Pulse spectrum pattern classification-based converter valve thyristor fault identification method and system

Abstract

The application provides a pulse spectrum classification-based converter valve thyristor fault identification method and system, which comprises the following steps: according to the design parameters of the converter valve in the actual project, a simulation model of a six-pulse thyristor converter valve is built; the size of the damping capacitor and the turn-off equivalent resistor in the simulation model is changed to simulate the fault condition of the thyristor stage, the return signal time sequence parameters of the thyristor stage under various normal or fault working conditions are collected, and the pulse spectrum of the thyristor return signal is made; the pulse spectrum data is analyzed, the pulse spectrum formed by the return signal sequence is made, and the input and output sample data set of the convolutional neural network model is used; the trained convolutional neural network model is used for the parameter monitoring of the actual project converter valve thyristor stage, the measured voltage data, current data and return signal pulse spectrum are taken as the input, and the working state of the thyristor stage is solved and obtained.

Description

Technical Field

[0001] This invention belongs to the field of high voltage direct current transmission technology, and particularly relates to a method and system for identifying faults in converter valve thyristors based on pulse spectrum classification. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] In high-voltage direct current (HVDC) transmission systems, thyristor converter valves play a crucial role. As key components for AC-DC power conversion and transmission, they directly impact the stability and reliability of the power system. Within this critical system, the thyristor, as the core component of the thyristor valve, is responsible for achieving precise current control and efficient conversion. However, over time and due to the unpredictability of external environmental factors, thyristors may age and fail, potentially leading to performance degradation in the power system and even the risk of power outages.

[0004] To ensure the stable operation of the power system, monitoring and managing the status of thyristors is crucial. Timely detection of thyristor aging and potential faults helps to take preventative maintenance measures and reduce the risk of unexpected failures.

[0005] However, current thyristor valve maintenance methods have some significant shortcomings. They mainly employ annual disassembly and power outage testing, which, while allowing for comprehensive component inspection, is time-consuming, labor-intensive, and wasteful of resources. Furthermore, they cannot monitor for failures in real time before they occur, thus failing to provide early warnings and effective equipment maintenance.

[0006] In addition, current methods for using neural networks to identify thyristor faults mainly rely on measuring the voltage, current, and impedance values ​​of key components and comparing them with actual measurements to obtain the status of the converter valve. This requires adding a status monitoring component to the key thyristor components, increasing design complexity, and it cannot determine the status of converter valves without such a component status monitoring component. Summary of the Invention

[0007] To overcome the shortcomings of the prior art, this invention provides a method for identifying thyristor faults in converter valves based on pulse spectrum classification, which achieves effective real-time thyristor fault detection through a model.

[0008] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0009] Firstly, a fault identification method for thyristors in converter valves based on pulse spectrum classification is disclosed, including:

[0010] Based on the design parameters of the converter valve in actual engineering, a simulation model of a six-pulse thyristor converter valve was built.

[0011] By changing the values ​​of the damping capacitor and the turn-off equivalent resistance in the simulation model, the fault conditions of the thyristor stage are simulated. The time series parameters of the feedback signals of each thyristor under various normal or fault operating conditions are collected and a pulse spectrum of the thyristor feedback signal is generated.

[0012] Analyze the pulse spectrum data to create pulse spectra formed by the return signal sequence, which are used as input and output sample datasets for convolutional neural network models;

[0013] The sample dataset is trained using a convolutional neural network (CNN) model. The CNN model is then used to classify pulse spectrum features. Once the required recognition accuracy is achieved, the weights of the trained CNN model are used as the initial values ​​for online learning, resulting in a well-trained CNN model.

[0014] The trained convolutional neural network model is used for parameter monitoring of the thyristor stage of the converter valve in actual engineering. The measured voltage data, current data and the pulse spectrum of the feedback signal are used as inputs to solve for the working state of the thyristor stage.

[0015] As a further technical solution, the valve arm of the six-pulse thyristor converter valve in the simulation model is mainly composed of multiple identical thyristor stages connected in series with a saturated reactor.

[0016] As a further technical solution, the thyristor stage includes a thyristor control unit, a DC voltage equalization resistor, a damping capacitor, and a damping resistor.

[0017] As a further technical solution, when simulating fault conditions of the thyristor stage, assuming that the number of thyristor stages in series in each thyristor valve is n, the parameters that need to be identified are: n damping resistors, n damping capacitors, and n DC voltage equalization resistors. By changing these electrical parameters, the working state of the thyristor stage under different fault conditions can be simulated.

