Heavy-load train group operation control sensing method and heavy-load train group operation control sensing system

Through graph convolutional neural network, the sensor data is converted into state diagrams is realized, and the automated perception supervision along the heavy-duty railway is solved, which is mainly civil defense and limited technical defense methods in the existing technology, and the safety risks of train operation are improved Monitoring and decision-making capabilities.

CN118953455AActive Publication Date: 2024-11-15CHINA SHENHUA ENERGY CO LTD +1

Patent Information

Application Number
CN202410923426.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-10
Publication Date
2024-11-15
Estimated Expiration
2044-07-10

AI Technical Summary

Technical Problem

The risk monitoring systems along the existing heavy-duty railways rely mainly on civil defense, with limited technical defense means and are unable to effectively monitor and respond to complex geological environments and natural disasters, resulting in threats to driving safety and lack of all-round holographic perception capabilities.

Method used

The graph convolution neural network is used to convert the train group operation status information collected by the sensor into a state diagram. The graph convolution neural network is used to automatically judge the estimated alarm level and alarm processing time, and establish data acquisition, graph conversion, model construction and Perception supervision unit to realize real-time operation control perceptual supervision.

Benefits of technology

Enhanced the automation and holographic perception and monitoring capabilities of train operation safety risks along heavy-duty railways, integrate big data mining and digital twin technology, improve the decision-making capabilities of risk analysis and safety situation deduction, and ensure the safety of heavy-duty train groups. run.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118953455A_ABST
    Figure CN118953455A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of rail transit operation control, in particular to a heavy haul train group operation control sensing method and system. Train group operation state information is collected through different sensors; the state information is converted into a state diagram, nodes in the diagram represent sampling points of the sensor, edges in the diagram represent time differences between latest sampling points of the sensor, and a proximity matrix in the diagram represents a connection relation between the edges of the nodes; establishing a graph convolutional neural network according to the state graph, wherein the graph convolutional neural network is used for estimating the alarm level and the alarm processing time of the train group operation state; training a graph convolutional neural network according to the historical state graph to obtain a trained graph convolutional neural network; and inputting the state diagram acquired in real time into the diagram convolutional neural network, and carrying out operation control perception supervision on the operation state of the train group. Through the technical means, the automatic and holographic sensing and monitoring capability of train operation safety risks along the heavy haul railway is enhanced, and a guarantee is provided for safe operation of a heavy haul train group.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the technical field of rail transit operation control, and in particular to a heavy-load train group operation control perception method and system. Background Art

[0002] Energy transportation is an important part of the modern comprehensive transportation system and a national strategic asset. Ensuring its operational safety plays a vital role in stabilizing the national energy supply and maintaining national economic and social stability. As my country's railway transportation safety has gradually expanded from the original signal safety protection to large-scale transportation safety protection for the perception of the train operation environment, the boundaries of transportation safety have been significantly expanded. At present, my country's railway train operation safety protection mainly relies on personnel protection, and technical defense means are limited, resulting in frequent accidents, which seriously affects the smooth flow of energy transportation channels and the safety of life and property.

[0003] However, along the heavy-duty railway lines in my country, the terrain and geology are extremely complex, and the environment is very fragile and sensitive. Under the harsh geological environment, continuous rainwater erosion and leakage, geological deformation, and large temperature differences can easily lead to landslides, mudslides, tunnel lining block loss, rail breakage and other line hazards, which directly threaten the safety of train driving. In recent years, due to the lack of fast and effective integrated risk perception methods, once a major disaster occurs, it will seriously affect the normal operation of the line, endanger driving safety, and cause huge casualties and immeasurable consequences and losses.

[0004] The existing heavy-haul railway lines mainly adopt the method of human defense as the main method and technical defense as the auxiliary method to monitor risks:

[0005] (1) It relies heavily on personnel monitoring, which results in a large workload and heavy safety responsibilities, and it is difficult to cover various complex scenarios on site around the clock and in all weather conditions;

[0006] (2) Technical prevention measures are limited, with relatively simple technical means and functions, and they are mainly used to monitor key areas;

[0007] (3) The existing monitoring information is not comprehensive, the information integration is low, and the monitoring results are managed separately by personnel from different departments. At present, a full-scale holographic perception monitoring system has not yet been formed;

[0008] (4) Existing risk monitoring remains at the perception monitoring level, and it is necessary to arrange on-duty personnel to judge the risk level of the event based on the monitoring results and take manual response.

