Motor train unit operation fault intelligent identification method
By using federated learning technology to collaboratively train the EMU operation image fault recognition system, the problems of decentralized deployment and data privacy protection of EMU operation fault detection systems have been solved, achieving efficient automatic fault recognition and improved accuracy.
Patent Information
- Application Number
- CN202510617505.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-14
- Publication Date
- 2025-10-28
AI Technical Summary
In existing technologies, the decentralized deployment of fault detection systems for high-speed trains results in insufficient data volume, making it difficult to train high-performance deep learning models. Furthermore, data privacy protection challenges hinder data sharing and integration among multiple railway bureaus.
By employing a federated learning approach, and through collaborative training of global parameter servers, railway bureau parameter servers, and EMU depot/station working nodes, intelligent identification of faults in EMU operation images is achieved, while ensuring data privacy and security during model parameter aggregation and updates.
While protecting user data privacy, the system has improved the accuracy of fault detection in high-speed train operation by 11.82%, optimized the performance of the global model, changed the traditional manual review method, and achieved automatic fault identification.
Smart Images
Figure CN120852831A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of railway safety monitoring technology, and in particular to an intelligent identification method for high-speed train operation faults based on federated learning. Background Technology
[0002] To ensure that defects and abnormalities in various components of high-speed trains can be quickly identified and addressed during operation, thereby effectively preventing safety hazards, a dynamic image detection system is currently used. This system captures real-time images of visible components on the bottom and sides of the train for detailed review by inspectors at the monitoring center. However, this method is not only time-consuming and labor-intensive, but also requires a high level of professional skill from the inspectors.
[0003] With the continuous advancement of artificial intelligence technology, especially breakthroughs in deep learning, deep learning-based image-based fault identification technology for high-speed trains is gradually becoming an effective way to solve this problem. However, in actual operating environments, these detection systems are often deployed in different areas of the railway network, facing two major challenges: first, the amount of data at each location is relatively small, making it difficult to independently train a high-performance deep learning model; second, due to network bandwidth limitations, centralized training of these image data is not only inefficient in real-time but also extremely costly.
[0004] Furthermore, the images of high-speed trains in operation contain a large amount of sensitive information, such as detailed data on components like bogies, floor plates, and skirts, which raises privacy concerns. As data security issues become increasingly prominent, relevant laws and regulations have imposed clear requirements on data privacy protection, leading to bottlenecks in both management and technology for data sharing between different railway bureaus.
[0005] Therefore, under the premise of ensuring that local data does not flow out of its region and ensuring the data security and privacy of multiple parties, it is particularly important and urgent to realize the sharing and fusion of EMU operation images from multiple railway bureaus, and then carry out fault identification work based on deep learning. Summary of the Invention
[0006] This invention provides a method for intelligent identification of EMU operation faults based on federated learning, which addresses the shortcomings of existing EMU operation fault dynamic image detection systems in terms of decentralized deployment and data privacy protection. It enables the sharing and fusion of multiple railway bureau EMU operation images to carry out EMU operation image fault identification based on deep learning.
[0007] This invention provides a method for intelligent identification of high-speed train operation faults based on federated learning, comprising the following steps: Step 1: Each train depot / station trains a local model based on the global fault identification model issued by the global parameter server, and uploads the model parameters to the parameter server of the railway bureau to which it belongs. Step 2: The parameter servers of each railway bureau aggregate the model parameters of their respective EMU depots / stations and broadcast the new model parameters to their respective working nodes; Step 3: The main data center parameter server extracts the model parameters of each railway bureau, performs aggregation and update based on the federated averaging method, and broadcasts the updated parameters to each railway bureau and distributes them to the corresponding EMU depots / stations.
[0008] Furthermore, the main data center deploys a global parameter server, each railway bureau deploys a local parameter server, and the EMU depots / stations under the railway bureaus deploy training servers as working nodes, with model parameters transmitted between the EMU depots / stations, railway bureaus, and the main data center.
[0009] Furthermore, step 1 also includes initializing the model parameters before training; Among them, the historical data of the dynamic image detection system for EMU operation faults is synchronized on the main data center server, and the YOLOv model is pre-trained on the bogie and floor related data to obtain the initial parameters of the model; the global fault identification model of the data center is initialized and distributed to each participating railway bureau as the initial model for EMU operation image fault identification, and the number of communication wheels is set.
