Hydrogen leakage fault composite diagnosis method, device and equipment of SOFC-GT system and storage medium

Abstract

The application discloses a SOFC-GT system hydrogen leakage fault composite diagnosis method, device, equipment and storage medium, and relates to the computer technical field. Wherein, the method comprises: obtaining the operation data of the SOFC-GT system and carrying out dimension reduction pretreatment through the uniform manifold approximation and projection method to obtain an initial data set; carrying out feature importance sorting on the initial data set through a bionic pulse feature weighting algorithm to obtain an optimal feature subset; processing the optimal feature subset through a composite diagnosis model combining a deep separable convolutional neural network and a gradient boosting decision tree to obtain a confusion matrix; changing the parameters of the composite diagnosis model, recording the fault diagnosis results obtained by the confusion matrix under different parameters and comparing them to obtain an optimal diagnosis model; and carrying out fault diagnosis on the SOFC-GT system through the optimal diagnosis model. The application faces the massive operation data of the SOFC-GT system, optimizes the fault feature extraction and classification ability, and improves the precision, efficiency and reliability of the fault diagnosis.

Description

Technical Field

[0001] This application relates to the field of computer technology, specifically to a composite diagnostic method, apparatus, equipment, and storage medium for hydrogen leakage faults in an SOFC-GT system. Background Technology

[0002] In practical engineering, the actual operating temperature of a solid oxide fuel cell (SOFC) is typically around 850°C, but theoretically it can reach 1000°C. Since the exhaust gas temperature is relatively high, in order to utilize the high-temperature exhaust gas, the fuel cell is usually connected to a gas turbine to form an SOFC-GT (Gas Turbine, abbreviated as GT) combined cycle system to achieve efficient energy utilization.

[0003] SOFC-GT (Solar Fiber Fuel Cell-Gateway) systems hold significant promise for application in the green energy sector due to their high power generation efficiency and clean, environmentally friendly characteristics. However, when operating at high temperatures of 973-1573K (approximately 700-1300℃), SOFC-GT systems are prone to various thermal management-related faults, severely impacting system reliability and lifespan. This indirectly increases system maintenance costs, leading to economic losses and potentially causing major safety accidents. Therefore, timely fault diagnosis in SOFC-GT systems to achieve better thermal management is crucial for reducing equipment maintenance costs and ensuring the safety of industrial activities.

[0004] Currently, machine learning algorithms can be used to build fault diagnosis models for intelligent diagnosis of SOFC-GT systems. However, due to the complexity of SOFC-GT's operating mechanism and the large amount of data that needs to be recorded for key components, coupled with subsequent human intervention, the resulting dataset is massive. Traditional machine learning algorithms have certain limitations in processing such data and are somewhat lacking in feature extraction capabilities. Furthermore, the SOFC-GT system requires the collection of a particularly large number of feature variables, which further increases the burden on the diagnostic model, resulting in poor diagnostic performance and unsatisfactory diagnostic results from traditional diagnostic models. Summary of the Invention

[0005] In view of this, the technical problem to be solved by this application is: how to optimize the fault feature extraction and classification capabilities for the massive operational data of SOFC-GT systems, so as to improve the accuracy, efficiency and reliability of fault diagnosis.

[0006] To solve the above-mentioned technical problems, this application provides the following technical solution.

[0007] In a first aspect, this application provides a composite diagnostic method for hydrogen leakage faults in SOFC-GT systems, comprising: acquiring operational data of the SOFC-GT system and performing dimensionality reduction preprocessing using a unified manifold approximation and projection method to obtain an initial dataset; ranking the feature importance of the initial dataset using a biomimetic pulse feature weighting algorithm to obtain an optimal feature subset; processing the optimal feature subset using a composite diagnostic model combining a deep separable convolutional neural network and a gradient boosting decision tree to obtain a confusion matrix; changing the parameters of the composite diagnostic model, recording and comparing the fault diagnosis results obtained using the confusion matrix under different parameters to obtain an optimal diagnostic model; and performing fault diagnosis on the SOFC-GT system using the optimal diagnostic model.

[0008] In one alternative embodiment of the first aspect, the step of obtaining the operating data of the SOFC-GT system and performing dimensionality reduction preprocessing using a unified manifold approximation and projection method to obtain an initial dataset further includes: obtaining the operating data of the SOFC-GT system and performing standardization and normalization. ; In the formula: This indicates the first [item] in the operating data of the SOFC-GT system. One characteristic variable, This represents the total number of the feature variables. This represents the mean of the characteristic variable. The standard deviation of the characteristic variable is represented. Indicates the first The standardized values ​​of the aforementioned feature variables; This represents the maximum value of the characteristic variable. This represents the minimum value of the characteristic variable. Indicates the first The normalized values ​​of the aforementioned feature variables.