[0009] In summary, the existing railway operation environment safety risk monitoring system has single monitoring information and single technical means, and requires on-duty personnel to judge the risk level of events based on the monitoring results, so as to take manual response. Taking the natural disaster and foreign body intrusion monitoring system as an example, it can only monitor, warn and alarm the wind, rain, snow, earthquake and foreign body intrusion along the railway in real time. When facing special natural disasters such as collapse of heavy-duty railways, landslides, mudslides, etc., it does not have effective monitoring coverage capabilities, and cannot meet the disaster monitoring needs of heavy-duty railways and effectively ensure driving safety. Summary of the invention

[0010] In view of the above problems, the present invention provides a heavy-load train group operation control perception method and system.

[0011] In a first aspect, a heavy-load train group operation control perception method is provided, the method comprising:

[0012] Collect train group operation status information through different sensors;

[0013] The state information is converted into a state graph, where nodes represent sensor sampling points, edges represent the time difference between the latest sensor sampling points, and proximity matrix represents the connection relationship between nodes.

[0014] A graph convolutional neural network is established based on the state diagram to estimate the alarm level and alarm processing time of the train group operation status;

[0015] Train the graph convolutional neural network according to the historical state graph to obtain a trained graph convolutional neural network;

[0016] The state diagram collected in real time is input into the graph convolutional neural network to perform operation control perception and supervision on the operation status of the train group.

[0017] Furthermore, the state information is converted into a state diagram, including:

[0018] G represents the state graph, and the node of graph G is V, which represents the data of the sensor, including: sampling number, sampling location, sampling time, sensor type, sensor data, and whether to alarm;

[0019] The edge of graph G is E, which represents the time difference between the latest sampling data of the sensor;

[0020] The proximity matrix A of the graph G represents the connection relationship between the edges of the nodes.

[0021] Furthermore, the graph G is (V, E, A), where the node V is an N×F matrix, N is the number of sensor nodes, and F is the number of data features of the sensor; the edge E is an N×N×P matrix, and P is the number of edge data features.

[0022] Furthermore, the information propagation model of sensor nodes in the graph convolutional neural network is:

[0023]

[0024] In the formula, represents the output feature of sensor node i in the (l+1)th layer of the graph convolutional neural network, and represents the output of sensor node i and sensor node j at layer l, σ is the activation function, R is the relationship set between sensor nodes, is the set of sensor nodes with a relationship of r among the one-hop neighboring nodes of sensor node i. In the present disclosure, relationship r means that there is an association relationship. i,r The value is The regularization constant of is the weight parameter matrix of the adjacent sensor nodes with relationship r in the l-th layer network, is the weight parameter matrix of sensor node i in the l-th layer network.

[0025] Furthermore, the propagation model of side information in graph convolutional neural networks is:

[0026] Where N i is the neighboring sensor node of sensor node i, α ij is the Softmax regularization function, W v and W ∈ are the node weight parameter matrix and edge weight parameter matrix respectively, ρ ij is the edge eigenvector, h i is the node feature vector, ‖ is the concatenation operation, It is the feature vector output after aggregating the edge feature information.

[0027] Furthermore, a graph convolutional neural network is established based on the state diagram, including:

[0028] The graph convolutional neural network includes L graph convolutional network layers GCN and multi-layer perceptron layers MLP. The activation functions of the GCN layer and LSTM layer are both ReLU (Rectified Linear Unit) functions, and the activation function of the MLP layer is tanh activation function.

[0029] The output of the graph convolutional neural network is O = (A i ,P i ,t i ,S i ,M i ,T i ), A i ,P i,t i ,S i ,M i ,T i They represent the alarm number, alarm location, alarm time, alarm type, alarm level and alarm processing time respectively. The alarm numbers are arranged in the order of alarms. The alarm location and alarm time are the sensor node location of the alarm and the sampling time of the alarm respectively. The alarm level is the estimated level after processing by the graph convolutional neural network. The alarm processing time is the time required for alarm processing estimated by the graph convolutional neural network.

[0030] Furthermore, the loss function of the graph convolutional neural network is as follows:

[0031]

[0032] In the formula, θ(S i ) is the weight of the alarm type set based on experience or iterative learning.

