Method and device for reducing number of steel rail insulation defect diagnosis training samples and storage medium

Abstract

The invention discloses a method for reducing the number of steel rail insulation defect diagnosis training samples, and the method comprises the steps: collecting the potential data of a station steel rail, carrying out the data preprocessing of the potential data, and obtaining the preprocessed potential data as an initial training sample; key features of the steel rail potential data are extracted from the preprocessed potential data; performing interval division on the key features to obtain a plurality of intervals; the similarity of the key features of the adjacent intervals is analyzed according to a similarity analysis algorithm, the adjacent intervals with the similar key features are combined into a combined interval, and samples in the combined interval serve as first training samples; forming a plurality of staggered overlapped sub-intervals in the merging interval according to a staggered division mode, and taking samples in the staggered overlapped sub-intervals as second training samples; and taking the first training sample and the second training sample as a training set. According to the method, the sample complexity is reduced, and meanwhile, the model diagnosis precision and generalization ability are guaranteed.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method, device, and storage medium for reducing the number of training samples for rail insulation defect diagnosis. Background Art

[0002] The vast majority of urban rail transit lines utilize a floating rail return DC traction power supply system. The rails serve as return conductors and are insulated and mounted on the roadbed using fasteners. In the early days of urban rail transit, rail insulation to the ground was excellent. However, with years of operation, contamination of the rail fasteners by tunnel water seepage, moisture, concrete, dust, oil, and metal debris gradually degrades the rail insulation, leading to stray current leakage. With the increase in urban rail transit operating mileage, increased network density, and expanded coverage, stray currents are causing significant problems, such as DC interference with surrounding power systems and damage to urban pipelines. These problems pose a serious threat to the safe operation of power systems and oil and gas pipelines.

[0003] Rail insulation defect diagnosis is a prerequisite for stray current assessment and control. Currently, rail-to-ground insulation defect diagnosis in continuous rail potential monitoring systems relies primarily on empirical experience and manual screening of monitoring data features, which can easily lead to missed or misjudgment of rail insulation defects. A knowledge- and data-driven approach can effectively address this issue. However, when using knowledge models to diagnose rail potential data, the number of station rail potential samples required for offline training of the network increases exponentially with the number of station intervals. Training models for long lines with multiple stations is prohibitively computationally demanding and time-consuming. Summary of the Invention

[0004] In response to the above situation, the embodiments of the present application propose a method, device and storage medium for reducing the number of training samples for rail insulation defect diagnosis, so as to overcome or at least partially overcome the shortcomings of the existing technology.

[0005] The method disclosed in the present application for reducing the number of training samples for rail insulation defect diagnosis comprises: Collecting rail potential data at the station, and performing data preprocessing on the potential data to obtain preprocessed potential data as initial training samples; Extract key features of rail potential data from preprocessed potential data; Divide the key features into intervals to obtain multiple intervals; Analyze the similarity of key features of adjacent intervals using a similarity analysis algorithm, merge adjacent intervals with similar key features, and use the samples in the merged intervals as the first training samples; forming a plurality of staggered overlapping subintervals in the merged interval according to a staggered partitioning method, wherein the plurality of staggered overlapping subintervals cover all key features, and using samples in the staggered overlapping subintervals as second training samples; The first training sample and the second training sample are used as training sets.

[0006] In one embodiment, a rail insulation defect diagnosis model is constructed, and a training set is input into the rail insulation defect diagnosis model for training. The diagnosis model is a hybrid model of random forest and convolutional neural network.

[0007] In one embodiment, a random forest algorithm model is used to process samples within the merged interval, and global feature classification is performed on the samples within the merged interval. A convolutional neural network model is used to extract local spatial features of samples within the misaligned and overlapping sub-intervals. The obtained global feature classification and local spatial features are fused through a weighted voting mechanism to obtain a diagnosis result.

[0008] In one embodiment, data enhancement and sorting are performed on the second training samples to enhance the diversity of the training samples.

[0009] In one embodiment, the similarity analysis algorithm includes a clustering algorithm or a similarity measurement method.

