Fault diagnosis method and system in starting process of pumped storage unit

Abstract

The invention relates to the technical field of pumped storage power station relay protection, and provides a pumped storage unit starting process fault diagnosis method and system, and the method comprises the steps: arranging an electrical quantity waveform collected in real time and a historically stored electrical quantity waveform according to a sequence, and then converting the electrical quantity waveform into a gray scale picture; based on gray-scale picture training and testing, generating a fault diagnosis model in the starting process of the pumped storage unit; the pumping and storage unit starting process fault diagnosis model carries out calculation fault judgment according to the input gray scale picture; the pumped storage unit start-up process fault diagnosis model outputs a fault judgment probability, judges that a fault occurs in the start-up process when a probability threshold value is met, and instantaneously triggers start-up fault remote signaling of a generator-transformer unit protection device; and after the generator-transformer unit protection device receives the starting fault remote signaling, the instantaneous outlet quickly cuts off the circuit breaker at the machine end of the pumped storage unit, so as to prevent the pumped storage unit from starting with a fault. According to the method, the accuracy of fault judgment can be effectively improved, and the method has wide applicability and relatively high practicability.

Description

Technical Field

[0001] The present invention relates to the technical field of relay protection for pumped storage power stations, and in particular to a method and system for diagnosing faults during the startup process of a pumped storage unit. Background Art

[0002] Pumped storage is the most technologically mature, most significant carbon emission reduction benefit over its lifecycle, most economically efficient, and most suitable for large-scale development. It is a green, low-carbon, clean, and flexible power source for the power system. It has peak-shaving, valley-filling, frequency regulation, phase regulation, and emergency standby functions, making it the most mature large-capacity electrical energy storage method. Pumped storage units operate in both power generation and pumping modes. Compared with conventional hydropower units, they have many special features, including diverse operating modes, special starting methods for pumping directions, and opposite phase sequences of electrical quantities during power generation and pumping operations. The startup process of a pumped storage unit is a complex one, involving the coordinated work of multiple systems and protection measures. If an alarm is not issued or the load is not removed in a timely manner when a fault occurs during the startup process of the pumped storage unit, irreversible damage will be caused to the generator itself, and in serious cases, the safe and stable operation of the entire power system may be affected. At present, the function of the protection device is mainly realized by digital filtering of the collected data, and then mathematical and logical operations are performed on the digital signal sampling Fourier algorithm, and analysis and judgment are performed, and finally a trip command, signal command or calculation result is output. During the startup process of the pumped storage unit, its frequency is also constantly changing. The protection device automatically adjusts the sampling frequency according to the actual frequency to adapt to the frequency changes of the pumped storage unit during the startup process, ensuring that the differential protection can operate accurately at different frequencies. This requires a certain amount of delayed protection to ensure correct exit action after a fault occurs during the startup process of the pumped storage unit, which may have a certain impact on the safe and stable operation of the pumped storage unit.

[0003] With the continuous advancement of artificial intelligence (AI) technology, many experts and scholars have applied AI to the field of fault diagnosis. The existing AI algorithms mainly use a large amount of collected labeled data to train AI models for fault diagnosis during the startup process of pumped storage units. On the one hand, during the startup process of pumped storage units, the frequency of faults is relatively low, and fault data is often difficult to obtain in large quantities. This may result in the model being unable to learn comprehensive and accurate fault characteristics due to insufficient data, affecting diagnostic performance. On the other hand, even if the model training is completed, in the actual online fault diagnosis of pumped storage units, it may take a certain amount of time to process and diagnose new data. In particular, the frequency changes accompanying the startup process of the pumped storage unit are constantly changing, and the collected data and its changes are subject to certain fluctuations, which may lead to misdiagnosis or missed diagnosis in actual operation. Summary of the Invention

[0004] The purpose of the present invention is to solve at least one technical problem in the background technology and to provide a method and system for diagnosing faults during the startup process of a pumped storage unit.

[0005] To achieve the above object, the present invention provides a method for diagnosing faults during the startup process of a pumped storage unit, comprising:

[0006] The protection device of the generator-transformer group of the pumped-storage power station collects the electrical quantity waveform of the pumped-storage unit, arranges the real-time collected electrical quantity waveform and the historically saved electrical quantity waveform in sequence, and then converts them into a grayscale image;

[0007] A regularly updated convolutional model knowledge base is established based on grayscale images. Gaussian noise is introduced to expand the grayscale image sample set. This expanded sample set is then input into a deep convolutional network model for training and testing, generating a fault diagnosis model for the startup process of pumped storage units.

