A comprehensive monitoring method and system for turnout working conditions based on feature extraction

Abstract

The present invention discloses a comprehensive monitoring method and system for turnout working conditions based on feature extraction, which relates to the technical field of rail transit. This solution overcomes the characteristic that traditional vibration spectrum analysis is difficult to reflect the change of electrical characteristics. By combining high-frequency vibration signals with contact resistance data, it realizes the monitoring of the target monitoring part of the turnout. While enriching the monitoring information, it also improves the comprehensive understanding of the turnout working conditions, improves the automation level of turnout monitoring, reduces the need for manual intervention, makes the maintenance and management of the target monitoring part of the turnout more scientific and systematic, and can make targeted maintenance decisions based on real-time monitoring data, thereby improving the operation efficiency and safety of the equipment.

Description

Technical Field

[0001] The present invention belongs to the technical field of rail transit, and particularly relates to a comprehensive monitoring method and system for turnout working conditions based on feature extraction. Background Art

[0002] Railway turnouts mainly consist of switch machines, frog hearts, switch rails, slide plates, connecting parts, etc. The switch machine includes stock rails, switch rails and connecting parts, and its main function is to guide the train to run on the correct track. The frog heart is a key component of the turnout, which is used to fix the turnout and keep it stable when the train passes through. The tightness between the switch rail and the stock rail is crucial for the running safety of the train. When the train approaches the turnout, the switch rail of the switch moves towards the selected track under the action of the repulsive force and the guidance of the stock rail. This process utilizes the principle of flange guidance. By controlling the position and movement of the flange, the train runs on the predetermined track. Currently, the commonly adopted on-line monitoring technology for railway turnouts is mainly sensor technology. Among them, vibration sensors are often used in sensor technology to obtain the vibration data of railway turnouts, and the vibration data is analyzed and processed to obtain the vibration spectrum. The vibration spectrum is used to reflect the dynamic characteristics of railway turnouts, such as whether the actions of the switch machine and the switch rail are smooth, and whether there are jams or wear. However, single-parameter monitoring has certain limitations. For example, the vibration spectrum can only reflect the dynamic characteristics of the switch machine and the switch rail, and cannot provide information on the change of electrical characteristics, resulting in difficulty in detecting early faults. Summary of the Invention

[0003] Aiming at the defects in the prior art, the present invention provides a comprehensive monitoring method and system for turnout working conditions based on feature extraction to solve the above technical problems.

[0004] A comprehensive monitoring method for turnout working conditions based on feature extraction, the steps of the method include:

[0005] Deploy a lightweight LSTM network to obtain multi-source feature data of the target monitoring part of the turnout, where the multi-source feature data includes high-frequency vibration signals and contact resistance data, and align and fuse the multi-source feature data based on the acquisition time;

[0006] Perform denoising processing, data conversion and dataset division on the aligned and fused multi-source feature data, so as to calculate the rank correlation coefficient of the multi-source feature data, and construct an association matrix based on the rank correlation coefficient by using graph theory;

[0007] Adopt a feature fusion algorithm to compare the association matrix with a preset fault template library, and output the association data exceeding the similarity with the preset threshold as a warning signal.

[0008] Among them, fusing multi-source data can more comprehensively reflect the operating state of the turnout, overcome the limitations of single-parameter monitoring, and make the identification of early faults more sensitive by combining electrical characteristics and dynamic characteristics, making the maintenance and management of the turnout more scientific and systematic.

[0009] Preferably, when calculating the rank correlation coefficient of the multi-source feature data, the following formula is specifically used: , ,

[0010] , ;

[0011] Among them, is the high-frequency vibration signal of the time series, is the resistance data collected synchronously, is the sliding window is the rank of the high-frequency vibration signal within the sliding window, is the acquisition time, is the half-width of the sliding window, is the rank of the resistance data corresponding to the sliding window within, is the time decay weight, is the decay coefficient, is the weighted average rank of the high-frequency vibration signal, is the weighted average rank of the high-frequency vibration signal.

