Sample augmentation based reactor system fault diagnosis method and device

Abstract

This invention provides a reactor system fault diagnosis method and apparatus based on sample augmentation, relating to the field of nuclear industry technology. In the sample augmentation-based reactor system fault diagnosis method, multiple rare fault diagnosis samples are constructed using target expert fault annotation text information and a generative network model corresponding to multiple rare faults. The initial fault diagnosis dataset is then enriched and expanded based on these rare fault diagnosis samples, achieving sample augmentation of the initial fault diagnosis dataset. This alleviates the problems of scarce reactor system fault data and imbalanced category distribution, thereby improving the representativeness and training effect of the target fault diagnosis dataset, and ultimately enhancing the fault diagnosis capability of the reactor system.

Description

Technical Field

[0001] This invention relates to the field of nuclear industry technology, and in particular to a method and apparatus for fault diagnosis of reactor systems based on sample enhancement. Background Technology

[0002] Miniature reactors (MMRs) are characterized by their small size, flexible power ratings, and diverse deployment methods, making them promising for applications in remote energy supply, deep space exploration, marine equipment, and emergency energy security. Compared to traditional large reactors, MMRs offer significant advantages in modular design and engineering adaptability. However, their highly integrated systems, complex operating conditions, and relatively limited safety margins mean that any anomalies or malfunctions can rapidly escalate into serious safety incidents. Safety has become a key factor restricting the engineering application and large-scale deployment of MMRs.

[0003] In situations where safety requirements are extremely high, accurate and timely diagnosis of the operating status and faults of micro-small reactors is crucial. However, fault diagnosis of micro-small reactors faces significant challenges, the core issue being the severe scarcity of fault data. On the one hand, the high potential risks and costs make it difficult to conduct extensive real-world fault tests; on the other hand, the limited operating time of existing devices and insufficient engineering experience result in a very small number of accumulated fault samples with uneven distribution of types, making it difficult to effectively train and validate data-driven diagnostic methods. Therefore, improving fault diagnosis capabilities in the context of scarce and unbalanced fault data for micro-small reactors is an urgent problem to be solved. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a method and apparatus for fault diagnosis of reactor systems based on sample enhancement, so as to alleviate the above-mentioned technical problems.

[0005] In a first aspect, embodiments of the present invention provide a reactor system fault diagnosis method based on sample augmentation. The method includes: constructing an initial fault diagnosis dataset for the reactor system under various fault scenarios; wherein the initial fault diagnosis dataset includes multiple fault diagnosis samples, each fault diagnosis sample including: a system operating parameter matrix, expert fault annotation text information, and a fault label; obtaining target expert fault annotation text information corresponding to multiple rare faults, and inputting the target expert fault annotation text information into a pre-trained generative network model so that the generative network model outputs a target system operating parameter matrix corresponding to the rare fault; wherein, rare faults are faults not included in the various fault scenarios; generating multiple rare fault diagnosis samples based on the target expert fault annotation text information, the target system operating parameter matrix, and the corresponding target fault labels, and determining a target fault diagnosis dataset based on the multiple rare fault diagnosis samples and the initial fault diagnosis dataset; training a fault diagnosis model based on the target fault diagnosis dataset, and performing fault diagnosis on the reactor system based on the trained fault diagnosis model.

[0006] Optionally, an initial fault diagnosis dataset for the reactor system under various fault scenarios is constructed, including: obtaining system operating parameters of the reactor system under various fault scenarios, and determining the corresponding system operating parameter matrix, fault labels, and expert fault annotation text information based on the system operating parameters; wherein, the expert fault annotation text information includes information on changes in faulty components and key system parameters; generating fault diagnosis samples based on the system operating parameter matrix, expert fault annotation text information, and fault labels, and constructing an initial fault diagnosis dataset based on multiple fault diagnosis samples.

[0007] Optionally, the generative network model includes an encoding layer, a mapping layer, and a decoding layer. Inputting the target expert fault annotation text information into the pre-trained generative network model so that the generative network model outputs the target system operating parameter matrix corresponding to the rare fault includes: inputting the target expert fault annotation text information into the encoding layer so that the encoding layer outputs the corresponding target semantic information; inputting the target semantic information into the mapping layer so that the mapping layer outputs the corresponding target latent semantic feature information; and inputting the target latent semantic feature information into the decoding layer so that the decoding layer outputs the corresponding target system operating parameter matrix.

[0008] Optionally, the target fault diagnosis dataset includes a training set, a validation set, and a test set. The fault diagnosis model is trained based on the target fault diagnosis dataset, including: training the fault diagnosis model using the training set and evaluating the model using the validation set during training to calculate the model accuracy; selecting the fault diagnosis model with the highest accuracy as the target fault diagnosis model; evaluating the performance of the target fault diagnosis model using the test set to calculate the target model accuracy; and if the target model accuracy is not less than a preset threshold, then the target fault diagnosis model is considered the trained fault diagnosis model.