[0010] In one embodiment, the staggered division method specifically includes: dividing according to different starting points and offsets to form multiple staggered overlapping sub-intervals.

[0011] In one embodiment, dividing the intervals according to different starting points and offsets to form multiple offset-overlapping subintervals further includes: Set the offset to 1-2 unmerged intervals; The starting point of the staggered overlapping subintervals is centered on the boundary of the merged interval and expands symmetrically to the left and right sides; Each merged interval generates at least three non-identical offset overlapping subintervals; The misaligned overlapping subintervals cover the potential characteristic mutation region within the range of ±Δ of the merged interval boundary, where Δ is the offset.

[0012] In one embodiment, the clustering algorithm is a K-means algorithm. When analyzing the feature similarity and correlation of adjacent intervals based on the K-means algorithm, cosine similarity is used as the clustering distance metric, and the number of clusters is dynamically determined according to the elbow rule of the interval feature distribution to ensure that the merged intervals have significant consistency in feature distribution.

[0013] The present application also provides a computer-readable storage medium storing a computer program, which implements the above method when executed by a processor.

[0014] The present application also provides a computing device for reducing the number of training samples for rail insulation defect diagnosis, comprising: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor. The instructions are executed by the at least one processor to enable the at least one processor to perform the above method.

[0015] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects: The application effectively reduces the total number of intervals by merging adjacent similar intervals, thereby reducing the number of samples.

[0016] This application introduces an additional model for offset partitioning intervals and designs an offset partitioning method. Based on the original merged interval, the partitioning starting point and offset are changed to re-divide the subintervals with a certain offset. The subintervals are offset and overlapping subintervals. Although these subintervals are fewer in number than the unmerged intervals, their offset design allows them to cover all the original potential key feature points, effectively compensating for the information loss that may occur when merging larger intervals.

[0017] At the same time, these additionally generated sub-interval data are used as additional training samples and combined with the merged interval data through ensemble learning or hybrid model methods to jointly improve the accuracy and robustness of the diagnostic model. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 Flowchart of a method for effectively reducing the number of training samples for rail insulation defect diagnosis.

[0019] Figure 2 The logical block diagram of a method for reducing the number of training samples for rail insulation defect diagnosis.

[0020] Figure 3 Schematic diagram of interval misalignment partitioning in a method for reducing the number of training samples for rail insulation defect diagnosis. DETAILED DESCRIPTION

[0021] To make the purpose, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the specific embodiments of this application and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0022] The present invention aims to provide a method that effectively reduces the number of training samples for rail insulation defect diagnosis. By using interval merging and offset partitioning strategies, the method reduces the complexity of training samples while maintaining or even improving the accuracy and generalization of the diagnostic model. Traditional methods are computationally intensive and inefficient when processing large amounts of fine-grained interval data, and can also affect diagnostic effectiveness due to overfitting. This invention achieves efficient utilization of training samples and improved diagnostic accuracy by optimizing interval partitioning and incorporating an additional offset partitioning model.

[0023] like Figure 1-Figure 2 As shown, the first embodiment of the present application discloses a method for reducing the number of training samples for rail insulation defect diagnosis, the method comprising: S100 , collecting rail potential data at the station, and performing data preprocessing on the potential data to obtain preprocessed potential data as initial training samples.

[0024] Specifically, the data preprocessing of the potential data further includes: First, the station rail potential data is cleaned and preprocessed to remove noise and outliers to ensure data quality.

[0025] The data is then subjected to fast Fourier transform to filter out high-frequency noise and low-frequency drift.

[0026] S200, extracting key features of the rail potential data from the preprocessed potential data.

[0027] S300: Divide the key features into intervals to obtain multiple intervals.

[0028] S400 , analyzing similarities of key features of adjacent intervals according to a similarity analysis algorithm, merging adjacent intervals with similar key features, and obtaining samples of the merged intervals as first training samples.

[0029] Specifically: The similarity analysis algorithm includes a clustering algorithm (such as K-means, DBSCAN, etc.) or a similarity measurement method (such as cosine similarity, Pearson correlation coefficient, etc.). For example, when analyzing the feature similarity and correlation of adjacent intervals based on the K-means algorithm, cosine similarity is used as the cluster distance metric, and the number of clusters is dynamically determined based on the elbow rule of the interval feature distribution to ensure that the merged intervals have significant consistency in feature distribution.