[0033] In a second aspect, a heavy-load train group operation control perception system includes: a data acquisition unit, a graph conversion unit, a model building unit and a perception supervision unit;

[0034] A data acquisition unit, used to collect train group operation status information through different sensors;

[0035] A graph conversion unit is used to convert the state information into a state graph, where the nodes in the graph represent the sampling points of the sensors, the edges in the graph represent the time differences between the latest sampling points of the sensors, and the proximity matrix A in the graph represents the connection relationship between the edges of the nodes;

[0036] A model building unit, used to build a graph convolutional neural network based on the state diagram, and used to estimate the alarm level and alarm processing time of the train group operation status;

[0037] The model building unit is further used to train the graph convolutional neural network according to the historical state graph to obtain a trained graph convolutional neural network;

[0038] The perception supervision unit is used to input the real-time collected state diagram into the graph convolutional neural network to perform operation control perception supervision on the operation status of the train group.

[0039] In a third aspect, an electronic device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus;

[0040] a memory storing a computer program;

[0041] The processor is used to implement the above-mentioned heavy-load train group operation control perception method when executing the computer program stored in the memory.

[0042] In a fourth aspect, a computer-readable storage medium stores a computer program, which, when executed by a processor, implements the above-mentioned heavy-load train group operation control perception method.

[0043] The present disclosure has at least the following beneficial effects:

[0044] This disclosure is aimed at the development needs of complex global environments for heavy-load train groups, short-distance train tracking, and automated train control systems. It uses technical means to enhance the automation and holographic perception and monitoring capabilities of train operation safety risks along heavy-load railways, integrates advanced technologies such as big data mining and digital twins, and conducts risk analysis, safety situation deduction, and decision-making on line perception information, to provide guarantees for the safe operation of heavy-load train groups.

[0045] Other features and advantages of the present disclosure will be described in the following description, and partly become apparent from the description, or be understood by implementing the present disclosure. The purpose and other advantages of the present disclosure can be realized and obtained by the structures indicated in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0047] Figure 1 A flow chart of the control perception method for running an embodiment of the present disclosure;

[0048] Figure 2 This is a schematic diagram of the architecture of the operation control perception system according to an embodiment of the present disclosure;

[0049] Figure 3 Schematic diagram of the graph convolutional neural network structure;

[0050] Figure 4 It is a schematic diagram of the structure of an electronic device;

[0051] Figure 5 This is a schematic diagram of the system hardware structure. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0053] In view of the development needs of complex global environments, short-distance train tracking, and automated train control systems, it is urgently necessary to enhance the holographic and automated perception capabilities of train safety operation through technical means, comprehensively improve the comprehensiveness of perception content and the level of cognitive reasoning, develop holographic perception and three-dimensional safety railway line security guarantees, and lead the technical research, development and construction of active safety protection of the world's advanced rail transit autonomous operation systems.

[0054] like Figure 1 As shown, a heavy-load train group operation control perception method, the method comprising:

[0055] S101, collecting train group running status information through different sensors;

[0056] S102, converting the state information into a state graph, where nodes in the graph represent sensor sampling points, edges in the graph represent time differences between the latest sampling points of the sensor, and a proximity matrix in the graph represents the connection relationship between the edges of the nodes;

[0057] S103, establishing a graph convolutional neural network according to the state diagram, for estimating the alarm level and alarm processing time of the train group operation status;

[0058] S104, training a graph convolutional neural network according to the historical state graph to obtain a trained graph convolutional neural network;

[0059] S105, inputting the real-time collected state diagram into the graph convolutional neural network to perform operation control perception supervision on the operation status of the train group.

[0060] The specific implementation is as follows:

[0061] When processing actual perception data, the sampling frequencies of different perception terminals are inconsistent, resulting in asynchronous sensor data. At the same time, when a sensor generates an alarm, the information of other related sensors may not be collected synchronously. It is necessary to use historical sensor data to determine the relevance of the sensor information and the degree of event risk.

[0062] Therefore, for the early warning analysis and processing of multi-source perception data, the present disclosure proposes a graph neural network prediction and classification algorithm for asynchronous spatiotemporal correlation analysis, which can process multi-source asynchronous sensor data and alarm data to determine the risk level of the event.

[0063] The nodes of the graph G that defines the perception data processing model are V, which contain sensor data (number, location, sampling time, sensor type, sensor data, whether to alarm, alarm level, and estimated alarm processing time). The edges of the graph G are E, which contain (the distance between sensors, the time difference between the latest sensor sampling data), and the proximity matrix A of the graph G represents the connection relationship between the edges of the nodes. The present disclosure obtains the position and time relationship between the data with the closest sensor sampling time by processing the edge E, obtains the asynchronous spatiotemporal relationship without the need for sensor synchronization, and obtains the correlation between the sensor data.