[0009] Optionally, the method further includes: if the accuracy of the target model is less than a preset threshold, updating the target fault diagnosis dataset and retraining the fault diagnosis model based on the updated target fault diagnosis dataset until the accuracy of the target model is not less than the preset threshold.

[0010] Optionally, fault diagnosis of the reactor system is performed based on the trained fault diagnosis model, including: obtaining the current system operating parameters of the reactor system in the current time window, and determining the current system operating parameter matrix based on the current system operating parameters; inputting the current system operating parameter matrix into the fault diagnosis model so that the fault diagnosis model outputs fault diagnosis information; wherein, the fault diagnosis information includes fault probability distribution information.

[0011] Secondly, embodiments of the present invention also provide a reactor system fault diagnosis device based on sample enhancement, the device comprising: The building module is used to construct an initial fault diagnosis dataset for the reactor system under various fault scenarios. The initial fault diagnosis dataset includes multiple fault diagnosis samples, each of which includes: a system operating parameter matrix, expert fault annotation text information, and fault labels. The acquisition module is used to acquire the target expert fault annotation text information corresponding to multiple rare faults, and input the target expert fault annotation text information into a pre-trained generative network model so that the generative network model outputs the target system operating parameter matrix corresponding to the rare faults; where rare faults are faults not included in multiple fault scenarios; The generation module is used to generate multiple scarce fault diagnosis samples based on the target expert fault annotation text information, the target system operating parameter matrix and the corresponding target fault labels, and to determine the target fault diagnosis dataset based on the multiple scarce fault diagnosis samples and the initial fault diagnosis dataset. The training module is used to train the fault diagnosis model based on the target fault diagnosis dataset, and to perform fault diagnosis on the reactor system based on the trained fault diagnosis model.

[0012] Optionally, fault diagnosis of the reactor system is performed based on the trained fault diagnosis model, including: obtaining the current system operating parameters of the reactor system in the current time window, and determining the current system operating parameter matrix based on the current system operating parameters; inputting the current system operating parameter matrix into the fault diagnosis model so that the fault diagnosis model outputs fault diagnosis information; wherein, the fault diagnosis information includes fault probability distribution information.

[0013] Thirdly, embodiments of the present invention also provide 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 steps of the method described in the first aspect.

[0014] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the method described in the first aspect.

[0015] The embodiments of the present invention bring the following beneficial effects: This invention provides a reactor system fault diagnosis method and apparatus based on sample augmentation. The method constructs an initial fault diagnosis dataset for the reactor system under various fault scenarios. The initial fault diagnosis dataset includes multiple fault diagnosis samples, each comprising: a system operating parameter matrix, expert fault annotation text information, and a fault label. Target expert fault annotation text information corresponding to multiple rare faults is obtained and input into a pre-trained generative network model, causing the model to output the target system operating parameter matrix corresponding to the rare fault. Rare faults are those not included in the various fault scenarios. Multiple rare fault diagnosis samples are generated based on the target expert fault annotation text information, the target system operating parameter matrix, and the corresponding target fault labels. A target fault diagnosis dataset is determined based on the multiple rare fault diagnosis samples and the initial fault diagnosis dataset. A fault diagnosis model is trained using the target fault diagnosis dataset, and the trained model is used to diagnose faults in the reactor system. The above method constructs multiple rare fault diagnosis samples by using the target expert fault annotation text information and generative network model corresponding to multiple rare faults. Based on these multiple rare fault diagnosis samples, the initial fault diagnosis dataset is enriched and expanded, thereby achieving sample augmentation of the initial fault diagnosis dataset. This alleviates the problems of scarce fault data and class distribution imbalance in reactor systems, thereby improving the representativeness and training effect of the target fault diagnosis dataset and ultimately enhancing the fault diagnosis capability of the reactor system.

[0016] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 A flowchart of a reactor system fault diagnosis method based on sample enhancement provided in an embodiment of the present invention; Figure 2 A flowchart of another reactor system fault diagnosis method based on sample enhancement provided in an embodiment of the present invention; Figure 3 A schematic diagram of the architecture of a generative network model provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the data distribution results before and after enhancement, provided as an embodiment of the present invention. Figure 5 This is a schematic diagram of the architecture of a fault diagnosis model provided in an embodiment of the present invention; Figure 6 A flowchart illustrating the training and verification process of a fault diagnosis model provided in an embodiment of the present invention; Figure 7 A schematic diagram of a reactor system fault diagnosis device based on sample enhancement provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] To facilitate understanding of this embodiment, the embodiments of the present invention will be described in detail below.

