. Background Art

[0002] As the backbone of modern transportation, the safety and reliability of the railway transportation system have always been the focus of attention of the railway industry in various countries. With the development of high-speed rail and high-speed railway, the overhead line suspension system, as a key component of the power supply network, undertakes the task of providing stable power for trains. The monitoring and maintenance of its operating status are particularly important. Any abnormality in the overhead line suspension system may lead to power outages and even cause serious safety accidents. Therefore, how to effectively monitor the state of the overhead line suspension and timely discover and warn potential faults has become an urgent problem to be solved in the railway industry. In the prior art, the monitoring of the overhead line suspension system mainly relies on regular manual inspections and simple monitoring equipment. For example, the physical parameters of the suspension system are obtained by installing displacement sensors and stress sensors on the overhead line suspension, and the state of the suspension system is monitored by combining video monitoring and other means. These methods have improved the monitoring efficiency to a certain extent, but they also have obvious limitations. First, the manual inspection method is inefficient and cannot achieve real-time monitoring. Especially in high-density railway networks, manual inspections are difficult to cover all monitoring points, which easily leads to the accumulation of hidden faults. Secondly, the existing monitoring equipment mostly uses a single sensor, and the data dimension is single, which makes it difficult to fully reflect the operating status of the overhead line suspension system. Since the failure of the overhead suspension system is often the result of the combined effect of multiple factors, such as mechanical stress, material fatigue, environmental changes, etc., a single monitoring data is difficult to reveal the full picture of the failure and is prone to false alarms or omissions.

[0003] In recent years, with the development of artificial intelligence and sensor technology, multimodal data fusion technology has gradually been applied to the field of industrial monitoring. Multimodal data fusion technology can provide more comprehensive and accurate system status information by integrating data from different types of sensors. For example, by combining lidar, spectral sensors and electromagnetic induction sensors, the morphological changes, material properties and electromagnetic properties of the contact network suspension can be monitored simultaneously, so as to more comprehensively evaluate its operating status. However, the application of existing multimodal data fusion systems in contact network suspension monitoring is still insufficient. On the one hand, the data processing complexity is high, and there is a lack of effective algorithms and models for comprehensive analysis of multimodal data; on the other hand, the performance of existing multimodal monitoring systems in anomaly detection still needs to be improved, especially when dealing with complex and changeable railway operating environments, the system's robustness and real-time performance are insufficient. Summary of the invention

[0004] In view of this, the main purpose of the present invention is to provide a multi-modal fusion intelligent detection system for monitoring the suspension of the contact network of railway tracks, which significantly improves the monitoring capability of the contact network suspension system. It not only overcomes the limitations of traditional methods, but also makes breakthrough progress in data processing, fault detection and early warning accuracy, providing strong technical support for the safe and stable operation of the railway system.

[0005] The technical solution adopted by the present invention is as follows:

[0006] A multimodal fusion intelligent detection system for monitoring the overhead contact network suspension of railway tracks, the system comprising: a multimodal data acquisition and processing unit, a data processing and analysis unit, and an abnormal state determination unit; the multimodal data acquisition and processing unit is used to uniformly select multiple sampling points on the overhead contact network suspension, and for each sampling point, use a multimodal data acquisition device to acquire multimodal data, and construct an initialization composite data matrix of the overhead contact network suspension based on the multimodal data; the data processing and analysis unit is used to perform nonlinear feature extraction and dimensionality reduction on the initialized composite data matrix to obtain a reduced dimensionality feature matrix; and perform multi-dimensional feature extraction and dimensionality reduction on the reduced dimensionality feature matrix. The invention discloses a method for determining the contact network suspension of a plurality of load-bearing networks by performing a multi-scale morphological analysis to obtain a multi-scale morphological feature matrix; the multi-scale morphological feature matrix is ​​used to perform tension estimation to obtain a tension estimation result and a displacement function; the abnormal state determination unit is used to calculate the abnormal state index of the contact network suspension according to the tension estimation result and the displacement function, and define an abnormal determination interval; when the abnormal state index is greater than the upper limit of the abnormal determination interval, it is determined that the contact network suspension is abnormal; when the abnormal state index is within the abnormal determination interval, it is determined that the probability of the contact network suspension being abnormal is between 0 and 1; when the abnormal state index is less than the lower limit of the abnormal determination interval, it is determined that the contact network suspension is not abnormal.

[0007] Furthermore, the multimodal data acquisition device includes: a laser radar, a multispectral camera and an electromagnetic induction sensor; The wavelength of the sampling point is ; The laser radar obtains the The three-dimensional coordinate vector of the sampling point is ; Multispectral camera in the The spectral vector obtained by sampling is ; Electromagnetic induction sensor in the The electromagnetic field intensity vector measured at the sampling points is ; The obtained initialized composite data matrix Use the following formula to express it:

[0008] ;

[0009] in, is the number of sampling points; Subscript index, value ranges from 1 to An integer.

[0010] Furthermore, the data processing and analysis unit performs nonlinear feature extraction and dimensionality reduction on the initialized composite data matrix by the following formula to obtain a reduced dimensionality feature matrix: :

[0011] ;

[0012] in, Represents element-wise multiplication; yes A matrix mask of ; is the eigenvalue vector; is the feature vector; is the regularized centralization kernel matrix; is a nonlinear mapping function.

[0013] Furthermore, nonlinear feature extraction and dimensionality reduction are performed on the initialized composite data matrix to obtain the reduced dimensionality feature matrix In the process, the data processing and analysis unit defines the Gaussian kernel function through the following formula: As a nonlinear mapping:

[0014] ;

[0015] in, is the bandwidth parameter of the kernel function; is the first variable of the Gaussian kernel function; is the second variable of the Gaussian kernel function; construct the kernel matrix , in the kernel matrix, Row, No. The column elements are ; The centralized kernel matrix is ​​obtained through the following formula :

[0016] ;

[0017] in, The elements are all of matrix;

[0018] The eigenvalue problem is solved by the following formula: ;in, is the eigenvalue, is the corresponding eigenvector; sort the eigenvalues ​​in descending order and select the first The eigenvalues ​​form the eigenvalue vector and the corresponding eigenvector :

[0019] ;

[0020] ;

[0021] The kernel principal component is calculated using the following formula: ; The regularization term is introduced through the following formula: ;in, is a set positive constant, is the identity matrix.

[0022] Furthermore, the data processing and analysis unit performs multi-scale morphological analysis on the dimension reduction feature matrix to obtain a multi-scale morphological feature matrix, and defines an adaptive multi-scale morphological operator by the following formula: :

[0023] ;

[0024] in, is the maximum number of scales defined; is the current scale index; For the A set of structural elements of different scales; For improved expansion operation, is the expansion strength parameter; For improved corrosion performance, is the corrosion intensity parameter; It is an adaptive opening operation; It is an adaptive closing operation; is a weighted sum operation; It is a weighted difference operation.

