A multi-system fusion monitoring system based on intelligent rail transit

Abstract

The present invention discloses a multi-system fusion monitoring system based on intelligent rail transit, including: a subsystem acquisition module, a subsystem vector module, which is used to perform feature splicing after feature standardization of N state monitoring features to generate a state monitoring vector representing the state of the subsystem; a train vector module, which is used to obtain the state monitoring vectors of M subsystems of the train; a state monitoring matrix construction module, which is used to align the state monitoring vectors of M subsystems and construct them into a state monitoring matrix; wherein the matrix row vector is the state monitoring vector of each subsystem; a query module, which is used to extract the state monitoring features after multi-system fusion from the state monitoring matrix based on the query vector of the train state; the present invention closely arranges the subsystem state monitoring vectors of similar features through cosine similarity, thereby reducing the amount of calculation during query. The system can respond to the current state changes of the train more quickly, and improves the real-time performance and efficiency of the query.

Description

Technical Field

[0001] The present invention relates to a multi-system fusion monitoring system, and in particular to a multi-system fusion monitoring system based on smart rail transit. Background Art

[0002] In modern rail transit systems, multiple monitoring subsystems (such as obstacle monitoring, track monitoring, and running parts monitoring) work together to ensure the safe operation of trains. These subsystems work independently and provide a large amount of monitoring data. The traditional approach is to fuse the data of these subsystems to fully reflect the operating status of the train. The patent document with patent publication number CN119018200A discloses a rail transit vehicle multi-monitoring system fusion integration system, which realizes the deep integration of multiple vehicle monitoring systems. Under the premise of multiple front-end devices, it reduces the number of detection hosts and reduces the difficulty of integrating the vehicle with the monitoring system. Through this fusion method, the data of multiple monitoring systems are aligned, and multi-system data fusion and correlation analysis are realized.

[0003] Then, although the above patent and the traditional multi-system data fusion method can integrate information from different subsystems, the traditional fusion method is to directly splice or merge the state monitoring vectors of the subsystems into a matrix. This simple data arrangement method cannot effectively optimize the query process. When querying data, the system must traverse the entire matrix and calculate each data point, which will lead to redundant calculations and reduce query efficiency.

[0004] In addition, in the multi-system integration of rail transit, the obstacle monitoring system cannot effectively and dynamically adjust the detection accuracy according to the train's operating environment (such as changes in speed and lighting conditions), resulting in the inability to fully utilize detection resources in certain environments, reducing the accuracy of the monitoring system. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides a multi-system fusion monitoring system based on intelligent rail transit, which solves the technical problems raised in the background technology by reconstructing the dynamic grid edge length and the train status matrix.

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

[0007] A multi-system fusion monitoring system based on intelligent rail transit, the monitoring system comprising:

[0008] Subsystem acquisition module, used to obtain N status monitoring features of each subsystem of the train;

[0009] The subsystem vector module is used to perform feature standardization on N state monitoring features and then perform feature concatenation to generate a state monitoring vector representing the state of the subsystem;

[0010] The train vector module is used to obtain the status monitoring vectors of M subsystems of the train;

[0011] A condition monitoring matrix construction module is used to align the condition monitoring vectors of M subsystems and construct them into a condition monitoring matrix; wherein the matrix row vector is the condition monitoring vector of each subsystem;

[0012] The query module is used to extract the multi-system fused state monitoring features from the state monitoring matrix based on the query vector of the train state.

[0013] In some embodiments, N state monitoring features are normalized and then concatenated to generate a state monitoring vector representing the state of the subsystem, including:

[0014] S2-1, obtain N state monitoring features, perform Z-Score standardization on them, and generate N standardized features for each subsystem;

[0015] S2-2. Concatenate the N standardized features of each subsystem in the order of the size of the feature values ​​to generate a state monitoring vector that represents the corresponding subsystem; wherein the state monitoring vector is a one-dimensional vector with a flat structure.

[0016] In some of the embodiments, the M subsystems of the train include: an obstacle monitoring subsystem, a pantograph-catwalk monitoring subsystem, a track monitoring subsystem, and a running gear monitoring subsystem; wherein the obstacle monitoring subsystem is used to obtain obstacle images in the direction of the train's advance.