[0022] Example 1 This invention provides a sample-enhanced method for fault diagnosis of reactor systems, wherein the reactor system includes at least one micro-reactor. For example... Figure 1 As shown, the method includes the following steps: Step S102: Construct an initial fault diagnosis dataset for the reactor system under various fault scenarios.

[0023] Specifically, reactor system failure tests and simulation studies under multiple failure scenarios are carried out simultaneously to construct an initial fault diagnosis dataset for the reactor system under multiple failure scenarios. The initial fault diagnosis dataset includes multiple fault diagnosis samples, each of which includes: a system operating parameter matrix, expert fault annotation text information, and fault labels.

[0024] In practical applications, due to the limited fault data used in fault testing and simulation studies, the initial fault diagnosis dataset constructed suffers from a scarcity of fault data and an imbalance in class distribution. If the initial fault diagnosis dataset is directly used to train the fault diagnosis model, the fault diagnosis capability of the trained fault diagnosis model will be limited, thus failing to meet the actual fault diagnosis needs of the reactor system.

[0025] Step S104: Obtain the target expert fault annotation text information corresponding to multiple rare faults, and input the target expert fault annotation text information into the pre-trained generative network model so that the generative network model outputs the target system operating parameter matrix corresponding to the rare faults.

[0026] Specifically, rare faults are faults not included in various fault scenarios. Based on experts' knowledge and experience of rare faults in reactor systems, this embodiment of the invention sets corresponding target expert fault annotation text information for multiple rare faults and inputs the target expert fault annotation text information into a pre-trained generative network model so that the generative network model outputs the target system operating parameter matrix corresponding to the rare fault. This allows multiple rare fault diagnostic samples to be generated based on the target expert fault annotation text information and the target system operating parameter matrix. In this way, the initial fault diagnosis dataset is enriched and expanded by multiple rare fault diagnostic samples, that is, the number of samples of different rare faults is expanded in the initial fault diagnosis dataset. This realizes the sample augmentation of the initial fault diagnosis dataset and alleviates the problem of scarce fault data and class distribution imbalance in reactor systems.

[0027] Step S106: Generate multiple scarce fault diagnosis samples based on the target expert fault annotation text information, the target system operating parameter matrix and the corresponding target fault labels, and determine the target fault diagnosis dataset based on the multiple scarce fault diagnosis samples and the initial fault diagnosis dataset.

[0028] Step S108: Train the fault diagnosis model based on the target fault diagnosis dataset, and perform fault diagnosis on the reactor system based on the trained fault diagnosis model.

[0029] The reactor system fault diagnosis method based on sample augmentation provided in this invention constructs multiple rare fault diagnosis samples by using target expert fault annotation text information and generative network models corresponding to multiple rare faults. The initial fault diagnosis dataset is then enriched and expanded based on these rare fault diagnosis samples, thereby achieving sample augmentation of the initial fault diagnosis dataset. This alleviates the problems of scarce reactor system fault data and class distribution imbalance, thereby improving the representativeness and training effect of the target fault diagnosis dataset and ultimately enhancing the fault diagnosis capability of the reactor system.

[0030] In one implementation, an initial fault diagnosis dataset for a reactor system under multiple fault scenarios is constructed, including: acquiring system operating parameters of the reactor system under multiple fault scenarios, and determining the corresponding system operating parameter matrix, fault labels, and expert fault annotation text information based on the system operating parameters; wherein, the expert fault annotation text information includes information on changes in faulty components and key system parameters; generating fault diagnosis samples based on the system operating parameter matrix, expert fault annotation text information, and fault labels, and constructing an initial fault diagnosis dataset based on multiple fault diagnosis samples.

[0031] Specifically, fault tests and simulations of the reactor system under various fault scenarios are conducted simultaneously, and system operating parameters of the reactor system under various fault scenarios are collected within a given time window. These system operating parameters are those related to reactor system fault diagnosis, including but not limited to: nuclear power, maximum fuel temperature, average emitter temperature, average receiver temperature, output current, output voltage, output power, average core coolant inlet temperature, average core coolant outlet temperature, and average fin temperature. The specific types and quantities of system operating parameters can be set according to actual conditions and are collected using corresponding sensors and other measuring equipment.

[0032] The given time window can be understood as a preset duration, such as 10 seconds per given time window. Within the given time window, the measuring device collects data every 1 second. If the system operating parameters include 10 types of parameters, such as nuclear power, maximum fuel temperature, average emitter temperature, average receiver temperature, output current, output voltage, output power, average core coolant inlet temperature, average core coolant outlet temperature, and average fin temperature, then the corresponding system operating parameter matrix can be determined based on the system operating parameters. The system operating parameter matrix is ​​a 10×10 time sequence matrix.