[0025] Further, improved expansion operation The calculation process is expressed by the following formula:

[0026] ;

[0027] in, For expansion operation; is the dimension-reduced feature matrix Elements of is the F-norm operation; Absolute value operation; improved corrosion operation Use the following formula to express it:

[0028] ;

[0029] in, For corrosion operation.

[0030] Furthermore, the multi-scale morphological feature matrix Calculated using the following formula:

[0031] ;

[0032] in, and Respectively Expansion strength parameters and corrosion strength parameters of each scale; is the mean parameter of Gaussian weight; Adjust parameters for preset matrix features; is the trace of the matrix.

[0033] Furthermore, the tension estimation result is obtained by using the scale morphological feature matrix through the following formula: and displacement function :

[0034] ;

[0035] ;

[0036] in, is the density of the overhead wire suspension; is the acceleration due to gravity; is the span of the overhead contact network suspension; is the midpoint sag of the overhead contact network suspension; is the correction factor set to adjust the effect of morphological features on tension estimation; is the horizontal distance along the span; is the number of terms in the Fourier series; For the Fourier coefficients, given by Sure: ;in It is with The weight matrix corresponding to the Fourier components; is the preset attenuation parameter.

[0037] Furthermore, the abnormal state determination unit calculates the abnormal state index of the contact network suspension according to the tension estimation result and the spectrum analysis result by the following formula: :

[0038] ;

[0039] in, is the standard stress value, is the set value; is the dimension-reduced feature matrix Dimensions; is the maximum value of the displacement function; is the minimum value of the displacement function.

[0040] By adopting the above technical scheme, the present invention produces the following beneficial effects: the present invention overcomes the limitations of the single sensor monitoring method in the prior art by introducing a multimodal data acquisition and processing unit. Traditional monitoring methods usually rely on a single type of sensor, such as a displacement sensor or a stress sensor. This method has limited data dimensions and is difficult to fully reflect the complex state of the overhead line suspension system. The present invention can simultaneously obtain multi-dimensional data such as the geometric shape, material properties and electromagnetic properties of the suspension system through a multimodal data acquisition device, including a laser radar, a multispectral camera and an electromagnetic induction sensor. These multimodal data provide the system with comprehensive and rich information, which can more accurately reflect the true state of the suspension system and avoid the problem of false alarms or omissions caused by a single data source. Secondly, the present invention introduces nonlinear feature extraction and dimensionality reduction technology in the data processing and analysis unit. Due to the complexity and high dimensionality of multimodal data, traditional linear feature extraction methods are difficult to capture the complex relationships within it. The present invention effectively maps the original high-dimensional data into a low-dimensional feature space by using nonlinear feature extraction techniques such as kernel principal component analysis (KPCA), thereby extracting key features that can reflect the state of the suspension system. The feature matrix after dimensionality reduction not only retains the important information in the original data, but also greatly reduces the computational complexity and improves the real-time performance and processing efficiency of the system. The application of this feature extraction method enables the present invention to accurately extract the core information of the system operation status when facing complex multimodal data, significantly improving the monitoring accuracy. Third, the present invention adopts a multi-scale morphological analysis method to further process the feature matrix after dimensionality reduction. The faults of the contact network suspension system are often manifested in multiple different spatial scales, such as local material wear, overall structural deformation, etc. Multi-scale morphological analysis can capture these morphological changes from different scales to ensure that the system can accurately identify potential anomalies at different levels. The present invention realizes comprehensive analysis and fine detection of the contact network suspension system through adaptive multi-scale morphological operators. Multi-scale analysis can not only find obvious structural problems, but also capture subtle and early signs of faults, thereby achieving earlier fault warnings and improving the safety and reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A schematic diagram of the system structure of a multi-modal fusion intelligent detection system for railway track contact network suspension monitoring provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] All features disclosed in this specification, or steps in all methods or processes disclosed, except mutually exclusive features and / or steps, can be combined in any manner.

[0043] Any feature disclosed in this specification (including any additional claims and abstract), unless otherwise stated, may be replaced by other alternative features that are equivalent or have similar purposes. That is, unless otherwise stated, each feature is only an example of a series of equivalent or similar features.

[0044] Example 1: Reference Figure 1 A multimodal fusion intelligent detection system for monitoring the overhead contact network suspension of railway tracks, the system comprising: a multimodal data acquisition and processing unit, a data processing and analysis unit, and an abnormal state determination unit; the multimodal data acquisition and processing unit is used to uniformly select multiple sampling points on the overhead contact network suspension, and for each sampling point, use a multimodal data acquisition device to acquire multimodal data, and construct an initialization composite data matrix of the overhead contact network suspension based on the multimodal data; the data processing and analysis unit is used to perform nonlinear feature extraction and dimensionality reduction on the initialized composite data matrix to obtain a reduced dimensionality feature matrix; and perform multi-dimensional feature extraction and dimensionality reduction on the reduced dimensionality feature matrix. The invention discloses a method for determining the contact network suspension of a plurality of load-bearing networks by performing a multi-scale morphological analysis to obtain a multi-scale morphological feature matrix; the multi-scale morphological feature matrix is ​​used to perform tension estimation to obtain a tension estimation result and a displacement function; the abnormal state determination unit is used to calculate the abnormal state index of the contact network suspension according to the tension estimation result and the displacement function, and define an abnormal determination interval; when the abnormal state index is greater than the upper limit of the abnormal determination interval, it is determined that the contact network suspension is abnormal; when the abnormal state index is within the abnormal determination interval, it is determined that the probability of the contact network suspension being abnormal is between 0 and 1; when the abnormal state index is less than the lower limit of the abnormal determination interval, it is determined that the contact network suspension is not abnormal.