[0064] Then the graph G is (V, E, A), where the nodes are N×F matrices, N is the number of sensor nodes, and in this disclosure, F=8 is the number of sensor data features. E is an N×N×P matrix, and in this disclosure, P=2 is the number of edge data features. ij ∈R(i=1,2,...,R;j=1,2,...,F) represents the jth channel of the feature vector of the i-th sensor node in the graph G, E ijp ∈R(i,j=1,2,...,R;p=1,2,...,P) represents the pth channel of the feature vector of the edge between the i-th sensor node and the j-th sensor node in the graph G.

[0065] In the graph convolutional neural network in this disclosure, the information propagation model of the sensor node is:

[0066]

[0067] In the formula, represents the output feature of sensor node i in the (l+1)th layer of the graph convolutional neural network, and represents the output of sensor node i and sensor node j at layer l, σ is the activation function, R is the relationship set between sensor nodes, is the set of sensor nodes with a relationship of r among the one-hop neighboring nodes of sensor node i. In the present disclosure, relationship r means that there is an association relationship. i,r The value is The regularization constant of is the weight parameter matrix of the adjacent sensor nodes with relationship r in the l-th layer network, is the weight parameter matrix of sensor node i in the l-th layer network.

[0068] The propagation model of side information is:

[0069]

[0070] In the formula, E j→i represents the eigenvector of edge (j→i), represents the feature vector of sensor node i in the tth graph convolution layer, M represents the information transfer function, U represents the sensor node information update function, Indicates an information collection operation. Specifically, one implementation form of the above formula may be:

[0071]

[0072] Where N i is the neighboring sensor node of sensor node i, α ij is the Softmax regularization function, W v and W ∈ are the node weight parameter matrix and edge weight parameter matrix respectively, ρ ij is the edge eigenvector, h i is the node feature vector, ‖ is the concatenation operation, It is the feature vector output after aggregating the edge feature information.

[0073] In the present disclosure, an L-layer graph convolutional neural network can be constructed as needed to After L iterations, LSTM and MLP layers can be stacked to After processing, the final output O is obtained, which is (alarm number, alarm location, alarm time, alarm type, alarm level, alarm processing time), where the alarm numbers are arranged in the order of alarms, the alarm location and alarm time are the sensor node location of the alarm and the sampling time of the alarm respectively, the alarm level is the estimated level after processing by the graph convolutional neural network, and the alarm processing time is the time required for alarm processing estimated by the graph convolutional neural network.

[0074] like Figure 2 As shown, a heavy-load train group operation control perception system includes: a data acquisition unit 201, a graph conversion unit 202, a model construction unit 203 and a perception supervision unit 204;

[0075] The data collection unit 201 is used to collect the train group running status information through different sensors;

[0076] A graph conversion unit 202 is used to convert the state information into a state graph, where nodes represent sensor sampling points, edges represent time differences between the latest sensor sampling points, and a proximity matrix A represents the connection relationship between the edges of the nodes;

[0077] A model building unit 203, used to establish a graph convolutional neural network according to the state diagram, for estimating the alarm level and alarm processing time of the train group operation status;

[0078] The model building unit 203 is further used to train the graph convolutional neural network according to the historical state graph to obtain a trained graph convolutional neural network;

[0079] The perception supervision unit 204 is used to input the real-time collected state diagram into the graph convolutional neural network to perform operation control perception supervision on the operation status of the train group.

[0080] When implementing it, Figure 3 As shown, the graph convolutional neural network in the present disclosure includes L layers of graph convolution network (GCN), an optional long short-term memory (LSTM) layer, and a multilayer perceptron (MLP) layer. The activation functions of the GCN layer and the LSTM layer are both ReLU (Rectified Linear Unit) functions, and the activation function of the MLP layer is the tanh activation function.

[0081] The output result O is (alarm number, alarm location, alarm time, alarm type, alarm level, alarm processing time), set the variable to O = (A ii ,P i ,t i ,S i ,M i ,T i ,), then the loss function is:

[0082]

[0083] In the formula, θ(S i ) is the weight of the alarm type set based on experience or iterative learning.