[0008] The generator-transformer protection device inputs the real-time generated grayscale image into the pumped storage unit startup process fault diagnosis model, which then calculates the fault diagnosis based on the input grayscale image.

[0009] The fault diagnosis model for the pumped storage unit startup process outputs the fault judgment probability and sets a probability threshold. When the probability threshold is met, it is judged that a fault has occurred during the startup process and the startup fault telesignal of the generator-transformer unit protection device is instantly triggered.

[0010] When the generator-transformer protection device receives the remote signal of the starting fault, the instantaneous output quickly cuts off the circuit breaker at the pumped storage unit end to prevent the pumped storage unit from starting with a fault.

[0011] According to one aspect of the present invention, the update cycle of the convolution model knowledge base is in years. During the update, the electrical quantity waveforms collected and recorded in real time and stored historically are converted into grayscale images and added to the convolution model knowledge base. At the same time, the judgment results obtained by the fault diagnosis model of the pumped storage unit startup process are converted into grayscale images and added to the convolution model knowledge base. Then, the fault diagnosis model of the pumped storage unit startup process is updated based on the updated convolution model knowledge base.

[0012] According to one aspect of the present invention, the training process of the deep convolutional network model includes:

[0013] The grayscale images are divided into training set and test set at a ratio of 80% and 20%;

[0014] The grayscale images of the training set are input into the deep convolutional network model. The deep convolutional network model uses a sliding window. The window size is the length of the data collected in 10ms at a sampling rate of 48 points per cycle, and the step size is the length of the interval between two adjacent sampling points.

[0015] By introducing an optimization algorithm to optimize the learning rate and number of hidden layer neurons of the convolutional neural network, and setting up a loss function, the deep convolutional network image recognition model is trained. When the model training accuracy reaches more than 90%, the training is stopped and the training parameters are saved.

[0016] The grayscale images of the test set are input into the trained deep convolutional network model to determine the accuracy of model recognition. When the accuracy is greater than 90%, the model is saved to obtain a fault diagnosis model for the startup process of the pumped storage unit.

[0017] According to one aspect of the present invention, a grayscale image is input into a fault diagnosis model for the startup process of a pumped storage unit. The model outputs a recognition probability, and the probability threshold is set to 0.85. When the model output probability value is greater than 0.85, it is determined that a fault has occurred during the startup process of the pumped storage unit.

[0018] According to one aspect of the present invention, when the pumped storage unit startup process fault diagnosis model determines that the pumped storage unit operating condition is a pumped storage unit startup process fault, the startup fault telesignal of the generator-transformer group protection device is instantaneously triggered and the telesignal is set to 1; when the pumped storage unit startup process fault diagnosis model determines that the pumped storage unit operating condition is a pumped storage unit startup process non-fault condition, the startup fault telesignal is set to 0.

[0019] According to one aspect of the present invention, the generator-transformer group protection device receives the start-up fault telesignaling point information in real time. When the generator-transformer group protection device receives the start-up fault telesignaling point and sets it to 1, the instantaneous outlet quickly cuts off the pumped storage unit end circuit breaker. When the generator-transformer group protection device receives the start-up fault telesignaling point and sets it to 0, the generator-transformer group protection device locks the start-up and shutdown related protection outlet components until the start-up process is completed.

[0020] According to one aspect of the present invention, it also includes: recording and saving the electrical quantities collected on-site and the model discrimination results, judging whether the model discrimination results are consistent with the actual operating conditions of the pumped storage unit startup process; if the actual operating conditions of the pumped storage unit startup process are consistent with the model discrimination results, then the discrimination is deemed correct; if the actual operating conditions of the pumped storage unit startup process are inconsistent with the model discrimination results, then the discrimination results are corrected according to the actual operating conditions, and then used to update the convolution model knowledge base.

[0021] To achieve the above object, the present invention further provides a pumped storage unit startup process fault diagnosis system, comprising:

[0022] Grayscale image acquisition module: The protection device of the generator-transformer group of the pumped-storage power station collects the electrical quantity waveform of the pumped-storage unit, and arranges the real-time collected electrical quantity waveform and the historically saved electrical quantity waveform in sequence and converts them into grayscale images;

[0023] The fault diagnosis model generation module establishes a regularly updated convolutional model knowledge base based on grayscale images. It also introduces Gaussian noise to expand the grayscale image sample set. This expanded sample set is then input into a deep convolutional network model for training and testing, generating a fault diagnosis model for the pumped storage unit startup process.