[0045] Specifically, during the data collection process, the system first selects multiple representative sampling points on the contact network suspension. The selection of these sampling points is usually based on an in-depth analysis of the contact network suspension structure to ensure that they cover all possible key points, especially those areas that are prone to wear, fatigue or other structural problems. By evenly distributing the sampling points, the system can monitor the overall state of the contact network suspension in real time and integrate the data of these sampling points into an initialized composite data matrix. Each row in the composite data matrix represents a sampling point, and each column represents a physical quantity. This structured data storage method not only facilitates subsequent data processing and analysis, but also ensures the integration and consistency of the data. The processing of multimodal data is to preprocess these multi-dimensional data to meet the subsequent advanced analysis requirements. The preprocessing process includes steps such as data filtering, denoising, and normalization. The purpose is to eliminate the influence of environmental noise and sensor errors on the data, and at the same time unify the data of different physical quantities to the same standard or scale to facilitate subsequent composite analysis. For example, the sensor noise is eliminated by filtering technology, the data anomalies caused by external interference are eliminated by denoising algorithms, and the data of different magnitudes are adjusted to the same scale by normalization methods to ensure that the data in the composite data matrix can be effectively fused. It does not rely solely on a single sensor or a single data source, but integrates the data of multiple sensors through multimodal fusion. This fusion greatly improves the system's perception accuracy of the state of the contact network suspension. By integrating data of different modes, the system can capture more state information, thereby better reflecting the actual operating state of the contact network suspension. For example, the use of displacement sensors alone may only capture some static or low-frequency dynamic information of the contact network, but after combining the data of vibration sensors, the system can simultaneously capture high-frequency dynamic changes, thereby fully understanding the health of the contact network. In addition, the data of temperature sensors can provide the system with information on the impact of ambient temperature on the contact network suspension, which is crucial when judging material stress and deformation. In order to further improve the effectiveness of the data, the multimodal data acquisition and processing unit integrates the multimodal data of each sampling point through the construction of a composite data matrix. This process ensures the systematicity and structure of the data. The composite data matrix is ​​not only a simple data storage structure, but also the basis for subsequent data processing and analysis. Based on this matrix, the system can use advanced nonlinear feature extraction algorithms to perform in-depth feature analysis on multimodal data, and then extract key features closely related to the suspension state of the contact network. This method can reveal complex relationships that are difficult to capture with single modal data, such as the nonlinear relationship between stress and displacement, vibration and temperature, thereby providing a more accurate basis for the detection of abnormal conditions.

[0046] The first major task of the data processing and analysis unit is to perform nonlinear feature extraction and dimensionality reduction on the initialized composite data matrix. The data of the overhead line suspension system usually has complex nonlinear characteristics, which are difficult to be effectively captured by traditional linear methods. Therefore, the system adopts advanced nonlinear feature extraction techniques, which can capture the hidden nonlinear relationships in the original data by mapping the data into a higher-dimensional feature space. In this high-dimensional feature space, the system can better identify the key patterns and trends in the data. However, the data dimension in the high-dimensional space is usually very high. Directly using these high-dimensional data for analysis is not only computationally intensive, but also may contain a large amount of redundant information. Therefore, the dimensionality reduction process is particularly important here. Through dimensionality reduction, the system can compress the data in the high-dimensional space into a lower-dimensional space while retaining the important feature information in the original data as much as possible. This process ensures that the key information of the data is retained to the greatest extent while reducing the computational complexity, laying the foundation for subsequent morphological analysis. After completing the dimensionality reduction, the data processing and analysis unit then performs multi-scale morphological analysis. The core idea of ​​this step is to extract the morphological information that best reflects the state of the suspension system by observing and analyzing the morphological characteristics of the overhead line suspension data at different scales. The multiscale morphological analysis method can detect and describe the structural features of the data at different resolutions, thereby capturing important patterns at various scales in the data. For example, some local anomalies may only appear at a smaller scale, while the overall trend or morphological features may be more obvious at a larger scale. By analyzing at multiple scales, the system can comprehensively capture the morphological changes of the overhead contact suspension system, ensuring that any potential anomalies or faults will not be ignored. This multiscale analysis method is particularly suitable for processing data with complex structures, such as the data of the overhead contact suspension system, whose morphological features often involve multi-level and multi-scale changes. After obtaining the multiscale morphological feature matrix, the data processing and analysis unit is also responsible for tension estimation and displacement function calculation. Tension is a key indicator reflecting the state of the overhead contact suspension system. Through the information in the morphological feature matrix, the system can accurately estimate the tension of the overhead contact suspension. The displacement function describes the displacement changes of the overhead contact suspension at each sampling point, which is crucial for judging the overall health of the overhead contact suspension. The principle of tension estimation and displacement function calculation is to use the local features and global trends in the morphological feature matrix to derive the most realistic tension and displacement distribution through a series of mathematical models and algorithms. These results not only directly reflect the current status of the overhead line suspension system, but also provide an important reference for abnormality detection of the system.

[0047] The core task of the abnormal state judgment unit is to calculate an abnormal state index, which is a quantitative expression of the system's operating state of the contact network suspension and can comprehensively reflect the current health status of the system. The calculation of the abnormal state index is not a simple weighted average or linear combination, but is achieved through a complex relationship between nonlinear functions and multiple parameters. In this process, the system will consider the tension estimation results, the trend of the displacement function, and the key information in the multi-scale morphological feature matrix, and integrate these factors in a highly nonlinear and dynamic way to generate a value reflecting the overall state of the contact network suspension. This index value is not just a static judgment criterion, but a combination of historical data and current data comparison. By dynamically adjusting the model parameters, the abnormal state index can reflect the operating status of the system in real time. This method ensures that the system can still accurately judge the state of the contact network suspension when facing a complex and changeable railway environment. After the calculation of the abnormal state index is completed, the abnormal state judgment unit will compare it with the pre-defined abnormal judgment interval to determine the specific operating state of the contact network suspension. The abnormality judgment interval is obtained through statistical analysis of a large amount of experimental data and actual monitoring results. It represents the normal range of system operation, the critical range of the probability of abnormality between 0 and 1, and the dangerous range of obvious abnormality. The system will judge whether there is a potential abnormality in the current overhead line suspension based on the position of the abnormal state index in the judgment interval.

[0048] Embodiment 2: The multimodal data acquisition device comprises: a laser radar, a multispectral camera and an electromagnetic induction sensor; The wavelength of the sampling point is ; The laser radar obtains the The three-dimensional coordinate vector of the sampling point is ; Multispectral camera in the The spectral vector obtained by sampling is ; Electromagnetic induction sensor in the The electromagnetic field intensity vector measured at the sampling points is ; The obtained initialized composite data matrix Use the following formula to express it:

[0049] ;

[0050] in, is the number of sampling points; Subscript index, value ranges from 1 to An integer.

[0051] Specifically, laser radar is a high-precision three-dimensional measurement device that obtains the three-dimensional coordinates of the target by emitting a laser beam and measuring the time difference or phase difference of the reflected beam. In the present invention, laser radar is used to obtain the three-dimensional coordinate vector of each sampling point in the overhead line suspension system. These coordinate vectors It not only reflects the actual geometric position of the overhead wire suspension in space, but also provides the system with accurate information about structural deformation or displacement. This three-dimensional spatial information is crucial for judging the overall health of the overhead wire suspension, especially when identifying possible structural problems, such as stretching or shrinking of the suspension. Next, the role of the multispectral camera in the system is to capture the reflectance spectral information of the overhead wire suspension under different spectra. Spectral vector Represents the The light reflection characteristics of a sampling point at multiple wavelengths can reveal changes in the physical and chemical properties of the material, such as corrosion, wear or temperature changes. The reflectivity of the material will be different in different spectral ranges, especially in the infrared or ultraviolet spectrum, where microstructural changes in some materials can be detected more sensitively. Therefore, the spectral vector It can help the system to monitor the state of the overhead line suspension more comprehensively, especially when detecting early minor damage. This method has extremely high sensitivity. In addition, the electromagnetic induction sensor is used to detect the electromagnetic field strength, which is particularly important for the electrical performance monitoring of the overhead line suspension system. Provides information about the electromagnetic environment at the overhead line suspension, such as electromagnetic interference caused by current passing through, poor grounding or other electrical anomalies.