[0017] In some embodiments, the step of acquiring the obstacle image includes:

[0018] A1. Obtain video frame images in the direction of the train's advance, and preprocess them to generate images to be processed for model input;

[0019] A2, receiving the image to be processed as input of the YOLO general model;

[0020] A3. Obtain the dynamic grid side length L;

[0021] A4. Divide the image to be processed into a number of L*L grids according to the side length L of the dynamic grid;

[0022] A5. Perform forward propagation on the plurality of grids, and output obstacle images in the foreground direction of the train at the current moment.

[0023] In some embodiments, obtaining the dynamic grid side length L includes:

[0024] A3-1. Define the standard running time of trains;

[0025] A3-2. Obtain the average speed of the train during the standard running time and the light intensity in the video frame image;

[0026] A3-3. Calculate the dynamic grid side length within the standard running time according to the average speed within the standard running time and the light intensity in the video frame image;

[0027] The calculation expression of the dynamic grid side length is:

[0028] ;

[0029] Where L represents the edge length of the dynamic mesh, represents the basic mesh edge length, It represents the average speed of the train during the standard running time. Represents the light intensity in the video frame image, is the proportional coefficient of the average speed, ranging from 0.001 to 0.01; is the proportional coefficient of light intensity, ranging from 0.001 to 0.05;

[0030] In some of the embodiments, forward propagation is performed on a plurality of the grids to output an obstacle image in the direction of the train at the current moment, including:

[0031] A5-1. Perform forward propagation for each grid and output several predicted bounding box vectors in each grid;

[0032] The features of the bounding box vector include: bounding box coordinates, category labels of the objects in the bounding box, and confidence of the category labels;

[0033] A5-2. Calculate the intersection-and-union ratio of each pair of bounding boxes to characterize the degree of overlap according to the bounding box coordinates of the bounding box vector;

[0034] A5-3. According to a preset intersection-over-union ratio threshold, select a number of bounding box pairs below the intersection-over-union ratio threshold, and define them as a candidate bounding box set;

[0035] A5-4. Select a target bounding box corresponding to the maximum confidence from the set of candidate bounding boxes;

[0036] A5-5. Extract the target label in the target bounding box and output it as an obstacle image in the direction of the train's advance at the current moment.

[0037] In some embodiments, the state monitoring vectors of the M subsystems are aligned to form a state monitoring matrix, including:

[0038] S4-1. Construct an initial permutation matrix according to the state monitoring vectors of the M subsystems;

[0039] S4-2, rearranging a plurality of state monitoring vectors according to the cosine similarity between adjacent row vectors of the initial arrangement matrix until the state monitoring matrix is ​​generated;

[0040] In some of the embodiments, an initial arrangement matrix is ​​constructed according to the state monitoring vectors of the M subsystems, including:

[0041] S4-1-1. Calculate the sum of the eigenvalues ​​of each subsystem according to the state monitoring vector of each subsystem;

[0042] S4-1-2, according to the absolute value of the sum of the eigenvalues ​​of each subsystem, arrange them in descending order and assign independent matrix row numbers to each subsystem;

[0043] S4-1-3. Arrange the state monitoring vector of each subsystem up and down according to the matrix row number to generate the initial arrangement matrix.

[0044] In some of the embodiments, a plurality of state monitoring vectors are rearranged according to the cosine similarity between adjacent row vectors of the initial arrangement matrix until the state monitoring matrix is ​​generated, including:

[0045] S4-2-1. Calculate the cosine similarities between adjacent row vectors in the initial arrangement matrix to obtain M-1 cosine similarities;

[0046] S4-2-2, selecting the state monitoring vector pair corresponding to the maximum cosine similarity;

[0047] S4-2-3, based on the selected state monitoring vector pairs, rearrange a number of state monitoring vectors in descending order according to the absolute value of the sum of their eigenvalues ​​to generate matrix row vector pairs;

[0048] S4-2-4. Select the state monitoring vector pair corresponding to the maximum cosine similarity from the remaining state monitoring vector pairs, and repeat S4-2-3 until the state monitoring matrix is ​​obtained by arrangement.

[0049] In some of the embodiments, the train query vector is a feature vector of the current train state.