[0024] In the fault identification module, the generator-transformer protection device inputs the grayscale image generated in real time into the pumped storage unit startup process fault diagnosis model. The pumped storage unit startup process fault diagnosis model calculates the fault identification based on the input grayscale image.

[0025] The fault diagnosis model of the pumped storage unit startup process outputs the fault judgment probability and sets a probability threshold. When the probability threshold is met, it is judged that a fault has occurred during the startup process and the startup fault telesignal of the generator-transformer unit protection device is instantly triggered.

[0026] The pumped storage unit control module, when the generator-transformer unit protection device receives the remote signal of the starting fault, the instantaneous output quickly cuts off the pumped storage unit end circuit breaker to prevent the pumped storage unit from starting with a fault.

[0027] To achieve the above-mentioned objectives, the present invention also provides an electronic device, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the above-mentioned method for diagnosing faults during the startup process of the pumped storage unit.

[0028] To achieve the above-mentioned object, the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the fault diagnosis method for the startup process of the pumped storage unit as described above is implemented.

[0029] According to the present invention, Gaussian noise is introduced when establishing the convolutional model knowledge base. This not only expands the sample set of grayscale images in the model knowledge base, but also, because environmental factors such as temperature, humidity, and electromagnetic interference may vary within the pumped-storage unit, these changes may affect the data features collected by sensors and, consequently, the diagnostic accuracy of the artificial intelligence model. In machine learning, Gaussian noise can simulate real-world uncertainty and noise, helping the model learn more features, improving its generalization ability and effectively preventing overfitting. Changes in the operating environment and operating conditions of the pumped-storage unit also lead to the emergence of new fault types. This solution utilizes field feedback data to establish a regularly updated convolutional model knowledge base, continuously iteratively updating the model and making the model more adaptable. Furthermore, using grayscale images for convolutional neural network model training, compared to directly processing collected electrical quantities for model training, grayscale images can better highlight information such as the trends and extent of voltage and current changes during the pumped-storage unit's startup and shutdown processes, making it easier for the convolutional neural network to learn these important features. At the same time, it can also effectively reduce the data dimension and computational complexity. Due to the low data dimension of grayscale images, when the convolutional neural network processes grayscale images, the number of model parameters is relatively small, the storage space required for training is smaller, and the transmission bandwidth occupancy is lower, thereby improving the efficiency of data processing. At the same time, the reduction of model parameters can effectively reduce the complexity of the model, making it easier for the model to converge to the global optimal solution, reducing the risk of overfitting, and improving the model's generalization ability on unknown data. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A flowchart schematically illustrates a method for diagnosing faults during the startup process of a pumped storage unit according to an embodiment of the present invention;

[0031] Figure 2 This is a diagram showing the electrical quantity ranking at the pumped storage unit end of Example 1. DETAILED DESCRIPTION

[0032] The present invention will now be discussed with reference to exemplary embodiments. It should be understood that the embodiments discussed are only intended to enable those skilled in the art to better understand and implement the present invention, rather than to imply any limitation on the scope of the present invention.

[0033] As used herein, the term "including" and variations thereof are to be interpreted as open-ended terms meaning "including, but not limited to." The term "based on" is to be interpreted as "based, at least in part, on." The terms "one embodiment" and "an embodiment" are to be interpreted as "at least one embodiment."

[0034] Figure 1The flowchart schematically shows a method for diagnosing faults during the startup process of a pumped storage unit according to an embodiment of the present invention. Figure 1 As shown, in this embodiment, the fault diagnosis method during the startup process of the pumped storage unit includes:

[0035] The protection device of the generator-transformer group of the pumped-storage power station collects the electrical quantity waveforms at the pumped-storage unit end, arranges the real-time collected electrical quantity waveforms and the historically saved electrical quantity waveforms in sequence, and then converts them into grayscale images (for example, using the Pillow library);

[0036] A regularly updated convolutional model knowledge base is established based on grayscale images. Gaussian noise is introduced to expand the grayscale image sample set. This expanded sample set is then input into a deep convolutional network model for training and testing, generating a fault diagnosis model for the startup process of pumped storage units.

[0037] The generator-transformer protection device inputs the real-time generated grayscale image into the pumped storage unit startup process fault diagnosis model, which then calculates the fault diagnosis based on the input grayscale image.

[0038] The fault diagnosis model for the startup process of the pumped storage unit outputs a fault judgment probability, sets a probability threshold value (for example, manually), and when the probability threshold value is met, it is judged that a fault has occurred during the startup process, and the startup fault telesignal of the generator-transformer unit protection device is instantaneously triggered;

[0039] When the generator-transformer protection device receives the remote signal of the starting fault, the instantaneous output quickly cuts off the circuit breaker at the pumped storage unit end to prevent the pumped storage unit from starting with a fault.