[0052] Embodiment 3: The data processing and analysis unit performs nonlinear feature extraction and dimensionality reduction on the initialized composite data matrix by the following formula to obtain a reduced dimensionality feature matrix :

[0053] ;

[0054] in, Represents element-wise multiplication; yes A matrix mask of ; is the eigenvalue vector; is the feature vector; is the regularized centralization kernel matrix; is a nonlinear mapping function.

[0055] Specifically, the nonlinear mapping function mentioned in the formula plays an important role, it will initialize the composite data matrix The original data in the image is mapped to a higher-dimensional feature space. Through this mapping, the system can capture nonlinear relationships that may be hidden in the original data. These relationships may be complex and difficult to directly observe in the multimodal data of the overhead line suspension. After mapping to a high-dimensional space, these nonlinear features are revealed, providing a basis for subsequent feature extraction. Next, the regularized central kernel matrix It plays a vital role in the formula. The kernel matrix is ​​a tool for measuring the similarity between data points in the feature space. By calculating the similarity between any two data points in the feature space, the structural characteristics of the data in the high-dimensional space can be effectively captured. However, directly using the unprocessed kernel matrix may cause the model to overfit when processing noisy data or high-dimensional data. Therefore, Additional penalty terms are introduced through regularization processing to suppress the influence of noise and overfitting. At the same time, the data bias is eliminated through centralization operation, making the feature extraction process more stable and robust. is another key element in the formula, which is multiplied by element and The role of the matrix mask is to selectively retain or suppress specific data points or features to ensure that only those truly important features are retained and participate in subsequent calculations. This step is similar to a data filtering operation to a certain extent, which further enhances the accuracy of feature extraction by filtering out irrelevant or noisy data. This process is particularly important for railway track contact network suspension systems, because the system needs to extract the most representative features from a large amount of complex and multimodal data to determine the operating status of the contact network. Subsequently, the eigenvector involved in the formula and the eigenvalue vector It is an indispensable part of the feature extraction process. Eigenvalue vector It represents the weight or importance of the features extracted from the kernel matrix. Generally, the larger the eigenvalue, the greater the role of the corresponding feature in the data. It defines the direction of the feature in the feature space and determines how the data is projected into the low-dimensional space after dimensionality reduction. This process ensures that important information in the data is retained as much as possible during dimensionality reduction, without losing key information due to dimensionality reduction.

[0056] The overall calculation logic of the formula is to use the regularized centralized kernel matrix to capture the characteristic structure in the high-dimensional space after the original multimodal data is nonlinearly mapped, and then selectively screen the matrix mask and combine the dimensionality reduction operations of eigenvalues ​​and eigenvectors to finally obtain a reduced-dimensional feature matrix. . This feature matrix is ​​much smaller than the original data in dimension, but the information it contains is more concentrated and effective, and can better reflect the true state of the contact network suspension system. The present invention realizes the conversion from original multimodal data to key feature information. This conversion not only improves the efficiency of data analysis, but more importantly, provides a more reliable basis for the state monitoring of the contact network suspension system. Traditional methods usually rely on linear feature extraction or simple dimensionality reduction techniques, which are difficult to cope with the challenges of complex nonlinear characteristics and multimodal data fusion in the contact network suspension system. The method proposed in the present invention successfully overcomes these challenges by utilizing the nonlinear mapping ability of the kernel method, the stability of the regularization processing, and the information retention ability in the dimensionality reduction process, and significantly improves the accuracy and reliability of the system in practical applications.

[0057] Embodiment 4: Perform nonlinear feature extraction and dimensionality reduction on the initialized composite data matrix to obtain a reduced dimensionality feature matrix In the process, the data processing and analysis unit defines the Gaussian kernel function through the following formula: As a nonlinear mapping:

[0058] ;

[0059] in, is the bandwidth parameter of the kernel function; is the first variable of the Gaussian kernel function; is the second variable of the Gaussian kernel function; construct the kernel matrix , in the kernel matrix, Row, No. The column elements are ; The centralized kernel matrix is ​​obtained through the following formula :

[0060] ;

[0061] in, The elements are all of matrix;

[0062] The eigenvalue problem is solved by the following formula: ;in, is the eigenvalue, is the corresponding eigenvector; sort the eigenvalues ​​in descending order and select the first The eigenvalues ​​form the eigenvalue vector and the corresponding eigenvector :

[0063] ;

[0064] ;

[0065] The kernel principal component is calculated using the following formula: ; The regularization term is introduced through the following formula: ;in, is a set positive constant, is the identity matrix.

[0066] Specifically, Example 4 uses the Gaussian kernel function As a nonlinear mapping tool, it maps the original multimodal data from the input space to a high-dimensional feature space. In this high-dimensional space, the similarity between data points is redefined, capturing complex nonlinear relationships that may not be apparent in the original low-dimensional space. The form of the Gaussian kernel function Indicates that similarity is measured by the Euclidean distance between data points. It is called the bandwidth parameter of the kernel function, which controls the decay rate of the similarity. Specifically, the smaller The similarity decreases rapidly as the distance increases, while the larger This will cause the similarity to decay more slowly. In overhead line suspension monitoring, this high-dimensional mapping can effectively reveal the complex relationships between different sampling points, such as the relationship between displacement and stress, the pattern of material properties changing with temperature, etc., thus laying the foundation for subsequent feature extraction. After calculating the Gaussian kernel function, the system constructs a kernel matrix , where each element Represents the sample and Similarity in feature space. This kernel matrix is ​​essentially a symmetric matrix that summarizes the similarity information between all sample points. In the multimodal monitoring of railway track contact network suspension, the construction of the kernel matrix enables the system to uniformly process multi-source data from different sensors and integrate the intrinsic relationship of these data through similarity measurement. However, directly using the unprocessed kernel matrix for feature extraction may introduce deviations due to the uneven distribution of data. Therefore, Example 4 adopts centralized processing, through the formula Correct the kernel matrix. This process eliminates the row and column means in the matrix so that all data points are symmetrically distributed relative to the center in the high-dimensional space, thereby reducing the bias of the data and ensuring that feature extraction is more stable and robust. Centralized kernel matrix After the calculation of is completed, the next step is to solve the eigenvalue problem, which is crucial because it is directly related to which features are extracted from the high-dimensional space for dimensionality reduction. , we can get the eigenvalue and its corresponding eigenvector . Eigenvalue It is a measure of the importance of a specific feature in a data set. A larger eigenvalue usually means that the corresponding eigenvector captures the main pattern or trend in the data. In practical applications, these main features may include deformation patterns of overhead wire suspension under different environmental conditions, changes in electromagnetic induction, etc., which are particularly important for system condition monitoring and fault prediction. Once all eigenvalues ​​are and the eigenvector Once they are found, the system will sort them according to the size of the eigenvalues ​​and select the top eigenvalues ​​and their corresponding eigenvectors constitute the eigenvalue vector and the eigenvector matrix These selections are made to ensure that the data retained after dimensionality reduction can still fully represent the main information in the original data. In the monitoring of the railway track contact network suspension system, this feature selection process is crucial because it determines whether the feature subset finally used by the system can accurately reflect the operating status of the contact network. Next, by calculating the formula , the system projects the data into the reduced-dimensional feature space, a process known as the final step of kernel principal component analysis (KPCA). Through this dimensionality reduction operation, the system can transform complex data in a high-dimensional space into a smaller feature set while retaining the key information in the data. The reduced-dimensional feature matrix is a concise but more information-dense representation that can be effectively used for monitoring system status assessment and anomaly detection. In overhead line suspension monitoring, this dimensionality reduction process helps reduce the computational burden, especially when faced with massive data, while ensuring that the system can respond quickly when anomaly detection occurs. In order to further enhance the robustness of the model, Example 4 also introduces a regularization term, using the formula Regularization is performed on the centralized kernel matrix. The introduction of regularization is mainly to prevent the model from overfitting when processing noisy data or data sets with a small number of samples. Overfitting will cause the model to perform well on training data but poorly on new data. Therefore, the introduction of regularization terms can effectively improve the generalization ability of the model, making the system more reliable and stable in practical applications. In railway track monitoring systems, the setting of regularization terms is particularly important, because the data of the overhead line suspension system is usually interfered by environmental noise, or the number of samples is limited. Through regularization, it can be ensured that the system still maintains a high accuracy in these complex situations.