[0050] The present invention provides a multi-system fusion monitoring system based on intelligent rail transit, which has the following beneficial effects:

[0051] The present invention optimizes the arrangement order of subsystem data by rearranging the state monitoring matrix. In the traditional method, since the state monitoring vectors of each subsystem are not effectively sorted, a large amount of redundant information needs to be processed during query, resulting in heavy computational burden and low query efficiency. However, the present invention closely arranges the subsystem state monitoring vectors with similar characteristics through cosine similarity, reducing the amount of computation during query. The system can respond to the current state changes of the train more quickly, improving the real-time and efficiency of the query.

[0052] Furthermore, the present invention dynamically adjusts the grid length according to the speed of the train and the light intensity. Specifically, when the speed is fast, the grid length is reduced to better capture details; when the light intensity is strong, the image quality is better, and the grid length can be appropriately increased to reduce the computational burden. Dynamically adjusting the grid length makes the target detection process more adaptable to different environmental conditions, thereby improving the accuracy and efficiency of detection. The dynamic grid length ensures accurate identification of obstacles in complex environments and improves the performance of the rail transit monitoring system. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a structural block diagram of a multi-system fusion monitoring system based on intelligent rail transit in the present invention;

[0054] Figure 2 A fusion schematic diagram of a multi-system fusion monitoring system based on intelligent rail transit according to the present invention;

[0055] Figure 3 This is a schematic diagram of the dynamic grid side length calculation process of the present invention;

[0056] Figure 4 A schematic diagram of the initial arrangement matrix generation process of the present invention;

[0057] Figure 5 The figure is a schematic diagram of the rearrangement process of the condition monitoring matrix of the present invention. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] First, the prior art and related concepts involved in the embodiments of the present invention are described:

[0060] YOLO general model: YOLO (You Only Look Once) is a very popular deep learning target detection model that aims to quickly detect targets in images through a single forward propagation.

[0061] Example 1: Please refer to Figures 1 to 3 The present invention provides a multi-system fusion monitoring system based on intelligent rail transit, the monitoring system comprising:

[0062] Subsystem acquisition module, used to obtain N status monitoring features of each subsystem of the train;

[0063] The subsystem vector module is used to perform feature standardization on N state monitoring features and then perform feature concatenation to generate a state monitoring vector representing the state of the subsystem;

[0064] The train vector module is used to obtain the status monitoring vectors of M subsystems of the train;

[0065] A condition monitoring matrix construction module is used to align the condition monitoring vectors of M subsystems and construct them into a condition monitoring matrix; wherein the matrix row vector is the condition monitoring vector of each subsystem;

[0066] The query module is used to extract the multi-system fused state monitoring features from the state monitoring matrix based on the query vector of the train state.

[0067] This embodiment integrates the monitoring features of different subsystems by fusing the state monitoring vectors of multiple subsystems of the train. A unified state monitoring vector is generated by feature standardization and splicing, and the monitoring vectors of multiple subsystems are aligned to form a state monitoring matrix. Finally, based on the query vector, the state monitoring features of multiple systems after fusion are extracted from the matrix, which improves the overall perception of the current state of the train.

[0068] The step of generating a state monitoring vector by the subsystem vector module includes:

[0069] S2-1, obtain N state monitoring features, perform Z-Score standardization on them, and generate N standardized features for each subsystem;

[0070] The Z-Score standardization method is used for each condition monitoring feature to ensure that the mean of each feature is 0 and the standard deviation is 1, thereby eliminating the dimensional differences between different features and making each feature have the same scale.

[0071] S2-2. Concatenate the N standardized features of each subsystem in the order of the size of the feature values ​​to generate a state monitoring vector that represents the corresponding subsystem; wherein the state monitoring vector is a one-dimensional vector with a flat structure.

[0072] In this embodiment, the M subsystems of the train include: an obstacle monitoring subsystem, a pantograph-catwalk monitoring subsystem, a track monitoring subsystem, and a running gear monitoring subsystem; wherein the obstacle monitoring subsystem is used to obtain obstacle images in the direction of the train's advance.

[0073] Specifically, the obstacle image is a real-time image of other vehicles, pedestrians, objects, etc., which is captured by sensors (such as radar, lidar or camera) and timely alarms are issued when potential danger occurs to ensure the safe operation of the train.