[0009] In one alternative of the first aspect, the operating data of the SOFC-GT system includes at least one characteristic variable among: stack voltage, stack current, output power, stack hydrogen inlet and outlet temperature, stack air inlet and outlet temperature, heat exchanger hydrogen inlet and outlet temperature, and heat exchanger air inlet and outlet temperature.

[0010] In one alternative of the first aspect, the step of processing the optimal feature subset using a composite diagnostic model combining a deep separable convolutional neural network and a gradient boosting decision tree to obtain a confusion matrix includes: performing convolution, batch normalization, and pooling processing on the optimal feature subset using a deep separable convolutional neural network, and inputting the processing results into a fully connected layer; and classifying the processing results in the fully connected layer using a gradient boosting decision tree to obtain a confusion matrix.

[0011] In one alternative embodiment of the first aspect, the depthably separable convolutional neural network includes: a convolutional layer, an activation function layer, a pooling layer, a fully connected layer, and an output layer, wherein: the convolutional kernel size of the convolutional layer is 4×4; the activation function layer uses the Swish activation function; the pooling layer is of the average pooling type and has a size of 2×2; and the number of neurons in the fully connected layer matches the number of feature variables in the operating data of the SOFC system.

[0012] In one alternative of the first aspect, the gradient boosting decision tree employs Gaussian radial Gaussian kernel and grid search cross-validation.

[0013] In one optional embodiment of the first aspect, changing the parameters of the composite diagnostic model, recording and comparing the fault diagnosis results obtained using the confusion matrix under different parameters, and obtaining the optimal diagnostic model includes: changing the hyperparameters of the deep separable convolutional neural network in the composite diagnostic model, recording the fault diagnosis accuracy, false alarm rate, and hit rate obtained using the confusion matrix under different hyperparameters of the deep separable convolutional neural network; comparing the fault diagnosis accuracy, false alarm rate, and hit rate, and selecting the composite diagnostic model corresponding to the confusion matrix with the best comparison result as the optimal diagnostic model.

[0014] Secondly, this application provides a composite diagnostic device for hydrogen leakage faults in an SOFC-GT system, comprising: a data acquisition unit for acquiring operating data of the SOFC-GT system and performing dimensionality reduction preprocessing using a unified manifold approximation and projection method to obtain an initial dataset; a feature selection unit for ranking the feature importance of the initial dataset using a biomimetic pulse feature weighting algorithm to obtain an optimal feature subset; a composite training unit for processing the optimal feature subset using a composite diagnostic model combining a deep separable convolutional neural network and a gradient boosting decision tree to obtain a confusion matrix; a comprehensive optimization unit for changing the parameters of the composite diagnostic model, recording and comparing the fault diagnosis results obtained using the confusion matrix under different parameters to obtain an optimal diagnostic model; and a fault diagnosis unit for performing fault diagnosis on the SOFC-GT system using the optimal diagnostic model.

[0015] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method provided by the first aspect of this application or any implementation thereof.

[0016] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method provided by the first aspect of this application or any implementation thereof.

[0017] This application uses multiple feature variables of the SOFC-GT system as input, covering a wide range. First, the UMAP method is used for data dimensionality reduction preprocessing, then the SNN-Relief algorithm is used to rank feature importance, retaining only key feature variables, which helps improve fault diagnosis efficiency. A composite diagnostic model combining a deep separable convolutional neural network and a gradient boosting decision tree is trained, and the synergy between their feature extraction and classification capabilities improves the accuracy and efficiency of fault diagnosis. By comprehensively comparing fault evaluation indicators under different hyperparameter settings, the optimal hyperparameter configuration is obtained, improving the reliability of fault diagnosis. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart of a composite diagnostic method for hydrogen leakage faults in an SOFC-GT system provided in this application embodiment.

[0020] Figure 2 This is a flowchart of a composite training method provided in an embodiment of this application.

[0021] Figure 3 This is a schematic diagram of the structure of a depth-separable convolutional neural network provided in an embodiment of this application.

[0022] Figure 4 This is a flowchart illustrating the processing of a composite diagnostic model provided in an embodiment of this application.

[0023] Figure 5 A flowchart of a comprehensive optimization method provided in an embodiment of this application.