[0012] Here, the rank correlation coefficient is obtained through the vibration signal and the resistance data and then data processing is carried out. Compared with the linear relationship processing method used in general data processing, it can effectively capture the non-linear relationship and monotonic relationship between variables, is not sensitive to outliers, and can still provide a reliable correlation evaluation in the case of noise or extreme values in the data.

[0013] Preferably, when aligning and fusing the multi-source feature data based on the acquisition time, the following steps are specifically included:

[0014] Construct a three-dimensional coordinate system, where the three dimensions respectively represent the instantaneous frequency, energy concentration degree, and frequency change rate of the high-frequency vibration signal;

[0015] Perform clustering analysis on the data coordinate points in the three-dimensional coordinate system to form several signal clustering clusters, and eliminate the data points far from these signal clustering clusters according to the K-means clustering algorithm;

[0016] Adopt the IEEE 1588 precision clock protocol to align the time stamps of the high-frequency vibration signal and the contact resistance data.

[0017] By constructing a three-dimensional coordinate system, the instantaneous frequency, energy concentration degree, and frequency change rate of high-frequency vibration signals can be considered simultaneously. When the K-means clustering algorithm is subsequently used to perform clustering analysis on data coordinate points, different modes and characteristics of the signals can be effectively identified, and the influence of noise on the analysis results can be reduced.

[0018] Preferably, when clustering the coordinate points in the three-dimensional coordinate system to obtain several signal clustering clusters, the following steps are specifically included:

[0019] Obtain the Euclidean distance between any data coordinate point and other data coordinate points;

[0020] Adopt the ISODATA clustering algorithm to determine the within-class distance based on the obtained multiple Euclidean distances;

[0021] Integrate the data coordinate points in the three-dimensional coordinate system into several signal clusters according to the within-class distance.

[0022] Quantifying the similarity between data points using the Euclidean distance can accurately reflect the true distribution of the data, lay a foundation for the subsequent process of the ISODATA clustering algorithm to adjust the clustering quantity and shape, and avoid information loss when fixing the clustering number.

[0023] Preferably, when adopting the ISODATA clustering algorithm to determine the within-class distance based on the obtained multiple Euclidean distances, use the metric loss function to determine the within-class distance;

[0024] The metric loss function is specifically as follows: , ;

[0025] Among them, is the overall loss value of the th iteration, is the number of clusters in the current iteration, is the number of high-frequency vibration signals in the th clustering cluster, is the th sample vector in the th clustering cluster, is the mean of the sample vectors in the th clustering cluster, is the centroid of the th class in the current iteration, is the within-class scatter threshold, is the regularization coefficient, is the number of samples in the th class, is the The sample quantity weight of the class is the standard deviation of the within-class distance.

[0026] By using a metric loss function to determine the within-class distance, the data points within the same cluster can be made closer, while the distance between different clusters becomes more obvious, reducing the overlap between clusters and meeting the requirements of different distributions.

[0027] Preferably, when removing the data points far from these signal clusters according to the K-means clustering algorithm, the following steps are specifically included:

[0028] After normalizing the multi-source feature data, use the elbow method to determine the number of clusters K, and obtain K cluster centers and the cluster labels of each multi-source feature data according to the K-means clustering algorithm;

[0029] The within-class distance determines the screening threshold for data points, and the data points exceeding the threshold are removed according to the screening threshold.

[0030] First, by normalizing the multi-source feature data, the dimensional difference between different feature data can be eliminated, thereby assisting the K-means clustering algorithm to quickly identify the structure and pattern of the data, quickly removing the abnormal data points that are too far from the cluster center, and ensuring the stability and consistency of the clustering results.

[0031] Preferably, when performing denoising processing, data conversion, and dataset partitioning on the aligned and fused multi-source feature data, the following steps are specifically included:

[0032] Select Daubechies wavelet to perform wavelet decomposition on the multi-source feature data to obtain the detail coefficients at multiple frequencies, perform soft threshold processing on the detail coefficients to remove noise, and then perform wavelet reconstruction on the denoised detail coefficients;

[0033] Use standardization processing means to process the multi-source data features after wavelet reconstruction, and scale the multi-source data features to the interval [0, 1];

[0034] Partition the multi-source data features after interval partitioning according to a set ratio.