[0018] As a further technical solution, during the process of creating the sample dataset, each parameter is selected to degrade at a set interval to obtain the number of sample datasets, corresponding to different fault states, and fault modes that may occur in the thyristor control unit are added, including abnormal temperature and fiber optic loss faults.

[0019] As a further technical solution, the generated reward signal pulse spectrum is obtained by superimposing all the reward signal spectra and using different gray levels to distinguish the density of pulse signals occurring at different times.

[0020] As a further technical solution, the input sample dataset is the characteristic spectrum of the return signal sequence; the output sample dataset is the working or fault state of each thyristor under each aging condition.

[0021] As a further technical solution, gradient descent is used to adjust the weights of the neural network, enabling it to learn online and output results that are closer to reality.

[0022] Secondly, a fault identification system for converter valve thyristors based on pulse spectrum classification is disclosed, including:

[0023] The model building module is configured to build a simulation model of a six-pulse thyristor converter valve based on the design parameters of the converter valve in actual engineering.

[0024] The pulse spectrum generation module is configured to: change the magnitude of the damping capacitor and the turn-off equivalent resistance in the simulation model, simulate the fault conditions of the thyristor stage, collect the time series parameters of the feedback signals of each stage of the thyristor under various normal or fault operating conditions, and generate a pulse spectrum of the thyristor feedback signal.

[0025] The convolutional neural network model training module is configured to: analyze pulse spectrum data, generate pulse spectra formed by the return signal sequence, and use them as the input and output sample dataset for the convolutional neural network model;

[0026] The sample dataset is trained using a convolutional neural network (CNN) model. The CNN model is then used to classify pulse spectrum features. Once the required recognition accuracy is achieved, the weights of the trained CNN model are used as the initial values ​​for online learning, resulting in a well-trained CNN model.

[0027] The thyristor-level operating status identification module is configured to: use a trained convolutional neural network model for parameter monitoring of the thyristor level in actual engineering converter valves, and use measured voltage data, current data and feedback signal pulse spectrum as inputs to solve for the operating status of the thyristor level.

[0028] The above one or more technical solutions have the following beneficial effects:

[0029] The fault identification method based on neural networks in the technical solution of this invention can provide early warning of faults, which can help with timely equipment maintenance, thereby reducing power outage time and maintenance costs.

[0030] Traditional manual disassembly and step-by-step inspection methods may require significant manpower and resources, while neural network-based methods can reduce costs and improve efficiency. For large-scale thyristor datasets, neural networks can process and analyze large numbers of trigger timing sequences more quickly, thereby improving efficiency.

[0031] Furthermore, neural network models can adapt to new data distributions through transfer learning or be updated through online learning, thus better adapting to changes in equipment. Neural network-based methods can provide early warnings of faults, facilitating timely equipment maintenance and reducing power outage time and repair costs.

[0032] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0033] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0034] Figure 1 This is a flowchart of a method according to an embodiment of the present invention;

[0035] Figure 2 The simulation model diagram of the six-pulse thyristor-based converter valve is shown in the embodiment.

[0036] Figure 3 This is a schematic diagram of the thyristor voltage and feedback signal during one cycle in an embodiment.

[0037] Figure 4 The following is a timing diagram of the negative feedback signal of the healthy and degraded thyristor stage feedback within one cycle of the embodiment;

[0038] Figure 5 This is a comparison diagram of power supply voltage and feedback signal timing for an example. Detailed Implementation

[0039] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0040] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0041] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0042] Example 1

[0043] This embodiment discloses a method for identifying faults in thyristors of converter valves based on pulse spectrum classification, including:

[0044] like Figure 1As shown in this embodiment, a method for identifying faults in a converter valve thyristor based on pulse spectrum classification includes the following specific steps:

[0045] A. Based on the design parameters of the converter valve in actual engineering, including the thyristor control unit, DC equalizing resistor, damping capacitor, and damping resistor, etc., the thyristor converter valves currently used mainly include six-pulse converters and twelve-pulse converter valves. Since a 12-pulse converter can be regarded as two 6-pulse converters in series, the 6-pulse converter is mostly used as the main research object of thyristor converter valves in most theoretical studies and analyses. A simulation model of the six-pulse thyristor converter valve is built in simulation software, such as... Figure 2 As shown.

[0046] By changing the values ​​of the damping capacitor and the turn-off equivalent resistance in the simulation model, fault conditions at the thyristor stage can be simulated, mainly including insulation resistance aging, RC branch aging, thyristor short circuit, and open circuit. Under different fault conditions, the thyristor stage feedback signal and the voltage and current waveforms across the converter valve will change accordingly. Therefore, different fault conditions of the thyristor converter valve can be simulated by adjusting the DC voltage equalization circuit and the parameters of the RC components. By collecting the time series parameters of the feedback signals of each thyristor stage under various normal or fault operating conditions, a pulse spectrum of the thyristor feedback signal can be created. This allows a large amount of data, such as the time sequence signals and signal amplitudes of multiple thyristor stages, to be aggregated in a single pulse spectrum, providing more comprehensive time-domain and frequency-domain information and facilitating faster computer processing.