[0010] Furthermore, step 2 also includes: after the railway bureau receives the model parameters and data volume of the working nodes of its affiliated EMU depots / stations after training, it performs local parameter aggregation through the federated averaging method and sends the aggregated model parameters and data volume to the main data center.
[0011] Furthermore, step 3 also includes: the main data center receives the aggregated model parameters and total data volume from each railway bureau, performs global model aggregation using the federated averaging method, calculates the contribution of each railway bureau using the contribution measurement method, and calculates the updated model parameter values for each railway bureau based on the contribution and sends them to each railway bureau.
[0012] Furthermore, step 3 also includes: each railway bureau receiving the model parameter update value issued by the main data center parameter server, downloading the allocated model parameter update amount, and obtaining the final updated model for each railway bureau in this round.
[0013] Furthermore, step 3, which involves aggregating the global model using the federated averaging method, also includes calculating the aggregated data of the global model using the following formula: ,in, This represents the number of work nodes in a specific railway bureau's EMU depot / station. In order to receive the model parameters trained by the working node of the corresponding EMU depot / station For data volume.
[0014] Furthermore, step 3, which involves calculating the updated model parameters for each railway bureau based on their contribution and sending them to each bureau, also includes calculating the contribution using the following formula: ,in, The amount of data owned by each railway bureau For the number of data categories.
[0015] Compared with existing technologies, this invention applies federated learning to a dynamic image detection system for high-speed train operation faults, changing the traditional manual inspection method and achieving automatic fault identification. Federated learning can optimize and improve the overall model performance while ensuring user data privacy and security, enabling participating entities to achieve a detection accuracy increase of up to 11.82% compared to models trained solely on their own data. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0017] Figure 1 This is a flowchart of the intelligent identification method for high-speed train operation faults based on federated learning provided by the present invention.
[0018] Figure 2 This is a flowchart illustrating the intelligent identification method for EMU operation faults based on federated learning provided in this application embodiment. Detailed Implementation
[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0020] Before providing a detailed description of the embodiments of this application, some of the nouns and terms involved in the embodiments of this application will be explained first. The nouns and terms involved in the embodiments of this application are subject to the following interpretations.
[0021] The YOLOv model is a classic version of the YOLO series of algorithms, and includes three loss terms: Classification loss (cls_loss): Determines whether the model can accurately identify objects in an image and classify them into the correct category; Bounding box loss: measures the difference between the bounding box predicted by the model and the true bounding box; Confidence loss (obj_loss): The degree to which the model predicts the bounding box covers the object.
[0022] The total loss is the weighted sum of the above three losses. By changing the weights, the degree of attention paid to the losses of the three can be adjusted.
[0023] The federal average algorithm is as follows: Let there be a total of Each client machine initializes model parameters on the central server, executing several rounds, with each round selecting at least one and at most one client machine. Several client machines participate in the training. Next, each selected client machine simultaneously trains locally based on the current round of commands issued by the server. Wheel model Train your own model using your own data. The data is then uploaded back to the server. The server aggregates the models collected from each client using a weighted average based on the number of samples from each party, resulting in the next round of models. : in, For the client The number of samples on This represents the total number of samples from all selected clients.
[0024] Figure 1 This is an optional flowchart of the intelligent identification method for EMU operation faults based on federated learning provided in the embodiments of this application. Figure 1 The method described herein, when applied to a terminal, may include, but is not limited to, steps S101 to S103.
[0025] Step S101: Each train depot / station trains a local model based on the global fault identification model issued by the global parameter server, and uploads the model parameters to the parameter server of the railway bureau to which it belongs.
[0026] Specifically, historical data from the dynamic image detection system for EMU operation faults are synchronized on the main data center server of China State Railway Group. The YOLOv model is pre-trained on bogie and floor data to obtain initial model parameters. The global fault identification model at the China State Railway Group end is then initialized and distributed to each participating railway bureau as the initial model for EMU operation image fault identification.
[0027] Furthermore, the model's loss function includes classification loss, localization loss, and target confidence loss, which are weighted and summed to obtain the final total loss, used to guide the model's training and optimization. Where N is the number of detection layers, , , These are the weights for the three types of losses.
