Pantograph arcing anomaly detection and early warning method based on fusion of ultraviolet light and infrared imaging

Abstract

The invention relates to the technical field of anomaly detection, in particular to a pantograph arcing anomaly detection and early warning method based on fusion of ultraviolet light and infrared imaging, and the method comprises the steps: collecting an ultraviolet spectrum signal of a pantograph region in real time, and extracting arcing feature data; acquiring infrared temperature distribution images of the pantograph and the overhead line system, and extracting temperature abnormal region data; timestamp synchronization and space coordinate mapping are carried out on the arcing characteristic data and the temperature anomaly region data, and a fusion data set with time-space alignment is generated; on the basis of the fused data set, calculating the space-time similarity between the arcing feature data and the temperature anomaly region data, and outputting an arcing-temperature correlation index; constructing a Gaussian mixture model, and dynamically judging an abnormal level; and when the abnormal grade exceeds a preset threshold value, a grading early warning signal is triggered, and a fault positioning report including an arcing position and a temperature abnormal range is generated. The method has high engineering practicability, and has important significance for guaranteeing safe and stable operation of a rail transit power supply system.

Description

Technical Field

[0001] The present invention relates to the technical field of anomaly detection, and particularly to a method for detecting and warning pantograph arc ignition anomalies by fusing ultraviolet light and infrared imaging. Background Art

[0002] In the rail transit power supply system, the contact state between the pantograph and the catenary directly affects the power continuity of train operation and the safety of equipment. Affected by factors such as complex operating environments, current fluctuations, and contact pressure changes, the pantograph is prone to arc discharge during operation, forming a high-temperature arc ignition area and accompanied by local thermal anomalies. Such arc ignition events not only accelerate the wear and ablation of the contact surface, but may also induce safety accidents such as power supply interruptions and catenary damage. Therefore, carrying out accurate detection and warning of arc ignition anomalies has become a key link in ensuring the reliability of the electrical system.

[0003] In the existing pantograph fault detection technologies, arc ignition detection mostly relies on a single ultraviolet or current signal, which is vulnerable to environmental interference and difficult to quantify the arc ignition intensity; although infrared imaging can monitor temperature, it is not deeply correlated with arc ignition characteristics, resulting in a high false alarm rate. In addition, the existing methods lack a dynamic analysis of the coupling relationship between arc ignition and temperature anomalies and cannot predict potential faults. The innovation of the present invention lies in: through the spatio-temporal synchronous acquisition of ultraviolet and infrared data, combined with arc ignition feature extraction and temperature field modeling, a dynamic threshold judgment model is constructed to achieve the fusion decision-making of multi-modal data, and solve the problems of inaccurate arc ignition detection and late warning in the existing technologies. Summary of the Invention

[0004] The present invention provides a method for detecting and warning pantograph arc ignition anomalies by fusing ultraviolet light and infrared imaging.

[0005] A method for detecting and warning pantograph arc ignition anomalies by fusing ultraviolet light and infrared imaging includes the following linearly progressive steps:

[0006] S1: Real-time collect the ultraviolet spectral signals in the pantograph area through an ultraviolet sensor array, and extract arc ignition feature data, where the arc ignition feature data includes arc ignition duration, spectral peak value, and frequency domain energy ratio;

[0007] S2: Use an infrared thermal imager to obtain the infrared temperature distribution images of the pantograph and the catenary, and extract the temperature anomaly area data based on edge detection and morphological closing operation to exclude environmental heat source interference;

[0008] S3: Synchronize the time stamps and map the spatial coordinates of the arc ignition feature data in S1 and the temperature anomaly area data in S2 to generate a spatio-temporally aligned fusion data set;

[0009] S4: Based on the fused dataset, calculate the spatio-temporal similarity between the arcing feature data and the temperature anomaly region data through weighted Euclidean distance, and output the arcing-temperature correlation index;

[0010] S5: Construct a Gaussian mixture model according to the historical fault case library, and dynamically judge the anomaly level by inputting the arcing-temperature correlation index, temperature gradient and historical fault probability;

[0011] S6: When the anomaly level exceeds the preset threshold, trigger a hierarchical warning signal and generate a fault location report including the arcing position and the temperature anomaly range.