[0035] By performing soft threshold processing on the detail coefficients using Daubechies wavelet, noise can be effectively removed, the main features of the signal can be retained, the change characteristics of the signal at different frequencies can be captured, and combined with the subsequent step of scaling the data to the interval [0, 1], it helps to eliminate the dimensional difference between different detail coefficients.

[0036] Preferably, the lightweight LSTM network uses a stacked multi-layer LSTM network and a fully connected layer.

[0037] The first - layer LSTM network is used to obtain the time - series features when collecting the multi - source feature data;

[0038] The second - layer LSTM network obtains further abstract time - dependence features based on the time - series features;

[0039] The fully - connected layer is used to convert the abstract time - output sequence output by the second - layer LSTM network into a fixed - length vector.

[0040] The first - layer LSTM network can effectively capture the dynamic patterns of data changing over time. Based on the extraction results of the first - layer LSTM network, the feature - abstraction process of the second - layer LSTM network enables the model to more deeply understand the complex patterns in the data. Through the combination of two layers of LSTM, the model can process more complex time - series data. Through layer - by - layer abstraction and conversion, the model can effectively extract and retain key information, reduce information loss. Finally, the fully - connected layer is used to convert the abstract time - series output by the second - layer LSTM network into a fixed - length vector, and this fixed - length output form simplifies the subsequent processing.

[0041] A comprehensive monitoring system for turnout working conditions based on feature extraction, comprising:

[0042] A lightweight LSTM network, a multi - source feature data acquisition module, a data cleaning module, and a data processing module;

[0043] The multi - source feature data acquisition module is used to collect high - frequency vibration signals and contact - resistance data of the turnout target monitoring part;

[0044] The lightweight LSTM network is used to process time - series data and capture the time - series dependence relationships in the multi - source feature data;

[0045] The data cleaning module constructs a correlation matrix based on the high - frequency vibration signals and contact - resistance data collected by the multi - source feature data acquisition module;

[0046] The data processing module is used to obtain the comparison result between the correlation matrix and a preset fault template library, and output the correlation data with a similarity exceeding the preset value as a warning signal.

[0047] Preferably, the data cleaning module further includes a data alignment and fusion module, a data - set division module, a rank - correlation coefficient acquisition module, and a matrix construction module;

[0048] The data alignment and fusion module is used to align and fuse the multi - source feature data based on the acquisition time;

[0049] The data - set division module is used to perform denoising processing, data conversion, and data - set division on the aligned and fused multi - source feature data;

[0050] The rank correlation coefficient obtaining module is used to calculate the rank correlation coefficient based on the divided dataset;

[0051] The matrix construction module is used to construct an association degree matrix by using the graph theory method based on the rank correlation coefficient.

[0052] The beneficial effects of the present invention are as follows: This solution overcomes the characteristic that traditional vibration spectrum analysis is difficult to reflect the changes in electrical characteristics. By combining high-frequency vibration signals with contact resistance data, it realizes the monitoring of the target monitoring part of the turnout. While enriching the monitoring information, it also improves the comprehensive understanding of the turnout working conditions, improves the automation level of turnout monitoring, reduces the need for manual intervention, makes the maintenance and management of the target monitoring part of the turnout more scientific and systematic, and can make targeted maintenance decisions based on real-time monitoring data, thereby improving the operation efficiency and safety of the equipment. Brief Description of the Drawings

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0054] Figure 1 It is a flowchart of a comprehensive turnout working condition monitoring method based on feature extraction provided by the present invention;

[0055] Figure 2 It is a structural schematic diagram of a comprehensive turnout working condition monitoring system based on feature extraction provided by the present invention. Detailed Embodiments

[0056] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention.

[0057] The following disclosure provides many different embodiments or examples for implementing different structures of the present invention. To simplify the disclosure of the present invention, the components and settings of specific examples are described below. Of course, they are only examples and are not intended to limit the present invention.

[0058] The embodiments of the invention will be described in detail below with reference to the accompanying drawings.

