Traction converter fault diagnosis method and device, computer device and storage medium

Abstract

The application relates to a traction converter fault diagnosis method and device, computer equipment, a storage medium and a computer program product. The method obtains working condition data of a traction converter; time-frequency features of the working condition data are extracted to obtain target feature data; the target feature data is input into a deep residual shrinkage network model for fault prediction to obtain a first fault prediction result, and the target feature data is input into a deep long short-term memory network model for fault prediction to obtain a second fault prediction result; the first fault prediction result and the second fault prediction result are weighted and fused to obtain a fault diagnosis result. The entire scheme trains the network model based on digital twin state sample data of multiple working conditions, is more accurate, then performs fault prediction based on the two network models, and performs weighted fusion based on the multi-dimensional fault prediction results to obtain a more accurate fault diagnosis result.

Description

Technical Field

[0001] This application relates to the field of railway technology, and in particular to a method, apparatus, computer equipment, storage medium and computer program product for diagnosing traction converter faults. Background Technology

[0002] With the construction of infrastructure and the rapid development of technology, fields such as power, intelligent vehicles, high-speed rail, urban rail transit, and subways have brought about rapid economic growth. However, while pursuing high speed, it is also essential to ensure the safety and reliability of equipment; safety is always the most important research topic. Therefore, PHM (Prognostics and Health Management) technologies for different fields are crucial for the healthy development of each sector. Traction converters are power conversion and traction devices that play a vital role in fields such as power, intelligent vehicles, high-speed rail, urban rail transit, and subways.

[0003] Taking the application of traction converters in high-speed rail as an example, the electric traction system of high-speed railways is subjected to frequent internal and external shocks, such as power grid oscillations, distortion, harmonics, and human error, as well as internal factors such as system aging and cumulative damage. In severe cases, this can cause system failures or even system damage. The traction converter is a key subsystem of electric locomotives, ensuring the safe and stable operation of trains. As a crucial subsystem for high-speed rail, urban rail, subway, and hybrid trains, a failure in such systems often results in enormous losses, or even catastrophic consequences.

[0004] Currently, the fault diagnosis methods for traction converters cannot achieve accurate fault diagnosis. Summary of the Invention

[0005] Therefore, it is necessary to provide an accurate method, device, computer equipment, computer-readable storage medium, and computer program product for traction converter fault diagnosis to address the aforementioned technical problems.

[0006] Firstly, this application provides a method for diagnosing faults in a traction converter. The method includes:

[0007] Obtain operating data of the traction converter;

[0008] Extract the time-frequency features of the operating condition data to obtain the target feature data;

[0009] The target feature data is input into the deep residual shrinkage network model for fault prediction to obtain the first fault prediction result, and the target feature data is input into the deep long short-term memory network model for fault prediction to obtain the second fault prediction result; the deep residual shrinkage network model and the deep long short-term memory network model are trained based on the digital twin state sample data of the traction converter under different operating conditions.

[0010] The fault diagnosis result is obtained by weighted fusion of the first fault prediction result and the second fault prediction result.

[0011] In one embodiment, the deep residual shrinkage network model is trained as follows: digital twin state sample data of the traction converter under different operating conditions and an initial deep residual shrinkage network model are acquired; the digital twin state sample data carries fault type labels; the digital twin state sample data is input into the initial deep residual shrinkage network model for fault diagnosis to obtain a first sample fault diagnosis result; loss is calculated based on the first sample fault diagnosis result and the fault type label to obtain a first loss value; the parameters of the initial deep residual shrinkage network model are updated based on the first loss value until the latest first loss value is less than a preset first loss threshold to obtain the deep residual shrinkage network model.

[0012] In one embodiment, the deep long short-term memory network model is trained as follows: digital twin state sample data of the traction converter under different operating conditions and an initial deep long short-term memory network model are acquired; the digital twin state sample data carries fault type labels; the digital twin state sample data is input into the initial deep long short-term memory network model for fault diagnosis to obtain a second sample fault diagnosis result; loss is calculated based on the second sample fault diagnosis result and the fault type label to obtain a second loss value; the parameters of the initial deep long short-term memory network model are updated based on the second loss value until the latest second loss value is less than a preset second loss threshold to obtain the deep long short-term memory network model.

[0013] In one embodiment, the digital twin state sample data includes sample operating condition data and the fault types corresponding to the operating condition data; obtaining the digital twin state sample data of the traction converter under different operating conditions includes: based on the digital twin model, obtaining the operating condition data of the traction converter under different operating conditions and the fault types corresponding to each operating condition data.

[0014] In one embodiment, acquiring operating condition data of the traction converter under different operating conditions and the corresponding fault types based on the digital twin model includes: acquiring state data of the traction converter and an initial digital twin model under different operating conditions; inputting the state data into the initial digital twin model for fault simulation to obtain fault simulation results; performing entity simulation based on the state data to obtain entity fault results; calculating the error based on the entity fault results and fault simulation results to obtain an error value; adjusting the parameters of the initial digital twin model based on the error value until the latest error value is less than a preset loss threshold to obtain the digital twin model; and generating operating condition data of the traction converter under different operating conditions and the corresponding fault types based on the digital twin model.

[0015] In one embodiment, the digital twin model includes a virtual simulation model and a hardware-in-the-loop simulation model; inputting state data into the initial digital twin model to perform fault simulation and obtain fault simulation results includes: inputting state data into the virtual simulation model to perform fault simulation and obtain fault simulation results; performing physical simulation based on the state data to obtain physical fault results includes: inputting state data into the hardware-in-the-loop simulation model to perform fault simulation and obtain physical fault results.

[0016] In one embodiment, the fault diagnosis result is obtained by weighted fusion of the first fault prediction result and the second fault prediction result, including: obtaining the prediction accuracy of the deep residual shrinkage network model to obtain the first fault weight, and obtaining the prediction accuracy of the deep long short-term memory network model to obtain the second fault weight; and performing weighted calculation based on the first fault weight and the first fault prediction result, as well as the second fault weight and the second fault prediction result, to obtain the fault diagnosis result.

[0017] Secondly, this application also provides a traction converter fault diagnosis device. The device includes:

[0018] The acquisition module is used to acquire the operating condition data of the traction converter;

[0019] The extraction module is used to extract the time-frequency features of the operating condition data to obtain the target feature data;

[0020] The prediction module is used to input the target feature data into the deep residual shrinkage network model for fault prediction to obtain the first fault prediction result, and input the target feature data into the deep long short-term memory network model for fault prediction to obtain the second fault prediction result; the deep residual shrinkage network model and the deep long short-term memory network model are trained based on the digital twin state sample data of the traction converter under different operating conditions;

[0021] The fusion module is used to perform weighted fusion based on the first fault prediction result and the second fault prediction result to obtain the fault diagnosis result.