[0012] Optionally, the S1 includes:

[0013] S11, Collect the ultraviolet spectral signals in the pantograph area through the ultraviolet sensor arrays set on the top and both sides of the pantograph. The wavelength range covered by the ultraviolet sensor arrays is from 200nm to 400nm, and the sampling frequency is not less than 100Hz;

[0014] S12, Preprocess the ultraviolet spectral signals, including background noise filtering and outlier removal;

[0015] S13, Identify the start and end time periods of arcing based on the dynamic threshold of the background ultraviolet intensity, and calculate the time length continuously exceeding this threshold as the arcing duration. The dynamic threshold is jointly determined by the background mean and the weighted standard deviation;

[0016] S14, Spectral peak and frequency domain feature extraction: During the arcing duration, extract the maximum value of the ultraviolet intensity as the spectral peak;

[0017] S15, Calculate the frequency domain energy ratio: Perform a fast Fourier transform on the ultraviolet signal during the arcing period, and calculate the proportion of the high-frequency energy in the total energy.

[0018] Optionally, the S2 includes:

[0019] S21, Infrared image acquisition: Continuously obtain the temperature distribution images of the pantograph and the catenary area through the infrared thermal imager arranged on the working path of the pantograph;

[0020] S22, Extract the temperature anomaly region: Based on the temperature image obtained by the infrared thermal imager, use the Canny edge detection algorithm to calculate the temperature gradient map, extract the boundary information of the hot area, and perform morphological closing operation on the edge image to obtain a temperature anomaly region with clear contours;

[0021] S23, Exclude environmental heat source interference: Combine the static reference background temperature model to exclude the interference of constant heat sources, and only retain the abnormal temperature rise regions related to motion or mutations.

[0022] Optionally, the S22 includes:

[0023] S221, High-temperature edge extraction: Based on the temperature gradient of the infrared image, apply the Canny edge detection algorithm to extract the boundary contour of the temperature mutation region and identify potential high-temperature abnormal regions;

[0024] S222, Morphological closing operation processing: Perform morphological closing operation on the extracted edge image to fill the region holes and eliminate the noise breakpoints.

[0025] Optionally, the S23 includes:

[0026] S231, Construct background temperature model: Collect the infrared thermal image data sequence in the static state of the pantograph or the non-arc-running state for a period of time;

[0027] Calculate the statistical features of the temperature distribution of each pixel point, including the mean value and the standard deviation;

[0028] S232, Screen out constant heat sources: Compare the dynamic temperature map with the background model, perform differential analysis on the infrared image frame during operation and the background temperature mean value map, and define the dynamic temperature rise mask.

[0029] Optionally, the S3 includes:

[0030] S31, Timestamp synchronization: Align the arc ignition feature data collected by the ultraviolet sensor with the temperature image frame sequence collected by the infrared thermal imager, and use linear interpolation to complete the asynchronous signal alignment;

[0031] S32, Spatial coordinate mapping: Based on the relatively fixed installation geometric relationship between the ultraviolet and infrared sensors, construct a spatial mapping function to map the ultraviolet feature point coordinates to the infrared image coordinate system;

[0032] S33, Fusion dataset generation: Merge the arc ignition features after time synchronization and the temperature anomaly data under the mapped coordinates to form a fusion dataset under the unified spatio-temporal reference system.

[0033] Optionally, the S4 includes:

[0034] S41, Feature vector construction: Based on the corresponding time and space positions in the spatio-temporal aligned fusion dataset, extract the arc ignition feature vector and the infrared temperature feature vector respectively;

[0035] S42, Spatio-temporal similarity calculation and correlation index output: Calculate the similarity between the normalized ultraviolet and infrared feature vectors using the weighted Euclidean distance, and convert the similarity distance into an "arc ignition - temperature correlation index" through an inverse function.

[0036] Optionally, the S42 includes:

[0037] S421, Normalization: Use Min-Max normalization to map each feature to the interval;

[0038] S422, Weighted Euclidean distance calculation: Use the weighted Euclidean distance after normalization to measure the difference between the arc ignition feature and the temperature feature;

[0039] S423, Correlation index calculation: Reflect the spatio-temporal similarity as the arc ignition-temperature correlation index.

[0040] Optionally, the S5 includes:

[0041] S51, Gaussian mixture model training: Based on the historical pantograph arc ignition fault case library, construct a Gaussian mixture model with the arc ignition-temperature correlation index, temperature gradient, and historical fault probability as the feature inputs;

[0042] S52, Abnormal level dynamic determination: Input the three-dimensional features at the current monitoring moment into the trained Gaussian mixture model, and comprehensively output the abnormal level at this moment by calculating the posterior probabilities of each Gaussian component;

[0043] According to the risk level label of the belonging Gaussian component, output the abnormal level .