[0059] As Figure 1 shown, a comprehensive monitoring method for turnout working conditions based on feature extraction, the steps of the method include:

[0060] Deploy a lightweight LSTM network to obtain multi-source feature data of the target monitoring part of the turnout, where the multi-source feature data includes high-frequency vibration signals and contact resistance data, and align and fuse the multi-source feature data based on the acquisition time;

[0061] Perform denoising processing, data conversion and dataset division on the aligned and fused multi-source feature data, thereby calculating the rank correlation coefficient of the multi-source feature data, and constructing an association matrix based on the rank correlation coefficient using the graph theory method;

[0062] Adopt a feature fusion algorithm to compare the similarity between the association matrix and a preset fault template library, and output the association data exceeding the similarity with a preset threshold as a warning signal.

[0063] This solution overcomes the characteristic that traditional vibration spectrum analysis is difficult to reflect the change of electrical characteristics. By combining high-frequency vibration signals and contact resistance data, it realizes the monitoring of the target monitoring part of the turnout. While enriching the monitoring information, it also improves the comprehensive understanding of the turnout working conditions, improves the automation level of turnout monitoring, reduces the need for manual intervention, makes the maintenance and management of the target monitoring part of the turnout more scientific and systematic, and can make targeted maintenance decisions based on real-time monitoring data, thereby improving the operation efficiency and safety of the equipment.

[0064] More specifically, when calculating the rank correlation coefficient of the multi-source feature data, the following formula is specifically used: , ,

[0065] , ;

[0066] Among them, is the time series high-frequency vibration signal, is the resistance data collected synchronously, is the sliding window the rank of the high-frequency vibration signal within, is the acquisition time, is the half-width of the sliding window, corresponds to the sliding window the rank of the internal resistance data within, is the time decay weight, is the decay coefficient, is the weighted average rank of the high-frequency vibration signal, is the weighted average rank of the high-frequency vibration signal.

[0067] By calculating the rank correlation coefficient, the high-frequency vibration signal and the contact resistance data are integrated into a correlation matrix, revealing the subtle changes between the vibration signal and the resistance data, thereby improving the sensitivity of early fault detection. This multi-dimensional data fusion method helps to comprehensively evaluate the health status of the monitored parts of the turnout, overcoming the limitations of single-parameter monitoring. This data-driven decision support system can improve maintenance efficiency, optimize resource allocation, simplify the data analysis process, make the monitoring system more efficient and automated, and reduce the need for manual intervention. Among them, takes 0.01.

[0068] More specifically, when aligning and fusing the multi-source feature data based on the acquisition time, the following steps are specifically included:

[0069] Construct a three-dimensional coordinate system, where the three dimensions respectively represent the instantaneous frequency, energy concentration, and frequency change rate of the high-frequency vibration signal;

[0070] Perform clustering analysis on the data coordinate points in the three-dimensional coordinate system to form several signal clustering clusters, and remove the data points far from these signal clustering clusters according to the K-means clustering algorithm;

[0071] Adopt the IEEE 1588 Precision Clock Protocol to align the time stamps of the high-frequency vibration signal and the contact resistance data.

[0072] The multi-dimensional feature representation method of constructing a three-dimensional coordinate system provides a more comprehensive signal feature analysis, can better capture the dynamic changes and complexity of the signal, and transcends the limitations of traditional single-parameter analysis. High-precision time stamp alignment such as the IEEE 1588 Precision Clock Protocol can improve the accuracy of data fusion and ensure the reliability of the analysis results.

[0073] More specifically, when clustering the coordinate points in the three-dimensional coordinate system to obtain several signal clustering clusters, the following steps are specifically adopted:

[0074] Obtain the Euclidean distance between any data coordinate point and other data coordinate points;

[0075] Adopt the ISODATA clustering algorithm to determine the within-class distance based on the obtained multiple Euclidean distances.

[0076] Integrate the data coordinate points in the three-dimensional coordinate system into several signal clusters according to the within-class distance.