[0030] By performing similarity analysis on adjacent or similar intervals, intervals with similar key features are identified, and adjacent intervals with similar features are merged into a larger interval to reduce the number of intervals, thereby reducing the number of samples and reducing data complexity.

[0031] This application performs data preprocessing such as filtering and fast Fourier transform on the station rail potential data, extracts key features, and merges adjacent similar intervals, effectively reducing the total number of intervals and thus the number of samples.

[0032] S500 : forming a plurality of staggered overlapping sub-intervals in a staggered partitioning manner within the merged interval, and obtaining samples within the staggered overlapping sub-intervals as second training samples.

[0033] Specifically, based on the merged intervals, new training samples are generated using a staggered partitioning method. The staggered partitioning method specifically includes partitioning based on different starting points and offsets to form multiple staggered overlapping subintervals. Each starting point and offset is a different offset method, and each offset method generates a submodel. Each seed model represents a series of staggered overlapping subintervals, ensuring that all generated submodels can cover the rail potential data of all stations.

[0034] like Figure 2 As shown, the specific steps of forming multiple staggered overlapping sub-intervals according to the staggered division method include: Set the starting point and offset; specifically: The offset is 1-2 unmerged intervals, and the starting point of the offset overlapping subinterval is centered on the merged interval boundary and extends symmetrically to the left and right sides; each merged interval generates at least 3 non-completely identical offset overlapping subintervals.

[0035] The misaligned overlapping subintervals cover the potential characteristic mutation region within the range of ±Δ of the merged interval boundary, where Δ is the offset.

[0036] In this application, each offset method generates a sub-model. Under the condition that "each merged interval generates at least 3 non-completely identical offset overlapping sub-intervals", all generated sub-models can cover the rail potential data of all stations. The minimum number of sub-models that all generated sub-models can cover the rail potential data of all stations is the number of sub-models that need to be determined.

[0037] The union of the inputs of all sub-models can cover the rail potential data of all stations, ensuring that all feature points are covered.

[0038] Based on the merged interval, this scheme re-divides multiple refined sub-intervals through a staggered partitioning strategy with predefined starting points and offsets to generate additional training samples. This staggered partitioning can increase the diversity of samples and help improve the generalization ability of the model.

[0039] S500 further includes: S510: Perform data enhancement and sorting on the samples in the sub-interval to enhance the diversity of training samples.

[0040] Specifically: Appropriate data enhancement processing is performed on the divided subintervals, such as adding noise, transforming eigenvalues, etc., to further increase the diversity of samples and prevent overfitting.

[0041] S600: Use samples in the merged interval and samples in the offset overlapping sub-intervals as training sets.

[0042] S700: Build a rail insulation defect diagnosis model, input a training set into the rail insulation defect diagnosis model for training, wherein the diagnosis model is a hybrid model of random forest and convolutional neural network (CNN).

[0043] Specifically: The random forest algorithm is used to process samples within the merged interval, and global feature classification is performed on the samples in the merged interval. The convolutional neural network (CNN) is used to extract the local spatial features of samples in the misaligned and overlapping sub-intervals. The outputs of the two (global feature classification, local spatial features) are fused through a weighted voting mechanism to obtain diagnostic results and achieve high-precision diagnosis.

[0044] To further illustrate the solution of the present application and its beneficial effects, the following examples are given to further illustrate: In the traditional rail potential insulation defect diagnosis sample generation process, the minimum diagnostic interval is generally the line between two traction substations. The insulation conditions of each interval are then traversed to generate all samples. Taking a line with m traction substations as an example, assuming that each interval has only two conditions: good rail insulation and defective rail insulation, the number of samples considering only the presence of insulation defects is: .

[0045] like Figure 3As shown in the figure, suppose a line has 11 stations numbered 1-11. After feature fusion and interval merging, an interval merging model is obtained. The intervals between stations 2, 3, and 4 with similar key features are merged into a larger interval A; the intervals between stations 6, 7, and 8 are merged into a larger interval B; and the intervals between stations 9, 10, and 11 are merged into a larger interval C.