[0044] Optionally, the S6 includes:

[0045] S61, Abnormal level judgment and classification warning trigger: Based on the abnormal level output by the Gaussian mixture model, trigger the corresponding classification warning signals respectively;

[0046] S62, Fault location report generation: According to the frame corresponding to the abnormal level, extract the arc ignition spatial position in the corresponding ultraviolet image and the boundary box of the temperature abnormal area in the infrared image , and automatically generate a location report .

[0047] Advantages of the present invention:

[0048] By integrating ultraviolet and infrared dual-modal perception means, the present invention can extract key features from multi-dimensional data such as arc ignition duration, spectral features, and thermal abnormal areas, construct a unified spatio-temporal alignment data set, significantly improve the recognition accuracy and response speed of arc ignition abnormal events, and compared with traditional detection methods based on a single signal or image source, the present invention has stronger anti-interference ability and fault location ability in complex environments, and is especially suitable for real-time monitoring under high-speed operation conditions.

[0049] The present invention further introduces the modeling of the arc ignition - temperature correlation index, and constructs a Gaussian mixture model in combination with historical fault data to achieve intelligent determination of the abnormal level and hierarchical early warning. When the abnormal level exceeds the threshold, the system can automatically generate a fault location report including the arc ignition position and the abnormal temperature range, which has high engineering practicability and popularization value and is of great significance for ensuring the safe and stable operation of the rail transit power supply system. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only those of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative efforts.

[0051] Figure 1 is the flowchart of the method according to the embodiment of the present invention;

[0052] Figure 2 is the data acquisition flowchart according to the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0053] The present invention will be described in detail below in combination with the drawings and specific embodiments. At the same time, it should be noted here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments. For some well - known technologies, those skilled in the art can also adopt other alternative methods for implementation; moreover, the drawings are only for more specifically describing the embodiments and are not intended to specifically limit the present invention.

[0054] It should be pointed out that in the specification, when referring to "an embodiment", "embodiment", "exemplary embodiment", "some embodiments", etc., it indicates that the described embodiment may include specific features, structures or characteristics, but not necessarily every embodiment includes such specific features, structures or characteristics. Additionally, when combining an embodiment to describe a specific feature, structure or characteristic, implementing such a feature, structure or characteristic in combination with other embodiments (whether explicitly described or not) should be within the knowledge of those skilled in the relevant art.

[0055] Generally, the terms can be understood at least in part from their use in the context. For example, at least in part depending on the context, the term "one or more" used herein can be used to describe any feature, structure or characteristic in a singular sense, or can be used to describe a combination of features, structures or characteristics in a plural sense. Additionally, the term "based on" can be understood as not necessarily aiming to convey a set of exclusive factors, but rather, at least in part depending on the context, allowing for the existence of other factors that may not be explicitly described.

[0056] Such as Figure 1 - Figure 2As shown in the figure, a method for detecting and warning abnormal arcing of a pantograph by fusing ultraviolet light and infrared imaging includes the following linearly progressive steps:

[0057] S1: Real-time collect the ultraviolet spectral signals in the pantograph area through an ultraviolet sensor array, and extract the arcing characteristic data, where the arcing characteristic data includes the arcing duration, spectral peak value, and frequency-domain energy ratio;

[0058] S2: Use an infrared thermal imager to obtain the infrared temperature distribution images of the pantograph and the catenary, and extract the temperature anomaly area data based on edge detection and morphological closing operation to exclude environmental heat source interference;

[0059] S3: Synchronize the time stamps and map the spatial coordinates of the arcing characteristic data in S1 and the temperature anomaly area data in S2 to generate a spatio-temporally aligned fusion data set;

[0060] S4: Based on the fusion data set, calculate the spatio-temporal similarity between the arcing characteristic data and the temperature anomaly area data through weighted Euclidean distance, and output the arcing-temperature correlation index;

[0061] S5: Construct a Gaussian mixture model according to the historical fault case base, and dynamically judge the abnormal level by inputting the arcing-temperature correlation index, temperature gradient, and historical fault probability;

[0062] S6: When the abnormal level exceeds the preset threshold, trigger a hierarchical warning signal and generate a fault location report including the arcing position and the temperature anomaly range.