[0067] Embodiment 5: When the data processing and analysis unit performs multi-scale morphological analysis on the dimension reduction feature matrix to obtain the multi-scale morphological feature matrix, an adaptive multi-scale morphological operator is defined by the following formula: :

[0068] ;

[0069] in, is the maximum number of scales defined; is the current scale index; For the A set of structural elements of different scales; For improved expansion operation, is the expansion strength parameter; For improved corrosion performance, is the corrosion intensity parameter; It is an adaptive opening operation; It is an adaptive closing operation; is a weighted sum operation; It is a weighted difference operation.

[0070] Specifically, the adaptive multi-scale morphological operator mentioned in the formula is a composite operator that combines multiple morphological operations such as dilation, erosion, opening, and closing. These operations are basic tools in morphological analysis and are often used in image processing and data analysis to highlight the structural features of data. Specifically, the dilation operation and corrosion operations They are used to expand and reduce the bright and dark areas in the data, respectively. They adjust the shape of the image or data so that the system can more sensitively capture the structural changes in the overhead line suspension. Here, and is the intensity parameter of dilation and erosion, which controls the degree of operation. By precisely adjusting these parameters, the system can perform detailed morphological analysis on the data of overhead wire suspension at different scales, ensuring that possible abnormal features are captured at multiple levels. In addition, opening and closing operations are combined operations in morphology, which are used to remove small noise in the data and fill small holes in the data, respectively. In the adaptive multi-scale morphological operator, these operations are designed to be based on the current scale. Dynamic adjustment, so that it can adapt to different morphological characteristics at different scales. Adaptability is reflected in the fact that the parameters of the operation change with the scale, which means that at a larger scale, the system will pay more attention to changes in the overall structure, while at a smaller scale, the system can capture more subtle changes in details. For railway track contact network suspension systems, this multi-scale analysis is particularly important, because the state of the contact network suspension system may be affected by many factors, and these influences may show different characteristics at different scales. For example, large-scale morphological changes may reflect the overall relaxation or deformation of the contact network, while small-scale changes may suggest local wear or corrosion. Through the combination of the above morphological operations, the adaptive multi-scale morphological operator At each scale, the reduced feature matrix The analysis is performed and the analysis results of each scale are integrated to form a comprehensive multi-scale morphological feature matrix. This feature matrix not only contains the morphological information of the data at each scale, but also can more comprehensively reflect the state of the overhead line suspension by integrating the results of multiple scales. The element-by-element multiplication operation mentioned in the formula and weighted sum operation It is a tool for combining the results of different morphological operations. These operations ensure that the results of different morphological transformations can be effectively integrated to generate the final multi-scale morphological feature matrix. In specific applications, through element-by-element multiplication, the system can combine the results of expansion, corrosion, and opening and closing operations to form a comprehensive evaluation of the contact network suspension system; and the weighted summation adjusts the weights of different operation results so that the system can focus on certain specific morphological features as needed. This flexible operation mode gives the system a strong adaptability under complex conditions, allowing it to more effectively capture various potential anomalies of the contact network suspension system. In addition, the maximum scale number in the formula and the current scale index Together, they define the analysis scope of the system at different scales. By gradually increasing the scale, the system can analyze the data comprehensively from details to the whole, thereby ensuring that any potential problems, whether small local anomalies or obvious global changes, can be detected in time. This multi-scale analysis strategy is particularly critical for the railway track contact network suspension system, because the system must face complex and changeable environmental conditions, such as temperature changes, mechanical stress fluctuations, etc., which may show different characteristics at different time and space scales. In general, the adaptive multi-scale morphological operator in Example 5 realizes the in-depth analysis of the multimodal data of the contact network suspension system. The operator can not only capture subtle changes in the data, but also integrate the data features at multiple scales to provide a comprehensive morphological feature matrix for subsequent anomaly detection. This multi-scale analysis method has significant advantages in railway track monitoring. It not only improves the sensitivity and accuracy of the system, but also enhances the robustness of the system in dealing with complex environmental changes. This method ensures that the state of the contact network suspension system under different operating conditions can be monitored and accurately evaluated in real time, thereby providing strong technical support for the safe operation of the railway system. Through this innovative morphological analysis method, the system can comprehensively monitor the contact network suspension at multiple levels and angles, greatly improving the safety and reliability of railway operations.

[0071] Example 6: Improved expansion operation The calculation process is expressed by the following formula:

[0072] ;

[0073] in, For expansion operation; is the dimension reduction feature matrix Elements of is the F-norm operation; Absolute value operation; improved corrosion operation Use the following formula to express it:

[0074] ;

[0075] in, For corrosion operation.

[0076] Specifically, the improved expansion operation It is an enhanced version of the traditional morphological dilation operation. The basic idea of ​​the dilation operation is to amplify and emphasize the prominent features in the data by expanding the range of the bright area in the data. In this formula, the core part of the operation is , which means that in the set of structural elements Each element of Up, take and This process can be regarded as the maximum value of the structural element in the data matrix Slide on the data, taking the maximum value in the local area each time, thereby amplifying the highlights in the data at different scales. The second part of the formula This is an improvement on the expansion operation. is the expansion strength parameter, which controls the expansion enhancement degree. As the expansion increases, the impact of the expansion operation also increases. It is expressed as the ratio of the current scale to the maximum scale. In addition, is a normalization factor, express The Frobenius norm of , which provides the matrix The overall magnitude information of is the maximum absolute value of the elements in the matrix. The introduction of this normalization factor ensures that the scale of the expansion operation is not affected by the overall size of the data, so that the expansion operation can maintain consistent effects under different data scales.