[0040] Furthermore, according to an embodiment of the present invention, the electrical quantities collected by the protection device of the generator-transformer group of the pumped-storage power station include three-phase current and three-phase voltage waveforms at the terminal of the pumped-storage unit.

[0041] The electrical quantity waveforms collected by the protection device of the generator-transformer group of the pumped-storage power station are, in order from top to bottom, phase A voltage, phase B voltage, phase C voltage, phase A current, phase B current and phase C current, and the collected waveforms are converted into grayscale images.

[0042] Furthermore, according to one embodiment of the present invention, the update cycle of the convolution model knowledge base is in years. During the update, the electrical quantity waveforms collected and recorded in real time and stored historically are converted into grayscale images and added to the convolution model knowledge base. At the same time, the judgment results obtained through the fault diagnosis model of the pumped storage unit startup process are converted into grayscale images and added to the convolution model knowledge base. Then, the fault diagnosis model of the pumped storage unit startup process is updated based on the updated convolution model knowledge base.

[0043] Furthermore, according to one embodiment of the present invention, the training process of the deep convolutional network model includes:

[0044] The grayscale images are divided into training set and test set at a ratio of 80% and 20%;

[0045] The grayscale images of the training set are input into the deep convolutional network model. The deep convolutional network model uses a sliding window. The window size is the length of the data collected in 10ms at a sampling rate of 48 points per cycle, and the step size is the length of the interval between two adjacent sampling points.

[0046] By introducing an optimization algorithm to optimize the learning rate and number of hidden layer neurons of the convolutional neural network, and setting up a loss function, the deep convolutional network image recognition model is trained. When the model training accuracy reaches more than 90%, the training is stopped and the training parameters are saved.

[0047] The grayscale images of the test set are input into the trained deep convolutional network model to determine the accuracy of model recognition. When the accuracy is greater than 90%, the model is saved to obtain a fault diagnosis model for the startup process of the pumped storage unit.

[0048] Furthermore, according to one embodiment of the present invention, a grayscale image is input into a fault diagnosis model for the startup process of the pumped storage unit, and the model outputs a recognition probability. The probability threshold value is set to 0.85. When the model output probability value is greater than 0.85, it is determined that a fault has occurred in the startup process of the pumped storage unit.

[0049] Furthermore, according to one embodiment of the present invention, when the pumped storage unit startup process fault diagnosis model determines that the pumped storage unit operating condition is a pumped storage unit startup process fault, the startup fault telesignal of the generator-transformer group protection device is instantaneously triggered and the telesignal is set to 1; when the pumped storage unit startup process fault diagnosis model determines that the pumped storage unit operating condition is a pumped storage unit startup process non-fault condition, the startup fault telesignal is set to 0.

[0050] Furthermore, according to one embodiment of the present invention, the generator-transformer group protection device receives the start-up fault telesignaling point information in real time. When the generator-transformer group protection device receives the start-up fault telesignaling point and sets it to 1, the instantaneous outlet quickly cuts off the pumped storage unit end circuit breaker. When the generator-transformer group protection device receives the start-up fault telesignaling point and sets it to 0, the generator-transformer group protection device locks the start-up and shutdown related protection outlet components until the start-up process is completed.

[0051] Furthermore, according to one embodiment of the present invention, the present invention also includes: recording and saving the on-site collected electrical quantities and the model judgment results, judging whether the model judgment results are consistent with the actual operating conditions of the pumped storage unit startup process; if the actual operating conditions of the pumped storage unit startup process are consistent with the model judgment results, then the judgment is deemed correct; if the actual operating conditions of the pumped storage unit startup process are inconsistent with the model judgment results, then the judgment results are corrected according to the actual operating conditions, and then used to update the convolution model knowledge base.