[0063] S1 includes:

[0064] S11: Collect the ultraviolet spectral signals in the pantograph area through the ultraviolet sensor array arranged on the top and both sides of the pantograph. The ultraviolet sensor array covers a wavelength range of 200 nm to 400 nm, and the sampling frequency is not less than 100 Hz;

[0065] S12: Preprocess the ultraviolet spectral signals, including background noise filtering (using the sliding window average method) and outlier removal. The filtered signal is expressed as:

[0066] is the odd window length,

[0067] where, is the original ultraviolet intensity sequence, is the index offset within the local window, is the ultraviolet intensity value after filtering;

[0068] S13. Identify the start and end time periods of arcing based on the dynamic threshold of the background ultraviolet intensity, calculate the time length of continuous exceeding this threshold as the arcing duration, and the dynamic threshold is jointly determined by the background mean and the weighted standard deviation, expressed as:

[0069] ;

[0070] Among them, is the dynamic threshold, is the mean value of the ultraviolet intensity in the background stage, is the background standard deviation, is the coefficient, empirically set to , which is used to enhance the anomaly discrimination;

[0071] S14. Spectral peak and frequency-domain feature extraction: During the arcing duration, extract the maximum value of the ultraviolet intensity as the spectral peak, expressed as:

[0072] ;

[0073] Among them, is the time period when arcing occurs, is the starting time point of arcing, that is, the time when the ultraviolet signal first exceeds the arcing determination threshold, is the end time point of arcing, that is, the time when the ultraviolet signal is higher than the threshold for the last time, is the ultraviolet wavelength;

[0074] S15. Calculation of the frequency-domain energy ratio: Perform a fast Fourier transform (FFT) on the ultraviolet signal during the arcing period, and calculate the ratio of the high-frequency energy in the total energy, expressed as:

[0075] ;

[0076] Among them, is the frequency component, is the spectral amplitude, is the high-frequency starting threshold, empirically set based on the typical arcing frequency (above 25 Hz).

[0077] S2 includes:

[0078] S21. Infrared image acquisition: Continuously obtain the temperature distribution image of the pantograph and the catenary area through an infrared thermal imager deployed on the working path of the pantograph , represents the spatial pixel coordinates in the image, with the unit of temperature ;

[0079] S22, Temperature anomaly region extraction: Based on the temperature image obtained by the infrared thermal imager, the Canny edge detection algorithm is used to calculate the temperature gradient map, extract the boundary information of the hot region, and perform morphological closing operation on the edge image to obtain a temperature anomaly region with clear contours;

[0080] S23, Excluding environmental heat source interference: Combining with the static reference background temperature model, exclude the interference of constant heat sources, and only retain the abnormal temperature rise regions related to movement or mutations.

[0081] S22 includes:

[0082] S221, High-temperature edge extraction: Based on the temperature gradient of the infrared image, the Canny edge detection algorithm is used to extract the boundary contour of the temperature mutation region and identify potential high-temperature anomaly regions, expressed as:

[0083] ;

[0084] Among them, is the edge intensity of the temperature gradient, respectively represent the gradients of the temperature map in the horizontal and vertical directions, is the edge intensity of the corresponding pixel point, and the larger the value, the more obvious the edge;

[0085] S222, Morphological closing operation processing: Perform morphological closing operation (dilation followed by erosion) on the extracted edge image to fill the region holes and eliminate the noise breakpoints, expressed as:

[0086] ;

[0087] Among them, represents the dilation operation, represents the erosion operation, is the structure element (a circular kernel with a radius of 3 is selected), is the high-temperature region mask obtained after the closing operation.

[0088] S23 includes:

[0089] S231, Constructing the background temperature model (static modeling): Collect the infrared thermal image data sequence in the state where the pantograph is stationary or there is no arcing operation for a period of time, expressed as where represents the frame image, and the temperature value at the pixel ;

[0090] Calculate the statistical characteristics of the temperature distribution of each pixel point, including the mean and the standard deviation , expressed as:

[0091] ;

[0092]

[0093] S232, Screening out constant heat sources: Comparing the dynamic temperature map with the background model, and performing differential analysis on the infrared image frames during operation and the background temperature mean map, and defining a dynamic temperature rise mask, expressed as:

[0094] ;

[0095] ;

[0096] Among them, is the anomaly threshold, , where is the empirical coefficient (value range 2 - 3), is the difference between the real-time temperature and the background temperature, is the binary mask of the temperature rise area, 1 means reserved, 0 means excluded.