[0022] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps:

[0023] Obtain operating data of the traction converter;

[0024] Extract the time-frequency features of the operating condition data to obtain the target feature data;

[0025] The target feature data is input into the deep residual shrinkage network model for fault prediction to obtain the first fault prediction result, and the target feature data is input into the deep long short-term memory network model for fault prediction to obtain the second fault prediction result; the deep residual shrinkage network model and the deep long short-term memory network model are trained based on the digital twin state sample data of the traction converter under different operating conditions.

[0026] The fault diagnosis result is obtained by weighted fusion of the first fault prediction result and the second fault prediction result.

[0027] Fourthly, this application also provides a computer-readable storage medium. This computer-readable storage medium stores a computer program thereon, which, when executed by a processor, performs the following steps:

[0028] Obtain operating data of the traction converter;

[0029] Extract the time-frequency features of the operating condition data to obtain the target feature data;

[0030] The target feature data is input into the deep residual shrinkage network model for fault prediction to obtain the first fault prediction result, and the target feature data is input into the deep long short-term memory network model for fault prediction to obtain the second fault prediction result; the deep residual shrinkage network model and the deep long short-term memory network model are trained based on the digital twin state sample data of the traction converter under different operating conditions.

[0031] The fault diagnosis result is obtained by weighted fusion of the first fault prediction result and the second fault prediction result.

[0032] Fifthly, this application also provides a computer program product. This computer program product includes a computer program that, when executed by a processor, performs the following steps:

[0033] Obtain operating data of the traction converter;

[0034] Extract the time-frequency features of the operating condition data to obtain the target feature data;

[0035] The target feature data is input into the deep residual shrinkage network model for fault prediction to obtain the first fault prediction result, and the target feature data is input into the deep long short-term memory network model for fault prediction to obtain the second fault prediction result; the deep residual shrinkage network model and the deep long short-term memory network model are trained based on the digital twin state sample data of the traction converter under different operating conditions.

[0036] The fault diagnosis result is obtained by weighted fusion of the first fault prediction result and the second fault prediction result.

[0037] The aforementioned traction converter fault diagnosis method, device, computer equipment, storage medium, and computer program products acquire the operating condition data of the traction converter; extract the time-frequency features of the operating condition data to obtain target feature data; input the target feature data into a deep residual contraction network model for fault prediction to obtain a first fault prediction result, and input the target feature data into a deep long short-term memory network model for fault prediction to obtain a second fault prediction result; the deep residual contraction network model and the deep long short-term memory network model are trained based on digital twin state sample data of the traction converter under different operating conditions; and the first fault prediction result and the second fault prediction result are weighted and fused to obtain the fault diagnosis result. The entire scheme first trains the deep residual contraction network model and the deep long short-term memory network model based on digital twin state sample data. The fault diagnosis model trained with digital twin state sample data under multiple operating conditions is more accurate. Then, fault prediction is performed on the state data of the traction converter based on the two fault diagnosis models, and the multi-dimensional fault prediction results are weighted and fused to obtain a more accurate fault diagnosis result. Attached Figure Description

[0038] Figure 1 This is a diagram illustrating the application environment of a traction converter fault diagnosis method in one embodiment.

[0039] Figure 2 This is a flowchart illustrating a traction converter fault diagnosis method in one embodiment;

[0040] Figure 3 This is a flowchart illustrating the training process of a deep residual shrinkage network model in one embodiment.

[0041] Figure 4 This is a schematic diagram of the structure of a deep residual shrinkage network model in one embodiment;

[0042] Figure 5 This is a flowchart illustrating a traction converter fault diagnosis method in another embodiment;

[0043] Figure 6 This is a structural block diagram of a traction converter fault diagnosis device in one embodiment;

[0044] Figure 7 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0046] PHM (Prognostics and Health Management) was developed to meet the requirements of autonomous maintenance and self-diagnosis. It is an upgrade of CBM (condition-based maintenance). It emphasizes condition awareness in asset and equipment management, monitors equipment health status, and identifies areas and cycles of frequent failures. Through data monitoring and analysis, it predicts the occurrence of failures, thereby significantly improving operational efficiency.

[0047] Currently, research on PHM technology in China is in its early exploratory stage, especially in the railway sector where research is extremely limited. With the rapid construction and development of railway engineering in my country in recent years, high-speed rail, urban rail, and subways have become crucial to the national economy and development. While pursuing high speeds, it is also essential to ensure the safety and reliability of trains. Safety is always the most important research topic; therefore, PHM technology urgently needs further exploration and research in the rail transit field.

[0048] High-speed railway electric traction systems are subjected to frequent internal and external shocks, such as power grid oscillations, distortions, harmonics, and human error, as well as internal factors like system aging and cumulative damage. These shocks can lead to system failures or even complete system damage. If the factors influencing system damage can be analyzed and a system health assessment or fault prediction model established, combined with real-time information on key train characteristics, model analysis and reasoning can be used to implement external safety control measures to prevent potential system failures and maximize the safe and stable operation of the system. The traction converter is a critical subsystem of electric locomotives, ensuring the safe and stable operation of trains. As a key subsystem of high-speed rail, urban rail, subway, and hybrid trains, the traction converter typically consists of numerous components with multiple structural layers and complex, highly coupled relationships between different parts. Its complex structure, difficult fault diagnosis and maintenance, and nonlinear, coupled, and stochastic characteristics make the system prone to failure and require high reliability and safety. Failures in such systems often result in enormous losses or even catastrophic consequences.

[0049] Under long-term operation in complex and harsh environments, the performance of key power devices in traction converters degrades or even fails. These failures or performance degradations significantly impact the converter. While some scholars have used reliability modeling theories such as Markov models and Bayesian methods, accurately determining the state transitions and inference probabilities of probabilistic statistical and inference models is difficult, hindering accurate and real-time assessments of system reliability and lifespan. Traditional fault diagnosis or prediction methods lack the ability to integrate various system state information; relying on single information or methods results in poor accuracy. With the development of digital twin and artificial intelligence technologies, PHM (Prognostics and Health Management) based on digital twin technology, combined with artificial intelligence, can achieve the fusion of multi-source state information and virtual-real models. This allows for more real-time and accurate correction and feedback of the twin, continuously approximating the system's true physical state. Therefore, this application proposes a method based on the fusion of digital twin models and deep learning algorithms to achieve accurate fault prediction and health management of traction converters.

[0050] Digital twin (DT) is a simulation process that fully utilizes data from physical models, sensor updates, and operational history to integrate multiple disciplines, physical quantities, scales, and probabilities. This process is mapped in virtual space to reflect the entire lifecycle of the corresponding physical equipment. A digital twin is a concept that transcends reality; it can be viewed as a digital mapping system of one or more important, interdependent equipment systems.