[0084] Network training process:

[0085] Dataset construction. First, collect the alarm data of a certain section of railway within one year, including sensor data (number, location, sampling time, sensor type, sensor data, whether to alarm, alarm level, estimated alarm processing time) and process the sensor data to obtain side information, including (the distance between sensors, the time difference between the latest sensor sampling data). In addition, based on expert experience, set the output of the network, including (alarm number, alarm location, alarm time, alarm type, alarm level, alarm processing time), among which (alarm number, alarm location, alarm time, alarm type) is obtained according to the actual alarm data of the sensor, the alarm level is set according to expert experience, and the alarm processing time is obtained by subtracting a random number from the longest processing time required by the system, and the range of random numbers is set according to expert experience. Then divide the dataset into training set, test set and validation set.

[0086] The designed network is trained on the training set, and the internal parameters of the model are updated based on the loss function to obtain the trained network model. The network model is then tested and verified through the test set and validation set.

[0087] like Figure 4 As shown, the present disclosure provides an electronic device, including a processor 401, a communication interface 402, a memory 403 and a communication bus 404, wherein the processor 401, the communication interface 402 and the memory 403 communicate with each other through the communication bus 404;

[0088] Memory 403, storing computer programs;

[0089] The processor 401 is configured to implement the above method when executing the computer program stored in the memory 403 .

[0090] The present disclosure provides a computer-readable storage medium storing a computer program, wherein the computer program implements the above method when executed by a processor.

[0091] The computer-readable storage medium may be included in the device / apparatus described in the above embodiment; or it may exist independently without being assembled into the device / apparatus. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0092] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0093] In order to enable those skilled in the art to better understand the present disclosure, the principles of the present disclosure are described as follows in conjunction with the accompanying drawings:

[0094] In the existing methods, the means of perceiving and monitoring the safety risks of train operation are single, and the safety risk judgment mainly comes from a single source, lacks mutual cross-checking, has high risk false alarm and missed alarm rates, and needs to rely on manual judgment of event risks. The holographic perception system designed in this disclosure intersects and integrates multi-source and multi-dimensional perception information, and can greatly improve the reliability of risk warning through safety strategies.

[0095] The present invention proposes a graph neural network prediction and classification algorithm for asynchronous spatiotemporal correlation analysis. By processing edge feature information, the position and time relationship between data with the closest sensor sampling time is obtained. Without the need for sensor synchronization, the asynchronous spatiotemporal relationship is obtained, the correlation between sensor data is obtained, and the risk level of an event can be judged.

[0096] By adopting the deep mining technology of perception data and building an integrated three-dimensional holographic perception system and platform, multi-source and multi-dimensional perception information is intersected and integrated, and through cross-checking, the safety situation deduction and risk identification technology of heavy-load train group operation is realized.

[0097] System architecture:

[0098] The present invention discloses a holographic perception system for the operation control of heavy-loaded train groups. The data analysis, decision-making and other functions of the holographic perception system are all supported by a large amount of perception information. Based on the access of massive perception data, the representativeness and extensiveness of the perception data are enhanced to ensure the accuracy and comprehensiveness of information mining. At the same time, through the deep mining and digital twin of multi-level, multi-scale and multi-modal complex data based on the holographic perception network, the safety situation perception and deduction information of each factor is output to improve the anti-interference ability of the perception data.

[0099] The system adopts a two-level architecture of on-site and holographic perception center:

[0100] On-site: It consists of a series of perception monitoring terminals to achieve high-quality collection of basic perception data and interface with the holographic perception center, including infrastructure perception terminals, operating environment perception terminals, contact network foreign body monitoring, rail break monitoring terminals, equipment monitoring, etc.

[0101] Holographic Perception Center: Analyze and process the perception data. Provide data support services in the data analysis stage, including data extraction, data conversion, data inspection, data storage, etc., to complete data aggregation and intersection; provide train operation risk warning, train operation safety situation deduction, data deep mining and digital twin services, data multi-dimensional analysis, data visualization, data reporting, etc. in the data processing stage, to realize the heavy-load train group operation safety situation deduction and risk warning based on perception data.

[0102] Hardware platform structure:

[0103] The hardware platform of the holographic perception system adopts a secure computer platform, including a power supply unit, a computing unit, an internal network unit, a communication interface unit, a maintenance management unit and a storage unit. The system hardware structure is as follows: Figure 5 shown.

[0104] 1) Safe computer platform: The holographic perception system uses the DS6-80 safe computer platform independently developed by the company, providing a two-by-two safety architecture based on task-level synchronization. The safe computer platform has been independently certified by the corresponding project at SIL4 level.