[0077] The ISODATA clustering algorithm is adopted to make the clustering result adapt to the actual distribution of the data, better conform to the data characteristics, avoid the information loss caused by a fixed number of clusters, and at the same time, it can also integrate the data coordinate points in the three-dimensional coordinate system into several signal clusters, effectively identify and extract the potential patterns in the data. This enhanced data analysis ability helps to deeply understand the data characteristics and provides support for subsequent fault detection and early warning.

[0078] More specifically, when using the ISODATA clustering algorithm to determine the within-class distance based on the obtained multiple Euclidean distances, a metric loss function is used to determine the within-class distance;

[0079] The metric loss function is specifically as follows: , ;

[0080] where is the overall loss value of the th iteration, is the number of clusters in the current iteration, is the number of high-frequency vibration signals in the th cluster, is the rd vector of the th sample in the th cluster, is the mean of the sample vectors in the th cluster, is the centroid of the th class in the current iteration, is the within-class dispersion threshold, is the regularization coefficient, is the th class sample number, is the sample number weight of the th class, is the standard deviation of the within-class distance.

[0081] Using the metric loss function to determine the within-class distance can dynamically adjust the distance metric according to the characteristics of the data, improving the accuracy of clustering. This self-adaptability makes the algorithm perform better when dealing with complex data such as high-frequency vibration signals and contact resistance data, effectively reducing the impact of outliers on the calculation of the within-class distance, improving the overall performance of the clustering model, and making the model perform better in classification and prediction tasks.

[0082] More specifically, when removing data points far from these signal clustering clusters according to the K-means clustering algorithm, the following steps are further included:

[0083] After normalizing the multi-source feature data, the elbow method is used to determine the number of clusters K, and K cluster centers and the cluster labels of each multi-source feature data are obtained according to the K-means clustering algorithm;

[0084] The within-class distance determines the screening threshold for data points, and data points exceeding the threshold are removed according to the screening threshold.

[0085] By analyzing the relationship between the clustering error and the number of clusters, a suitable K value is found to optimize the clustering effect. By obtaining K cluster centers and the cluster labels of each multi-source feature data, the structure and pattern of the data can be quickly identified, the efficiency of data analysis can be improved, and the clustering fluctuation caused by outliers can be further reduced, improving the interpretability of the results.

[0086] More specifically, when performing denoising processing, data transformation, and dataset partitioning on the aligned and fused multi-source feature data, the following steps are further included:

[0087] Select Daubechies wavelet to perform wavelet decomposition on the multi-source feature data to obtain detail coefficients at multiple frequencies, perform soft threshold processing on the detail coefficients to remove noise, and then perform wavelet reconstruction on the denoised detail coefficients;

[0088] Use standardization processing means to process the multi-source data features after wavelet reconstruction, and scale the multi-source data features to the interval [0, 1];

[0089] Partition the multi-source data features after interval partitioning according to a set ratio.

[0090] The Daubechies wavelet processing method performs excellently in processing non-stationary signals such as high-frequency vibration signals and contact resistance data, which can greatly improve the quality of data. At the same time, the multi-scale analysis ability of the Daubechies wavelet processing method itself enables a deeper understanding of complex signals, can better identify potential fault patterns, and the processes of wavelet decomposition and wavelet reconstruction ensure the integrity and accuracy of the signals, enabling subsequent analysis to be based on high-quality data, ensuring the stability of model training. This flexibility makes the dataset partitioning more in line with actual needs and improves the generalization ability of the model.

[0091] More specifically, the lightweight LSTM network uses a stacked multi-layer LSTM network and a fully connected layer,

[0092] The first-layer LSTM network is used to obtain the time-series features when collecting the multi-source feature data;

[0093] The second-layer LSTM network obtains further abstract time-dependent features based on the time-series features;

[0094] The fully connected layer is used to convert the abstract time output sequence output by the second-layer LSTM network into a fixed-length vector.