[0051] The traction converter fault diagnosis method provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or placed on a cloud or other network server. Terminal 102 sends a fault detection command to the server. Server 104 obtains the operating condition data of the traction converter according to the fault detection command; extracts the time-frequency features of the operating condition data to obtain target feature data; inputs the target feature data into a deep residual contraction network model for fault prediction to obtain a first fault prediction result, and inputs the target feature data into a deep long short-term memory network model for fault prediction to obtain a second fault prediction result; the deep residual contraction network model and the deep long short-term memory network model are trained based on digital twin state sample data of the traction converter under different operating conditions; the first fault prediction result and the second fault prediction result are weighted and fused to obtain a fault diagnosis result. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can be smart speakers, smart TVs, smart air conditioners, smart vehicle devices, etc. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. Server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0052] In one embodiment, such as Figure 2 As shown, a fault diagnosis method for a traction converter is provided, which is applied to... Figure 1 Taking the server in the example, the following steps are included:

[0053] Step 202: Obtain the operating condition data of the traction converter.

[0054] The traction converter, a key component of the train, is installed at the bottom of the vehicle. Its main function is to convert electrical energy between DC and AC systems, transforming 1500V DC power from the overhead contact line into 0-1150V three-phase AC power. Through voltage and frequency regulation, it controls the starting, braking, and speed regulation of the AC traction motors. The traction converter's operating data includes three-phase electrical data and DC-side electrical data. Specifically, the three-phase electrical data includes three-phase voltage, three-phase current, and motor data; the DC-side electrical data includes DC bus voltage and DC bus current; and the motor data includes motor speed and torque. The traction converter operates under various conditions, including no-load, light-load, and heavy-load. Each condition includes different fault information, such as degradation faults in key components like capacitors, aging or open / short circuits in power devices. These faults can cause varying degrees of change in the voltage or current output of the high-speed railway's electric traction system.

[0055] Specifically, the server receives fault detection commands from the terminal and, based on these commands, acquires the operating condition data of the traction converter collected by the terminal. This operating condition data is the latest, real-time data collected at the moment. The server can also periodically acquire the latest current operating condition data from the terminal for real-time fault detection. The terminal can collect real-time operating condition data of the traction converter through data acquisition sensors and transmit it to the server.

[0056] Step 204: Extract the time-frequency features of the operating condition data to obtain the target feature data.

[0057] Among these, time-frequency features can be the waveform features of the operating data, describing the frequency information of the operating data over a period of time. Given a discrete-time signal x(n), a two-dimensional representation x(n, f) can be obtained through time-frequency feature extraction methods; it is a complex function of time and frequency. Therefore, it can provide information on how the amplitude and phase of different frequency components of the operating data change over time. The time-frequency representation of the operating data, whether in terms of amplitude or phase, can be a set of features, which can be directly presented as a 2D image using novel deep learning-based classifiers such as convolutional neural networks.

[0058] A common limitation of time-domain and frequency-domain feature extraction methods is that important, high-resolution features are discarded because information is calculated from only one domain. For example, time-domain features do not provide oscillation information, and in the case of frequency analysis, detailed information on how the spectral signal changes over time is not provided, even though these are important concerns in signal analysis research. This limitation can be addressed through time-frequency analysis, such as the short-time Fourier transform (STFT), Morlet wavelet transform (MW), and filter-based Hilbert transform (FHT).

[0059] Specifically, after acquiring the current operating condition data of the traction converter, the server performs noise reduction on the data to minimize noise interference, obtaining denoised operating condition data. Then, it performs time-frequency feature analysis and extraction on the denoised operating condition data to obtain target feature data. Further, the server performs time-frequency feature analysis and extraction on the denoised operating condition data based on short-time Fourier transform to obtain target feature data; alternatively, the server performs time-frequency feature analysis and extraction on the denoised operating condition data based on Morlet wavelet transform to obtain target feature data; or alternatively, the server performs time-frequency feature analysis and extraction on the denoised operating condition data based on Hilbert transform of the filter to obtain target feature data.

[0060] Step 206: Input the target feature data into the deep residual shrinkage network model for fault prediction to obtain the first fault prediction result, and input the target feature data into the deep long short-term memory network model for fault prediction to obtain the second fault prediction result.

[0061] Among them, the Deep Residual Shrinkage Network (DRSN) and Deep Long Short-Term Memory (LSTM) network models are trained based on digital twin state sample data of traction converters under different operating conditions. DRSN is a novel improvement on Deep Residual Networks, introducing soft thresholding as a nonlinear layer into the ResNet network structure, which can improve the feature learning performance of deep learning methods on noisy or complex data. LSTM is a type of recurrent neural network specifically designed to address the long-term dependency problem of general RNNs (Recurrent Neural Networks). All RNNs have a chain-like form of repeating neural network modules.

[0062] Specifically, the server inputs the target feature data into a deep residual shrinking network model for fault prediction, obtaining a first fault prediction result. Simultaneously, it inputs the target feature data into a deep long short-term memory network model for fault prediction, obtaining a second fault prediction result. The target feature data undergoes parallel processing by the deep residual shrinking network model and the deep long short-term memory network model to obtain the prediction results of the two models, namely the first fault prediction result and the second fault prediction result.

[0063] Step 208: Perform a weighted fusion of the first fault prediction result and the second fault prediction result to obtain the fault diagnosis result.

[0064] Specifically, the server obtains the weight coefficients of the deep residual shrinking network model and the deep long short-term memory network model, and performs a weighted calculation on the first fault prediction result and the second fault prediction result based on the weighted average algorithm and the weight coefficients of the deep residual shrinking network model and the deep long short-term memory network model to obtain the fault diagnosis result.

[0065] In the aforementioned traction converter fault diagnosis method, the operating condition data of the traction converter is acquired; the time-frequency features of the operating condition data are extracted to obtain target feature data; the target feature data is input into a deep residual contraction network model for fault prediction to obtain a first fault prediction result, and then input into a deep long short-term memory network model for fault prediction to obtain a second fault prediction result; the deep residual contraction network model and the deep long short-term memory network model are trained based on digital twin state sample data of the traction converter under different operating conditions; the first fault prediction result and the second fault prediction result are weighted and fused to obtain the fault diagnosis result. The entire scheme first trains a deep residual contraction network model and a deep long short-term memory network model based on digital twin state sample data. The fault diagnosis model trained with digital twin state sample data under multiple operating conditions is more accurate. Then, fault prediction is performed on the state data of the traction converter based on the two fault diagnosis models, and the multi-dimensional fault prediction results are weighted and fused to obtain a more accurate fault diagnosis result.

[0066] In an optional embodiment, such as Figure 3 As shown, the deep residual shrinkage network model is trained in the following way:

[0067] Step 302: Obtain digital twin state sample data of the traction converter under different operating conditions and the initial deep residual shrinkage network model.

[0068] The digital twin state sample data carries fault type labels. This data is generated from the digital twin model. It includes various fault data and fault types for the traction converter under different operating conditions; the fault types are accurately determined by experts based on analysis of different fault data.

[0069] Specifically, the server uses a digital twin model to obtain digital twin state sample data of the traction converter under different operating conditions, as well as an untrained initial deep residual shrinkage network model.

[0070] Step 304: Input the digital twin state sample data into the initial deep residual shrinkage network model for fault diagnosis to obtain the first sample fault diagnosis result.