[0046] The input for each model training is the rail potential data at the stations at both ends of the interval. Specifically, the input data for the interval-merging model is the rail potential data at stations 1, 2, 4, 5, 6, 8, 9, and 11. Therefore, the network cannot learn the potential features of intervals such as Station 2-3 and Station 3-4, affecting the accuracy of insulation defect diagnosis.

[0047] By using staggered partitioning, the interval information that was missed due to merging is revealed. The specific operation is as follows: for intervals A, B, and C, with stations 2, 6, and 9 as the offset starting points, the offset Δ is one unmerged interval, and the intervals are shifted to the left to obtain staggered overlapping sub-intervals A', B', and C', as shown in sub-model 1. Since interval C is at the end and cannot be shifted to the right, only intervals A and B are shifted to the right with stations 2 and 6 as the starting points, and the offset Δ is 1 unmerged interval, resulting in the offset overlapping sub-intervals A'' and B'', as shown in sub-model 2; For intervals B and C, with stations 6 and 9 as the offset starting points, the offset Δ is two unmerged intervals, and the offset is to the left, resulting in staggered overlapping sub-intervals B''' and C'''. For interval A, when the offset Δ is two unmerged intervals, it exceeds the first section of the line, so Δ is made less than 2 to obtain A''', as shown in sub-model 3. After the above division, each merged interval has three offset overlapping subintervals. For sub-model 1, the input data are the rail potential data at stations numbered 1, 3, 4, 5, 7, 8, 10, and 11; For sub-model 2, the input data is rail potential data at stations numbered 1, 2, 3, 5, 6, 7, 9, and 11; for sub-model 3, the input data is rail potential data at stations numbered 1, 3, 4, 6, 7, 9, 10, and 11. With this division, the three sub-models cover rail potential data for all stations, ensuring that the network learns the potential characteristics of rail potential data for all stations and their adjacent stations.

[0048] like Figure 3 As shown in Figure 2, when there are 11 traction substations in the line, the number of samples required by the traditional diagnosis method is 2^(11-1)=1024. Figure 3After the staggered partitioning, the number of samples required for the interval merging model is S1 = 2^(8-1) = 128; the number of samples required for sub-model 1 is S2 = 2^(8-1) = 128; the number of samples required for sub-model 2 is S3 = 2^(8-1) = 128; and the number of samples required for sub-model 3 is S4 = 2^(8-1) = 128. The total number of samples, S0 = S1 + S2 + S3 + S4 = 512, reducing the number of samples to 50% of the original method. This ratio can be further reduced when the number of traction substations on the line increases.

[0049] In order to further ensure that the accuracy of diagnosis does not decrease due to the reduction in the number of intervals, this application introduces an additional model for staggered partitioning of intervals and designs a staggered partitioning method, that is, on the basis of the original merged interval, by changing the starting point and offset of the division, the sub-intervals (staggered overlapping sub-intervals) are re-divided with a certain amount of staggering. Although the number of these sub-intervals is less than the intervals before the merger, due to their staggered design, they can cover all the original potential key feature points, effectively compensating for the information loss that may be caused by the merger of larger intervals. At the same time, these additionally generated sub-interval data are used as additional training samples, and are combined with the merged interval data through ensemble learning or hybrid model methods to jointly improve the accuracy and robustness of the diagnostic model.

[0050] The beneficial effects of the present invention are: 1. Improve sample generation efficiency: Through the intelligent interval merging strategy, the present invention significantly reduces the number of intervals in the training sample, thereby reducing the complexity of data processing and model training, and improving the overall training efficiency; 2. Maintaining diagnostic accuracy: Although the number of intervals is reduced, the present invention generates additional refined sub-intervals as training samples by introducing a staggered partitioning strategy, which effectively compensates for the information loss that may be caused by the merging of larger intervals.

[0051] 3. Enhanced model generalization ability: Since the training samples contain diversified sub-interval data obtained through staggered partitioning, these data can cover all potential key feature change points, thereby enhancing the generalization ability of the model.