[0047] The generated reward signal pulse spectrum is created by superimposing the spectrum of all reward signals caused by the same trigger signal, and using different gray levels to distinguish the density of pulse signals occurring at different times, so as to analyze the pulse spectrum data.

[0048] B. Constructing a sample dataset: Assume n is the number of thyristors connected in series in each single valve. Then, the electrical parameters to be identified are n damping capacitance values, n damping resistance values, and n DC equalizing resistance values, for a total of 3n parameters. When constructing the sample dataset, the more datasets the better to fully characterize various typical degradation states of the thyristor stage. Here, each parameter is degraded at 5% intervals, and the degradation range is set to [0, 45%]. Therefore, the sample dataset contains 10... n ×10 n ×10 n The system is divided into groups corresponding to different aging states. When the parameter degradation reaches 40%, a parameter aging fault is predicted to occur. Other possible fault modes of the thyristor control unit are also included, such as abnormal temperature and fiber optic loss. Depending on the circuit fault, for example, fiber optic loss can cause a decrease in the trigger signal amplitude, which will be reflected in the report signal pulse spectrum.

[0049] C. Create a sample input dataset. Select the time series spectrum of the thyristor stage's return signal at each sampling time within a period (0.02s), and overlay all the spectra. After all the return signal spectra overlap, use different gray levels to distinguish the density of the pulse signal at different times. Collect the loop current and the voltage images across the reference thyristor stage, and use them together with the return signal pulse spectrum as the input dataset for the convolutional neural network.

[0050] D. Establish a convolutional neural network model and train it offline on the collected thyristor current and voltage data. The neural network is a 6-layer nonlinear neural network, including an input layer, convolutional layer, pooling layer, activation function layer, fully connected layer, and output layer. First, the input layer receives the raw image data; the convolutional layer extracts features from the image and outputs a feature map; the pooling layer is used for downsampling and reducing the size of the feature map; the activation function layer introduces nonlinearity, enabling the network to learn more complex patterns and representations; the fully connected layer learns weights to combine the features from the previous layer, providing input for the final classification or regression task; and the output layer produces the final output result. The activation function used in the activation function layer is the PReLU function.

[0051]

[0052] f(x) = max(0,x) + α·min(0,x)

[0053] In the formula, x represents the weighted sum of the input signals, n is the total number of neurons in the pooling layer, and w i For connection weights, a i The output of the pooling layer is the value at certain locations in the feature map of the previous pooling layer, capturing different features of the image. b is the bias term, and the parameter α is learnable and can be automatically adjusted according to the data, introducing a learnable negative slope in the negative value region of the function.

[0054] E. Use a convolutional neural network model to train the sample dataset offline, use the convolutional neural network to classify the pulse spectrum features, and after achieving the required recognition accuracy, use the weights of the convolutional neural network obtained from offline training as the initial values ​​for online learning of the convolutional neural network to obtain the trained convolutional neural network model.

[0055] By using offline sample data and gradient descent with varying learning rate to adjust the weights of the neural network, the system is trained and adjusted online so that the output of the neural network is close to the actual value. When the loss function L converges to its minimum value, the weights obtained from offline training are used as the initial weights for online monitoring of the neural network.

[0056]

[0057] In the formula, iL,j with u VTr,j i represents the predicted value of the data in the neural network model. Lm,j with u VTrm,j The values ​​represent the actual measured current and voltage in the six-pulse thyristor converter model, where N is the number of sampling points and t is the voltage. pi,j With t pim,j These are the timing sequences of the feedback signals in the neural network model and the actual model, respectively. 'a' represents the number of thyristors in a single bridge arm, and 'n' represents the sequence of feedback signals from each thyristor stage within one power frequency cycle. F. The trained 6-layer convolutional neural network model is used for parameter monitoring of the thyristor stage in an actual engineering converter valve. Measured voltage and circuit current data across the reference thyristor are used, combined with the measured timing spectrum of the feedback signal pulses as input. The input and output layers have the same variable dimensions. Finally, the parameters of each thyristor stage and the fault identification results are obtained.

[0058] The method for processing the pulse spectrum of the measured return signal is consistent with that of the simulation sample, while keeping the structure of the convolutional neural network unchanged.