[0097] S3 includes:

[0098] S31, Timestamp synchronization: Aligning the arc ignition feature data collected by the ultraviolet sensor with the sequence of temperature image frames collected by the infrared thermal imager, and completing the asynchronous signal alignment using linear interpolation, expressed as:

[0099] ;

[0100] Among them, is the temperature value of the abnormal area in the infrared image, is the adjacent sampling moment of the infrared image, is the time point of the ultraviolet arc ignition event, is the infrared temperature value obtained by interpolation for aligning with the arc ignition feature, is at the infrared temperature image data at the moment, is at the infrared temperature image data at the moment;

[0101] S32, Spatial coordinate mapping: Based on the relatively fixed installation geometric relationship between the ultraviolet and infrared sensors, constructing a spatial mapping function , and mapping the ultraviolet feature point coordinates to the infrared image coordinate system, expressed as:

[0102] ;

[0103] Among them, is the pixel coordinate where the arc ignition occurs in the ultraviolet image, is the coordinate in the infrared image, is the 2D homography matrix obtained by camera calibration, is the mapping function for image registration;

[0104] S33, Fusion dataset generation: The arc ignition characteristics after time synchronization and the temperature anomaly data under the mapped coordinates are merged to form a fusion dataset in a unified spatio-temporal reference system, denoted as:

[0105] ;

[0106] where, is the fusion dataset with spatio-temporal alignment, is the ultraviolet arc ignition feature vector (including duration, spectral peak, frequency domain energy ratio).

[0107] S4 includes:

[0108] S41, Feature vector construction: Based on the corresponding time and space positions in the fusion dataset with spatio-temporal alignment, the arc ignition feature vector (including arc ignition intensity, duration, frequency domain energy ratio) and the infrared temperature feature vector (including maximum temperature, temperature gradient, temperature variance) are extracted respectively, denoted as:

[0109] ;

[0110] where, is the ultraviolet arc ignition peak intensity, is the arc ignition duration, is the high-frequency energy ratio of the arc ignition frequency domain, is the highest temperature in the temperature anomaly region, is the temperature gradient (edge slope), is the variance within the temperature region, is the ultraviolet feature vector (arc ignition feature), is the infrared feature vector (temperature feature);

[0111] S42, Spatio-temporal similarity calculation and correlation index output: The weighted Euclidean distance is used to calculate the similarity between the normalized ultraviolet and infrared feature vectors, and the similarity distance is converted into an "arc ignition - temperature correlation index" through an inverse function.

[0112] S42 includes:

[0113] S421, Normalization processing: Use Min - Max normalization to map each feature to the interval, denoted as:

[0114] ;

[0115] ;

[0116] ;

[0117] Among them, is the original eigenvalue, is the feature the minimum and maximum values in all samples, is the normalized eigenvalue, is the normalized peak intensity of ultraviolet arcing, is the normalized arcing duration, is the proportion of high-frequency energy in the normalized arcing frequency domain, is the highest temperature in the normalized temperature anomaly region, is the normalized temperature gradient (edge slope), is the variance within the normalized temperature region, is the normalized ultraviolet feature vector (arcing feature), is the normalized infrared feature vector (temperature feature);

[0118] S422, Weighted Euclidean distance calculation: The weighted Euclidean distance after normalization is used to measure the difference degree between the arcing feature and the temperature feature, expressed as:

[0119] ;

[0120] Among them, is the weighted Euclidean distance, which is the similarity between the ultraviolet and infrared feature vectors, is the weighting coefficient of the ;

[0121] S423, Correlation index calculation: The spatio-temporal similarity is reflected as the arcing-temperature correlation index , expressed as:

[0122] ;

[0123] Among them, is the arcing-temperature correlation index, , the larger the value, the higher the matching degree between the arcing and the temperature anomaly.

[0124] S5 includes:

[0125] S51, Gaussian mixture model training: Based on the historical pantograph arcing fault case library, a Gaussian mixture model with the arcing-temperature correlation index, temperature gradient, and historical fault probability as feature inputs is constructed, expressed as:

[0126] ;

[0127] ;

[0128] in, For the The arc-temperature correlation index of the samples, is the normalized temperature gradient, For the The historical failure probability corresponding to each sample (obtained from historical annotation statistics), is the number of mixed Gaussian distributions (i.e., the number of clusters), which is generally 2-5. is a multidimensional Gaussian probability density function, For the The covariance matrix of the Gaussian distribution describes the correlation of features in each dimension;

[0129] S52, dynamic determination of abnormal level: input the three-dimensional features of the current monitoring moment into the trained Gaussian mixture model, calculate the posterior probability of each Gaussian component, and comprehensively output the abnormal level of the moment, expressed as:

[0130] ;

[0131] in, is the new input feature vector, , is the corresponding posterior probability in the GMM model;

[0132] According to the Gaussian component Danger level label, output abnormal level , expressed as:

[0133] .