[0028] In step S102, after receiving the model parameters and data volume of the working nodes of its affiliated EMU depots / stations, the railway bureau performs local parameter aggregation using the federated averaging method and sends the aggregated model parameters and data volume to the main data center.
[0029] Specifically, each railway bureau's EMU depot / station utilizes its local EMU operation image dataset. Conduct training and set the learning rate. Calculate the gradient The local model updates parameters in this round. The updated values of the model parameters are calculated. And update the model parameter values. and the size of local data Send to the relevant railway bureau.
[0030] Furthermore, each railway bureau receives the model parameters trained by the work nodes of its affiliated EMU depots / stations. and data volume Aggregation of local parameters is performed using the federated averaging method. ,in, q represents the number of working nodes of a certain railway bureau's EMU depot / station, t represents the number of railway bureaus, and t represents the number of communication rounds.
[0031] In step S103, the main data center parameter server extracts the model parameters of each railway bureau, performs aggregation and update based on the federated averaging method, and broadcasts the updated parameters to each railway bureau and distributes them to the corresponding EMU depots / stations.
[0032] Specifically, China State Railway Group receives the aggregated model parameters from various railway bureaus. and total data volume Through the federal average method Perform global model aggregation.
[0033] Furthermore, China State Railway Group Co., Ltd. calculated the contribution of each railway bureau using a contribution measurement method. ,in The amount of data owned by each railway bureau This refers to the number of data categories. Based on contribution, the updated model parameter values obtained by each railway bureau are calculated. Send to the railway bureau .
[0034] Furthermore, the amount of model parameter updates downloaded and allocated by each railway bureau. The final updated models for each railway line in this round were obtained. .
[0035] Furthermore, each railway bureau will distribute the updated model parameters indiscriminately to its affiliated EMU depots / stations for the next round of local working node model training, repeating the above process until the preset stopping condition is met or the model converges.
[0036] Steps S101 to S103, as illustrated in this embodiment, integrate federated learning technology into the dynamic image detection system for high-speed train operation faults, thereby changing the traditional manual review process and achieving intelligent automatic fault identification. While ensuring full protection of user data privacy, it also achieves optimization and significant improvement in global model performance. Through joint training, the detection accuracy of participating parties can increase by 11.82% compared to models trained solely on their own limited data.
[0037] The following is combined Figure 2 The process of one embodiment of the present invention is described.
[0038] Figure 2 This is a flowchart illustrating an embodiment of an intelligent identification method for high-speed train operation faults based on federated learning provided by the present invention. Figure 2 As shown, the method includes the following: Step ① Synchronize historical data from the dynamic image detection system for train operation faults on the main data center server of China State Railway Group, and pre-train the YOLOv model on bogie and floor data. This involves setting hyperparameters such as initial learning rate, weight decay, and data augmentation, and training the model on the dataset to obtain initial model parameters. .
[0039] A global fault identification model is initialized at the China State Railway Group level and distributed to participating railway bureaus as the initial model for fault identification in EMU operation images. The model's loss function includes classification loss, localization loss, and target confidence loss. These are weighted and summed to obtain the final total loss, which guides model training and optimization. Where N is the number of detection layers, , , These are the weights for the three types of losses.
[0040] Step 2: Deploy a global parameter server in the main data center of China State Railway Group, deploy local parameter servers in each railway bureau, and deploy training servers as working nodes in the EMU depots / stations under the railway bureaus.
[0041] The parameters will be transmitted between the EMU depot / station, the railway bureau, and the China State Railway Group's main data center. Each EMU depot / station is responsible for training its local model and uploading the model parameters to its respective railway bureau. The parameter servers of each railway bureau are responsible for aggregating the model parameters of its EMU depot / station and broadcasting the new model parameters to its respective working nodes. The parameter servers of the main data center are responsible for pulling the model parameters of each railway bureau, aggregating and updating them, and broadcasting the updated parameters to each railway bureau, which then broadcasts them to its respective EMU depot / station.
[0042] Step ③ Data set of local EMU operation images used by each railway bureau's EMU depot / station Conduct training and set the learning rate. Calculate the gradient The local model updates parameters in this round. The updated values of the model parameters are calculated. And update the model parameter values. and the size of local data Send to the relevant railway bureau.
