Train communication network fault diagnosis method and device of multi-view hypergraph convolutional network

Abstract

The invention discloses a train communication network fault diagnosis method and device based on a multi-view hypergraph convolutional network, and the method comprises the following steps: firstly, collecting various high-order heterogeneous fault data of a train communication network under multiple working conditions, multiple time periods and multiple communication modes, then carrying out the analysis and preprocessing of the collected original data, and carrying out the fault diagnosis of the train communication network. The method comprises the steps of preprocessing data, dividing the preprocessed data into a training set and a test set according to a certain proportion, then fully mining potential relations among nodes from multiple dimensions through multi-view joint modeling, then building a train communication network fault diagnosis MVHGNN model, and finally performing training and parameter adjustment on the MVHGNN model through the training set. And sending the test set into the trained MVHGNN model to identify various network faults. According to the method, a complex heterogeneous high-order structure in the train communication network is analyzed through multiple views, the classification capability of the model on various types of space-time faults is improved, and intelligent diagnosis of the train communication network can be realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent diagnostic technology for train communication networks, and in particular to a method and apparatus for diagnosing faults in train communication networks using multi-view hypergraph convolutional networks. Background Technology

[0002] As the core information exchange platform of the rail transit system, the health status of the train communication network directly affects the safe operation and control efficiency of the train. Existing fault diagnosis methods mainly rely on single-view signal feature analysis or traditional machine learning algorithms, which do not adequately consider the complex functional coordination relationships between nodes, differences in operational behavior, and spatial topological characteristics.

[0003] Meanwhile, with the increasing scale and complexity of train communication networks, nodes exhibit heterogeneity and high-order correlation characteristics. Existing methods have limitations in uncovering potential connections between nodes, especially when dealing with minority-class faults and marginal samples, resulting in low diagnostic accuracy. Furthermore, traditional methods struggle to simultaneously integrate multi-dimensional information for joint modeling, limiting their ability to perceive fault characteristics.

[0004] Therefore, existing technologies cannot meet the requirements of high-precision and high-reliability fault diagnosis for train communication networks. There is an urgent need for a fault diagnosis method for train communication networks based on multi-view hypergraph convolution to effectively perceive high-order correlation features between heterogeneous nodes and improve the diagnostic accuracy of minority and marginal samples. Summary of the Invention

[0005] The purpose of this invention is to provide a method and apparatus for diagnosing train communication network faults using multi-view hypergraph convolutional networks, addressing the aforementioned problems. This solves the issues of existing methods' difficulty in perceiving high-order correlation features between heterogeneous nodes and their low accuracy in diagnosing minority and marginal samples.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: According to one aspect of the present invention, a method and apparatus for fault diagnosis of train communication networks using multi-view hypergraph convolutional networks are provided, comprising the following steps: S01. Collect various high-order heterogeneous fault data of the train communication network under multiple operating conditions, multiple time periods, and multiple communication modes; S02. The collected high-order heterogeneous data is parsed and preprocessed, and the preprocessed data is divided into training set and test set according to a certain ratio. S03. Modeling through multi-view joint modeling; S04: Build a train communication network fault diagnosis MVHGNN model; S05: Train and adjust the parameters of the MVHGNN model using the training set, and then feed the test set into the trained MVHGNN model to identify various network faults.

[0007] Preferably, in step S02, the specific steps for parsing and preprocessing the original data are as follows: S0201, Data Parsing: Parse the raw data collected from the train communication network, extract node identifiers, communication message information and performance indicators, and convert them into a structured data format. S0202, Data Cleaning and Feature Standardization: The parsed data is deduplicated, denoised, and missing values ​​are filled in. Abnormal data is removed to ensure data integrity. Numerical features are normalized or standardized, and categorical features are encoded. S0203, Time Series Slicing: The data is divided into multiple time series samples according to a preset time window and sliding step size. Slicing is processed by the following formula: in, Indicates the window length; Indicates the sliding step size; Indicates the first 1 sequence sample.