[0059] This embodiment uses measured voltage and current data combined with measured feedback signal pulse timing spectrum as input. If the feedback signal amplitude weakens, problems such as fiber loss and abnormal temperature occur in the trigger module. If the thyristor voltage amplitude increases steadily and an abnormal feedback signal occurs at a certain level in the feedback signal pulse, a unipolar thyristor short circuit problem occurs. By using a trained neural network to identify and judge the feature spectrum and voltage and current characteristics under various fault conditions, the actual working state is classified and identified to obtain the working state of the thyristor level.

[0060] Neural network-based methods can provide early warnings of faults, enabling timely equipment maintenance and thus reducing power outage time and repair costs.

[0061] Example 2

[0062] The purpose of this embodiment is to provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the above-described method.

[0063] Example 3

[0064] The purpose of this embodiment is to provide a computer-readable storage medium.

[0065] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the steps of the above-described method.

[0066] Example 4

[0067] The purpose of this embodiment is to provide a thyristor fault identification system for converter valves based on pulse spectrum classification, including:

[0068] The model building module is configured to build a simulation model of a six-pulse thyristor converter valve based on the design parameters of the converter valve in actual engineering.

[0069] The pulse spectrum generation module is configured to: change the magnitude of the damping capacitor and the turn-off equivalent resistance in the simulation model, simulate the fault conditions of the thyristor stage, collect the time series parameters of the feedback signals of each stage of the thyristor under various normal or fault operating conditions, and generate a pulse spectrum of the thyristor feedback signal.

[0070] The convolutional neural network model training module is configured to: analyze pulse spectrum data, generate pulse spectra formed by the return signal sequence, and use them as the input and output sample dataset for the convolutional neural network model;

[0071] The sample dataset is trained using a convolutional neural network (CNN) model. The CNN model is then used to classify pulse spectrum features. Once the required recognition accuracy is achieved, the weights of the trained CNN model are used as the initial values ​​for online learning, resulting in a well-trained CNN model.

[0072] The thyristor-level operating status identification module is configured to: use a trained convolutional neural network model for parameter monitoring of the thyristor level in actual engineering converter valves, and use measured voltage data, current data, and feedback signal pulse spectrum as inputs to solve for the operating status of the thyristor level.

[0073] The steps and methods involved in the apparatuses of Embodiments 2, 3, and 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0074] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0075] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for fault identification of thyristors in converter valves based on pulse spectrum classification, characterized in that, include: Based on the design parameters of the converter valve in actual engineering, a simulation model of a six-pulse thyristor converter valve was built. By changing the values ​​of the damping capacitor and the turn-off equivalent resistance in the simulation model, the fault conditions of the thyristor stage are simulated. The time series parameters of the feedback signals of each thyristor under various normal or fault operating conditions are collected and a pulse spectrum of the thyristor feedback signal is generated. The generated reward signal pulse spectrum is obtained by superimposing all the reward signal spectra and using different gray levels to distinguish the density of pulse signal occurrences at different times. Analyze the pulse spectrum data to create pulse spectra formed by the return signal sequence, which are used as input and output sample datasets for convolutional neural network models; The sample dataset is trained using a convolutional neural network (CNN) model. The CNN model is then used to classify pulse spectrum features. Once the required recognition accuracy is achieved, the weights of the trained CNN model are used as the initial values ​​for online learning, resulting in a well-trained CNN model. The trained convolutional neural network model is used for parameter monitoring of the thyristor stage of the converter valve in actual engineering. The measured voltage data, current data and the pulse spectrum of the feedback signal are used as inputs to solve for the working state of the thyristor stage. The neural network is a 6-layer nonlinear neural network, comprising an input layer, convolutional layers, pooling layers, activation function layers, fully connected layers, and an output layer. First, the input layer receives the raw image data; the convolutional layers extract features from the image and output a feature map; the pooling layers are used for downsampling and reducing the size of the feature map; the activation function layers introduce nonlinearity, enabling the network to learn more complex patterns and representations; the fully connected layers learn weights to combine features from the previous layer, providing input for the final classification or regression task. The final output result is then generated at the output layer.

2. The method for thyristor fault identification in a converter valve based on pulse spectrum classification as described in claim 1, characterized in that, The valve arm of the six-pulse thyristor converter valve in the simulation model is mainly composed of multiple identical thyristor stages connected in series with a saturated reactor.

3. The method for thyristor fault identification in a converter valve based on pulse spectrum classification as described in claim 1, characterized in that, The thyristor stage includes a thyristor control unit, a DC equalizing resistor, a damping capacitor, and a damping resistor.