[0033] Then, human experts label the system operating parameters for faults, generating corresponding expert fault labeling text information. The expert fault labeling text information includes information on faulty components and changes in key system parameters, and is represented using a binary semantic structure: faulty component—changes in key system parameters. Here, the faulty component is used to identify the object in the reactor system that has an anomaly, and the change in key system parameters is used to describe the direction and magnitude of changes in system operating parameters related to the faulty component.

[0034] For example, if a heat pipe failure occurs in the reactor system, the generated expert fault labeling text will describe the faulty components as "Normal components: heat pipe, cesium vapor generator, control drum; Faulty component: electromagnetic pump"; and the system key parameter change information will be described as "Nuclear power: normal, low noise, good continuity; average emitter temperature: normal, low noise, good continuity; average receiver temperature: normal, medium noise, good continuity; moderator: normal, medium noise, medium continuity; output voltage: normal, medium noise, medium continuity; output current: normal, medium noise, medium continuity; mass flow rate: normal, medium noise, medium continuity; average core coolant inlet temperature: abnormal, medium noise, poor continuity; average core coolant outlet temperature: abnormal, medium noise, poor continuity; average fin temperature: abnormal, medium noise, medium continuity".

[0035] It should be noted that the expert fault annotation text information can be entered by human experts through an interactive interface, or it can be obtained from a database that pre-stores human expert fault analysis data. This database stores expert fault annotation text information for each time window.

[0036] Therefore, fault diagnosis samples are generated based on the system operating parameter matrix, expert fault annotation text information, and corresponding fault labels. That is, "system operating parameter matrix - expert fault annotation text information - fault labels" is taken as a complete fault diagnosis sample. Fault diagnosis samples under different fault scenarios are collected to construct the initial fault diagnosis dataset.

[0037] In one implementation, the generative network model includes an encoding layer, a mapping layer, and a decoding layer. Inputting the target expert fault annotation text information into the pre-trained generative network model, so that the generative network model outputs a target system operating parameter matrix corresponding to the rare fault, includes: inputting the target expert fault annotation text information into the encoding layer, so that the encoding layer outputs corresponding target semantic information; inputting the target semantic information into the mapping layer, so that the mapping layer outputs corresponding target latent semantic feature information; and inputting the target latent semantic feature information into the decoding layer, so that the decoding layer outputs the corresponding target system operating parameter matrix.

[0038] Specifically, due to the scarcity of fault diagnosis samples and the imbalance in class distribution in the initial fault diagnosis dataset, this embodiment of the invention generates corresponding target expert fault annotation text information for multiple scarce faults by human experts. This target expert fault annotation text information is then input into a pre-trained generative network model. Since the generative network model includes an encoding layer, a mapping layer, and a decoding layer arranged sequentially, the target expert fault annotation text information is first input into the encoding layer, causing the encoding layer to output the corresponding target semantic information. Next, the target semantic information is input into the mapping layer, causing the mapping layer to output the corresponding target latent semantic feature information. Finally, the target latent semantic feature information is input into the decoding layer, causing the decoding layer to output the corresponding target system operating parameter matrix. This generates multiple scarce fault diagnosis samples based on the target expert fault annotation text information and the target system operating parameter matrix. These samples enrich and expand the initial fault diagnosis dataset, achieving sample augmentation of the initial fault diagnosis dataset. This alleviates the problems of scarce fault data and imbalanced class distribution in reactor systems, improves the representativeness and training effect of the target fault diagnosis dataset, and ultimately enhances the fault diagnosis capability of the reactor system.

[0039] In one implementation, a target fault diagnosis dataset is determined based on multiple scarce fault diagnosis samples and an initial fault diagnosis dataset. The target fault diagnosis dataset includes a training set, a validation set, and a test set. Training a fault diagnosis model using the target fault diagnosis dataset includes: training the fault diagnosis model using the training set and evaluating the model using the validation set during training to calculate the model accuracy; selecting the fault diagnosis model with the highest accuracy as the target fault diagnosis model; evaluating the target fault diagnosis model's performance using the test set to calculate the target model accuracy; and if the target model accuracy is not less than a preset threshold, then the target fault diagnosis model is considered the trained fault diagnosis model.

[0040] In addition, if the accuracy of the target model is less than the preset threshold, the target fault diagnosis dataset is updated, and the fault diagnosis model is retrained based on the updated target fault diagnosis dataset until the accuracy of the target model is not less than the preset threshold.

[0041] Therefore, by enriching and expanding the initial fault diagnosis dataset with multiple scarce fault diagnosis samples, a target fault diagnosis dataset with a larger number of samples is generated. When the fault diagnosis model is trained based on the target fault diagnosis dataset, the fault diagnosis model's accuracy and generalization ability for different fault types are improved, thereby enhancing the fault diagnosis capability of the reactor system.