[0024] Figure 6 This is a schematic diagram of the structure of a composite diagnostic device for hydrogen leakage faults in an SOFC-GT system, provided in an embodiment of this application.

[0025] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0027] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or modules is not limited to the steps or modules listed, but may optionally include steps or modules not listed, or may optionally include other steps or modules inherent to such process, method, product, or apparatus.

[0028] See Figure 1 This application provides a composite diagnostic method for hydrogen leakage faults in an SOFC-GT system, comprising the following steps.

[0029] S100: Obtain the running data of the SOFC-GT system and perform dimensionality reduction preprocessing using the Uniform Manifold Approximation and Projection (UMAP) method to obtain the initial dataset.

[0030] In this embodiment, the principle of the UMAP method is as follows.

[0031] Construction of fuzzy complexes in high-dimensional space: In the formula: and These represent the first and second parts of the SOFC-GT system's operating data. The and the first One sample point, Indicates the distance between the two. Represents sample points Distance to its nearest neighbor Represents the normalization parameter. Represents the maximum value function. Represents a logarithmic function; Represents sample points and The connection weights between the two are used to measure the degree of similarity between them.

[0032] Layout optimization in low-dimensional space: In the formula: and These represent the true and predicted distributions of the SOFC-GT system's operational data, respectively, and are used to characterize the minimized high-dimensional graph. Represents the true distribution The Middle The probability of running data of a class. Represents the predicted distribution The Middle The probability of running data of a class. Represents the set of categories of the running data; Represents the true distribution and predicted distribution The cross-entropy is used to measure the difference between the two.

[0033] In practical applications, the preferred time interval for collecting operating data from the SOFC-GT system is 30 seconds.

[0034] In some implementations, S100 further includes: acquiring the operating data of the SOFC-GT system and performing standardization and normalization: ; In the formula: This indicates the first [item] in the operating data of the SOFC-GT system. One characteristic variable, This represents the total number of feature variables. Represents the mean of the characteristic variable. The standard deviation of the characteristic variable, Indicates the first Standardized values ​​of the characteristic variables; This represents the maximum value of the characteristic variable. This represents the minimum value of the characteristic variable. Indicates the first The normalized values ​​of the feature variables.

[0035] Based on the above technical solution, the standardization and normalization process ensures that the scales of various feature variables in the initial dataset are relatively consistent, providing reliable data support for subsequent processing.

[0036] S200 uses the biomimetic pulse feature weighting (SNN-Relief) algorithm to rank the features of the initial dataset by importance, and obtains the optimal feature subset.

[0037] In this embodiment, a sample is randomly selected from the initial dataset, and its similar and dissimilar neighbors are identified. Then, based on the distance changes of the feature variables between the sample and its neighbors, the weight value of each feature variable is calculated. If the value of a sample on a certain feature variable is close to that of its similar neighbors and far away from that of its dissimilar neighbors, the weight of that feature is increased; otherwise, it is decreased. After multiple iterations, the feature variables are finally sorted according to their weights, and the feature variables with weights greater than a threshold (preferably 0.2) are selected to form the optimal feature subset.

[0038] Based on the above technical solution, the SNN-Relief algorithm is used to calculate the importance of each feature variable in the operating data of the SOFC-GT system, identify and remove features that contribute the least to the diagnostic performance, which is beneficial to improving the efficiency of fault diagnosis.

[0039] In some implementations, the operating data of the SOFC-GT system includes at least one of the following characteristic variables: stack voltage, stack current, output power, stack hydrogen inlet and outlet temperatures, stack air inlet and outlet temperatures, heat exchanger hydrogen inlet and outlet temperatures, and heat exchanger air inlet and outlet temperatures.

[0040] Understandably, when using the aforementioned multiple feature variables as the initial dataset to construct a composite diagnostic model for training, high-dimensional sample data will lead to excessively long training times and overly complex computations.

[0041] Based on the above technical solution, the UMAP method is first used for data dimensionality reduction preprocessing, and then the SNN-Relief algorithm is used to sort the feature variables and remove the feature variables with lower importance, thus effectively overcoming the above defects.

[0042] S300 uses a composite diagnostic model that combines a deep separable convolutional neural network (DSCNN) with a gradient boosting decision tree (GBDT) to process the optimal feature subset and obtain the confusion matrix.

[0043] In this embodiment, the optimal feature subset is used as input, and the structure and parameters of DSCNN and GBDT are adjusted to match it, with the fault diagnosis result as the output.