[0052] According to the above scheme of the present invention, the present invention uses the three-phase current and three-phase voltage waveforms of the pumped storage unit terminal collected by the protection device of the pumped storage power station generator transformer group, arranges the three-phase current and voltage waveforms in a top-down order, and finally generates an image file. The image adopts a grayscale image to reduce the amount of calculation, and inputs the image into the deep convolutional network model. Gaussian noise is introduced during training to generate a new sequence. Under the premise of not affecting the overall properties and label information of the sequence, the sample data set is expanded. The deep convolutional network adopts a small-scale trained model suitable for mobile terminals or embedded devices. The model adopts a sliding window mode, and the window size is set to a conventional preservation The sampling rate achieved by the protection device is 48 points per cycle, with the length of data collected in 10ms. The step length is the length of the interval between two adjacent sampling points. By setting an appropriate recognition probability threshold, a deep convolutional network is used for fault identification. The identification result is used to instantly trigger the pumped storage unit startup fault telesignal. When the generator-transformer protection device receives the startup fault telesignal, the instantaneous output quickly opens the pumped storage unit's terminal circuit breaker to prevent the pumped storage unit from starting with a fault. This will further improve the safety and stability of the pumped storage unit. At the same time, by combining the model identification results with actual site conditions, a regularly updated convolutional model knowledge base is established, forming a closed-loop feedback loop to improve the accuracy and robustness of identification. This method has broad applicability and high practicality and can be widely applied to the generator-transformer protection devices of pumped storage power plants.

[0053] Furthermore, to achieve the above-mentioned object, the present invention also provides a pumped storage unit startup process fault diagnosis system, comprising:

[0054] Grayscale image acquisition module: The protection device of the generator-transformer group of the pumped-storage power station collects the electrical quantity waveform of the pumped-storage unit, and arranges the real-time collected electrical quantity waveform and the historically saved electrical quantity waveform in sequence and converts them into grayscale images;

[0055] The fault diagnosis model generation module establishes a regularly updated convolutional model knowledge base based on grayscale images. It also introduces Gaussian noise to expand the grayscale image sample set. This expanded sample set is then input into a deep convolutional network model for training and testing, generating a fault diagnosis model for the pumped storage unit startup process.

[0056] In the fault identification module, the generator-transformer protection device inputs the grayscale image generated in real time into the pumped storage unit startup process fault diagnosis model. The pumped storage unit startup process fault diagnosis model calculates the fault identification based on the input grayscale image.

[0057] The fault diagnosis model of the pumped storage unit startup process outputs the fault judgment probability and sets a probability threshold. When the probability threshold is met, it is judged that a fault has occurred during the startup process and the startup fault telesignal of the generator-transformer unit protection device is instantly triggered.

[0058] The pumped storage unit control module, when the generator-transformer unit protection device receives the remote signal of the starting fault, the instantaneous output quickly cuts off the pumped storage unit end circuit breaker to prevent the pumped storage unit from starting with a fault.

[0059] Furthermore, according to an embodiment of the present invention, the electrical quantities collected by the protection device of the generator-transformer group of the pumped-storage power station include three-phase current and three-phase voltage waveforms at the terminal of the pumped-storage unit.

[0060] The electrical quantity waveforms collected by the protection device of the generator-transformer group of the pumped-storage power station are, in order from top to bottom, phase A voltage, phase B voltage, phase C voltage, phase A current, phase B current and phase C current, and the collected waveforms are converted into grayscale images.

[0061] Furthermore, according to one embodiment of the present invention, the update cycle of the convolution model knowledge base is in years. During the update, the electrical quantity waveforms collected and recorded in real time and stored historically are converted into grayscale images and added to the convolution model knowledge base. At the same time, the judgment results obtained through the fault diagnosis model of the pumped storage unit startup process are converted into grayscale images and added to the convolution model knowledge base. Then, the fault diagnosis model of the pumped storage unit startup process is updated based on the updated convolution model knowledge base.

[0062] Furthermore, according to one embodiment of the present invention, the training process of the deep convolutional network model includes:

[0063] The grayscale images are divided into training set and test set at a ratio of 80% and 20%;

[0064] The grayscale images of the training set are input into the deep convolutional network model. The deep convolutional network model uses a sliding window. The window size is the length of the data collected in 10ms at a sampling rate of 48 points per cycle, and the step size is the length of the interval between two adjacent sampling points.

[0065] By introducing an optimization algorithm to optimize the learning rate and number of hidden layer neurons of the convolutional neural network, and setting up a loss function, the deep convolutional network image recognition model is trained. When the model training accuracy reaches more than 90%, the training is stopped and the training parameters are saved.

[0066] The grayscale images of the test set are input into the trained deep convolutional network model to determine the accuracy of model recognition. When the accuracy is greater than 90%, the model is saved to obtain a fault diagnosis model for the startup process of the pumped storage unit.

[0067] Furthermore, according to one embodiment of the present invention, a grayscale image is input into a fault diagnosis model for the startup process of the pumped storage unit, and the model outputs a recognition probability. The probability threshold value is set to 0.85. When the model output probability value is greater than 0.85, it is determined that a fault has occurred in the startup process of the pumped storage unit.