[0008] The final sample set is: in, A set representing time-series samples; S0204. Sample labeling: Label the time series samples according to the actual operating status or fault injection results.

[0009] Preferably, in step S2, the ratio of the training set to the test set is 7:3.

[0010] Preferably, in step S03, the multi-view joint modeling includes the following steps: S0301. Multi-view construction: Based on the functional attributes, operational behavior characteristics, and spatial topological relationships of nodes in the train communication network, establish corresponding view structures and generate hyperedge sets. S0302, Hypergraph Inclusion Matrix Generation: The view structure is transformed into a node-hyperedge inclusion matrix to form a multi-view hypergraph structure representation; wherein the hypergraph inclusion matrix... The assignment rules satisfy the following formula: in, Indicates the first In each sample, the node The corresponding hyperedge contains its Top-k similar nodes. .

[0011] S0303, Multi-view Joint Modeling: Input the hypergraph association matrix generated under different views into a multi-branch hypergraph convolutional network to jointly represent the potential relationships between nodes.

[0012] Preferably, in step S04, establishing the train communication network fault diagnosis MVHGNN model specifically includes the following steps: S0401. Construction of Multi-view Hypergraph Convolution Module: Construct a multi-view hypergraph convolution module, which is used to perform convolution operations on hypergraph structures from different perspectives to extract high-order correlation features between nodes. S0402, Construction of Temporal Information Extraction Network: Construct a temporal information extraction network, which is used to perform temporal modeling on the fused feature sequence and obtain the dynamic features of node state changes over time; S0403. Improved loss function: Set the LDAM-FL combined loss function, which is used to handle the class imbalance problem and optimize the classification boundary during model training.

[0013] Preferably, the LDAM-FL combined loss function is: in, The corrected probability is the predicted probability of the correct class. , For the number of categories, The parameter is used to control the degree to which the model pays attention to difficult-to-classify samples.

[0014] Preferably, a fault diagnosis device for train communication network based on multi-view hypergraph convolutional network includes a train communication network fault data acquisition module, a data processing module, a multi-view modeling module, a fault diagnosis module, and a diagnosis display and guidance module; The train communication network fault data acquisition module is used to collect operating data and fault data of the train communication network under multiple working conditions, multiple time periods, and multiple communication modes. The data processing module is used to parse, clean and preprocess the collected raw data, and divide the sorted data into training set and test set according to a preset ratio; The multi-view modeling module is used to model the relationships between nodes from multiple dimensions, construct a multi-view hypergraph structure, and generate a node-hyperedge association matrix. The fault diagnosis module is used to build and run a fault diagnosis model for the train communication network. The diagnostic display and guidance module is used to visualize the results output by the fault diagnosis module and generate corresponding operation and maintenance guidance information based on the diagnostic results.

[0015] Preferably, the train communication network data acquisition module includes a managed three-layer switch, a two-layer switch, a network tester, a mobile client, and an AC / DC transformer.

[0016] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention integrates the high-order and heterogeneous nature of train communication network nodes, and introduces spatial and temporal elements into the data, thereby increasing the comprehensiveness and complementarity of information.

[0017] 2. This invention enhances the model's feature extraction capability in spatial information by constructing a multi-view hypergraph convolution, and deeply explores the high-order heterogeneous correlations between nodes.

[0018] 3. This invention sets up a temporal feature extraction network that integrates one-dimensional convolution and bidirectional attention LSTM. The one-dimensional convolution extracts short-term sporadic temporal patterns between nodes, while the bidirectional attention LSTM perceives global temporal information by introducing global context variables and attention mechanisms.

[0019] 4. This invention constructs the LDAM-FL loss function, which enhances the ability to pay attention to minority classes and difficult-to-classify samples while strengthening the adjustment of class boundaries, effectively improving the generalization performance of the model under imbalanced data conditions. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the fault diagnosis method of the present invention.