[0044] Based on the above implementation method, see Figure 2 This application provides a composite training method, which, as a preferred embodiment of S300, specifically includes the following steps.

[0045] S301 uses a deep separable convolutional neural network to perform convolution, batch normalization, and pooling on the optimal feature subset, and then inputs the processing results into a fully connected layer.

[0046] S302 uses a gradient boosting decision tree to classify the processing results in the fully connected layer, resulting in a confusion matrix.

[0047] Figure 3 This is a schematic diagram of the structure of a depth-separable convolutional neural network provided in an embodiment of this application.

[0048] In some implementations, see Figure 3 A depthwise separable convolutional neural network specifically includes: convolutional layers, activation function layers, pooling layers, fully connected layers, and an output layer.

[0049] In this embodiment, the kernel size of the convolutional layer is 4×4; the activation function layer uses the Swish activation function; the pooling layer is of the average pooling type and has a size of 2×2; the number of neurons in the fully connected layer matches the number of feature variables in the running data of the SOFC system.

[0050] Figure 4 This is a flowchart illustrating the processing of a composite diagnostic model provided in an embodiment of this application.

[0051] See Figure 4 In the convolution operation, the convolution kernel slides across the relevant data in the optimal subset of input features to extract local features and generate a new feature map. In the subsequent pooling operation, a window slides across the feature map and takes the maximum or average value to reduce the size of the feature map while retaining the main features. Convolution and pooling operations are performed alternately to progressively extract and compress features.

[0052] In practical applications, the preferred learning rate for DSCNN is 0.01, the preferred batch size is 32, the preferred number of training epochs is 50, the preferred weight initialization is He initialization, and the preferred regularization method is Dropout initialization.

[0053] In some implementations, the gradient boosting decision tree employs a Gaussian radial Gaussian kernel and grid search cross-validation. The regularization parameter of the GBDT is preferably 0.5, and the gamma parameter is preferably 0.1.

[0054] Understandably, GBDT introduces the idea of ​​gradient descent, using the negative gradient of the loss function in the current composite diagnostic model as an approximation of the residual, thus enabling it to flexibly handle various linear and nonlinear loss functions and demonstrating excellent generalization ability in classification tasks.

[0055] Based on the above technical solution, the DSCNN and GBDT classifiers are combined. By fully synergizing the feature extraction capabilities of DSCNN and the classification capabilities of GBDT, the accuracy and efficiency of fault diagnosis are improved.

[0056] S400: Change the parameters of the composite diagnostic model, record the fault diagnosis results obtained by using the confusion matrix under different parameters, and compare them to obtain the optimal diagnostic model.

[0057] Based on the above implementation method, see Figure 5 This application provides a comprehensive optimization method, which, as a preferred implementation of S400, specifically includes the following steps.

[0058] S401, change the hyperparameters of the deep separable convolutional neural network in the composite diagnostic model, and record the fault diagnosis accuracy, fault diagnosis false alarm rate and fault diagnosis hit rate obtained by using the confusion matrix under different hyperparameters of the deep separable convolutional neural network.

[0059] S402 compares the overall fault diagnosis accuracy, fault diagnosis false alarm rate, and fault diagnosis hit rate, and selects the composite diagnostic model corresponding to the confusion matrix with the best comparison result as the best diagnostic model.

[0060] In this embodiment, the output confusion matrix is ​​4×4 in size. The confusion matrix is ​​used to compare the detailed classification results provided by the true and predicted values ​​to evaluate the performance of the composite diagnostic model. The formulas for calculating the fault diagnosis accuracy (CR), fault diagnosis false alarm rate (FAR), and fault diagnosis hit rate (HR) are shown below.

[0061] ; ; .

[0062] In the formula: , , , The times represent the actual state as follows: , , and ; , , , The time represents the predicted state as follows: , , and ; Indicates the first The true state diagnosis is the first The number of samples for each predicted state.

[0063] It should be noted that, considering the optimal configuration between the pooling layer structure and the number of hidden neurons in DSCNN, the preferred hyperparameters of DSCNN in S401 are the size of the pooling layer and the number of neurons in the fully connected layer.

[0064] Based on the above technical solution, by comprehensively comparing the fault evaluation indicators under different hyperparameter settings, the optimal configuration of hyperparameters is obtained, and thus the optimal diagnostic model for hydrogen leakage faults in SOFC systems is obtained, thereby improving the reliability of fault diagnosis.

[0065] The S500 uses the optimal diagnostic model to diagnose faults in the SOFC-GT system.