[0095] The first-layer LSTM network focuses on obtaining the time-series features of multi-source feature data. This time-series feature extraction ability enables the model to understand the timing information of the data. The second-layer LSTM network, based on the time-series features extracted by the first layer, further obtains abstract time-dependent features. This hierarchical feature abstraction process enables the lightweight LSTM network to more deeply understand the complex patterns in non-stationary signal data such as high-frequency vibration signals and contact resistance data, improving the modeling ability for non-linear relationships and long-term and short-term dependencies. Through layer-by-layer abstraction and conversion, finally the fully connected layer converts the abstract time series output by the second-layer LSTM network into a fixed-length vector, facilitating subsequent classification or regression tasks. The model can effectively extract and retain key information and reduce information loss.

[0096] As Figure 2 shown, a comprehensive turnout condition monitoring system based on feature extraction includes:

[0097] A lightweight LSTM network, a multi-source feature data acquisition module, a data cleaning module, and a data processing module;

[0098] The multi-source feature data acquisition module is used to collect high-frequency vibration signals and contact resistance data of the turnout target monitoring part;

[0099] The lightweight LSTM network is used to process time-series data and capture the timing dependency relationships in multi-source feature data;

[0100] The data cleaning module constructs a correlation matrix based on the high-frequency vibration signals and contact resistance data collected by the multi-source feature data acquisition module;

[0101] The data processing module is used to obtain the comparison result between the correlation matrix and the preset fault template library, and output the correlation data with a similarity exceeding the preset value as a warning signal.

[0102] Among them, the multi-source data acquisition module uses an acceleration sensor to collect high-frequency vibration signals of the turnout target monitoring part, uses a contact resistance measuring instrument to measure the contact resistance of the turnout part, and a data acquisition card is also configured in the multi-source data acquisition module to convert the analog signals collected by the acceleration sensor into digital signals.

[0103] More specifically, the data cleaning module further includes a data alignment and fusion module, a data set division module, a rank correlation coefficient acquisition module, and a matrix construction module;

[0104] The data alignment and fusion module is used to align and fuse the multi-source feature data based on the acquisition time;

[0105] The data set division module is used to perform denoising processing, data conversion, and data set division on the aligned and fused multi-source feature data;

[0106] The rank correlation coefficient acquisition module is used to calculate the rank correlation coefficient based on the divided data set;

[0107] The matrix construction module is used to construct an association matrix based on the rank correlation coefficient by using the graph theory method.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered by the scope of the claims and the specification of the present invention.

Claims

1. A comprehensive monitoring method for turnout working conditions based on feature extraction, characterized in that: The steps of the method include: Deploy a lightweight LSTM network to obtain multi-source feature data of the target monitoring part of the turnout, wherein the multi-source feature data includes high-frequency vibration signals and contact resistance data, and align and fuse the multi-source feature data based on the acquisition time; De-noising, data conversion and data set partitioning are performed on the aligned and fused multi-source feature data, thereby calculating the rank correlation coefficient of the multi-source feature data, and constructing a correlation matrix based on the rank correlation coefficient using a graph theory method; Comparing the correlation matrix with a preset fault template library for similarity, and outputting correlation data with a similarity exceeding a preset threshold as a warning signal; When aligning and fusing the multi-source feature data based on the acquisition time, the following steps are specifically included: A three-dimensional coordinate system is constructed, in which the three dimensions represent the instantaneous frequency, energy concentration and frequency change rate of the high-frequency vibration signal respectively; Perform cluster analysis on the data coordinate points in the three-dimensional coordinate system to form several signal clusters, and remove data points far away from these signal clusters based on the K-means clustering algorithm; Adopt IEEE 1588 precision clock protocol to align the high-frequency vibration signal and contact resistance data; The lightweight LSTM network uses a stacked multi-layer LSTM network and a fully connected layer. The first layer of LSTM network is used to obtain the time series features when collecting the multi-source feature data; The second layer of LSTM network obtains further abstract time-dependent features based on the time series features; The fully connected layer is used to convert the abstract time output sequence output by the second layer LSTM network into a fixed-length vector.