[0105] 2) Power supply unit: Provides stable and reliable DC power supply for other devices in the holographic perception system (the secure computing unit itself contains a power supply module). Its input is AC220, the output is DC24V, and it is equipped with UPS as a backup power supply device.

[0106] 3) Operation unit: The operation unit is responsible for the logic operation function. The holographic perception system contains multiple operation units, each of which is composed of two symmetrical CPU modules. Each CPU module is a fully functional embedded minimum system, which integrates CPU, memory, CF card storage, Ethernet communication and other functions. The CPUs use Ethernet ports to exchange data, thereby realizing the cross-comparison function of dual CPU data within the system and the synchronous comparison function of data between systems.

[0107] 4) Internal network unit: It undertakes the switching function of the internal network of the holographic perception system and is implemented by multiple switches.

[0108] 5) Communication interface unit: used to realize the communication between the host unit of the holographic perception system and external devices, and is a non-security-related unit.

[0109] 6) Maintenance management unit: includes independent monitoring and maintenance software and hardware. Its main function is to monitor and record the real-time working status and alarm information of the system. It is a non-safety-related unit.

[0110] 7) Storage unit: provides data storage services and runs as a relatively independent process. It not only provides storage services for external businesses, including perception information and data information collected by signal equipment, but also provides access services for the log data generated by the holographic perception system itself.

[0111] The system functions mainly include business functions and management functions:

[0112] (1) Business functions,

[0113] 1) Perception data reception,

[0114] The system can receive monitoring data, alarm data, etc. from the perception monitoring terminal;

[0115] 2) Perception data filtering,

[0116] The system can process the information received from the sensing monitoring terminal, mark abnormal data such as those that are not within the measurement range and garbled as suspicious data, and prompt an abnormal data alarm;

[0117] 3) Perception information display,

[0118] The holographic sensing system can display various monitoring information of the sensing monitoring points in the form of text, graphics, etc., and the specific display content can be dynamically configured according to user needs;

[0119] 4) Perception information storage and backup,

[0120] The holographic perception system can store and back up the monitoring information uploaded by the perception monitoring terminal;

[0121] 5) Perception information alarm,

[0122] When the perception monitoring terminal uploads the alarm information, the holographic perception center can issue an alarm prompt and provide the alarm information to the CTC and the on-site data center;

[0123] 6) The perception information alarm is lifted,

[0124] The holographic sensing system supports manual confirmation to release the alarm;

[0125] 7) Perception historical data query,

[0126] The holographic perception system can provide real-time, specified time period perception monitoring information and alarm information in the form of charts, curves, etc., and provide the output and printing functions of query result files;

[0127] 8) Signal equipment collects data for storage,

[0128] The holographic perception system can store the data collected by the device;

[0129] 9) Processing of sensor equipment monitoring data,

[0130] The holographic perception system has the function of analyzing and processing the monitoring data of a single perception terminal and generating event warnings and alarms.

[0131] 10) Multi-source fusion perception data mining,

[0132] The holographic perception system has the function of matching the data correlation of monitoring data from different perception terminals, and performs multi-source data fusion, situation deduction, risk prediction and early warning protection for perception data with correlation.

[0133] (2) Business functions,

[0134] 1) Configuration management,

[0135] The holographic perception system can provide functions such as alarm parameter configuration, terminal function configuration, and user role configuration;

[0136] 2) Fault information management,

[0137] The holographic perception system can receive fault information from the perception monitoring terminal and classify, count and analyze the faults;

[0138] 3) Performance management,

[0139] The holographic perception system can provide real-time monitoring of server CPU usage, I / O performance, memory usage, and storage device usage;

[0140] 4) User authentication management,

[0141] When users log in to the holographic perception system, they should use password authentication (specific scenarios need to be listed);

[0142] 5) Permission management,

[0143] The holographic perception system can configure corresponding software function module permissions and data information item permissions according to different monitoring terminals;

[0144] 6) Log management,

[0145] The holographic perception system log should include operation log (operation traces), quality log (error log, alarm log, abnormal data and other information records) and login log (user, time and other information records of each login to the system);

[0146] 7) Network status management,

[0147] The holographic perception system can monitor the network connection status of the interface device in real time, and generate network alarm information in real time when the network connection is abnormal;

[0148] 8) Time synchronization,

[0149] The holographic perception system has clock synchronization function.