[0042] In one implementation, for a trained fault diagnosis model, fault diagnosis of the reactor system is performed based on the trained fault diagnosis model, including: obtaining the current system operating parameters of the reactor system in the current time window, and determining the current system operating parameter matrix based on the current system operating parameters; inputting the current system operating parameter matrix into the fault diagnosis model so that the fault diagnosis model outputs fault diagnosis information; wherein, the fault diagnosis information includes fault probability distribution information.

[0043] In summary, the reactor system fault diagnosis method based on sample augmentation provided in this invention has good universality. It is not only applicable to fault diagnosis scenarios of micro-small reactors, but can also be extended to other industrial systems where fault data is scarce. Furthermore, addressing the problem that some system operating parameters are difficult to measure directly, this invention uses a real-time simulation program to invert the system state, thereby obtaining complete and usable system operating parameters. Simultaneously, it sets up expert fault annotation text information with a binary semantic structure, balancing qualitative description and quantitative modeling needs. By mapping the qualitative changes in system operating parameters to numerical ranges and sampling them, it achieves an effective transformation from semantic annotation to numerical samples. Therefore, the embodiments of the present invention fully utilize expert knowledge and experience, and can generate fault diagnosis samples in a targeted manner for specific fault types. In particular, for scarce faults, corresponding target expert fault annotation text information is constructed, and a corresponding target system operating parameter matrix is ​​generated based on the target expert fault annotation text information and the generative network model. This allows for the generation of multiple scarce fault diagnosis samples, thereby enriching and expanding the initial fault diagnosis dataset. This achieves sample augmentation of the initial fault diagnosis dataset, effectively alleviating the problems of scarce reactor system fault data and imbalanced category distribution, thereby improving the representativeness and training effect of the target fault diagnosis dataset, and ultimately enhancing the fault diagnosis capability of the reactor system.

[0044] Example 2 Based on the above-described method embodiments, this invention provides another sample-enhanced reactor system fault diagnosis method. This method uses the heat pipe improved TOPAZ-II reactor system as an example, and conducts system-level simulations of the system under various fault conditions based on a previously developed simulation program for the heat pipe improved TOPAZ-II reactor system. The considered fault scenarios cover four single fault types: heat pipe failure, electromagnetic pump failure, abnormal cesium steam temperature, and control drum shaft jamming, as well as composite faults formed by the coupling of multiple single faults, totaling 13 operating conditions and 4410 noise-free fault diagnosis samples. These are expanded to 11205 fault diagnosis samples by injecting noise of different intensities. Furthermore, the fault labels employ multi-thermal coding to characterize the multiple fault types coexisting in a single fault diagnosis sample. Specifically, each fault type is assigned a binary bit; if the fault type exists, it is set to 1; otherwise, it is set to 0.

[0045] like Figure 2 As shown, the method includes the following steps: Step S202: Simultaneously conduct reactor system fault tests and simulation studies, collect system measurement data and expert fault annotation text information; at the same time, correct the simulation model coefficients and invert the system state parameters to construct an initial fault diagnosis dataset.

[0046] Specifically, reactor system failure tests and simulations under various failure scenarios are conducted simultaneously. System measurement data (i.e., system operating parameters) are collected within a given time window (e.g., 10 seconds), such as nuclear power, output current, output voltage, output electrical power, average core coolant inlet temperature, average core coolant outlet temperature, and average fin temperature. Human experts then annotate the data, generating corresponding expert fault annotation text information. This expert fault annotation text information uses a binary semantic structure: "faulty component—system key parameter change information." Here, "faulty component" identifies the object experiencing an anomaly, and the system key parameter change information describes the direction and magnitude of the changes in relevant system parameters.

[0047] Furthermore, based on the system measurement data collected within the given time window, the simulation model coefficients of the reactor system are corrected in real time to ensure that the simulation results are consistent with the online measurement results, thereby retrieving the system state in digital space. The expression for the correction coefficient F is F= C 1(X1-X2), where X1 is the measurement result and X2 is the simulation result. C 1 represents the relaxation factor.

[0048] For example, based on the inverted state space, system operating parameters within the current time window are extracted, including: nuclear power, maximum fuel temperature, average emitter temperature, average receiver temperature, output current, output voltage, output power, average core coolant inlet temperature, average core coolant outlet temperature, and average fin temperature. The maximum fuel temperature, average emitter temperature, average receiver temperature, and average fin temperature are inverted data, constructing a 10×10 system operating parameter matrix. Furthermore, corresponding expert fault annotation text information and fault tags are obtained. Using "system operating parameter matrix—expert fault annotation text information—fault tags" as a complete fault diagnosis sample, an initial fault diagnosis dataset is constructed by aggregating fault diagnosis samples under different fault scenarios. For example, the fault tags are shown in Table 1 below: Table 1

[0049] Step S204: Build a generative network model, take expert fault annotation text information as input and system operation parameter matrix as output, and train it using a phased training strategy.