[0134] S6 includes:

[0135] S61, abnormal level judgment and graded warning triggering: Based on the abnormal level output by the Gaussian mixture model, the corresponding graded warning signal is triggered respectively , expressed as:

[0136] ;

[0137] S62, fault location report generation: according to the frame corresponding to the abnormal level , extract the arcing spatial position in the corresponding UV image Bounding box of temperature anomaly area in infrared image , automatically generate positioning reports , expressed as:

[0138] 。

[0139] The present invention encompasses any alternatives, modifications, equivalent methods, and schemes made within the spirit and scope of the present invention. To enable the public to have a thorough understanding of the present invention, specific details are set forth in the following preferred embodiments of the present invention. However, those skilled in the art can fully understand the present invention even without the description of these details. Additionally, well-known methods, processes, procedures, components, and circuits are not described in detail to avoid unnecessary confusion to the essence of the present invention.

[0140] The above description is only a preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as within the protection scope of the present invention.

Claims

1. A method for detecting and warning of abnormal arcing of a pantograph by fusing ultraviolet light and infrared imaging, characterized in that, Including the following linearly progressive steps: S1: Real-time collect the ultraviolet spectral signals in the pantograph area through an ultraviolet sensor array, and extract the arc ignition characteristic data, where the arc ignition characteristic data includes the arc ignition duration, spectral peak value, and frequency domain energy ratio; S2: Use an infrared thermal imager to obtain the infrared temperature distribution images of the pantograph and the catenary, and extract the temperature anomaly area data based on edge detection and morphological closing operation to exclude environmental heat source interference; S3: Synchronize the time stamps and map the spatial coordinates of the arc ignition characteristic data in S1 and the temperature anomaly area data in S2 to generate a spatio-temporally aligned fusion data set; S4: Based on the fusion data set, calculate the spatio-temporal similarity between the arc ignition characteristic data and the temperature anomaly area data through weighted Euclidean distance, and output the arc ignition-temperature correlation index; S5: Construct a Gaussian mixture model according to the historical fault case library, and dynamically judge the anomaly level by inputting the arc ignition-temperature correlation index, temperature gradient, and historical fault probability; S6: When the anomaly level exceeds the preset threshold, trigger a hierarchical warning signal and generate a fault location report including the arc ignition position and the temperature anomaly range.

2. The method for detecting and warning of abnormal arcing of a pantograph by using the fusion of ultraviolet light and infrared imaging according to claim 1, wherein, The S1 includes: S11, Collect the ultraviolet spectral signals in the pantograph area through the ultraviolet sensor arrays arranged on the top and both sides of the pantograph. The wavelength range covered by the ultraviolet sensor arrays is 200nm to 400nm, and the sampling frequency is not less than 100Hz; S12, Preprocess the ultraviolet spectral signals, including background noise filtering and outlier removal; S13, Identify the start and end time periods of arc ignition based on the dynamic threshold of the background ultraviolet intensity, and calculate the time length continuously exceeding this threshold as the arc ignition duration. The dynamic threshold is jointly determined by the background mean and weighted standard deviation; S14, Spectral peak and frequency domain feature extraction: During the arc ignition duration, extract the maximum value of the ultraviolet intensity as the spectral peak value; S15, Frequency domain energy ratio calculation: Perform a fast Fourier transform on the ultraviolet signals during the arc ignition period, and calculate the proportion of the high-frequency energy in the total energy.

3. A method for detecting and warning abnormal arcing of a pantograph by fusing ultraviolet light and infrared imaging according to claim 2, characterized in that, The S2 includes: S21, Infrared image acquisition: Continuously obtain the temperature distribution images of the pantograph and the catenary area through the infrared thermal imager arranged on the working path of the pantograph; S22, Temperature anomaly area extraction: Based on the temperature image obtained by the infrared thermal imager, use the Canny edge detection algorithm to calculate the temperature gradient map, extract the boundary information of the hot area, and perform a morphological closing operation on the edge image to obtain a temperature anomaly area with clear contours; S23, Environmental heat source interference exclusion: Combine the static reference background temperature model to exclude the interference of constant heat sources, and only retain the abnormal temperature rise areas related to motion or mutations.