[0150] Data processing,

[0151] When processing actual perception data, the sampling frequencies of different perception terminals are inconsistent, resulting in asynchronous sensor data. At the same time, when a sensor generates an alarm, the information of other related sensors may not be collected synchronously. It is necessary to use historical sensor data to determine the relevance of the sensor information and the degree of event risk.

[0152] Therefore, for the early warning analysis and processing of multi-source perception data, the present disclosure proposes a graph neural network prediction and classification algorithm for asynchronous spatiotemporal correlation analysis, which can process multi-source asynchronous sensor data and alarm data to determine the risk level of the event.

[0153] The nodes of the graph G that defines the perception data processing model are V, which contain sensor data (number, location, sampling time, sensor type, sensor data, whether to alarm, alarm level, and estimated alarm processing time). The edges of the graph G are E, which contain (the distance between sensors, the time difference between the latest sensor sampling data), and the proximity matrix A of the graph G represents the connection relationship between the edges of the nodes. The present disclosure obtains the position and time relationship between the data with the closest sensor sampling time by processing the edge E, obtains the asynchronous spatiotemporal relationship without the need for sensor synchronization, and obtains the correlation between the sensor data.

[0154] Then the graph G is (V, E, A), where the nodes are N×F matrices, N is the number of sensor nodes, and in this disclosure, F=8 is the number of sensor data features. E is an N×N×P matrix, and in this disclosure, P=2 is the number of edge data features. ij ∈R(i=1,2,...,R;j=1,2,...,F) represents the jth channel of the feature vector of the i-th sensor node in the graph G, E ijp ∈R(i,j=1,2,...,R;p=1,2,...,P) represents the pth channel of the feature vector of the edge between the i-th sensor node and the j-th sensor node in the graph G.

[0155] In the graph convolutional neural network in this disclosure, the information propagation model of the sensor node is:

[0156]

[0157] In the formula, represents the output feature of sensor node i in the (l+1)th layer of the graph convolutional neural network, and represents the output of sensor node i and sensor node j at layer l, σ is the activation function, R is the relationship set between sensor nodes, is the set of sensor nodes with a relationship of r among the one-hop neighboring nodes of sensor node i. In the present disclosure, relationship r means that there is an association relationship. i,r The value is The regularization constant of is the weight parameter matrix of the adjacent sensor nodes with relationship r in the l-th layer network, is the weight parameter matrix of sensor node i in the l-th layer network.

[0158] The propagation model of side information is:

[0159]

[0160] In the formula, E j→i represents the eigenvector of edge (j→i), represents the feature vector of sensor node i in the tth graph convolution layer, M represents the information transfer function, U represents the sensor node information update function, Indicates an information collection operation. Specifically, one implementation form of the above formula may be:

[0161]

[0162] Where N i is the neighboring sensor node of sensor node i, α ij is the Softmax regularization function, W v and W ∈ are the node weight parameter matrix and edge weight parameter matrix respectively, ρ ij is the edge eigenvector, h i is the node feature vector, ‖ is the concatenation operation, It is the feature vector output after aggregating the edge feature information.

[0163] In the present disclosure, an L-layer graph convolutional neural network can be constructed as needed to After L iterations, LSTM and MLP layers can be stacked to After processing, the final output O is obtained, which is (alarm number, alarm location, alarm time, alarm type, alarm level, alarm processing time), where the alarm numbers are arranged in the order of alarms, the alarm location and alarm time are the sensor node location of the alarm and the sampling time of the alarm respectively, the alarm level is the estimated level after processing by the graph convolutional neural network, and the alarm processing time is the time required for alarm processing estimated by the graph convolutional neural network.

[0164] Although the present disclosure has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure.

Claims

1. A heavy-load train group operation control perception method, characterized in that: The method comprises: Collect train group operation status information through different sensors; The state information is converted into a state graph, where nodes represent sensor sampling points, edges represent the time difference between the latest sensor sampling points, and the proximity matrix represents the connection relationship between the edges of the nodes. A graph convolutional neural network is established based on the state diagram to estimate the alarm level and alarm processing time of the train group operation status; Train the graph convolutional neural network according to the historical state graph to obtain a trained graph convolutional neural network; The state diagram collected in real time is input into the graph convolutional neural network to perform operation control perception and supervision on the operation status of the train group.