[0043] Step 4: Each railway bureau receives the trained model parameters from the work nodes of its affiliated EMU depots / stations. and data volume Aggregation of local parameters is performed using the federated averaging method. ,in, q represents the number of working nodes of a certain railway bureau's EMU depot / station, t represents the number of railway bureaus, and t represents the number of communication rounds.
[0044] Step 5: China State Railway Group receives the aggregated model parameters from various railway bureaus. and total data volume Through the federal average method Perform global model aggregation.
[0045] Step 6: China State Railway Group Co., Ltd. calculates the contribution of each railway bureau using the contribution measurement method. ,in The amount of data owned by each railway bureau This refers to the number of data categories. Based on contribution, the updated model parameter values obtained by each railway bureau are calculated. Send to the railway bureau .
[0046] Step 7: Download the model parameter update amount allocated to each railway bureau. The final updated models for each railway line in this round were obtained. .
[0047] Step 8: Each railway bureau will distribute the updated model parameters indiscriminately to its affiliated EMU depots / stations for the next round of local working node model training, repeating the above process until the preset stopping condition is met or the model converges.
[0048] On the other hand, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, is implemented to perform the above-described intelligent identification method for EMU operation faults based on federated learning.
[0049] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0050] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0051] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for intelligent identification of operational faults in high-speed trains, characterized in that, include: Step 1: Each train depot / station trains a local model based on the global fault identification model issued by the global parameter server, and uploads the model parameters to the parameter server of the railway bureau to which it belongs. Step 2: The parameter servers of each railway bureau aggregate the model parameters of their respective EMU depots / stations and broadcast the new model parameters to their respective working nodes; Step 3: The main data center parameter server extracts the model parameters of each railway bureau, performs aggregation and update based on the federated averaging method, and broadcasts the updated parameters to each railway bureau and distributes them to the corresponding EMU depots / stations.
2. The method according to claim 1, characterized in that, The main data center deploys a global parameter server, each railway bureau deploys a local parameter server, and the EMU depots / stations under the railway bureaus deploy training servers as working nodes. Model parameters are transmitted between the EMU depots / stations, railway bureaus, and the main data center.
3. The method according to claim 1, characterized in that, Step 1 further includes initializing the model parameters before training; Among them, the historical data of the dynamic image detection system for train operation faults were synchronized on the main data center server, and the YOLOv model was pre-trained on the bogie and floor related data to obtain the initial parameters of the model. Initialize the global fault identification model of the data center and distribute it to each participating railway bureau as the initial model for fault identification of EMU operation images, and set the number of communication rounds.
4. The method according to claim 1, characterized in that, Step 1 further includes: Each EMU depot / station uses its local EMU operation image dataset for training, calculates the updated model parameters, and sends the updated model parameters and local data volume to its respective railway bureau.
5. The method according to claim 1, characterized in that, Step 2 also includes: After receiving the model parameters and data volume trained by the working nodes of its affiliated EMU depots / stations, the railway bureau aggregates the local parameters using the federated averaging method and sends the aggregated model parameters and data volume to the main data center.
6. The method according to claim 1, characterized in that, Step 3 also includes: The main data center receives the aggregated model parameters and total data volume from each railway bureau, performs global model aggregation using the federated averaging method, and calculates the contribution of each railway bureau using the contribution measurement method. Based on the contribution level, the updated model parameter values obtained by each railway bureau are calculated and sent to each railway bureau.
7. The method according to claim 6, characterized in that, Step 3 also includes: Each railway bureau receives the updated model parameters from the main data center parameter server, downloads the allocated updated model parameters, and obtains the final updated model for this round for each railway bureau.
8. The method according to claim 6, characterized in that, Step 3, which involves global model aggregation using the federated averaging method, further includes: The aggregated data of the global model is obtained through the following calculation formula: , in, This represents the number of work nodes in a specific railway bureau's EMU depot / station. In order to receive the model parameters trained by the working node of the corresponding EMU depot / station For data volume.
9. The method according to claim 6, characterized in that, Step 3, which involves calculating the updated model parameters for each railway bureau based on their contribution and sending them to each bureau, further includes: The contribution is calculated using the following formula: ， in, The amount of data owned by each railway bureau For the number of data categories.
10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the intelligent identification method for EMU operation faults as described in any one of claims 1 to 9.