[0066] In summary, the embodiments of this application use multiple feature variables of the SOFC-GT system as inputs, which has a wide coverage and comprehensive consideration; after feature selection, only key feature variables are retained, improving fault diagnosis efficiency; the composite diagnostic model has reliable feature extraction and classification capabilities and supports posterior optimization, thus forming a fault diagnosis method that takes into account accuracy, efficiency and reliability.

[0067] See Figure 6 This application also provides a composite diagnostic device for hydrogen leakage faults in an SOFC-GT system. This device can be implemented as all or part of a terminal through software, hardware, or a combination of both, or it can be integrated as an independent module on a server. Specifically, the device includes a data acquisition unit 601, a feature selection unit 602, a composite training unit 603, a comprehensive optimization unit 604, and a fault diagnosis unit 605.

[0068] The data acquisition unit 601 is used to acquire the operating data of the SOFC-GT system and perform dimensionality reduction preprocessing through the unified manifold approximation and projection method to obtain the initial dataset.

[0069] The feature selection unit 602 is used to sort the feature importance of the initial dataset using a biomimetic pulse feature weighting algorithm to obtain the optimal feature subset.

[0070] The composite training unit 603 is used to process the optimal feature subset through a composite diagnostic model that combines a deep separable convolutional neural network with a gradient boosting decision tree to obtain a confusion matrix.

[0071] The comprehensive optimization unit 604 is used to change the parameters of the composite diagnostic model, record the fault diagnosis results obtained by using the confusion matrix under different parameters, and compare them to obtain the optimal diagnostic model.

[0072] The fault diagnosis unit 605 is used to diagnose faults in the SOFC-GT system using the optimal diagnostic model.

[0073] It should be noted that the apparatus provided in the above embodiments, when performing the SOFC-GT system hydrogen leak fault composite diagnosis method, is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the equipment can be divided into different functional modules to complete all or part of the functions described above. Furthermore, the apparatus provided in the above embodiments and the SOFC-GT system hydrogen leak fault composite diagnosis method embodiments belong to the same concept, and its implementation process is detailed in the method embodiments, which will not be repeated here.

[0074] See Figure 7 This application provides an electronic device 700, including a processor 701 and a memory 702.

[0075] In this embodiment, the processor 701 is the control center of the computer system, and can be a processor of a physical machine or a processor of a virtual machine. The processor 701 may include one or more processing cores, such as a 4-core processor or an 8-core processor. The processor 701 can be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array).

[0076] Processor 701 may also include a main processor and a coprocessor. The main processor is a processor used to process data in the wake-up state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor used to process data in the standby state.

[0077] The memory 702 may include one or more computer-readable storage media. The memory 702 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices.

[0078] In some embodiments, the computer-readable storage medium in memory 702 is used to store at least one instruction for execution by processor 701 to implement the method in the embodiments of this application.

[0079] In some embodiments, the electronic device 700 further includes a peripheral device interface 703 and at least one peripheral device 704. The processor 701, memory 702, and peripheral device interface 703 can be connected via a bus or signal line. Each peripheral device 704 can be connected to the peripheral device interface 703 via a bus, signal line, or circuit board. Specifically, the peripheral device 704 includes: a display screen, a camera, and audio circuitry. The peripheral device interface 703 can be used to connect at least one I / O (Input / Output) related peripheral device to the processor 701 and memory 702.

[0080] In some embodiments, the processor 701, memory 702, and peripheral device interface 703 are integrated on the same chip or circuit board; in other embodiments, any one or two of the processor 701, memory 702, and peripheral device interface 703 can be implemented on separate chips or circuit boards. This application does not specifically limit this aspect.

[0081] It should be noted that the structural schematic diagram of the electronic device shown in this embodiment does not constitute a limitation on the electronic device 700. The electronic device 700 may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0082] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the methods in any of the foregoing embodiments. The computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, DVDs, CD-ROMs, microdrives, as well as magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic cards or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.

[0083] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of software products. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0084] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A composite diagnostic method for hydrogen leakage faults in an SOFC-GT system, characterized in that, The method includes: The running data of the SOFC-GT system is obtained and preprocessed by dimensionality reduction using the unified manifold approximation and projection method to obtain the initial dataset. The initial dataset is sorted by feature importance using a biomimetic pulse feature weighting algorithm to obtain the optimal feature subset; The optimal feature subset is processed by a composite diagnostic model that combines a deep separable convolutional neural network with a gradient boosting decision tree to obtain a confusion matrix; By changing the parameters of the composite diagnostic model, recording and comparing the fault diagnosis results obtained using the confusion matrix under different parameters, the optimal diagnostic model can be obtained. The SOFC-GT system is used to diagnose faults using the optimal diagnostic model.