[0021] Figure 2 This is a schematic diagram of the topology of the train communication network of the present invention.

[0022] Figure 3 This is a schematic diagram of the high-order heterogeneity of the train communication network nodes of the present invention.

[0023] Figure 4 This is a schematic diagram of the train communication network fault data acquisition process of the present invention.

[0024] Figure 5 This is a schematic diagram of the train communication network fault diagnosis process of the present invention.

[0025] Figure 6 This is a structural block diagram of the train communication network fault diagnosis device of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and preferred embodiments. However, it should be noted that many details listed in the specification are merely to provide the reader with a thorough understanding of one or more aspects of the invention, and these aspects of the invention can be achieved even without these specific details.

[0027] Please see Figure 1-6 This invention provides a method and apparatus for fault diagnosis of train communication networks using multi-view hypergraph convolutional networks. The technical solution is as follows: like Figure 1 As shown, a fault diagnosis method for train communication networks using multi-view hypergraph convolutional networks is proposed. The method includes: S01. Collect high-order heterogeneous fault data of the train communication network under multiple operating conditions, time periods, and communication modes to provide comprehensive node characteristics and network status information for fault diagnosis. Specifically, data is collected from each node of the train communication network under different operating conditions, multiple time periods, and different communication modes. The collected data includes inter-node message transmission time, payload length, packet loss, retransmission count, communication delay, and SYN / ACK / RST message statistics. The data is recorded in real-time using the packet capture tool Wireshark to generate raw communication PCAP files.

[0028] S02. The collected raw data of the train communication network is parsed and preprocessed to meet the input requirements of the fault diagnosis model, and the processed data is divided into training set and test set for model training and verification.

[0029] The specific steps are as follows: S0201. Perform protocol parsing on the collected raw communication data to identify the data structure of different communication layers. In this embodiment, the raw PCAP file is parsed using the Scapy library in Python to identify the protocol fields of each layer. Key features extracted include timestamps, node addresses, message types, payload lengths, and other key fields. Feature information related to fault diagnosis is separated from the parsed data fields. This feature information includes communication latency, packet loss rate, retransmission count, number of abnormal messages, etc., and redundant information unrelated to the fault is removed.

[0030] S0202. Clean the extracted features, removing incomplete, redundant, or abnormal data packets, and correcting missing values ​​and erroneous records to ensure the validity and accuracy of the input data. Normalize or standardize feature data of different dimensions to give them a uniform numerical range, avoiding bias in model training caused by differences in dimensions. Normalization is performed by the following formula: in, Indicates the first The first sample One feature column, Represents the normalized eigenvalues. Indicates the first All feature values ​​of the column.

[0031] S0203. Based on the operating characteristics and diagnostic requirements of the train communication network, a fixed-length time window parameter is set. In this embodiment, the time window is 7 units, used to control the time range covered by each sample. A sliding step size is set between adjacent time windows to ensure the continuity of data in the time dimension and the differences between samples. In this embodiment, the sliding step size is 2 units. According to the preset time window and sliding step size, continuous communication data is sliced ​​to generate multiple time series samples. Slicing is performed using the following formula: in, Indicates the window length; Indicates the sliding step size; Indicates the first 1 sequence sample.

[0032] The final sample set satisfies:

[0033] S0204. Based on the needs of model training and evaluation, a preset ratio of training set to test set is established to ensure that the model can be fully trained while effectively verifying its generalization performance. In this embodiment, the obtained time-series samples are randomly divided into training set and test set in a ratio of 7:3, that is, the training set accounts for 70% and the test set accounts for 30%, for model training and verification.

[0034] S03. Construct a multi-view joint modeling mechanism to establish a potential connection model between nodes from multiple dimensions such as function, behavior and space, so as to enhance the ability to identify and judge the fault characteristics of heterogeneous nodes.

[0035] Specifically, building a multi-view joint modeling mechanism includes the following steps: S0301. Based on the characteristics of the train communication network, perspectives are constructed from different dimensions such as functional coordination, operational behavior, and spatial topology to comprehensively characterize the potential connections between nodes.