[0052] The embodiment of the present application also provides a computer-readable storage medium, which stores one or more programs, wherein the one or more programs include instructions, which, when executed by an electronic device including multiple application programs, can enable the electronic device to execute Figure 1 The method for reducing the number of rail insulation defect diagnosis training samples in the illustrated embodiment is specifically used to execute the aforementioned method.

[0053] An embodiment of the present application also provides a computing device for brain microwave imaging, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method for reducing the number of training samples for rail insulation defect diagnosis as described above.

[0054] At the hardware level, the computing device includes a processor and, optionally, an internal bus, a network interface, and memory. The memory may include internal memory, such as high-speed random-access memory (RAM), or non-volatile memory, such as at least one disk drive. Of course, the electronic device may also include other hardware required for its services.

[0055] The processor, network interface, and memory can be interconnected via an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus. The bus can be divided into an address bus, a data bus, a control bus, and the like.

[0056] The memory is used to store programs. Specifically, the program may include program code, which includes computer operating instructions. The memory may include internal memory and non-volatile memory, and provides instructions and data to the processor.

[0057] The processor reads the corresponding computer program from the non-volatile memory into the internal memory and then runs it, logically forming a computing device that implements a method for reducing the number of training samples for rail insulation defect diagnosis. The processor executes the program stored in the memory and is specifically configured to perform the aforementioned method.

[0058] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0059] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A method for reducing the number of training samples for rail insulation defect diagnosis, characterized in that: include: Collecting rail potential data at the station, and performing data preprocessing on the potential data to obtain preprocessed potential data as initial training samples; Extract key features of rail potential data from preprocessed potential data; Divide the key features into intervals to obtain multiple intervals; Analyze the similarity of key features of adjacent intervals using a similarity analysis algorithm, merge adjacent intervals with similar key features, and use the samples in the merged intervals as the first training samples; forming a plurality of staggered overlapping subintervals in the merged interval according to a staggered partitioning method, wherein the plurality of staggered overlapping subintervals cover all key features, and using samples in the staggered overlapping subintervals as second training samples; The first training sample and the second training sample are used as training sets.

2. The method according to claim 1, wherein A rail insulation defect diagnosis model is constructed, and a training set is input into the rail insulation defect diagnosis model for training. The diagnosis model is a hybrid model of random forest and convolutional neural network.

3. The method according to claim 1, wherein The random forest algorithm model is used to process the samples in the merged interval, and the global feature classification of the samples in the merged interval is performed. The convolutional neural network model is used to extract the local spatial features of the samples in the dislocated overlapping sub-intervals. The obtained global feature classification and local spatial features are fused through a weighted voting mechanism to obtain the diagnosis results.

4. The method according to claim 1, wherein Perform data enhancement and sorting on the second training samples to enhance the diversity of the training samples.

5. The method according to claim 1, wherein The similarity analysis algorithm includes a clustering algorithm or a similarity measurement method.

6. The method according to claim 1, wherein The staggered division method specifically includes: dividing according to different starting points and offsets to form multiple staggered overlapping sub-intervals.

7. The method according to claim 6, wherein The division according to different starting points and offsets to form multiple staggered overlapping sub-intervals further includes: Set the offset to 1-2 unmerged intervals; The starting point of the staggered overlapping subintervals is centered on the boundary of the merged interval and expands symmetrically to the left and right sides; Each merged interval generates at least three non-identical offset overlapping subintervals; The misaligned overlapping subintervals cover the potential characteristic mutation region within the range of ±Δ of the merged interval boundary, where Δ is the offset.

8. The method according to claim 5, wherein The clustering algorithm is the K-means algorithm. When analyzing the feature similarity and correlation of adjacent intervals based on the K-means algorithm, cosine similarity is used as the clustering distance metric, and the number of clusters is dynamically determined according to the elbow rule of the interval feature distribution to ensure that the merged intervals have significant consistency in feature distribution.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

10. A computing device for reducing the number of training samples for rail insulation defect diagnosis, characterized in that: include: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor. The instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 8.

Images