2. The composite diagnostic method for hydrogen leakage faults in SOFC-GT systems according to claim 1, characterized in that, The process of obtaining the SOFC-GT system's operational data and performing dimensionality reduction preprocessing using a unified manifold approximation and projection method to obtain the initial dataset also includes: Obtain the operating data of the SOFC-GT system and perform standardization and normalization: ； ； In the formula: This indicates the first [item] in the operating data of the SOFC-GT system. One characteristic variable, This represents the total number of the feature variables. This represents the mean of the characteristic variable. The standard deviation of the characteristic variable is represented. Indicates the first Standardized values ​​of the aforementioned feature variables; This represents the maximum value of the characteristic variable. This represents the minimum value of the characteristic variable. Indicates the first The normalized values ​​of the aforementioned feature variables.

3. The composite diagnostic method for hydrogen leakage faults in SOFC-GT systems according to claim 1, characterized in that, The operating data of the SOFC-GT system includes at least one characteristic variable among: stack voltage, stack current, output power, stack hydrogen inlet and outlet temperature, stack air inlet and outlet temperature, heat exchanger hydrogen inlet and outlet temperature, and heat exchanger air inlet and outlet temperature.

4. The composite diagnostic method for hydrogen leakage faults in SOFC-GT systems according to claim 1, characterized in that, The composite diagnostic model, which combines a deep separable convolutional neural network with a gradient boosting decision tree, processes the optimal feature subset to obtain a confusion matrix, including: The optimal feature subset is processed by convolution, batch normalization and pooling through a deep separable convolutional neural network, and the processing result is input into a fully connected layer. The processing results in the fully connected layer are classified using a gradient boosting decision tree to obtain a confusion matrix.

5. The composite diagnostic method for hydrogen leakage faults in SOFC-GT systems according to claim 1, characterized in that, The depthwise separable convolutional neural network includes: convolutional layers, activation function layers, pooling layers, fully connected layers, and an output layer, wherein: The kernel size of the convolutional layer is 4×4; The activation function layer uses the Swish activation function; The pooling layer is of type average pooling and has a size of 2×2; The number of neurons in the fully connected layer matches the number of feature variables in the operating data of the SOFC system.

6. The composite diagnostic method for hydrogen leakage faults in SOFC-GT systems according to claim 1, characterized in that, The gradient boosting decision tree uses Gaussian radial Gaussian kernel and grid search cross-validation.

7. The composite diagnostic method for hydrogen leakage faults in SOFC-GT systems according to claim 1, characterized in that, The process of changing the parameters of the composite diagnostic model, recording and comparing the fault diagnosis results obtained using the confusion matrix under different parameters, and obtaining the optimal diagnostic model includes: By changing the hyperparameters of the deep separable convolutional neural network in the composite diagnostic model, the fault diagnosis accuracy, fault diagnosis false alarm rate, and fault diagnosis hit rate obtained by using the confusion matrix under different hyperparameters of the deep separable convolutional neural network are recorded. By comparing the fault diagnosis accuracy, the fault diagnosis false alarm rate, and the fault diagnosis hit rate, the composite diagnostic model corresponding to the confusion matrix with the best comparison result is taken as the optimal diagnostic model.

8. A composite diagnostic device for hydrogen leakage faults in an SOFC-GT system, characterized in that, The apparatus for performing the SOFC-GT system hydrogen leakage fault composite diagnostic method according to any one of claims 1-7 comprises: The data acquisition unit is used to acquire the operating data of the SOFC-GT system and perform dimensionality reduction preprocessing using the unified manifold approximation and projection method to obtain the initial dataset. The feature selection unit is used to rank the features of the initial dataset by importance using a biomimetic pulse feature weighting algorithm to obtain the optimal feature subset; A composite training unit is used to process the optimal feature subset through a composite diagnostic model that combines a deep separable convolutional neural network with a gradient boosting decision tree to obtain a confusion matrix; The comprehensive optimization unit is used to change the parameters of the composite diagnostic model, record and compare the fault diagnosis results obtained by using the confusion matrix under different parameters, and obtain the optimal diagnostic model. The fault diagnosis unit is used to perform fault diagnosis on the SOFC-GT system using the optimal diagnostic model.

9. An electronic 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 computer program, it implements the steps of the method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-7.

Images