[0050] Specifically, such as Figure 3 As shown, the generative network model consists of three parts: the encoding layer, the mapping layer, and the decoding layer. Each part is described below: (a) Encoding Layer: Expert fault annotation text information is mapped to the semantic embedding space to generate textual semantic feature information. On one hand, the textual semantic feature information is encoded using the BERT (Bidirectional Encoder Representations from Transformers) model to obtain the semantic representation information of the faulty components. On the other hand, for the changing characteristics of system variables, a numerical mapping mechanism is constructed based on the statistical information of the fault images. By statistically analyzing the mean and variance of variable changes in similar data, interval parameters are generated to achieve a numerical expression of semantic distribution information. Finally, semantic information is determined based on the semantic representation information and semantic distribution information.

[0051] (b) Mapping Layer: This layer represents the conditional mapping relationship between text semantics and the generated latent space. It employs a Transformer structure and models the global semantic dependencies in the semantic embedding space through a self-attention mechanism, mapping the text embeddings to latent feature representations (e.g., 2×4×4×4) for the system's operating parameter matrix. Simultaneously, it combines an upsampling network to progressively recover the spatial resolution of the latent features to meet the input requirements of subsequent decoding stages. Therefore, the input to the mapping layer is the speech information output from the encoding layer, and the output is the corresponding latent semantic feature information.

[0052] (c) Decoding Layer: This layer maps latent semantic feature information to the system operating parameter matrix. A vector quantization variational autoencoder is used as the decoding structure, modeling the latent semantic feature information by introducing discrete latent variables. In practical applications, the mapping layer outputs a continuous vector. ( D express The dimension is used to represent latent semantic feature information; to reduce the continuous redundancy of the latent representation, a codebook mechanism is introduced. The discretization mapping is expressed as follows: ,in, , k =1,2,…, K , with corresponding code words As discrete latent variables Decoding network with As input, the latent semantic features are decoded to generate a system operating parameter matrix consistent with the fault state, expressed as: ;in, M This represents the total number of monitored variables, that is, the number of monitored operating parameters. S Indicates the size of the monitoring time window. T Indicates time.

[0053] For the aforementioned generative network model, based on the constructed initial fault diagnosis dataset, the model is trained using expert fault annotation text information as input and the system operating parameter matrix as output. In practical applications, the training of the generative network model includes four main steps: ① First, train the VQ-VAE (Vector Quantized Variational Autoencoder) model of the encoding layer; ② Second, train the upsampling network; ③ Third, train the WordPiece word segmenter; ④ Finally, train the Transformer network. The specific training process of each network can be referred to existing technologies, and will not be described in detail in this embodiment of the invention.

[0054] In the training process of the generative network model, the VQ-VAE network is used to reconstruct the system's operating parameter matrix, and its decoding network and codebook mechanism are selected as the decoding layer of the generative network model. The loss function of VQ-VAE considers the reconstruction error. Quantization error and promised losses The specific expression is: (1) in, This indicates stopping the gradient operation. R Indicates the input value. This represents the reconstructed value.

[0055] Specifically, for other sub-models in the generative network model, the L2 loss function is used, expressed as follows: ;in, y Indicates the predicted value. Represents the actual value.

[0056] Step S206: Construct target expert fault annotation text information corresponding to multiple rare faults, and input the target expert fault annotation text information into the generative network model so that the generative network model outputs the target system operating parameter matrix corresponding to the rare faults.

[0057] To address the issues of scarce fault diagnosis samples and imbalanced category distribution in the initial fault diagnosis dataset, this invention generates standardized expert fault annotation text information corresponding to scarce faults based on expert knowledge and experience. To distinguish this from the expert fault annotation text information used in the aforementioned reactor system fault test and simulation studies, this expert fault annotation text information corresponding to scarce faults is referred to here as target expert fault annotation text information. Specifically, scarce faults are used here to characterize faults not included in the aforementioned reactor system fault test and simulation studies.

[0058] The aforementioned expert-annotated fault text information is input into a trained generative network model, which outputs the corresponding target system operating parameter matrix. Based on the expert-annotated fault text information, the target system operating parameter matrix, and the corresponding fault labels, a scarce fault diagnosis sample is constructed, consisting of "target system operating parameter matrix - expert-annotated fault text information - fault labels". Therefore, by constructing multiple scarce fault diagnosis samples through various scarce fault types, and using these samples along with the initial fault diagnosis dataset as the target fault diagnosis dataset, the number of fault diagnosis samples for different fault types is expanded and enriched. This achieves sample augmentation of the initial fault diagnosis dataset, alleviating the problems of scarce reactor system fault data and class imbalance, thereby improving the representativeness and training effect of the target fault diagnosis dataset.