[0074] The pantograph-catenary monitoring subsystem is used to monitor the contact status between the train and the catenary, especially during power transmission, to detect problems such as poor contact, pantograph-catenary detachment, and wear, to ensure the stability of the train power supply system.

[0075] The track monitoring subsystem is used to monitor the status and health of the train track, including the flatness, gauge, cracks, wear, etc. of the track to ensure the integrity of the track.

[0076] The running part monitoring subsystem is used to monitor the status of the train running parts, such as wheels, axles, transmission systems, etc., and to detect the wear, temperature, vibration and other indicators of the mechanical parts in real time.

[0077] In this embodiment, the step of acquiring the obstacle image includes:

[0078] A1. Obtain video frame images in the direction of the train's advance, and preprocess them to generate images to be processed for model input;

[0079] Among them, the preprocessing of the video frame image includes:

[0080] Adjust the image size to a fixed size (e.g. 416x416, 608x608, etc.)

[0081] Normalize the pixel values ​​in the image and scale the pixel values ​​from the range of 0-255 to between 0 and 1; if the color space of the input image is not RGB, convert it to RGB format.

[0082] A2, receiving the image to be processed as input of the YOLO general model;

[0083] A3. Obtain the dynamic grid side length L;

[0084] A4. Divide the image to be processed into a number of L*L grids according to the side length L of the dynamic grid;

[0085] A5. Perform forward propagation on the plurality of grids, and output obstacle images in the foreground direction of the train at the current moment.

[0086] This embodiment obtains obstacle images through the YOLO general model in the field of target detection. This process preprocesses the video frame image in the direction of the train's advance and uses the YOLO target detection model to identify and extract obstacles in front of the train at the current moment.

[0087] In this embodiment, the step of obtaining the dynamic grid side length L includes:

[0088] A3-1. Define the standard running time of trains;

[0089] A3-2. Obtain the average speed of the train during the standard running time and the light intensity in the video frame image;

[0090] In this embodiment, the average speed of the standard operating time is the speed of the train at a predetermined time period during operation; and the light intensity can be extracted by using the average brightness value or light intensity distribution in the video frame image through image processing methods.

[0091] A3-3. Calculate the dynamic grid side length within the standard running time according to the average speed within the standard running time and the light intensity in the video frame image;

[0092] The calculation expression of the dynamic grid side length is:

[0093] ;

[0094] Where L represents the edge length of the dynamic mesh, Indicates the basic grid side length, for example, 416 pixels or 608 pixels, the grid size in the initial state); It represents the average speed of the train during the standard operating time. The higher the speed, the smaller the grid side length should be, so the inverse relationship is adopted. It represents the light intensity in the video frame image. The stronger the light, the longer the grid side length, so a positive correlation is adopted. is the proportional coefficient of the average speed, ranging from 0.001 to 0.01; is the proportional coefficient of light intensity, ranging from 0.001 to 0.05;

[0095] Specifically, when the train speed increases, the images captured by the camera may appear motion blurred or blurry, and fast-moving objects are more difficult to detect accurately. In order to adapt to this change, the side length of the grid needs to be reduced, so that more small grids are used in the image to better capture details. Therefore, the faster the speed, the smaller the grid needs to be; under better lighting conditions, the image clarity is higher and the detection accuracy is better. The side length of the grid can be appropriately increased, which can reduce the amount of calculation. Therefore, the stronger the light, the larger the grid can be. Under low light conditions, the clarity of the image is reduced and the target is not easy to identify. The side length of the grid needs to be reduced to ensure that more target information can be detected through a finer grid. Therefore, the weaker the light, the smaller the grid needs to be.

[0096] This embodiment calculates the appropriate dynamic grid edge length based on the standard running time and average speed of the train and the light intensity in the video frame image. The dynamic grid edge length adjustment method significantly improves the detection performance of the obstacle monitoring subsystem of the train in different environments, especially in high-speed or low-light conditions, ensuring accurate detection of obstacles and providing a robust real-time monitoring capability for the rail transit system.