[0036] S0302. From each perspective, the original features of the nodes are mapped to corresponding feature representations to capture the relational features of the nodes in that dimension. Based on the similarity or interaction relationships between nodes, a corresponding set of hyperedges is constructed to form a multi-view hypergraph structure. The hypergraph association matrix... The assignment rules satisfy the following formula: in, Indicates the first In each sample, the node The corresponding hyperedge contains its Top-k similar nodes. .

[0037] S0303. Joint modeling and weighted fusion of the hypergraph formed from multiple views are performed to integrate the feature information of different views, thereby improving the expressive ability of heterogeneous node fault features. The relationship between the branches of each view satisfies: in, For the first The weight coefficients corresponding to each perspective.

[0038] S04. Construct a train communication network fault diagnosis MVHGNN model. This model is based on a hypergraph convolutional neural network. It achieves comprehensive modeling of the relationship between nodes through multi-view feature weighted fusion and combines an improved loss function to deal with class imbalance and difficult-to-classify samples, thereby improving the ability to identify minority classes and marginal samples.

[0039] Specifically, establishing a train communication network fault diagnosis MVHGNN model includes the following steps: S0401. In each view, a hypergraph convolutional neural network is selected as the spatial information extraction network to extract higher-order heterogeneous relationships between nodes, wherein the hypergraph convolution process satisfies: in, It is a learnable weight matrix. It is a non-linear activation function.

[0040] S0402. A one-dimensional convolutional neural network and a bidirectional attention long short-term memory network BiALSTM (Bi-Attentive LSTM) are used to extract information about the fault in the time dimension. The one-dimensional convolution extracts the local patterns in the sudden fault on the time axis. BiALSTM uses a bidirectional LSTM structure to capture long-term temporal dependencies from both directions. At the same time, an attention mechanism is introduced to focus on key moments in periodic changes, so as to achieve accurate perception of periodic anomalies.

[0041] S0403. To address the imbalanced fault data problem in train communication networks, a combined LDAM-FL loss function is used instead of the ordinary cross-entropy loss function. This loss function combines the LDAM and FL functions, improving the model's diagnostic performance from the perspectives of class discriminability and sample difficulty, thereby enhancing its ability to identify minority and hard-to-classify samples. The final loss function is expressed as follows: in, The corrected probability is the predicted probability of the correct class. , For the number of categories, The parameter is used to control the degree to which the model pays attention to difficult-to-classify samples.

[0042] S05. Input high-order heterogeneous fault data with time-space characteristics into the MVHGNN model for training and testing. Use the training set to learn and optimize the model parameters, and use the test set to identify different types of train communication network faults.

[0043] Specifically, the weights of the MVHGNN model are initialized, iterative training is performed using the training set, and hyperparameters (learning rate, number of layers, node feature dimensions, etc.) are adjusted. Test set data is then input into the trained MVHGNN model for forward inference, outputting the node fault category for each time window. If the MVHGNN model can identify different fault types on the test set, it can be used for fault diagnosis in train communication networks, exhibiting good accuracy, stability, and F1 score. The F1 score expression is as follows: in, TN is the number of samples where the predicted value is 1 and the actual value is also 1; TN refers to the number of samples where the predicted value is 0 and the actual value is also 0. It is the number of samples where the predicted value is 1 but the actual value is 0; This is the number of samples where the predicted value is 0 but the actual value is 1; therefore, The maximum score is 1, and the minimum score is 0. In this case, it is specified that the model's fault diagnosis accuracy and When all scores are greater than 99%, the MVHGNN model is considered to have good fault diagnosis performance.