[0068] Furthermore, according to one embodiment of the present invention, when the pumped storage unit startup process fault diagnosis model determines that the pumped storage unit operating condition is a pumped storage unit startup process fault, the startup fault telesignal of the generator-transformer group protection device is instantaneously triggered and the telesignal is set to 1; when the pumped storage unit startup process fault diagnosis model determines that the pumped storage unit operating condition is a pumped storage unit startup process non-fault condition, the startup fault telesignal is set to 0.

[0069] Furthermore, according to one embodiment of the present invention, the generator-transformer group protection device receives the start-up fault telesignaling point information in real time. When the generator-transformer group protection device receives the start-up fault telesignaling point and sets it to 1, the instantaneous outlet quickly cuts off the pumped storage unit end circuit breaker. When the generator-transformer group protection device receives the start-up fault telesignaling point and sets it to 0, the generator-transformer group protection device locks the start-up and shutdown related protection outlet components until the start-up process is completed.

[0070] Furthermore, according to one embodiment of the present invention, the present invention also includes: recording and saving the on-site collected electrical quantities and the model judgment results, judging whether the model judgment results are consistent with the actual operating conditions of the pumped storage unit startup process; if the actual operating conditions of the pumped storage unit startup process are consistent with the model judgment results, then the judgment is deemed correct; if the actual operating conditions of the pumped storage unit startup process are inconsistent with the model judgment results, then the judgment results are corrected according to the actual operating conditions, and then used to update the convolution model knowledge base.

[0071] According to the above scheme of the present invention, the present invention uses the three-phase current and three-phase voltage waveforms of the pumped storage unit terminal collected by the protection device of the pumped storage power station generator transformer group, arranges the three-phase current and voltage waveforms in a top-down order, and finally generates an image file. The image adopts a grayscale image to reduce the amount of calculation, and inputs the image into the deep convolutional network model. Gaussian noise is introduced during training to generate a new sequence. Under the premise of not affecting the overall properties and label information of the sequence, the sample data set is expanded. The deep convolutional network adopts a small-scale trained model suitable for mobile terminals or embedded devices. The model adopts a sliding window mode, and the window size is set to a conventional preservation The sampling rate achieved by the protection device is 48 points per cycle, with the length of data collected in 10ms. The step length is the length of the interval between two adjacent sampling points. By setting an appropriate recognition probability threshold, a deep convolutional network is used for fault identification. The identification result is used to instantly trigger the pumped storage unit startup fault telesignal. When the generator-transformer protection device receives the startup fault telesignal, the instantaneous output quickly opens the pumped storage unit's terminal circuit breaker to prevent the pumped storage unit from starting with a fault. This will further improve the safety and stability of the pumped storage unit. At the same time, by combining the model identification results with actual site conditions, a regularly updated convolutional model knowledge base is established, forming a closed-loop feedback loop to improve the accuracy and robustness of identification. This method has broad applicability and high practicality and can be widely applied to the generator-transformer protection devices of pumped storage power plants.

[0072] Furthermore, to achieve the above-mentioned purpose, the present invention also provides an electronic device, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, the method for diagnosing faults during the startup process of the pumped storage unit as described above is implemented.

[0073] Furthermore, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the fault diagnosis method for the startup process of the pumped storage unit as described above is implemented.

[0074] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiment described herein is only an optimal embodiment of the present invention and is only used to explain the present invention and does not limit the scope of protection of the present invention. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0075] Example 1

[0076] A method for fault diagnosis during the startup process of a pumped storage unit using a deep convolutional network includes:

[0077] S1: If Figure 2 As shown in the figure, the electrical quantities collected by the protection device of the generator-transformer group of the pumped storage power station include the three-phase current and three-phase voltage at the terminal of the pumped storage unit. The collected electrical quantity waveforms are converted into grayscale images in the order of phase A voltage, phase B voltage, phase C voltage, phase A current, phase B current and phase C current from top to bottom;

[0078] S2: Use the prepared convolution model knowledge base to train and test the deep convolution network model. During training, Gaussian noise is introduced to generate new sequences. Under the premise of not affecting the overall properties of the sequence and label information, the sample data set is expanded. The grayscale image library is divided into a training set and a test set at a ratio of 80% and 20%. The training set images are input into the deep convolution network model. The model uses a sliding window. The window size is the length of the data collected in 10ms at a sampling rate of 48 points per cycle that can be achieved by conventional protection devices. The step length is the interval length between two adjacent sampling points. In order to solve problems such as slow convergence and overfitting, an optimization algorithm is introduced to optimize the learning rate and the number of hidden layer neurons of the convolutional neural network. By setting up a loss function, the model is trained. When the model training accuracy reaches more than 90%, the training is stopped, and the relevant parameters of the model are saved at the same time.