[0077] Improved corrosion operation It is the reverse process of the dilation operation, which aims to reduce and suppress the dark areas in the data. The erosion operation takes the set of structural elements. This means that The corresponding position in the local area takes the minimum value, thereby reducing the dark spots in the data. This operation can remove noise and small structures in the data and highlight the main morphological features of the data. Similar to the expansion operation, the improvement of the corrosion operation is reflected in the second half of the formula Here, is the corrosion intensity parameter, which controls the degree of corrosion inhibition. The effect of the corrosion operation will be adjusted accordingly. This ratio expresses the relative relationship between the current scale and the maximum scale. Normalization factor The scale independence of the operation is also ensured, where It is the minimum absolute value of the elements in the matrix, reflecting the strength of the weakest signal in the data. This part ensures that the corrosion operation can maintain the appropriate suppression effect under different data intensities.

[0078] By combining these improved dilation and erosion operations, the system is able to perform dimension reduction on the feature matrix at multiple scales. Perform detailed morphological analysis. The advantage of this analysis method is that it can adapt to different scale changes, thereby capturing various abnormal phenomena that may exist in the contact network suspension system. For example, at a smaller scale, the system can detect tiny local deformations or material fatigue; at a larger scale, the system can identify overall structural relaxation or large-scale morphological changes. The formula in Example 6 introduces dynamically adjusted expansion and corrosion operations, so that the system can more accurately analyze the state of the contact network suspension system when facing a complex and changeable railway environment. These operations not only enhance or suppress features at different levels of data, but also ensure the stability and robustness of the operation through normalization factors, avoiding analysis errors caused by changes in data magnitude. In railway track monitoring, this improved morphological analysis method greatly improves the sensitivity and accuracy of the system, allowing the system to detect potential faults earlier, thereby preventing possible safety accidents.

[0079] Example 7: Multi-scale morphological feature matrix Calculated using the following formula:

[0080] ;

[0081] in, and Respectively Expansion strength parameters and corrosion strength parameters of each scale; is the mean parameter of Gaussian weight; Adjust parameters for preset matrix features; is the trace of the matrix.

[0082] Specifically, the core idea of ​​the formula is to transform the adaptive multi-scale morphological operator The output results at each scale are weighted and summed to finally obtain a multi-scale morphological feature matrix The process starts by summing the feature matrix results of all scales, and the summed result is averaged and divided by the total number of scales , to ensure that the matrix The overall numerical level of is comparable at different scales. This step is achieved by To implement, it means that at each scale The output of the morphological operator is calculated on , and the results of all scales are combined. Then, a Gaussian weight factor is introduced into the formula , which is used to adjust the contribution of different scales. The form of Gaussian weight factor ensures that when the scale Close to the mean When , the contribution of this scale is higher, and as and As the gap between them increases, the contribution gradually decreases. This weight distribution method imitates the characteristics of Gaussian distribution, so that when the system processes multi-scale features, it can focus on those features close to the main scale, and moderately weaken the features at too large or too small scales, thereby avoiding noise interference caused by extreme scales. This design is very suitable for monitoring the suspension system of railway track contact network, because in actual applications, the anomalies of the suspension system are often most significant at a certain medium scale, while the features at extreme scales may be more affected by environmental noise or data instability. In addition, the formula also introduces a feature adjustment term , which further adjusts the contribution of the feature matrix. Here, represents the trace of the matrix, that is, the sum of the elements on the main diagonal of the matrix, The Frobenius norm of a matrix reflects the overall energy or magnitude of the matrix. By calculating the ratio of the trace and the norm, the system can assess the importance of the principal components of the matrix in the overall data at the current scale. Feature Adjustment Parameters This design ensures that when the main component of the matrix contributes more to the overall energy, its corresponding scale is larger in the final feature matrix. The influence of the matrix will also increase accordingly, thus highlighting the features that are particularly important at a specific scale. This multi-level, multi-parameter formula structure makes the multi-scale morphological feature matrix It can capture the characteristics of the overhead line suspension system at multiple scales at the same time, and through reasonable weighting and adjustment, ensure that the final feature matrix can not only fully reflect the state of the system, but also avoid misjudgment caused by abnormal characteristics at a certain scale. For example, in actual monitoring, some anomalies of the overhead line suspension may be particularly significant at the medium scale, but be masked by noise at smaller or larger scales. Through this formula, the system can automatically adjust the weights of different scales, so as to more accurately detect these medium-scale anomalies. Finally, the feature matrix The calculation of not only provides a comprehensive indicator for subsequent anomaly detection, but also ensures the robustness of the indicator in various situations through Gaussian weight and feature adjustment. Especially in the face of complex and changing railway operating environments, this multi-scale, multi-parameter processing method can effectively filter out environmental noise while highlighting those features that are truly valuable for diagnosis, enabling the system to warn of potential faults earlier and provide a basis for maintenance decisions.

[0083] Example 8: The tension estimation result is obtained by using the scale morphological feature matrix through the following formula: and displacement function :

[0084] ;

[0085] ;

[0086] in, is the density of the overhead wire suspension; is the acceleration due to gravity; is the span of the overhead contact network suspension; is the midpoint sag of the overhead contact network suspension; is the correction factor set to adjust the effect of morphological features on tension estimation; is the horizontal distance along the span; is the number of terms in the Fourier series; For the Fourier coefficients, given by Sure: ;in It is with The weight matrix corresponding to the Fourier components; is the preset attenuation parameter.