2. A heavy-load train group operation control perception method according to claim 1, characterized in that: Convert state information into a state diagram, including: G represents the state graph, and the node of graph G is V, which represents the data of the sensor, including: sampling number, sampling location, sampling time, sensor type, sensor data, and whether to alarm; The edge of graph G is E, which represents the time difference between the latest sampling data of the sensor; The proximity matrix A of the graph G represents the connection relationship between the edges of the nodes.

3. A heavy-load train group operation control perception method according to claim 2, characterized in that: The graph G is (V, E, A), where the node V is an N×F matrix, N is the number of sensor nodes, and F is the number of sensor data features; the edge E is an N×N×P matrix, and P is the number of edge data features.

4. A heavy-load train group operation control perception method according to claim 1, characterized in that: The information propagation model of sensor nodes in the graph convolutional neural network is: In the formula, represents the output feature of sensor node i in the l+1th layer of the graph convolutional neural network, and represents the output of sensor node i and sensor node j at layer l, σ is the activation function, R is the relationship set between sensor nodes, is the set of sensor nodes with a relationship of r among the one-hop neighboring nodes of sensor node i, where relationship r means there is an association relationship, c i,r The value is The regularization constant of is the weight parameter matrix of the adjacent sensor nodes with relationship r in the l-th layer network, is the weight parameter matrix of sensor node i in the l-th layer network.

5. A heavy-load train group operation control perception method according to claim 1, characterized in that: The propagation model of side information in graph convolutional neural networks is: Where N i is the neighboring sensor node of sensor node i, is the Softmax regularization function, W v and W ∈ are the node weight parameter matrix and edge weight parameter matrix respectively, is the edge eigenvector, h i is the node feature vector, ‖ is the concatenation operation, It is the feature vector output after aggregating the edge feature information.

6. A heavy-load train group operation control perception method according to claim 1, characterized in that: Establish a graph convolutional neural network based on the state diagram, including: The graph convolutional neural network includes L graph convolutional network layers GCN and multi-layer perceptron layers MLP. The activation functions of the GCN layer and LSTM layer are both ReLU functions, and the activation function of the MLP layer is tanh activation function. The output of the graph convolutional neural network is O = (A i ,P i ,t i ,S i ,M i ,T i ), A i ,P i ,t i ,S i ,M i ,T i They represent the alarm number, alarm location, alarm time, alarm type, alarm level and alarm processing time respectively. The alarm numbers are arranged in the order of alarms. The alarm location and alarm time are the sensor node location of the alarm and the sampling time of the alarm respectively. The alarm level is the estimated level after processing by the graph convolutional neural network. The alarm processing time is the time required for alarm processing estimated by the graph convolutional neural network.

7. A heavy-load train group operation control perception method according to claim 6, characterized in that: The loss function of the graph convolutional neural network is as follows: In the formula, θ(S i ) is the weight of the alarm type set based on experience or iterative learning.

8. A heavy-load train group operation control perception system, characterized in that: include: Data collection unit, graph conversion unit, model building unit and perception supervision unit; A data acquisition unit, used to collect train group operation status information through different sensors; A graph conversion unit is used to convert the state information into a state graph, where the nodes in the graph represent the sampling points of the sensors, the edges in the graph represent the time differences between the latest sampling points of the sensors, and the proximity matrix A in the graph represents the connection relationship between the edges of the nodes; A model building unit, used to build a graph convolutional neural network based on the state diagram, and used to estimate the alarm level and alarm processing time of the train group operation status; The model building unit is further used to train the graph convolutional neural network according to the historical state graph to obtain a trained graph convolutional neural network; The perception supervision unit is used to input the real-time collected state diagram into the graph convolutional neural network to perform operation control perception supervision on the operation status of the train group.

9. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; a memory storing a computer program; The processor is used to implement a heavy-load train group operation control perception method as described in any one of claims 1-7 when executing a computer program stored in a memory.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, a heavy-load train group operation control perception method described in any one of claims 1-7 is implemented.

Citation Information

Patent Citations

  • Suspended maglev train service state acquisition and data fusion method and system

    CN109109909A

  • Train autonomous scheduling deep reinforcement learning method and module

    CN111369181A

  • Subway passenger flow volume prediction method and system based on space-time diagram ordinary differential equation

    CN117592602A

  • Industrial process anomaly detection method based on graph neural network

    CN117724452A

  • Real-time metering data processing platform

    CN117725537A

Cited By

  • Medical record data quality monitoring method and system based on knowledge base

    CN119418845A