[0059] For example, such as Figure 4 As shown, the number of fault diagnosis samples in the initial fault diagnosis dataset is used as the data distribution before augmentation, and the number of fault diagnosis samples in the target fault diagnosis dataset is used as the data distribution after augmentation. It can be seen that by enriching and expanding the initial fault diagnosis dataset with multiple scarce fault diagnosis samples to generate the target fault diagnosis dataset, the sample augmentation of the initial fault diagnosis dataset is achieved, which effectively alleviates the problem of scarce reactor system fault data and class distribution imbalance, thereby improving the representativeness and training effect of the target fault diagnosis dataset.

[0060] Step S208: Divide the target fault diagnosis dataset into a training set, a validation set, and a test set. Train the fault diagnosis model based on the training set and determine the fault diagnosis model with the best performance on the validation set.

[0061] Specifically, for the aforementioned target fault diagnosis dataset, it is divided into a training set, a validation set, and a test set according to a preset ratio. Expert fault annotation text information and / or target expert fault annotation text information are used as inputs to the fault diagnosis model, and the corresponding fault probability distribution is used as the output of the fault diagnosis model. The fault diagnosis model is trained using the training set, and a validation set is introduced during the training process to evaluate and monitor the model performance. Based on the diagnostic performance (i.e., model accuracy) on the validation set, the optimal model parameters are selected and saved for subsequent testing and analysis.

[0062] The formula for calculating model accuracy is: Precision = TP / ( TP + FP );here TP This indicates the number of times a fault is diagnosed as a certain fault label and that fault label actually exists. FP This indicates the number of times a fault label is predicted to exist but does not actually exist.

[0063] Furthermore, the preferred fault diagnosis model is a FastVIT network, whose architecture is as follows: Figure 5 As shown in the diagram. Specifically, the fault diagnosis model comprises three stages: fault encoding, feature extraction, and feature mapping. The fault encoding stage includes a reparameter mixing module and an image patch embedding module; the feature extraction stage includes an attention mechanism module and a lightweight convolutional network; and the feature mapping stage includes an average pooling + connection module and an activation function. Based on this architecture, the fault diagnosis model works as follows: After rapid dimensionality reduction, the input signal (i.e., expert fault annotation text information and / or target expert fault annotation text information) first enters the fault encoding stage, where preliminary fault encoding is completed by the reparameter mixing module and the image patch embedding module. Then, it enters the feature extraction stage, where key fault features are further extracted through the attention mechanism module and the lightweight convolutional network. Finally, it enters the feature mapping stage, where the key fault features are processed by the average pooling + connection module and the activation function to output a fault type label.

[0064] Specifically, the image patch embedding module uses 7×7 depthwise convolution and 1×1 convolution during inference, while the attention mechanism module employs encoding, batch normalization, attention mechanism and convolutional feedforward network during inference. The specific inference process of the image patch embedding module and attention mechanism module can be referred to the prior art, and the embodiments of the present invention will not be described in detail here.

[0065] In addition, the BCELoss loss function is used during training, which can independently calculate the binary cross-entropy error of each fault label, as shown in the following expression: (2) in, Indicates the actual fault label. Indicates the predicted fault label, P Indicates the number of faulty parts. N This indicates the total number of fault diagnosis samples. C The number of fault types is indicated, and in this embodiment of the invention, it is preferably 13.

[0066] Therefore, the fault diagnosis model is trained using a training set, and a validation set is introduced during the training process to evaluate and monitor the model performance. The model accuracy is calculated to select the fault diagnosis model with the highest accuracy for subsequent testing and analysis.

[0067] Step S210: Evaluate the optimal model on the test set. If the predetermined index is met, the training ends. Otherwise, update the target fault diagnosis dataset and retrain the fault diagnosis model based on the updated target fault diagnosis dataset until the predetermined index is met.

[0068] Specifically, for the optimal model selected from the validation set, its performance is evaluated using a test set, with model accuracy as the primary evaluation metric. If the model accuracy is not less than a preset threshold, the optimal model is considered to meet the predetermined metric, training ends, meaning the optimal model's performance has converged, and the optimal model is used as the trained fault diagnosis model. Conversely, if the model accuracy is less than the preset threshold, the optimal model is considered to not meet the predetermined metric. In this case, the target fault diagnosis dataset is updated. For example, the initial fault diagnosis dataset may be expanded by introducing reactor system fault experiments or simulation examples, or by introducing new rare faults and expert fault annotation text information to expand the rare fault diagnosis samples. The fault diagnosis model is then retrained based on the updated target fault diagnosis dataset until the predetermined metric is met.