[0087] Specifically, considering the complexity of the contact network suspension system in the real environment, the tension estimation formula not only relies on the classical physical model, but also combines the features extracted from the multi-scale morphological feature matrix. The core part of this formula comes from the classical physical formula, and its goal is to reflect the basic tension of the suspension system under static load. However, the classical physical model is usually based on several assumptions, such as the idealized state and linear conditions of the system, which are often not fully met in complex practical applications. Therefore, relying solely on traditional physical formulas is not enough to accurately describe the actual state of the suspension system, especially in the face of multiple environmental factors and structural complexity. In order to meet these challenges, the formula is further combined with the results of the morphological feature matrix. Morphological feature matrix It is obtained through multi-scale morphological analysis, which captures the morphological characteristics of the overhead contact suspension system at different scales. Specifically, morphological analysis can reveal changes in the overhead contact suspension system at different time and space scales, such as local stress concentration, wear, or other morphological changes. These changes may be caused by external environmental influences, long-term fatigue, or the complexity of the structure itself. By incorporating these morphological characteristics into the tension estimation formula, the system can more accurately reflect the actual situation instead of relying on idealized assumptions. The term is a key innovation, which introduces the correction of the effect of morphological features on tension. is the Frobenius norm of the morphological feature matrix, which represents the overall "energy" or amplitude of the matrix. This norm provides a quantitative indicator to describe the overall intensity of the changes contained in the morphological feature matrix. The purpose of introducing this term is to enable the tension estimation to adaptively consider the influence of these morphological features, rather than just relying on simple geometric and physical parameters. The physical meaning of the Frobenius norm is that it integrates the amount of information in the entire morphological matrix, which may contain morphological changes at different scales. The morphological feature matrix can be regarded as a high-dimensional description of the state of the suspension system to some extent. By compressing this high-dimensional information into a scalar through the Frobenius norm, the system can simply and effectively introduce morphological features into the tension estimation. This design enables the system to flexibly respond to various complex real-world situations. For example, local wear or deformation may cause local tension changes, and these changes can be captured through morphological analysis and reflected in the final tension estimation results. In addition, the correction factor in the formula The influence of morphological features on tension estimation is controlled. The choice of this correction factor depends on the specific application scenario and empirical data, and the purpose is to ensure that the influence of morphological features is not overly amplified or weakened. Reasonable choice This allows the tension estimation to reflect the importance of morphological features without causing distortion in the tension estimation due to noise or errors in the morphological features. On this basis, the tension estimation formula also takes into account the morphological feature matrix Dynamic adjustment of tension estimation is implemented in the form of an exponential function. The exponential function ensures that when the morphological feature matrix When the norm of is large, the tension estimation result will be significantly adjusted, while when the norm is small, the adjustment is relatively mild. This nonlinear adjustment mechanism is based on the assumption that the greater the change in morphological characteristics, the higher the probability that the actual stress on the system deviates from the ideal situation, and therefore requires a larger correction. This mechanism enables the tension estimation formula to adaptively reflect the system state under different morphological characteristics, ensuring the accuracy and robustness of the estimation results.

[0088] After obtaining the tension estimation result After that, the system further calculates the displacement function , to describe the morphological changes of the contact network suspension under the action of force. The calculation of the displacement function also has important physical significance. It combines the density of the suspension system, the acceleration of gravity, and the tension estimation results to construct a mathematical model that can accurately describe the morphological changes of the suspension system. The first part of the displacement function is based on the classic formula of the catenary, which describes the displacement curve under uniform tension. This part of the formula reflects the smooth curve formed by the deadweight and tension of the suspension system under ideal conditions. In the mathematical description of the catenary, the hyperbolic cosine function Describes the typical shape of the suspension system without external disturbance. This description is very suitable for analyzing the basic behavior of the suspension system under constant tension. However, the overhead suspension system is affected by many complex factors in actual operation, including external loads, changes in environmental conditions (such as wind force, temperature changes) and the nonlinear characteristics of the system itself. Therefore, relying solely on the classical catenary formula is not enough to fully describe the displacement behavior of the suspension system. In order to more accurately describe these complexities, the formula introduces the Fourier series expansion to supplement and correct the displacement function. The introduction of the Fourier series is intended to capture periodic disturbances and local changes in the suspension system, which are often difficult to describe by simple linear models. Each term of the Fourier series corresponds to a sine wave of a specific frequency, which can describe the periodic vibrations and subtle local deformations in the suspension system. In this way, the system can capture the high-frequency components in the displacement, which are usually related to local mechanical stress concentration or material fatigue properties. The Gaussian attenuation factor introduced in the formula The high-order terms of the Fourier series are moderately attenuated to avoid overfitting high-frequency noise while retaining the main vibration modes that have an important impact on the system state. By combining the classical physical model and Fourier series correction, the displacement function It provides a complete description of the catenary suspension system. This description can not only reflect the overall morphology of the system, but also capture subtle local changes, thus providing a more detailed analysis tool for system status monitoring. This is of great significance for ensuring the safe operation of the railway system, because any slight abnormality in the catenary suspension system may indicate a potential major failure.

[0089] Embodiment 9: The abnormal state determination unit calculates the abnormal state index of the overhead line suspension according to the tension estimation result and the spectrum analysis result by the following formula: :

[0090] ;

[0091] in, is the standard stress value, is the set value; is the dimension-reduced feature matrix Dimensions; is the maximum value of the displacement function; is the minimum value of the displacement function.

[0092] Specifically, the first part of the formula Used to evaluate the tension of the current measurement With standard stress value The standard stress value here is It is the ideal stress value determined based on system design or long-term operation data, reflecting the tension level that the suspension system should have under normal operating conditions. The deviation from the standard stress value reflects the change in the stress state of the system, which may be caused by external load changes, material fatigue, or structural damage. , the formula quantifies the extent of this deviation. The exponent 0.25 is introduced to amplify the deviation nonlinearly so that even small deviations can be significantly reflected in the abnormal state index. This nonlinear treatment increases the sensitivity of the formula to small tension changes, especially in the overhead suspension system, which may indicate early failure or damage. The next section It is based on the dimension reduction feature matrix and initialize the composite data matrix Here, the formula is calculated by calculating the matrix Elements of With the reduced feature matrix The difference between them can further evaluate the abnormality of the system. Contains feature information after multi-scale morphological analysis and dimensionality reduction, which reflects the core features of the system after multimodal data processing. is the original data matrix obtained directly from the multimodal data acquisition unit. In this process, the formula is compared and The differences between corresponding elements in the data are used to detect deviations and anomalies between data. is the Frobenius norm of the morphological feature matrix, which plays a normalizing role in the formula. By normalizing the difference, the formula can avoid error amplification caused by data magnitude or noise. At the same time, The exponential processing is also to enhance the ability to respond to data deviations, where is the dimension-reduced feature matrix This part is designed to ensure that even in a complex multimodal data environment, the system can still accurately detect data deviations from the normal state. Through this comparison and standardization process, the formula can comprehensively evaluate the overall abnormality of multimodal data, thereby further improving the reliability of the abnormal state index. The last part of the formula It is an important factor to adjust the abnormal state index based on the dynamic range of the displacement function. It reflects the deformation of the contact network when it is suspended in different positions. and are the maximum and minimum values ​​of the displacement function, respectively. Their difference reflects the horizontal distance of the suspension system. The magnitude of deformation on the suspension. A larger displacement difference usually means that the suspension system is subjected to greater stress or has a more significant deformation, which may be an early signal of potential failure. By introducing this part, the formula can dynamically adjust the abnormal state index so that it can reflect the potential abnormality caused by the displacement change. The index processing here This makes the effect of displacement difference on abnormal state index show nonlinear growth. When the displacement difference is large, the output of the exponential function will increase rapidly, indicating that the system may be in an abnormal state. Preset parameters Control the sensitivity of this index and set it reasonably This allows the system to respond appropriately to displacement changes of varying degrees, thereby avoiding false alarms or missed alarms.