[0071] Building upon deep residual networks, deep residual shrinking networks introduce a small sub-network. This sub-network learns a set of thresholds to perform soft thresholding on each channel of the feature map. This process can be viewed as a trainable feature selection process. Specifically, the preceding convolutional layers transform important features into values ​​with larger absolute values ​​and features corresponding to redundant information into values ​​with smaller absolute values. The sub-network learns the boundary between these two values, and soft thresholding sets redundant features to zero while ensuring that important features have non-zero outputs.

[0072] like Figure 4 As shown, the deep residual shrinking network model includes convolutional layers, two residual modules, batch normalization, global mean pooling, and a fully connected output layer. Further, the residual module includes two batch normalization (BN) functions, two rectified linear unit activation functions (ReLU), two convolutional layers, the residual shrinking module, and an identity shortcut.

[0073] Specifically, the server inputs the digital twin state sample data into the initial deep residual shrinkage network model, and performs fault diagnosis on the digital twin state sample data through the initial deep residual shrinkage network to obtain the first sample fault diagnosis result.

[0074] Step 306: Calculate the loss based on the fault diagnosis results and fault type markings of the first sample to obtain the first loss value.

[0075] Specifically, the server obtains the fault type of the digital twin state sample data, calculates the loss based on the fault diagnosis result of the first sample and the fault type label, and obtains the first loss value.

[0076] Step 308: Update the parameters of the initial deep residual shrinkage network model based on the first loss value until the latest first loss value is less than the preset first loss threshold, and obtain the deep residual shrinkage network model.

[0077] Specifically, the server performs backpropagation based on the first loss value, continuously updating the parameters of the initial deep residual shrinkage network model until the latest first loss value is less than the preset first loss threshold, thus obtaining the deep residual shrinkage network model.

[0078] In this embodiment, soft thresholding based on a deep residual shrinking network is used to eliminate noise-related features. More importantly, the threshold is automatically learned in the residual shrinking module, thus enabling robust analysis, identification, and extraction of weak features of traction converter faults under strong noise interference. The deep residual shrinking network can be used not only for noisy data but also for noise-free data. This is because the threshold in the deep residual shrinking network is adaptively determined based on the sample conditions. In other words, if the sample does not contain redundant information and soft thresholding is not needed, the threshold can be trained to be very close to zero, thus soft thresholding is essentially unnecessary. Simultaneously, soft thresholding sets features in the interval [-τ, τ] to 0, causing other features farther from 0 to shrink towards 0 as well. If considered together with the bias b of the previous convolutional layer, this interval of zero becomes [-τ+b, τ+b]. Since τ and b are parameters that can be learned automatically, from this perspective, soft thresholding can actually set features in any interval to zero. It is a more flexible way to delete features in a certain range of values, and can also be understood as a more flexible nonlinear mapping.

[0079] In an optional embodiment, the deep long short-term memory network model is trained as follows: digital twin state sample data of the traction converter under different operating conditions and an initial deep long short-term memory network model are acquired; the digital twin state sample data are input into the initial deep long short-term memory network model for fault diagnosis to obtain a second sample fault diagnosis result; loss is calculated based on the second sample fault diagnosis result and fault type label to obtain a second loss value; the parameters of the initial deep long short-term memory network model are updated based on the second loss value until the latest second loss value is less than a preset second loss threshold to obtain the deep long short-term memory network model.

[0080] The digital twin state sample data carries fault type labels. RNN models suffer from long-term dependency issues and cannot effectively learn features from longer time series. Long short-term memory (LSTM) networks can effectively alleviate this problem. Compared to RNNs, LSTMs introduce an input gate i, a forget gate f, an output gate o, and an internal memory unit c.

[0081] Specifically, the server, based on a digital twin model, acquires digital twin state sample data of the traction converter under different operating conditions, as well as an untrained initial deep long short-term memory (DSL) network model. The server inputs the digital twin state sample data into the initial DSL network model, and performs fault diagnosis on the digital twin state sample data through the initial DSL network model to obtain a second sample fault diagnosis result. The server obtains the fault type of the digital twin state sample data, and calculates the loss based on the second sample fault diagnosis result and the fault type label to obtain a second loss value. The server performs backpropagation based on the second loss value, continuously updating the parameters of the initial DSL network model until the latest second loss value is less than a preset second loss threshold, thus obtaining the deep long short-term memory network model.

[0082] In this embodiment, the advantages of deep long short-term memory networks enable the memory storage of historical fault state information of converters and the deep feature analysis and extraction. By using dynamic difference methods to extract and enrich the feature information of converter faults under multimodal conditions, the problem that static features are insufficient to characterize the fault evolution trend of multimodal processes can be solved.

[0083] In an optional embodiment, obtaining digital twin state sample data of the traction converter under different operating conditions includes: based on the digital twin model, obtaining operating condition data of the traction converter under different operating conditions and the fault type corresponding to each operating condition data.

[0084] The digital twin status sample data includes sample operating condition data and the fault types corresponding to the operating condition data.

[0085] Specifically, from the perspective of electrical and physical models, virtual simulation is used to model the traction converter to obtain virtual data, while a semi-physical simulation test platform is used to obtain actual test data. These are integrated to form a digital twin model. Based on the digital twin model, operating condition data of the traction converter under different operating conditions and the corresponding fault types are generated.

[0086] In this implementation, to address the shortcomings of current traction converter fault data, such as insufficient information, limited diagnostic types, and poor visualization, digital twin state sample data generated by a digital twin model is used. This data includes data on different fault types under various operating conditions. Based on the digital twin state sample data, a deep residual shrinkage network model and a deep long short-term memory network model are trained, thereby improving the fault diagnosis performance of the model and realizing online intelligent state monitoring and fault diagnosis of the traction converter.

[0087] In an optional embodiment, acquiring operating condition data of the traction converter under different operating conditions and the corresponding fault types based on the digital twin model includes: acquiring state data of the traction converter and an initial digital twin model under different operating conditions; inputting the state data into the initial digital twin model for fault simulation to obtain fault simulation results; performing entity simulation based on the state data to obtain entity fault results; calculating the error based on the entity fault results and fault simulation results to obtain the error value; adjusting the parameters of the initial digital twin model based on the error value until the latest error value is less than a preset loss threshold to obtain the digital twin model; and generating operating condition data of the traction converter under different operating conditions and the corresponding fault types based on the digital twin model.

[0088] The state data of the traction converter under different operating conditions can be obtained by analyzing the historical state data of the traction converter. The digital twin model is a semi-virtual simulation model of the traction converter, and the structure of the traction converter in the digital twin model is consistent with that of the real traction converter.

[0089] Specifically, the server acquires the status data and initial digital twin model of the traction converter under different operating conditions. The server inputs the status data into the initial digital twin model for fault simulation, obtaining simulation waveform data under that condition; performs entity simulation based on the status data, obtaining entity waveform data; calculates the error based on the entity waveform data and simulation waveform data, obtaining the error value; adjusts the parameters of the initial digital twin model based on the error value until the latest error value is less than a preset loss threshold, obtaining the digital twin model; and generates operating condition data of the traction converter under different operating conditions and the corresponding fault types for each operating condition based on the digital twin model.