[0044] It should be noted that the topology of the train communication network is as follows: Figure 2As shown, according to the IEC 61375 series standards and typical railway applications, TCN is mainly divided into three layers: train-level network, vehicle-level network, and equipment-level network. The train-level network follows the IEC 61375-2-5 backbone network standard. The vehicle-level network follows the IEC 61375-3-4 marshalling network standard. The train backbone network consists of three layers of switches connected in series to achieve packet forwarding between different network segments. In the train marshalling network, a ring structure is used to form a network segment, achieving packet forwarding at the data link layer. Each layer 2 switch connects to multiple terminal devices, each terminal device representing a communication unit in the actual train. Here, ETBN represents an Ethernet train backbone network node, and ECNN represents an Ethernet marshalling network node.

[0045] like Figure 3 As shown, Figure 3 This diagram illustrates the high-order heterogeneity of nodes in a train communication network. Nodes in a train communication network exhibit heterogeneity, with fundamental differences in function, protocol, data, and data transmission methods. According to the IEC 61375-3-4 standard, train communication networks have five different communication data types, each with different communication modes, call and response mechanisms, frame formats, and transmission cycles. Node 1 and Node 3 transmit process data every 10ms via UDP, while Node 2 and Node 3 transmit message data intermittently via TCP, reflecting the differences in data types and transmission methods between nodes. Furthermore, although Node 1 and Node 2 are within the same network segment, they perform different functions in the system: Node 1 belongs to the communication and data synchronization module, primarily responsible for complex train operation log synchronization; Node 2 belongs to the carriage environment control module, used for regulating and controlling the air conditioning temperature.

[0046] like Figure 4 As shown, the process for obtaining fault data from the train communication network includes the following steps: Based on the actual structure of the train communication network, various nodes in the network topology are configured, including the Central Control Unit (CCU), Train Communication Unit (TCU), Fire Detection Unit (FDU), and Heating, Ventilation and Air Conditioning Controller (HVAC). By configuring the functional attributes, communication interfaces, and location distribution of the nodes, a network topology model that reflects the actual operating environment of the train is formed.

[0047] In the constructed topology, various service types are set according to the train operation scenario, such as control command transmission, status monitoring messages, passenger information dissemination, and environmental control data interaction. By configuring different communication services, the diverse interaction processes between nodes in real operation are simulated.

[0048] Based on the topology and service configuration, various typical failure modes are introduced for simulation, including abnormal link latency, packet loss, congestion, out-of-order delivery, and node failure. By adjusting the location, time, and intensity of the failures, high-order heterogeneous failure scenarios covering multiple operating conditions and modes are generated.

[0049] The network described above was run in the experiment to collect communication data and status information between nodes, including link latency, packet loss rate, transmission rate, and message interaction characteristics. This resulted in a raw dataset that can be used for subsequent training and testing of fault diagnosis models.

[0050] like Figure 5 As shown, Figure 5 This is a schematic diagram of the fault diagnosis process for train communication networks. It collects fault data from the train communication network and uses a pre-trained MVHGNN model to achieve intelligent diagnosis. Finally, it issues alarm information based on the model's classification results to guide professionals to carry out maintenance.

[0051] To verify the effectiveness of the train communication network fault diagnosis method of the present invention, comparative experiments were conducted with deep learning and spatiotemporal graph convolution methods. The deep learning and spatiotemporal graph convolution methods were CTNet, GCN-LSTM, MHA-1DCNN, MSRNet-GLAM, Res-BiGRU, TCN-MoH, TV-Net, and WKGNN, respectively. Table 1 shows the performance of each model in the comparative experiments.

[0052] Table 1 Comparison of experimental performance As can be seen from the experimental results in Table 1, the MVHGNN model of this invention exhibits the best classification performance in multiple repeated experiments compared to other baseline models, with an average test accuracy of 99.41%. Other baseline models show poor overall performance on the dataset; except for the MSRNet-GLAM model, the remaining models are all below 99%, significantly lagging behind the model of this invention. Furthermore, considering the fluctuation of accuracy over three rounds, the maximum fluctuation amplitude of the model of this invention is only 0.13%, far lower than methods such as Res-BiGRU, MHA-1DCNN, and TV-Net. The experimental results in Table 1 fully demonstrate that the MVHGNN model of this invention outperforms existing models in both average accuracy and test accuracy fluctuation, exhibiting excellent fault classification capabilities.