4. The method for thyristor fault identification in a converter valve based on pulse spectrum classification as described in claim 1, characterized in that, When simulating fault conditions of the thyristor stage, assuming that the number of thyristor stages in series in each thyristor valve is n, the parameters that need to be identified are: n damping resistors, n damping capacitors, and n DC voltage equalization resistors. By changing these electrical parameters, the working state of the thyristor stage under different fault conditions can be simulated.

5. The method for thyristor fault identification in a converter valve based on pulse spectrum classification as described in claim 1, characterized in that, During the creation of the sample dataset, each parameter is degraded at a set interval to obtain the number of sample datasets, corresponding to different fault states, and fault modes that may occur in the thyristor control unit are added, including abnormal temperature and fiber optic loss faults.

6. The method for fault identification of converter valve thyristors based on pulse spectrum classification as described in claim 1, characterized in that, input The sample dataset consists of the characteristic spectrum of the return signal sequence; the output sample dataset consists of the operating or fault state of each thyristor under each aging condition. By using gradient descent to adjust the weights of a neural network, enabling it to learn online, the output results can more closely approximate reality.

7. A thyristor fault identification system for converter valves based on pulse spectrum classification, characterized in that, include: The model building module is configured to build a simulation model of a six-pulse thyristor converter valve based on the design parameters of the converter valve in actual engineering. The pulse spectrum generation module is configured to: change the magnitude of the damping capacitor and the turn-off equivalent resistance in the simulation model to simulate the fault conditions of the thyristor stage, collect the time series parameters of the feedback signals of each stage of the thyristor under various normal or fault operating conditions, and generate a pulse spectrum of the thyristor feedback signal; the generated feedback signal pulse spectrum is generated by superimposing all the feedback signal spectra and using different gray levels to distinguish the density of the pulse signal at different times. The convolutional neural network model training module is configured to: analyze pulse spectrum data, generate pulse spectra formed by the return signal sequence, and use them as the input and output sample dataset for the convolutional neural network model; The sample dataset is trained using a convolutional neural network (CNN) model. The CNN model is then used to classify pulse spectrum features. Once the required recognition accuracy is achieved, the weights of the trained CNN model are used as the initial values ​​for online learning, resulting in a well-trained CNN model. The thyristor-level operating status identification module is configured to: use the trained convolutional neural network model for parameter monitoring of the thyristor level in actual engineering converter valves, and use measured voltage data, current data and feedback signal pulse spectrum as inputs to solve for the operating status of the thyristor level; The neural network is a 6-layer nonlinear neural network, comprising an input layer, convolutional layers, pooling layers, activation function layers, fully connected layers, and an output layer. First, the input layer receives the raw image data; the convolutional layers extract features from the image and output a feature map; the pooling layers are used for downsampling and reducing the size of the feature map; the activation function layers introduce nonlinearity, enabling the network to learn more complex patterns and representations; the fully connected layers learn weights to combine features from the previous layer, providing input for the final classification or regression task. The final output result is then generated at the output layer.

8. The converter valve thyristor fault identification system based on pulse spectrum classification as described in claim 7, characterized in that, The valve arm of the six-pulse thyristor converter valve in the simulation model is mainly composed of multiple identical thyristor stages connected in series with a saturated reactor.

9. The converter valve thyristor fault identification system based on pulse spectrum classification as described in claim 7, characterized in that, The thyristor stage includes a thyristor control unit, a DC equalizing resistor, a damping capacitor, and a damping resistor.

10. The converter valve thyristor fault identification system based on pulse spectrum classification as described in claim 7, characterized in that, When simulating fault conditions of the thyristor stage, assuming that the number of thyristor stages in series in each thyristor valve is n, the parameters that need to be identified are: n damping resistors, n damping capacitors, and n DC voltage equalization resistors. By changing these electrical parameters, the working state of the thyristor stage under different fault conditions can be simulated.

11. The converter valve thyristor fault identification system based on pulse spectrum classification as described in claim 7, characterized in that, During the creation of the sample dataset, each parameter is degraded at a set interval to obtain the number of sample datasets, corresponding to different fault states, and fault modes that may occur in the thyristor control unit are added, including abnormal temperature and fiber optic loss faults.

12. The converter valve thyristor fault identification system based on pulse spectrum classification as described in claim 7, characterized in that, input The sample dataset consists of the characteristic spectrum of the return signal sequence; the output sample dataset consists of the operating or fault state of each thyristor under each aging condition. By using gradient descent to adjust the weights of a neural network, enabling it to learn online, the output results can more closely approximate reality.

13. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method described in any one of claims 1-6.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it performs the steps of the method described in any one of claims 1-6 above.

Images