[0090] In this embodiment, in order to obtain the fault characteristics of high-speed trains under various faults under traction conditions, the waveforms of current, voltage, motor speed and torque under different faults are compared and analyzed, so as to complete the fault identification and diagnosis of different faults.

[0091] In an optional embodiment, inputting state data into an initial digital twin model to perform fault simulation and obtain fault simulation results includes: inputting state data into a virtual simulation model to perform fault simulation and obtain fault simulation results; performing entity simulation based on state data to obtain entity fault results includes: inputting state data into a semi-physical simulation model to perform fault simulation and obtain entity fault results.

[0092] The digital twin model includes both virtual simulation models and hardware-in-the-loop (HIL) simulation models. In the virtual simulation model, both the controller and the controlled object are virtual. The virtual simulation model is a purely digital virtual MATLAB electrical simulation model built using the MATLAB SIMULINK module, capable of simulating various operating conditions and fault modes of the traction converter. Furthermore, the virtual simulation model simulates the output three-phase voltage and current, DC-side voltage and current, and other data of the traction converter's power switching devices, capacitors, and other key components under different conditions such as open circuits, short circuits, and degradation.

[0093] The hardware-in-the-loop (HIL) simulation model is an improvement on the virtual simulation model. The controller still uses a virtual simulation model, while the main circuit of the controlled object uses a physical test bench. Fault injection or simulation is input from the virtual model of the controller, and the output waveform is collected from the test bench and fed back to the virtual control terminal to form a closed loop.

[0094] Specifically, the server inputs state data into a virtual simulation model to simulate faults and obtains simulated waveform data; it also inputs state data into a hardware-in-the-loop simulation model to simulate faults and obtains physical waveform data. Error calculations are performed based on the physical and simulated waveform data to obtain error values; the parameters of the initial digital twin model are adjusted based on these error values ​​until the latest error value is less than a preset loss threshold, thus obtaining the digital twin model; based on the digital twin model, operating condition data of the traction converter under different operating conditions and the corresponding fault types for each operating condition are generated.

[0095] In this embodiment, virtual modeling and fault data acquisition are achieved based on pure virtual simulation, while experimental data testing and verification are achieved based on semi-physical simulation. Through multiple means, a fault twin dataset driven by both simulation and experimental data is realized. Subsequently, diagnosis and fusion based on deep learning algorithms can more accurately and efficiently achieve in-depth fault feature mining, analysis, and fault diagnosis of the traction converter.

[0096] In an optional embodiment, the fault diagnosis result is obtained by weighted fusion of the first fault prediction result and the second fault prediction result, including: obtaining the prediction accuracy of the deep residual shrinkage network model to obtain the first fault weight, and obtaining the prediction accuracy of the deep long short-term memory network model to obtain the second fault weight; and performing weighted calculation based on the first fault weight and the first fault prediction result, as well as the second fault weight and the second fault prediction result, to obtain the fault diagnosis result.

[0097] Specifically, during the training process of the deep residual shrinking network model and the deep long short-term memory network model, when the error of the two models is lower than the corresponding preset model loss threshold, the server calculates the prediction accuracy of the model, obtaining the prediction accuracy of the deep residual shrinking network model and the prediction accuracy of the deep long short-term memory network model. The server obtains the prediction accuracy of the deep residual shrinking network model, uses it as the first fault weight, and multiplies the first fault weight with the first fault prediction result to obtain the first diagnostic result. The server obtains the prediction accuracy of the deep long short-term memory network model, uses it as the second fault weight, multiplies the second fault weight with the second fault prediction result to obtain the second diagnostic result, and adds the first diagnostic result to obtain the fault diagnosis result.

[0098] In this embodiment, weighted fusion of the prediction results of the two models can eliminate the error of the prediction of a single model. By combining the prediction results of the two models, the weight of the model with higher prediction accuracy is increased according to the model prediction accuracy, thereby improving the accuracy of the fault diagnosis results.

[0099] To facilitate understanding of the technical solutions provided in the embodiments of this application, such as Figure 5 As shown, the traction converter fault diagnosis method provided in this application embodiment is briefly described using a complete traction converter fault diagnosis process:

[0100] (1) Obtain the state data and initial digital twin model of the traction converter under different operating conditions; input the state data into the virtual simulation model to perform fault simulation and obtain the fault simulation results; input the state data into the hardware-in-the-loop simulation model to perform fault simulation and obtain the physical fault results; calculate the error based on the physical fault results and the fault simulation results and obtain the error value; adjust the parameters of the initial digital twin model based on the error value until the latest error value is less than the preset loss threshold and obtain the digital twin model.

[0101] (2) Based on the digital twin model, generate the operating condition data of the traction converter under different operating conditions and the fault type corresponding to each operating condition data.

[0102] (3) Obtain digital twin state sample data of the traction converter under different operating conditions and the initial deep residual shrinkage network model; the digital twin state sample data carries fault type labels; input the digital twin state sample data into the initial deep residual shrinkage network model for fault diagnosis to obtain the first sample fault diagnosis result; calculate the loss based on the first sample fault diagnosis result and the fault type label to obtain the first loss value; update the parameters of the initial deep residual shrinkage network model based on the first loss value until the latest first loss value is less than the preset first loss threshold to obtain the deep residual shrinkage network model.

[0103] (4) Obtain digital twin state sample data of the traction converter under different operating conditions and the initial deep long short-term memory network model; the digital twin state sample data carries fault type labels; input the digital twin state sample data into the initial deep long short-term memory network model for fault diagnosis to obtain the second sample fault diagnosis result; calculate the loss based on the second sample fault diagnosis result and the fault type label to obtain the second loss value; update the parameters of the initial deep long short-term memory network model based on the second loss value until the latest second loss value is less than the preset second loss threshold to obtain the deep long short-term memory network model.

[0104] (5) Obtain the operating condition data of the traction converter.

[0105] (6) Extract the time-frequency features of the operating condition data to obtain the target feature data.

[0106] (7) Input the target feature data into the deep residual shrinkage network model for fault prediction to obtain the first fault prediction result, and input the target feature data into the deep long short-term memory network model for fault prediction to obtain the second fault prediction result.

[0107] (8) The first fault prediction result and the second fault prediction result are weighted and fused to obtain the fault diagnosis result.

[0108] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0109] Based on the same inventive concept, this application also provides a traction converter fault diagnosis device for implementing the traction converter fault diagnosis method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more embodiments of the traction converter fault diagnosis device provided below can be found in the limitations of the traction converter fault diagnosis method described above, and will not be repeated here.

[0110] In one embodiment, such as Figure 6As shown, a traction converter fault diagnosis device is provided, including: an acquisition module 602, an extraction module 604, a prediction module 606, and a fusion module 608, wherein:

[0111] The acquisition module 602 is used to acquire the operating condition data of the traction converter.