[0053] Furthermore, in order to verify the effectiveness of the multi-view structure MV and the improved loss function LDAMF, an ablation experiment was conducted. The invention was verified through train communication network fault data. Table 2 verifies the performance of each model under the ablation experiment.

[0054] Table 2 Comparison of ablation test performance The experimental results are shown in Table 2. Compared with the complete model, each variant showed a certain degree of decline in performance across different evaluation metrics. The most significant performance degradation occurred after removing the multi-view structure, with all four metrics decreasing by approximately 0.70%, indicating that this module played a crucial role in capturing higher-order heterogeneity among nodes. Furthermore, removing the LDAMF loss function (using the ordinary cross-entropy loss function) also resulted in a significant performance decline, further validating the effectiveness of this function in handling imbalanced samples.

[0055] like Figure 6 As shown, the present invention also discloses a train communication network fault diagnosis device based on multi-scale spatiotemporal graph convolution, including a train communication network fault data acquisition module 601, a data processing module 602, a multi-view modeling module 603, a fault diagnosis module 604, and a diagnosis display and guidance module 605. The train communication network fault data acquisition module 601 connects to the mobile client via an RJ45 network cable. The client is equipped with Wireshark packet capture software to capture communication data in the network and stores the data packets in the form of PCAP files. The data processing module 602 uses Python with the Scapy library on the PC to parse the PCAP file, and counts feature columns for each target node, mainly including the number of packets sent and received, the number of bytes, and the transmission delay. Then, it cleans the statistical feature columns to eliminate outliers in the data, and then randomly divides the data into training set and test set in a 7:3 ratio. The multi-view modeling module 603, based on the processed data, establishes a relationship graph between nodes from three dimensions: functional collaboration, operational behavior, and spatial location, and further transforms it into a multi-view hypergraph association matrix. This module generates the hypergraph structure from various perspectives through a matrix operation program and performs weighted fusion, thereby achieving comprehensive modeling of high-order relationships between nodes. The fault diagnosis module 604 is used to train and optimize the parameters of the fault diagnosis MVHGNN module using the training set, and to test the trained model using the test set to achieve fault classification of the train communication network. Then, it outputs the corresponding predicted labels. The input data is judged according to the value of the predicted label. The label of normal data is 0, the label of IUC interface connection failure is 1, the label of blind zone communication failure (BZC) is 2, the label of multi-protocol bridging failure (MPBC) is 3, the label of multicast message parsing failure (MMP) is 4, the label of broadcast storm (BS) is 5, and the label of quality of service (QoS) is 6.

[0056] The diagnostic display and guidance module 605 is used to visualize diagnostic results. The system will output the identification categories of various faults and generate operation and maintenance guidance suggestions to help maintenance personnel quickly locate the fault location and take targeted measures in actual train operation.

[0057] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for fault diagnosis of train communication networks using multi-view hypergraph convolutional networks, characterized in that, Includes the following steps: S01. Collect various high-order heterogeneous fault data of the train communication network under multiple operating conditions, multiple time periods, and multiple communication modes; S02. The collected high-order heterogeneous data is parsed and preprocessed, and the preprocessed data is divided into training set and test set according to a certain ratio. S03. Modeling through multi-view joint modeling; S04: Build a train communication network fault diagnosis MVHGNN model; S05: Train and adjust the parameters of the MVHGNN model using the training set, and then feed the test set into the trained MVHGNN model to identify various network faults.