[0079] S3: Input the test set images into the trained deep convolutional network model to determine the model's recognition accuracy. If the accuracy is greater than 90%, save the model. If the accuracy is less than 90%, continue training the model using the training set until the model's accuracy on the test set reaches 90% or above.

[0080] S4: The grayscale images converted from the electrical waveforms collected at the project site are input into the trained and tested deep convolutional network model. The model outputs the recognition probability, and the probability threshold is 0.85. When the model output probability value is greater than 0.85, it is judged as a fault condition during the startup process of the pumped storage unit.

[0081] S5: When the model identifies the pumped-storage unit as operating under a startup fault condition, the startup fault telesignal of the generator-transformer group protection device is instantaneously triggered and set to "1." If the model identifies the pumped-storage unit as operating under a non-fault condition during startup, the startup fault telesignal remains unchanged and is set to "0." The generator-transformer group protection device receives the startup fault telesignal point information in real time. When the startup fault telesignal point is set to "1," the generator-transformer group protection device instantly opens the pumped-storage unit's terminal circuit breaker. When the startup fault telesignal point is set to "0," the generator-transformer group protection device locks the startup and shutdown related protection output components until the startup process is complete.

[0082] S6: On-site operation and maintenance personnel should promptly verify the correctness of the export, record and save the on-site collected electrical quantities and model discrimination results. If the actual operating conditions of the pumped storage unit are consistent with the model discrimination results, the discrimination can be deemed correct; if the actual operating conditions of the pumped storage unit are inconsistent with the model discrimination results, the discrimination data should be corrected according to the actual operating conditions, and a regularly updated convolutional model knowledge base should be established. With a two-year cycle, the on-site records and historically saved electrical quantity waveforms and discrimination results should be converted into grayscale images, added to the convolutional model knowledge base, and the deep convolutional network model should be updated to form a closed-loop feedback loop to improve the accuracy and robustness of recognition.

[0083] Those skilled in the art will appreciate that the modules and algorithm steps described in conjunction with the embodiments disclosed herein can be implemented using electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0084] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and equipment can refer to the corresponding processes in the aforementioned method implementation methods and will not be repeated here.

[0085] In the embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is merely a logical function division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0086] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the objectives of the embodiments of the present invention.

[0087] In addition, each functional module in the embodiment of the present invention may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0088] If the functions are implemented as software modules and sold or used as standalone products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the energy-saving signal transmission / reception method according to various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, ROM, RAM, a magnetic disk, or an optical disk.

[0089] The above description is merely a preferred embodiment of the present application and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but also encompasses other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the inventive concept. For example, a technical solution formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in this application.

[0090] It should be understood that the size of the serial numbers of each step in the content of the invention and the implementation methods of the present invention does not absolutely mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the implementation methods of the present invention.

Claims

1. A method for diagnosing faults during the startup process of a pumped storage unit, characterized in that: include: The protection device of the generator-transformer group of the pumped-storage power station collects the electrical quantity waveform of the pumped-storage unit, arranges the real-time collected electrical quantity waveform and the historically saved electrical quantity waveform in sequence, and then converts them into a grayscale image; A regularly updated convolutional model knowledge base is established based on grayscale images. Gaussian noise is introduced to expand the grayscale image sample set. This expanded sample set is then input into a deep convolutional network model for training and testing, generating a fault diagnosis model for the startup process of pumped storage units. The generator-transformer protection device inputs the real-time generated grayscale image into the pumped storage unit startup process fault diagnosis model, which then calculates the fault diagnosis based on the input grayscale image. The fault diagnosis model for the pumped storage unit startup process outputs the fault judgment probability and sets a probability threshold. When the probability threshold is met, it is judged that a fault has occurred during the startup process and the startup fault telesignal of the generator-transformer unit protection device is instantly triggered. When the generator-transformer protection device receives the remote signal of the starting fault, the instantaneous output quickly cuts off the circuit breaker at the pumped storage unit end to prevent the pumped storage unit from starting with a fault.

2. The method for diagnosing faults during the startup process of a pumped storage unit according to claim 1, characterized in that: The update cycle of the convolution model knowledge base is in years. During the update, the real-time collected and recorded electrical quantity waveforms and historically stored electrical quantity waveforms are converted into grayscale images and added to the convolution model knowledge base. At the same time, the judgment results obtained through the pumped storage unit startup process fault diagnosis model are converted into grayscale images and added to the convolution model knowledge base. Then, the pumped storage unit startup process fault diagnosis model is updated based on the updated convolution model knowledge base.