4. A method for detecting and warning abnormal arcing of a pantograph by fusing ultraviolet light and infrared imaging according to claim 3, characterized in that, The S22 includes: S221, Temperature edge extraction: Based on the temperature gradient of the infrared image, use the Canny edge detection algorithm to extract the boundary contour of the temperature mutation area and identify the temperature anomaly area; S222, Morphological closing operation processing: Perform a morphological closing operation on the extracted edge image to fill the area holes and eliminate the noise breakpoints.

5. The method for detecting and warning of abnormal arcing of a pantograph by fusing ultraviolet light and infrared imaging according to claim 3, characterized in that, The S23 includes: S231. Build a background temperature model: Collect a sequence of infrared thermal image data during a period when the pantograph is stationary or operating without arcing, and calculate the statistical characteristics of the temperature distribution of each pixel point, including the mean and standard deviation. S232. Screen out constant heat sources: Compare the dynamic temperature map with the background model, perform differential analysis on the infrared image frames during operation and the background temperature mean map, and define a dynamic temperature rise mask.

6. A method for detecting and warning of abnormal arcing of a pantograph by fusing ultraviolet light and infrared imaging according to claim 3, characterized in that, The said S3 includes: S31. Timestamp synchronization: Align the arcing feature data collected by the ultraviolet sensor with the sequence of temperature image frames collected by the infrared thermal imager in terms of time, and use linear interpolation to complete the alignment of asynchronous signals. S32. Spatial coordinate mapping: Based on the relatively fixed installation geometric relationship between the ultraviolet and infrared sensors, construct a spatial mapping function to map the coordinates of the ultraviolet feature points to the infrared image coordinate system. S33. Generation of a fused data set: Merge the arcing features after time synchronization and the temperature anomaly data under the mapped coordinates to form a fused data set in a unified spatio-temporal reference system.

7. The method for detecting and warning of abnormal arcing of a pantograph by using the fusion of ultraviolet light and infrared imaging according to claim 6, characterized in that, The said S4 includes: S41. Construction of feature vectors: Based on the corresponding time and space positions in the spatio-temporally aligned fused data set, extract the arcing feature vector and the infrared temperature feature vector respectively. S42. Calculation of spatio-temporal similarity and output of the correlation index: Use the weighted Euclidean distance to calculate the similarity between the normalized ultraviolet and infrared feature vectors, and convert the similarity distance into an "arcing-temperature correlation index" through an inverse function.

8. A method for detecting and warning of abnormal arcing of a pantograph by using the fusion of ultraviolet light and infrared imaging according to claim 7, characterized in that, The said S42 includes: S421, Normalization processing: Use Min-Max normalization to map each feature to the interval; S422. Calculation of the weighted Euclidean distance: Use the normalized weighted Euclidean distance to measure the degree of difference between the arcing feature and the temperature feature. S423. Calculation of the correlation index: Reflect the spatio-temporal similarity as the arcing-temperature correlation index.

9. The method for detecting and warning of abnormal arcing of a pantograph by fusing ultraviolet light and infrared imaging according to claim 8, characterized in that, The said S5 includes: S51. Gaussian mixture model training: Based on the historical pantograph arcing fault case library, construct a Gaussian mixture model with the arcing-temperature correlation index, temperature gradient, and historical fault probability as the feature inputs. S52. Dynamic determination of the anomaly level: Input the three-dimensional features at the current monitoring moment into the trained Gaussian mixture model, and comprehensively output the anomaly level at this moment by calculating the posterior probabilities of each Gaussian component. Output the abnormality level according to the risk level label of the Gaussian component to which it belongs .​ 10. A method for detecting and warning of abnormal arcing of a pantograph by using the fusion of ultraviolet light and infrared imaging according to claim 9, characterized in that, The said S6 includes: S61. Anomaly level judgment and triggering of hierarchical early warning: Based on the anomaly level output by the Gaussian mixture model, trigger the corresponding hierarchical early warning signals respectively. S62, Fault Location Report Generation: According to the frame corresponding to the abnormal level , extract the arc ignition spatial position in the corresponding ultraviolet image and the boundary box of the temperature anomaly area in the infrared image , and automatically generate a location report .

Images