[0112] The extraction module 604 is used to extract the time-frequency features of the operating condition data to obtain the target feature data.

[0113] The prediction module 606 is used to input the target feature data into the deep residual shrinkage network model for fault prediction to obtain the first fault prediction result, and input the target feature data into the deep long short-term memory network model for fault prediction to obtain the second fault prediction result; the deep residual shrinkage network model and the deep long short-term memory network model are trained based on the digital twin state sample data of the traction converter under different operating conditions.

[0114] The fusion module 608 is used to perform weighted fusion based on the first fault prediction result and the second fault prediction result to obtain the fault diagnosis result.

[0115] In one embodiment, the traction converter fault diagnosis device further includes a training module for acquiring digital twin state sample data of the traction converter under different operating conditions and an initial deep residual shrinkage network model; the digital twin state sample data carries a fault type label; the digital twin state sample data is input into the initial deep residual shrinkage network model for fault diagnosis to obtain a first sample fault diagnosis result; loss is calculated based on the first sample fault diagnosis result and the fault type label to obtain a first loss value; the parameters of the initial deep residual shrinkage network model are updated based on the first loss value until the latest first loss value is less than a preset first loss threshold to obtain the deep residual shrinkage network model.

[0116] In one embodiment, the training module is further used to acquire digital twin state sample data of the traction converter under different operating conditions and an initial deep long short-term memory network model; the digital twin state sample data carries fault type labels; the digital twin state sample data is input into the initial deep long short-term memory network model for fault diagnosis to obtain a second sample fault diagnosis result; loss is calculated based on the second sample fault diagnosis result and the fault type label to obtain a second loss value; the parameters of the initial deep long short-term memory network model are updated based on the second loss value until the latest second loss value is less than a preset second loss threshold to obtain the deep long short-term memory network model.

[0117] In one embodiment, the digital twin state sample data includes sample operating condition data and the fault types corresponding to the operating condition data; the training module is also used to obtain the operating condition data of the traction converter under different operating conditions and the fault types corresponding to each operating condition data based on the digital twin model.

[0118] In one embodiment, the training module is further configured to acquire state data and an initial digital twin model of the traction converter under different operating conditions; input the state data into the initial digital twin model for fault simulation to obtain fault simulation results; perform entity simulation based on the state data to obtain entity fault results; calculate the error based on the entity fault results and fault simulation results to obtain the error value; adjust the parameters of the initial digital twin model based on the error value until the latest error value is less than a preset loss threshold to obtain the digital twin model; and generate operating condition data of the traction converter under different operating conditions and the fault types corresponding to each operating condition data based on the digital twin model.

[0119] In one embodiment, the digital twin model includes a virtual simulation model and a hardware-in-the-loop simulation model; the training module is also used to input state data into the virtual simulation model to perform fault simulation and obtain fault simulation results; and to input state data into the hardware-in-the-loop simulation model to perform fault simulation and obtain physical fault results.

[0120] In one embodiment, the fusion module 608 is further configured to obtain the prediction accuracy of the deep residual shrinkage network model to obtain the first fault weight, and obtain the prediction accuracy of the deep long short-term memory network model to obtain the second fault weight; and perform weighted calculation based on the first fault weight and the first fault prediction result, as well as the second fault weight and the second fault prediction result, to obtain the fault diagnosis result.

[0121] Each module in the aforementioned traction converter fault diagnosis device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0122] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 7As shown, the computer device includes a processor, memory, and network interface connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores operating condition data of the traction converter under different operating conditions, as well as the corresponding fault types, digital twin models, deep long short-term memory network models, and deep residual shrinkage network models. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements a traction converter fault diagnosis method.

[0123] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0124] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0125] Obtain operating data of the traction converter;

[0126] Extract the time-frequency features of the operating condition data to obtain the target feature data;

[0127] The target feature data is input into the deep residual shrinkage network model for fault prediction to obtain the first fault prediction result, and the target feature data is input into the deep long short-term memory network model for fault prediction to obtain the second fault prediction result; the deep residual shrinkage network model and the deep long short-term memory network model are trained based on the digital twin state sample data of the traction converter under different operating conditions.

[0128] The fault diagnosis result is obtained by weighted fusion of the first fault prediction result and the second fault prediction result.

[0129] In one embodiment, when the processor executes the computer program, it further implements the following steps: the deep residual shrinkage network model is trained in the following manner: acquiring digital twin state sample data of the traction converter under different operating conditions and an initial deep residual shrinkage network model; the digital twin state sample data carries fault type labels; inputting the digital twin state sample data into the initial deep residual shrinkage network model for fault diagnosis to obtain a first sample fault diagnosis result; calculating the loss based on the first sample fault diagnosis result and the fault type label to obtain a first loss value; updating the parameters of the initial deep residual shrinkage network model based on the first loss value until the latest first loss value is less than a preset first loss threshold to obtain the deep residual shrinkage network model.

[0130] In one embodiment, when the processor executes the computer program, it further performs the following steps: the deep long short-term memory network model is trained in the following manner: acquiring digital twin state sample data of the traction converter under different operating conditions and an initial deep long short-term memory network model; the digital twin state sample data carries fault type labels; inputting the digital twin state sample data into the initial deep long short-term memory network model for fault diagnosis to obtain a second sample fault diagnosis result; calculating the loss based on the second sample fault diagnosis result and the fault type label to obtain a second loss value; updating the parameters of the initial deep long short-term memory network model based on the second loss value until the latest second loss value is less than a preset second loss threshold to obtain the deep long short-term memory network model.

[0131] In one embodiment, when the processor executes the computer program, it further implements the following steps: the digital twin state sample data includes sample operating condition data and the fault type corresponding to the operating condition data; obtaining the digital twin state sample data of the traction converter under different operating conditions includes: based on the digital twin model, obtaining the operating condition data of the traction converter under different operating conditions and the fault type corresponding to each operating condition data.

[0132] In one embodiment, when the processor executes the computer program, it further implements the following steps: based on the digital twin model, acquiring the operating condition data of the traction converter under different operating conditions and the fault types corresponding to each operating condition data includes: acquiring the state data of the traction converter and an initial digital twin model under different operating conditions; inputting the state data into the initial digital twin model for fault simulation to obtain fault simulation results; performing entity simulation based on the state data to obtain entity fault results; calculating the error based on the entity fault results and fault simulation results to obtain the error value; adjusting the parameters of the initial digital twin model based on the error value until the latest error value is less than a preset loss threshold to obtain the digital twin model; and generating the operating condition data of the traction converter under different operating conditions and the fault types corresponding to each operating condition data based on the digital twin model.

[0133] In one embodiment, when the processor executes the computer program, it further performs the following steps: the digital twin model includes a virtual simulation model and a hardware-in-the-loop simulation model; inputting state data into the initial digital twin model to perform fault simulation and obtain fault simulation results includes: inputting state data into the virtual simulation model to perform fault simulation and obtain fault simulation results; performing physical simulation based on the state data and obtain physical fault results includes: inputting state data into the hardware-in-the-loop simulation model to perform fault simulation and obtain physical fault results.