3. The method for diagnosing faults during the startup process of a pumped storage unit according to claim 1, characterized in that: The training process of the deep convolutional network model includes: The grayscale images are divided into training set and test set at a ratio of 80% and 20%; The grayscale images of the training set are input into the deep convolutional network model. The deep convolutional network model uses a sliding window. The window size is the length of the data collected in 10ms at a sampling rate of 48 points per cycle, and the step size is the length of the interval between two adjacent sampling points. By introducing an optimization algorithm to optimize the learning rate and number of hidden layer neurons of the convolutional neural network, and setting up a loss function, the deep convolutional network image recognition model is trained. When the model training accuracy reaches more than 90%, the training is stopped and the training parameters are saved. The grayscale images of the test set are input into the trained deep convolutional network model to determine the accuracy of model recognition. When the accuracy is greater than 90%, the model is saved to obtain a fault diagnosis model for the startup process of the pumped storage unit.

4. The method for diagnosing faults during the startup process of a pumped storage unit according to claim 1, characterized in that: The grayscale image is input into the fault diagnosis model of the pumped storage unit startup process. The model outputs the recognition probability, and the probability threshold is set to 0.

85. When the model output probability value is greater than 0.85, it is judged that a fault has occurred during the startup process of the pumped storage unit.

5. The method for diagnosing faults during the startup process of a pumped storage unit according to claim 1, characterized in that: When the pumped storage unit startup process fault diagnosis model determines that the pumped storage unit operating condition is a pumped storage unit startup process fault, the startup fault telesignal of the generator-transformer group protection device is instantaneously triggered and the telesignal is set to 1. When the pumped storage unit startup process fault diagnosis model determines that the pumped storage unit operating condition is a pumped storage unit startup process non-fault condition, the startup fault telesignal is set to 0.

6. The method for diagnosing faults during the startup process of a pumped storage unit according to claim 1, characterized in that: The generator-transformer group protection device receives the start-up fault telesignal point information in real time. When the generator-transformer group protection device receives the start-up fault telesignal point and sets it to 1, the instantaneous output quickly cuts off the pumped storage unit end circuit breaker. When the generator-transformer group protection device receives the start-up fault telesignal point and sets it to 0, the generator-transformer group protection device locks the start-up and shutdown related protection output components until the start-up process is completed.

7. The method for diagnosing faults during startup of a pumped storage unit according to any one of claims 1 to 6, characterized in that: Also includes: Record and save the on-site collected electrical quantities and the model discrimination results, and determine whether the model discrimination results are consistent with the actual pumped storage unit startup process conditions. If the actual pumped storage unit startup process conditions are consistent with the model discrimination results, the discrimination is considered correct; If the actual operating conditions of the pumped storage unit startup process do not match the model judgment results, the judgment results are corrected according to the actual operating conditions and then used to update the convolution model knowledge base.

8. Pumped storage unit startup process fault diagnosis system, characterized by: include: Grayscale image acquisition module: The protection device of the generator-transformer group of the pumped-storage power station collects the electrical quantity waveform of the pumped-storage unit, and arranges the real-time collected electrical quantity waveform and the historically saved electrical quantity waveform in sequence and converts them into grayscale images; The fault diagnosis model generation module establishes a regularly updated convolutional model knowledge base based on grayscale images. It also introduces Gaussian noise to expand the grayscale image sample set. This expanded sample set is then input into a deep convolutional network model for training and testing, generating a fault diagnosis model for the pumped storage unit startup process. In the fault identification module, the generator-transformer protection device inputs the grayscale image generated in real time into the pumped storage unit startup process fault diagnosis model. The pumped storage unit startup process fault diagnosis model calculates the fault identification based on the input grayscale image. The fault diagnosis model of the pumped storage unit startup process outputs the fault judgment probability and sets a probability threshold. When the probability threshold is met, it is judged that a fault has occurred during the startup process and the startup fault telesignal of the generator-transformer unit protection device is instantly triggered. The pumped storage unit control module, when the generator-transformer unit protection device receives the remote signal of the starting fault, the instantaneous output quickly cuts off the pumped storage unit end circuit breaker to prevent the pumped storage unit from starting with a fault.

9. An electronic device, characterized in that The method comprises a processor, a memory and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, the method for diagnosing faults during the startup process of a pumped storage unit as claimed in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the method for diagnosing faults during the startup process of a pumped storage unit according to any one of claims 1 to 7 is implemented.

Images