[0093] Although the specific embodiments of the present invention are described above, it should be understood by those skilled in the art that these specific embodiments are only illustrative, and those skilled in the art may omit, replace, and change the details of the above methods and systems in various ways without departing from the principles and essence of the present invention. For example, merging the above method steps so as to perform substantially the same functions in substantially the same manner to achieve substantially the same results is within the scope of the present invention. Therefore, the scope of the present invention is limited only by the appended claims.

Claims

1. Multi-modal fusion intelligent detection system for railway track contact network suspension monitoring, It is characterized in that The system comprises: a multimodal data acquisition and processing unit, a data processing and analysis unit, and an abnormal state determination unit; the multimodal data acquisition and processing unit is used to uniformly select a plurality of sampling points on the overhead line suspension, and for each sampling point, use a multimodal data acquisition device to acquire multimodal data, and construct an initialization composite data matrix of the overhead line suspension based on the multimodal data; the data processing and analysis unit is used to perform nonlinear feature extraction and dimensionality reduction on the initialization composite data matrix to obtain a reduced dimensionality feature matrix; perform multiscale morphological analysis on the reduced dimensionality feature matrix to obtain a multiscale morphological feature matrix; and use the scale morphological feature matrix to extract nonlinear features and reduce the dimensionality of the initialization composite data matrix. The tension of the array is estimated to obtain the tension estimation result and the displacement function; the abnormal state judgment unit is used to calculate the abnormal state index of the contact network suspension according to the tension estimation result and the displacement function, and define the abnormal judgment interval; when the abnormal state index is greater than the upper limit of the abnormal judgment interval, it is judged that the contact network suspension is abnormal; when the abnormal state index is within the abnormal judgment interval, it is judged that the probability of the contact network suspension being abnormal is between 0 and 1; when the abnormal state index is less than the lower limit of the abnormal judgment interval, it is judged that the contact network suspension is not abnormal; the multimodal data acquisition device includes: a laser radar, a multispectral camera and an electromagnetic induction sensor; set the first The wavelength of the sampling point is ; The laser radar obtains the The three-dimensional coordinate vector of the sampling point is ; Multispectral camera in the The spectral vector obtained by sampling is ; Electromagnetic induction sensor in the The electromagnetic field intensity vector measured at the sampling points is ; The obtained initialized composite data matrix Use the following formula to express it: ; in, is the number of sampling points; Subscript index, value range is 1 to The data processing and analysis unit performs nonlinear feature extraction and dimensionality reduction on the initialized composite data matrix through the following formula to obtain a reduced dimensionality feature matrix: : ; in, Represents element-wise multiplication; yes A matrix mask of ; is the eigenvalue vector; is the feature vector; is the regularized centralization kernel matrix; is a nonlinear mapping function.

2. The multi-modal fusion intelligent detection system for railway track contact network suspension monitoring according to claim 1, It is characterized in that After performing nonlinear feature extraction and dimensionality reduction on the initialized composite data matrix, the reduced dimensionality feature matrix is ​​obtained. In the process, the data processing and analysis unit defines the Gaussian kernel function through the following formula: As a nonlinear mapping: ; in, is the bandwidth parameter of the kernel function; is the first variable of the Gaussian kernel function; is the second variable of the Gaussian kernel function; construct the kernel matrix , in the kernel matrix, Row, No. The column elements are ; The centralized kernel matrix is ​​obtained through the following formula : ; in, The elements are all of matrix; The eigenvalue problem is solved by the following formula: ;in, is the eigenvalue, is the corresponding eigenvector; sort the eigenvalues ​​in descending order and select the first The eigenvalues ​​form the eigenvalue vector and the corresponding eigenvector : ; ; The kernel principal component is calculated using the following formula: ; The regularization term is introduced through the following formula: ;in, is a set positive constant, is the identity matrix.

3. The multi-modal fusion intelligent detection system for monitoring the suspension of railway track contact network as claimed in claim 2, It is characterized in that The data processing and analysis unit performs multi-scale morphological analysis on the dimension reduction feature matrix to obtain a multi-scale morphological feature matrix, and defines an adaptive multi-scale morphological operator by the following formula: : ; in, is the maximum number of scales defined; is the current scale index; For the A set of structural elements of different scales; For improved expansion operation, is the expansion strength parameter; For improved corrosion performance, is the corrosion intensity parameter; It is an adaptive opening operation; It is an adaptive closing operation; is a weighted sum operation; It is a weighted difference operation.

4. The multi-modal fusion intelligent detection system for monitoring the suspension of railway track contact network as claimed in claim 3, It is characterized in that Improved expansion operation The calculation process is expressed by the following formula: ; in, For expansion operation; is the dimension-reduced feature matrix Elements of is the F-norm operation; is the absolute value operation; Indicates A set of structural elements of different scales No. elements; Represented in the set of structural elements Each element of Up, take and The maximum value of the corresponding position elements in ; Representation Matrix The maximum absolute value of the elements in; Representation Matrix The minimum absolute value of the element in reflects the strength of the weakest signal in the data; Improved corrosion operation Use the following formula to express it: ; in, For corrosion operation.

5. The multi-modal fusion intelligent detection system for monitoring the suspension of railway track contact network as claimed in claim 4, It is characterized in that Multi-scale morphological feature matrix Calculated using the following formula: ; in, and Respectively Expansion strength parameters and corrosion strength parameters of each scale; is the mean parameter of Gaussian weight; Adjust parameters for preset matrix features; is the trace of the matrix.

6. The multi-modal fusion intelligent detection system for monitoring the suspension of railway track contact network according to claim 5, It is characterized in that The tension estimation result is obtained by using the scale morphological feature matrix to estimate the tension through the following formula: and displacement function : ; ; in, is the density of the overhead wire suspension; is the acceleration due to gravity; is the span of the overhead contact network suspension; is the midpoint sag of the overhead contact network suspension; is the correction factor set to adjust the effect of morphological features on tension estimation; is the horizontal distance along the span; is the number of terms in the Fourier series; For the Fourier coefficients, given by Sure: ;in It is with The weight matrix corresponding to the Fourier components; is the preset attenuation parameter.

7. The multi-modal fusion intelligent detection system for monitoring the suspension of railway track contact network according to claim 6, It is characterized in that The abnormal state determination unit calculates the abnormal state index of the contact network suspension according to the tension estimation result and the spectrum analysis result by the following formula: : ; in, is the standard stress value, is the set value; is the dimension-reduced feature matrix Dimensions; is the maximum value of the displacement function; is the minimum value of the displacement function.

Citation Information

Cited By

  • Track mileage calibration method and system based on track geometric dynamic and static waveform matching

    CN120632276A

  • Track mileage calibration method and system based on track geometry dynamic and static waveform matching

    CN120632276B