[0097] Furthermore, the step of outputting the obstacle image in the direction of the train at the current moment includes:

[0098] A5-1. Perform forward propagation for each grid and output several predicted bounding box vectors in each grid;

[0099] The features of the bounding box vector include: bounding box coordinates (usually the center coordinates, width, and height of the box), the category label of the target in the bounding box, and the confidence of the category label; the confidence of each bounding box vector represents the probability of the existence of the target in the box and the confidence of the classification of the target in the box.

[0100] A5-2. According to the bounding box coordinates of the bounding box vector, the intersection-and-union ratio of each pair of bounding boxes is calculated to characterize the degree of overlap between them. The intersection-and-union ratio refers to the ratio of the intersection area to the union area of ​​the two bounding boxes.

[0101] A5-3. According to a preset intersection-over-union ratio threshold, select a number of bounding box pairs below the intersection-over-union ratio threshold, and define them as a candidate bounding box set;

[0102] A5-4. Select a target bounding box corresponding to the maximum confidence from the set of candidate bounding boxes;

[0103] A5-5. Extract the target label in the target bounding box and output it as an obstacle image in the direction of the train's advance at the current moment.

[0104] This embodiment generates several predicted bounding box vectors by forward propagating each grid. These vectors contain the coordinates, category labels and confidence of the target. Then, based on the coordinates of the bounding boxes, the intersection-and-union ratio is calculated to measure the degree of overlap between different bounding boxes, and a set of candidate boxes with low intersection-and-union ratios is screened out by setting a threshold. Finally, by selecting the candidate bounding box with the highest confidence, the target category label is extracted, and an obstacle image of the train's forward direction at the current moment is generated.

[0105] Example 2: See Figures 4 to 5 The technical solution of this embodiment 2 is different from that of embodiment 1 in that the matrix construction steps of the condition monitoring matrix construction module in embodiment 1 are disclosed, and the construction steps include:

[0106] S4-1. Construct an initial permutation matrix according to the state monitoring vectors of the M subsystems;

[0107] S4-2, rearranging a plurality of state monitoring vectors according to the cosine similarity between adjacent row vectors of the initial arrangement matrix until the state monitoring matrix is ​​generated;

[0108] Furthermore, the step 4-1 specifically includes:

[0109] S4-1-1. Calculate the sum of the eigenvalues ​​of each subsystem according to the state monitoring vector of each subsystem;

[0110] S4-1-2, according to the absolute value of the sum of the eigenvalues ​​of each subsystem, arrange them in descending order and assign independent matrix row numbers to each subsystem;

[0111] S4-1-3. Arrange the state monitoring vector of each subsystem up and down according to the matrix row number to generate the initial arrangement matrix.

[0112] The construction step of the initial permutation matrix ensures the effective sorting and organization of the subsystem condition monitoring vectors. The monitoring vectors of each subsystem are assigned corresponding row numbers by calculating the sum of the eigenvalues ​​of the condition monitoring vectors of each subsystem and arranging them in descending order. This sorting method arranges the condition monitoring vectors according to the importance of their eigenvalues, which helps to align the vectors properly.

[0113] The initial arrangement matrix provides an orderly basis for subsequent matrix alignment and multi-system data fusion, making it easier to combine similar subsystem states, thereby improving the accuracy of the entire system status monitoring.

[0114] Furthermore, the step S4-2 specifically includes:

[0115] S4-2-1. Calculate the cosine similarities between adjacent row vectors in the initial arrangement matrix to obtain M-1 cosine similarities;

[0116] S4-2-2, selecting the state monitoring vector pair corresponding to the maximum cosine similarity;

[0117] S4-2-3, based on the selected state monitoring vector pairs, rearrange the plurality of state monitoring vectors in descending order according to the absolute value of the sum of their eigenvalues ​​to generate matrix row vector pairs;

[0118] S4-2-4. Select the state monitoring vector pair corresponding to the maximum cosine similarity from the remaining state monitoring vector pairs, and repeat S4-2-3 until the state monitoring matrix is ​​obtained by arrangement.

[0119] This embodiment can identify similar subsystem state monitoring vectors by calculating the cosine similarity between adjacent row vectors and arrange similar features together. Then, according to the order of cosine similarity, the vector pairs with the highest similarity are selected and rearranged according to the order of the sum of eigenvalues. This step actually brings similar features in the state monitoring matrix closer together, optimizing the fusion of state data of different subsystems.