[0134] In one embodiment, when the processor executes the computer program, it further performs the following steps: weighted fusion of the first fault prediction result and the second fault prediction result to obtain a fault diagnosis result, including: obtaining the prediction accuracy of the deep residual shrinkage network model to obtain the first fault weight, and obtaining the prediction accuracy of the deep long short-term memory network model to obtain the second fault weight; and performing weighted calculation based on the first fault weight and the first fault prediction result, as well as the second fault weight and the second fault prediction result, to obtain the fault diagnosis result.

[0135] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0136] Obtain operating data of the traction converter;

[0137] Extract the time-frequency features of the operating condition data to obtain the target feature data;

[0138] The target feature data is input into the deep residual shrinkage network model for fault prediction to obtain the first fault prediction result, and the target feature data is input into the deep long short-term memory network model for fault prediction to obtain the second fault prediction result; the deep residual shrinkage network model and the deep long short-term memory network model are trained based on the digital twin state sample data of the traction converter under different operating conditions.

[0139] The fault diagnosis result is obtained by weighted fusion of the first fault prediction result and the second fault prediction result.

[0140] In one embodiment, when the computer program is executed by the processor, it further implements the following steps: the deep residual shrinkage network model is trained in the following manner: acquiring digital twin state sample data of the traction converter under different operating conditions and an initial deep residual shrinkage network model; the digital twin state sample data carries fault type labels; inputting the digital twin state sample data into the initial deep residual shrinkage network model for fault diagnosis to obtain a first sample fault diagnosis result; calculating the loss based on the first sample fault diagnosis result and the fault type label to obtain a first loss value; updating the parameters of the initial deep residual shrinkage network model based on the first loss value until the latest first loss value is less than a preset first loss threshold to obtain the deep residual shrinkage network model.

[0141] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: the deep long short-term memory network model is trained in the following manner: acquiring digital twin state sample data of the traction converter under different operating conditions and an initial deep long short-term memory network model; the digital twin state sample data carries fault type labels; inputting the digital twin state sample data into the initial deep long short-term memory network model for fault diagnosis to obtain a second sample fault diagnosis result; calculating the loss based on the second sample fault diagnosis result and the fault type label to obtain a second loss value; updating the parameters of the initial deep long short-term memory network model based on the second loss value until the latest second loss value is less than a preset second loss threshold to obtain the deep long short-term memory network model.

[0142] In one embodiment, when the computer program is executed by the processor, it further implements the following steps: the digital twin state sample data includes sample operating condition data and the fault type corresponding to the operating condition data; obtaining the digital twin state sample data of the traction converter under different operating conditions includes: based on the digital twin model, obtaining the operating condition data of the traction converter under different operating conditions and the fault type corresponding to each operating condition data.

[0143] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: based on the digital twin model, acquiring the operating condition data of the traction converter under different operating conditions and the fault types corresponding to each operating condition data includes: acquiring the state data of the traction converter and an initial digital twin model under different operating conditions; inputting the state data into the initial digital twin model for fault simulation to obtain fault simulation results; performing entity simulation based on the state data to obtain entity fault results; calculating the error based on the entity fault results and fault simulation results to obtain the error value; adjusting the parameters of the initial digital twin model based on the error value until the latest error value is less than a preset loss threshold to obtain the digital twin model; and generating the operating condition data of the traction converter under different operating conditions and the fault types corresponding to each operating condition data based on the digital twin model.

[0144] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: the digital twin model includes a virtual simulation model and a hardware-in-the-loop simulation model; inputting state data into the initial digital twin model to perform fault simulation and obtain fault simulation results includes: inputting state data into the virtual simulation model to perform fault simulation and obtain fault simulation results; performing physical simulation based on the state data and obtaining physical fault results includes: inputting state data into the hardware-in-the-loop simulation model to perform fault simulation and obtain physical fault results.

[0145] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: weighted fusion of the first fault prediction result and the second fault prediction result to obtain a fault diagnosis result, including: obtaining the prediction accuracy of the deep residual shrinkage network model to obtain the first fault weight, and obtaining the prediction accuracy of the deep long short-term memory network model to obtain the second fault weight; and performing weighted calculation based on the first fault weight and the first fault prediction result, as well as the second fault weight and the second fault prediction result, to obtain the fault diagnosis result.

[0146] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0147] Obtain operating data of the traction converter;

[0148] Extract the time-frequency features of the operating condition data to obtain the target feature data;

[0149] The target feature data is input into the deep residual shrinkage network model for fault prediction to obtain the first fault prediction result, and the target feature data is input into the deep long short-term memory network model for fault prediction to obtain the second fault prediction result; the deep residual shrinkage network model and the deep long short-term memory network model are trained based on the digital twin state sample data of the traction converter under different operating conditions.

[0150] The fault diagnosis result is obtained by weighted fusion of the first fault prediction result and the second fault prediction result.

[0151] In one embodiment, when the computer program is executed by the processor, it further implements the following steps: the deep residual shrinkage network model is trained in the following manner: acquiring digital twin state sample data of the traction converter under different operating conditions and an initial deep residual shrinkage network model; the digital twin state sample data carries fault type labels; inputting the digital twin state sample data into the initial deep residual shrinkage network model for fault diagnosis to obtain a first sample fault diagnosis result; calculating the loss based on the first sample fault diagnosis result and the fault type label to obtain a first loss value; updating the parameters of the initial deep residual shrinkage network model based on the first loss value until the latest first loss value is less than a preset first loss threshold to obtain the deep residual shrinkage network model.

[0152] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: the deep long short-term memory network model is trained in the following manner: acquiring digital twin state sample data of the traction converter under different operating conditions and an initial deep long short-term memory network model; the digital twin state sample data carries fault type labels; inputting the digital twin state sample data into the initial deep long short-term memory network model for fault diagnosis to obtain a second sample fault diagnosis result; calculating the loss based on the second sample fault diagnosis result and the fault type label to obtain a second loss value; updating the parameters of the initial deep long short-term memory network model based on the second loss value until the latest second loss value is less than a preset second loss threshold to obtain the deep long short-term memory network model.

[0153] In one embodiment, when the computer program is executed by the processor, it further implements the following steps: the digital twin state sample data includes sample operating condition data and the fault type corresponding to the operating condition data; obtaining the digital twin state sample data of the traction converter under different operating conditions includes: based on the digital twin model, obtaining the operating condition data of the traction converter under different operating conditions and the fault type corresponding to each operating condition data.