2. The method for comprehensive monitoring of turnout working conditions based on feature extraction according to claim 1 is characterized in that: When calculating the rank correlation coefficient of the multi-source feature data, the following formula is specifically used: ， ， ， ； in, is a time series high frequency vibration signal, For the synchronously collected resistance data, For sliding window The rank of the internal high frequency vibration signal, is the collection time, is the half width of the sliding window, For the corresponding sliding window The rank of the internal resistance data, is the time decay weight, is the attenuation coefficient, is the weighted average rank of the high-frequency vibration signal, is the weighted average rank of the high-frequency vibration signal.

3. The method for comprehensive monitoring of turnout working conditions based on feature extraction according to claim 1 is characterized in that: When clustering the coordinate points in the three-dimensional coordinate system to obtain a number of signal clusters, the following steps are specifically adopted: Get the Euclidean distance between any data coordinate point and other data coordinate points; Using the ISODATA clustering algorithm, determining the intra-class distance based on the obtained multiple Euclidean distances; The data coordinate points in the three-dimensional coordinate system are integrated into several signal clusters according to the intra-cluster distance.

4. The method for comprehensive monitoring of turnout working conditions based on feature extraction according to claim 3 is characterized in that: Using the ISODATA clustering algorithm, when determining the intra-class distance based on the obtained multiple Euclidean distances, a metric loss function is used to determine the intra-class distance; The metric loss function is specifically as follows: ， ； in, For the The total loss value of the iteration is is the number of clusters in the current iteration, For the The number of high-frequency vibration signals in clusters, For the The first A vector of samples, For the The mean of the sample vectors in the clusters, For the the centroid of the current iteration of the class, is the intra-class dispersion threshold, is the regularization coefficient, For the The number of samples of the class, For the The sample size weight of each class.

5. The method for comprehensive monitoring of turnout working conditions based on feature extraction according to claim 1 is characterized in that: When performing denoising, data conversion and data set division on the aligned and fused multi-source feature data, the following steps are specifically included: Select Daubechies wavelet to perform wavelet decomposition on the multi-source feature data to obtain detail coefficients at multiple frequencies, perform soft threshold processing on the detail coefficients to remove noise, and then perform wavelet reconstruction on the denoised detail coefficients; The multi-source data features after wavelet reconstruction are processed by a standardized processing method, and the multi-source data features are scaled to the interval [0, 1]; The multi-source data features after interval division are divided into data sets according to the set ratio.

6. A comprehensive monitoring system for turnout working conditions based on feature extraction, characterized in that: include: Lightweight LSTM network, multi-source feature data acquisition module, data processing module; The multi-source characteristic data acquisition module is used to collect high-frequency vibration signals and contact resistance data of the target monitoring part of the turnout; The lightweight LSTM network is used to process time series data and capture the temporal dependencies in multi-source feature data; The data processing module is used to obtain the comparison result between the correlation matrix and the preset fault template library, and output the correlation data whose similarity exceeds the preset value as a warning signal; It also includes data alignment and fusion module, data set partitioning module, rank correlation coefficient acquisition module, and matrix construction module; The data alignment and fusion module is used to align and fuse the multi-source feature data based on the acquisition time; The data set division module is used to perform denoising, data conversion and data set division on the aligned and fused multi-source feature data; The rank correlation coefficient acquisition module is used to calculate the rank correlation coefficient based on the divided data set; The matrix construction module is used to construct a correlation matrix based on the rank correlation coefficient using a graph theory method; When aligning and fusing the multi-source feature data based on the acquisition time, the following steps are specifically included: A three-dimensional coordinate system is constructed, in which the three dimensions represent the instantaneous frequency, energy concentration and frequency change rate of the high-frequency vibration signal respectively; Perform cluster analysis on the data coordinate points in the three-dimensional coordinate system to form several signal clusters, and remove data points far away from these signal clusters based on the K-means clustering algorithm; Adopt IEEE 1588 precision clock protocol to align the high-frequency vibration signal and contact resistance data; The lightweight LSTM network uses a stacked multi-layer LSTM network and a fully connected layer. The first layer of LSTM network is used to obtain the time series features when collecting the multi-source feature data; The second layer of LSTM network obtains further abstract time-dependent features based on the time series features; The fully connected layer is used to convert the abstract time output sequence output by the second layer LSTM network into a fixed-length vector.

Images