[0120] During the query process, similar features are arranged together, reducing the search space and computational complexity of the query. Therefore, the system can query in this optimized order, thereby improving the efficiency of the query.

[0121] Therefore, by constructing a multi-system fusion state monitoring matrix, based on the arrangement optimization of multi-system state monitoring vectors, the system realizes the fusion of similar features, and can more efficiently process and extract relevant features in subsequent queries and data extraction. This makes the multi-system fusion monitoring state monitoring matrix able to improve the accuracy of queries and the efficiency of data calculation.

[0122] In this embodiment, the train query vector is a feature vector of the current train state. Specifically, it is usually composed of multi-dimensional features such as the train's operating state, environmental information, and monitoring data. The query vector includes but is not limited to:

[0123] The train’s current speed, direction, position and other operating information;

[0124] The current environmental conditions, such as temperature, humidity, light, etc.

[0125] Monitoring data of each subsystem, such as obstacle detection results, power system status, track health, etc.

[0126] After the matrix is ​​constructed, the train query vector is used to extract relevant features that match the current train status in the condition monitoring matrix, thereby supporting real-time monitoring of the train system. Specifically, the train query vector is similarity-calculated with the row vector of the condition monitoring matrix to identify and extract the most relevant subsystem monitoring data.

[0127] In summary, in the process of multi-system integration of rail transit, the present invention dynamically adjusts the grid side length of the YOLO model according to the obtained train running speed and light intensity. Specifically, when the running speed is faster, the grid side length is reduced to better capture details; when the light intensity is strong, the image quality is better, and the grid side length is appropriately increased to reduce the computational burden. Dynamically adjusting the grid side length makes the target detection process more adaptable to different environmental conditions, thereby improving the accuracy and efficiency of detection. The dynamic grid side length ensures that obstacles can be identified more accurately in complex environments (such as low light or high-speed driving conditions), and improves the performance of the rail transit monitoring system. In addition, in the present invention, the subsystem state monitoring vectors with similar features are closely arranged through the cosine similarity between the state monitoring vectors of each subsystem, reducing the amount of calculation during query. It can respond to the current state changes of the train more quickly, improving the real-time and efficiency of the query.

[0128] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.) means.

[0129] The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that includes one or more available media. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium may be a solid-state hard disk.

[0130] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division of a waterway underwater terrain change analysis system and method. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0131] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be covered by the protection scope of the present application.

Claims

1. A multi-system fusion monitoring system based on intelligent rail transit, characterized in that: The monitoring system comprises: Subsystem acquisition module, used to obtain N status monitoring features of each subsystem of the train; The subsystem vector module is used to perform feature standardization on N state monitoring features and then perform feature concatenation to generate a state monitoring vector representing the state of the subsystem; The train vector module is used to obtain the status monitoring vectors of M subsystems of the train; A condition monitoring matrix construction module is used to align the condition monitoring vectors of M subsystems and construct them into a condition monitoring matrix; wherein the matrix row vector is the condition monitoring vector of each subsystem; The query module is used to extract the multi-system fused state monitoring features from the state monitoring matrix based on the query vector of the train state.

2. According to claim 1, a multi-system fusion monitoring system based on intelligent rail transit is characterized in that: After the N state monitoring features are standardized, feature concatenation is performed to generate a state monitoring vector representing the state of the subsystem, including: S2-1, obtain N state monitoring features, perform Z-Score standardization on them, and generate N standardized features for each subsystem; S2-2. Concatenate the N standardized features of each subsystem in the order of the size of the feature values ​​to generate a state monitoring vector that represents the corresponding subsystem; wherein the state monitoring vector is a one-dimensional vector with a flat structure.

3. According to claim 1, a multi-system fusion monitoring system based on intelligent rail transit is characterized in that: The M subsystems of the train include: an obstacle monitoring subsystem, a pantograph-catwalk monitoring subsystem, a track monitoring subsystem and a running part monitoring subsystem; wherein the obstacle monitoring subsystem is used to obtain obstacle images in the direction of the train's advance.