[0154] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: based on the digital twin model, acquiring the operating condition data of the traction converter under different operating conditions and the fault types corresponding to each operating condition data includes: acquiring the state data of the traction converter and an initial digital twin model under different operating conditions; inputting the state data into the initial digital twin model for fault simulation to obtain fault simulation results; performing entity simulation based on the state data to obtain entity fault results; calculating the error based on the entity fault results and fault simulation results to obtain the error value; adjusting the parameters of the initial digital twin model based on the error value until the latest error value is less than a preset loss threshold to obtain the digital twin model; and generating the operating condition data of the traction converter under different operating conditions and the fault types corresponding to each operating condition data based on the digital twin model.

[0155] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: the digital twin model includes a virtual simulation model and a hardware-in-the-loop simulation model; inputting state data into the initial digital twin model to perform fault simulation and obtain fault simulation results includes: inputting state data into the virtual simulation model to perform fault simulation and obtain fault simulation results; performing physical simulation based on the state data and obtaining physical fault results includes: inputting state data into the hardware-in-the-loop simulation model to perform fault simulation and obtain physical fault results.

[0156] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: weighted fusion of the first fault prediction result and the second fault prediction result to obtain a fault diagnosis result, including: obtaining the prediction accuracy of the deep residual shrinkage network model to obtain the first fault weight, and obtaining the prediction accuracy of the deep long short-term memory network model to obtain the second fault weight; and performing weighted calculation based on the first fault weight and the first fault prediction result, as well as the second fault weight and the second fault prediction result, to obtain the fault diagnosis result.

[0157] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0158] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0159] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A traction inverter failure diagnosis method characterized by, The method comprises: acquiring working condition data of a traction converter; extracting time-frequency features of the working condition data to obtain target feature data; inputting the target feature data into a deep residual shrinkage network model to perform fault prediction and obtain a first fault prediction result, and inputting the target feature data into a deep long short-term memory network model to perform fault prediction and obtain a second fault prediction result; the deep residual shrinkage network model and the deep long short-term memory network model are trained based on digital twin state sample data of the traction converter under different working conditions; the digital twin state sample data comprises working condition data of the traction converter under different working conditions generated based on a digital twin model and fault types corresponding to each working condition data, and the digital twin model comprises a virtual simulation model and a semi-physical simulation model; the step of obtaining the digital twin model comprises: acquiring state data of the traction converter under different working conditions and an initial digital twin model, inputting the state data into the virtual simulation model in the initial digital twin model to perform fault simulation and obtain a fault simulation result, inputting the state data into the semi-physical simulation model in the initial digital twin model to perform entity simulation and obtain an entity fault result, performing error calculation based on the entity fault result and the fault simulation result to obtain an error value, adjusting parameters of the initial digital twin model based on the error value until a latest error value is less than a preset loss threshold to obtain a digital twin model; performing weighted fusion based on the first fault prediction result and the second fault prediction result to obtain a fault diagnosis result; the step of obtaining the fault diagnosis result comprises: acquiring a prediction accuracy of the deep residual shrinkage network model to obtain a first fault weight, and acquiring a prediction accuracy of the deep long short-term memory network model to obtain a second fault weight; performing weighted calculation based on the first fault weight and the first fault prediction result and the second fault weight and the second fault prediction result to obtain a fault diagnosis result.

2. The method of claim 1, wherein, The deep residual shrinkage network model is trained in the following manner: acquiring digital twin state sample data of the traction converter under different working conditions and an initial deep residual shrinkage network model; the digital twin state sample data carries a fault type label; inputting the digital twin state sample data into the initial deep residual shrinkage network model to perform fault diagnosis and obtain a first sample fault diagnosis result; performing loss calculation based on the first sample fault diagnosis result and the fault type label to obtain a first loss value; updating parameters of the initial deep residual shrinkage network model based on the first loss value until a latest first loss value is less than a preset first loss threshold to obtain a deep residual shrinkage network model.

3. The method of claim 1, wherein, The deep long short-term memory network model is trained in the following manner: acquiring digital twin state sample data of the traction converter under different working conditions and an initial deep long short-term memory network model; the digital twin state sample data carries a fault type label; inputting the digital twin state sample data into the initial deep long short-term memory network model for fault diagnosis to obtain a second sample fault diagnosis result; performing loss calculation according to the second sample fault diagnosis result and the fault type label to obtain a second loss value; updating parameters of the initial deep long short-term memory network model based on the second loss value until a latest second loss value is less than a preset second loss threshold to obtain a deep long short-term memory network model.

4. The method of claim 1, wherein, The inputting the state data into the initial digital twin model for fault simulation to obtain a fault simulation result includes: inputting the state data into the virtual simulation model for fault simulation to obtain a fault simulation result; The performing entity simulation according to the state data to obtain an entity fault result includes: inputting the state data into the semi-physical simulation model for fault simulation to obtain an entity fault result.

5. The method of claim 1, wherein, The extracting time-frequency features of the working condition data to obtain target feature data includes: performing noise reduction on the working condition data to obtain noise-reduced working condition data; performing time-frequency feature analysis and extraction on the noise-reduced working condition data to obtain target feature data.

6. The method of claim 1, wherein, The working condition data of the traction converter includes three-phase side electrical data and DC side electrical data.

7. The method of claim 6, wherein, The three-phase side electrical data includes three-phase voltage data, three-phase current data, and motor data; The DC side electrical data includes DC bus voltage and DC bus current.

8. A traction inverter failure diagnosis device characterized by comprising: The device includes: an acquisition module configured to acquire working condition data of a traction converter; an extraction module configured to extract time-frequency features of the working condition data to obtain target feature data; a prediction module configured to input the target feature data into a deep residual shrinkage network model for fault prediction to obtain a first fault prediction result, and input the target feature data into a deep long short-term memory network model for fault prediction to obtain a second fault prediction result; the deep residual shrinkage network model and the deep long short-term memory network model are trained based on digital twin state sample data of the traction converter under different working conditions; the digital twin state sample data includes working condition data of the traction converter under different working conditions generated based on a digital twin model and corresponding fault types of each working condition data; the digital twin model includes a virtual simulation model and a semi-physical simulation model; the steps of obtaining the digital twin model include: acquiring state data of the traction converter under different working conditions and an initial digital twin model, inputting the state data into the virtual simulation model in the initial digital twin model for fault simulation to obtain a fault simulation result, inputting the state data into the semi-physical simulation model in the digital twin model for entity simulation to obtain an entity fault result, performing error calculation according to the entity fault result and the fault simulation result to obtain an error value, adjusting parameters of the initial digital twin model based on the error value until a latest error value is less than a preset loss threshold to obtain a digital twin model; The fusion module is configured to perform weighted fusion according to the first fault prediction result and the second fault prediction result to obtain a fault diagnosis result, and the step of obtaining the fault diagnosis result comprises: obtaining a prediction accuracy of the deep residual shrinkage network model to obtain a first fault weight, and obtaining a prediction accuracy of the deep long short-term memory network model to obtain a second fault weight; and performing weighted calculation according to the first fault weight and the first fault prediction result and the second fault weight and the second fault prediction result to obtain the fault diagnosis result. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The computer program, when executed by the processor, implements the steps of the method of any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by the processor, implements the steps of the method of any one of claims 1 to 7.

Images