2. The fault diagnosis method for train communication networks using multi-view hypergraph convolutional networks according to claim 1, characterized in that: In step S02, the specific steps for parsing and preprocessing the raw data are as follows: S0201, Data Parsing: Parse the raw data collected from the train communication network, extract node identifiers, communication message information and performance indicators, and convert them into a structured data format. S0202, Data Cleaning and Feature Standardization: The parsed data is deduplicated, denoised, and missing values ​​are filled in. Abnormal data is removed. Numerical features are normalized or standardized, and categorical features are encoded. S0203, Time Series Slicing: The data is divided into multiple time series samples according to a preset time window and sliding step size. Slicing is processed by the following formula: in, Indicates the window length; Indicates the sliding step size; Indicates the first 1 sequence sample. The final sample set is: in, A set representing time-series samples; S0204. Sample labeling: Label the time series samples according to the actual operating status or fault injection results.

3. The fault diagnosis method for train communication networks using multi-view hypergraph convolutional networks according to claim 1, characterized in that: In step S2, the ratio of the training set to the test set is 7:

3.

4. The method for fault diagnosis of train communication network using a multi-view hypergraph convolutional network according to claim 1, characterized in that: In step S03, the multi-view joint modeling includes the following steps: S0301. Multi-view construction: Based on the functional attributes, operational behavior characteristics, and spatial topological relationships of nodes in the train communication network, establish corresponding view structures and generate hyperedge sets. S0302, Hypergraph Inclusion Matrix Generation: The view structure is transformed into a node-hyperedge inclusion matrix to form a multi-view hypergraph structure representation; wherein the hypergraph inclusion matrix... The assignment rules satisfy the following formula: in, Indicates the first In each sample, the node The corresponding hyperedge contains its Top-k similar nodes. ; S0303, Multi-view Joint Modeling: Input the hypergraph association matrix generated under different views into a multi-branch hypergraph convolutional network to jointly represent the potential relationships between nodes.

5. The method for fault diagnosis of train communication network using a multi-view hypergraph convolutional network according to claim 1, characterized in that: In step S04, establishing the train communication network fault diagnosis MVHGNN model specifically includes the following steps: S0401. Construction of Multi-view Hypergraph Convolution Module: Construct a multi-view hypergraph convolution module, which is used to perform convolution operations on hypergraph structures from different perspectives to extract high-order correlation features between nodes. S0402, Construction of Temporal Information Extraction Network: Construct a temporal information extraction network, which is used to perform temporal modeling on the fused feature sequence and obtain the dynamic features of node state changes over time; S0403. Improved loss function: Set the LDAM-FL combined loss function, which is used to handle the class imbalance problem and optimize the classification boundary during model training.

6. The fault diagnosis method for train communication networks using multi-view hypergraph convolutional networks according to claim 5, characterized in that: The LDAM-FL combined loss function is: in, The corrected probability is the predicted probability of the correct class. ; Number of categories; The parameter is used to control the degree to which the model pays attention to difficult-to-classify samples.

7. A fault diagnosis device for train communication networks using multi-view hypergraph convolutional networks, applied to the fault diagnosis method for train communication networks using multi-view hypergraph convolutional networks as described in any one of claims 1-6, characterized in that: It includes a train communication network fault data acquisition module, a data processing module, a multi-view modeling module, a fault diagnosis module, and a diagnostic display and guidance module; The train communication network fault data acquisition module is used to collect operating data and fault data of the train communication network under multiple working conditions, multiple time periods, and multiple communication modes. The data processing module is used to parse, clean and preprocess the collected raw data, and divide the sorted data into training set and test set according to a preset ratio; The multi-view modeling module is used to model the relationships between nodes from multiple dimensions, construct a multi-view hypergraph structure, and generate a node-hyperedge association matrix. The fault diagnosis module is used to build and run a fault diagnosis model for the train communication network. The diagnostic display and guidance module is used to visualize the results output by the fault diagnosis module and generate corresponding operation and maintenance guidance information based on the diagnostic results.

8. The fault diagnosis device for train communication network based on multi-view hypergraph convolutional network according to claim 7, characterized in that: The train communication network data acquisition module includes a managed three-layer switch, a two-layer switch, a network tester, a mobile client, and an AC / DC transformer.

Images