4. According to claim 3, a multi-system fusion monitoring system based on intelligent rail transit is characterized in that: The step of acquiring the obstacle image comprises: A1. Obtain video frame images in the direction of the train's advance, and preprocess them to generate images to be processed for model input; A2, receiving the image to be processed as input of the YOLO general model; A3. Obtain the dynamic grid side length L; A4. Divide the image to be processed into a number of L*L grids according to the side length L of the dynamic grid; A5. Perform forward propagation on the plurality of grids, and output obstacle images in the foreground direction of the train at the current moment.

5. According to claim 4, a multi-system fusion monitoring system based on intelligent rail transit is characterized in that: Get the dynamic mesh edge length L, including: A3-1. Define the standard running time of trains; A3-2. Obtain the average speed of the train during the standard running time and the light intensity in the video frame image; A3-3. Calculate the dynamic grid side length within the standard running time according to the average speed within the standard running time and the light intensity in the video frame image; The calculation expression of the dynamic grid side length is: ； Where L represents the length of the dynamic mesh edge, represents the basic mesh edge length, It represents the average speed of the train during the standard running time. Represents the light intensity in the video frame image, is the proportional coefficient of the average speed, ranging from 0.001 to 0.01; It is the proportional coefficient of light intensity, and its value range is 0.001 to 0.

05.

6. According to claim 5, a multi-system fusion monitoring system based on intelligent rail transit is characterized in that: Perform forward propagation on the plurality of grids, and output obstacle images in the direction of the train at the current moment, including: A5-1. Perform forward propagation for each grid and output several predicted bounding box vectors in each grid; The features of the bounding box vector include: bounding box coordinates, category labels of the objects in the bounding box, and confidence of the category labels; A5-2. Calculate the intersection-and-union ratio of each pair of bounding boxes to characterize the degree of overlap according to the bounding box coordinates of the bounding box vector; A5-3. According to a preset intersection-over-union ratio threshold, select a number of bounding box pairs below the intersection-over-union ratio threshold, and define them as a candidate bounding box set; A5-4. Select a target bounding box corresponding to the maximum confidence from the set of candidate bounding boxes; A5-5. Extract the target label in the target bounding box and output it as an obstacle image in the direction of the train's advance at the current moment.

7. The multi-system fusion monitoring system based on intelligent rail transit according to claim 1 is characterized in that: Align the condition monitoring vectors of the M subsystems and construct them into a condition monitoring matrix, including: S4-1. Construct an initial permutation matrix according to the state monitoring vectors of the M subsystems; S4-2. Rearrange a plurality of state monitoring vectors according to the cosine similarity between adjacent row vectors of the initial arrangement matrix until the state monitoring matrix is ​​generated.

8. The multi-system fusion monitoring system based on intelligent rail transit according to claim 7 is characterized in that: According to the state monitoring vectors of M subsystems, an initial arrangement matrix is ​​constructed, including: S4-1-1. Calculate the sum of the eigenvalues ​​of each subsystem according to the state monitoring vector of each subsystem; S4-1-2, according to the absolute value of the sum of the eigenvalues ​​of each subsystem, arrange them in descending order and assign independent matrix row numbers to each subsystem; S4-1-3. Arrange the state monitoring vector of each subsystem up and down according to the matrix row number to generate the initial arrangement matrix.

9. The multi-system fusion monitoring system based on intelligent rail transit according to claim 8 is characterized in that: According to the cosine similarity between adjacent row vectors of the initial arrangement matrix, a plurality of state monitoring vectors are rearranged until the state monitoring matrix is ​​generated, including: S4-2-1. Calculate the cosine similarities between adjacent row vectors in the initial arrangement matrix to obtain M-1 cosine similarities; S4-2-2, selecting the state monitoring vector pair corresponding to the maximum cosine similarity; S4-2-3, based on the selected state monitoring vector pairs, rearrange the plurality of state monitoring vectors in descending order according to the absolute value of the sum of their eigenvalues ​​to generate matrix row vector pairs; S4-2-4. Select the state monitoring vector pair corresponding to the maximum cosine similarity from the remaining state monitoring vector pairs, and repeat S4-2-3 until the state monitoring matrix is ​​obtained by arrangement.

10. The multi-system fusion monitoring system based on intelligent rail transit according to claim 1 is characterized in that: The train query vector is a feature vector of the current train state.

Images