[0069] Therefore, as Figure 6As shown, the training and validation process of the fault diagnosis model is as follows: First, an initial fault diagnosis dataset is constructed based on the reactor system fault experiment and simulation program. Then, target expert fault annotation text information corresponding to multiple rare faults is constructed, and the target system operating parameter matrix is ​​determined based on the target expert fault annotation text information and the generative network model, so as to construct rare fault diagnosis samples based on the target expert fault annotation text information and the target system operating parameter matrix. Next, the target fault diagnosis dataset is determined based on the multiple rare fault diagnosis samples and the initial fault diagnosis dataset. Finally, the target fault diagnosis dataset is divided into a training set, a validation set, and a test set. The fault diagnosis model is trained using the training set, and the model is validated and evaluated using the validation set during training to select the optimal model. The optimal model is tested and evaluated using the test set. If the predetermined indicators are met, training ends; otherwise, the target fault diagnosis dataset is updated, and the fault diagnosis model is retrained based on the updated target fault diagnosis dataset until the predetermined indicators are met.

[0070] For example, under the same test set and training configuration, the fault diagnosis models trained on the initial fault diagnosis dataset before augmentation and the target fault diagnosis dataset after augmentation achieved model accuracies of 54.61% and 97.32% on the test set, respectively. Therefore, compared to the unaugmented initial fault diagnosis dataset, the data augmentation strategy significantly improved the fault diagnosis accuracy of the fault diagnosis model. This indicates that the constructed fault diagnosis sample generation and augmentation scheme can effectively improve the performance limitations caused by the scarcity of fault diagnosis samples and class imbalance, thereby enhancing the fault diagnosis model's ability to identify different fault types and its generalization performance.

[0071] Example 3 Corresponding to the above method embodiments, this invention also provides a reactor system fault diagnosis device based on sample enhancement, such as... Figure 7 As shown, the device includes: Module 71 is used to build an initial fault diagnosis dataset for the reactor system under various fault scenarios. The initial fault diagnosis dataset includes multiple fault diagnosis samples, each of which includes: a system operating parameter matrix, expert fault annotation text information, and fault labels. The acquisition module 72 is used to acquire the target expert fault annotation text information corresponding to multiple rare faults, and input the target expert fault annotation text information into the pre-trained generative network model so that the generative network model outputs the target system operating parameter matrix corresponding to the rare faults; wherein, the rare faults are faults not included in multiple fault scenarios; The generation module 73 is used to generate multiple scarce fault diagnosis samples based on the target expert fault annotation text information, the target system operating parameter matrix and the corresponding target fault labels, and to determine the target fault diagnosis dataset based on the multiple scarce fault diagnosis samples and the initial fault diagnosis dataset. Training module 74 is used to train the fault diagnosis model based on the target fault diagnosis dataset, and to perform fault diagnosis on the reactor system based on the trained fault diagnosis model.

[0072] The reactor system fault diagnosis device based on sample augmentation provided in this invention constructs multiple rare fault diagnosis samples by using target expert fault annotation text information and generative network models corresponding to multiple rare faults. The initial fault diagnosis dataset is enriched and expanded based on the multiple rare fault diagnosis samples, thereby realizing sample augmentation of the initial fault diagnosis dataset. This alleviates the problems of scarce reactor system fault data and class distribution imbalance, thereby improving the representativeness and training effect of the target fault diagnosis dataset, and ultimately enhancing the fault diagnosis capability of the reactor system.

[0073] Optionally, the construction module 71 is further configured to: obtain system operating parameters of the reactor system under various failure scenarios, and determine the corresponding system operating parameter matrix, fault labels and expert fault annotation text information based on the system operating parameters; wherein, the expert fault annotation text information includes information on changes in faulty components and key system parameters; generate fault diagnosis samples based on the system operating parameter matrix, expert fault annotation text information and fault labels, and construct an initial fault diagnosis dataset based on multiple fault diagnosis samples.

[0074] Optionally, the generative network model includes an encoding layer, a mapping layer, and a decoding layer. Inputting the target expert fault annotation text information into the pre-trained generative network model so that the generative network model outputs the target system operating parameter matrix corresponding to the rare fault includes: inputting the target expert fault annotation text information into the encoding layer so that the encoding layer outputs the corresponding target semantic information; inputting the target semantic information into the mapping layer so that the mapping layer outputs the corresponding target latent semantic feature information; and inputting the target latent semantic feature information into the decoding layer so that the decoding layer outputs the corresponding target system operating parameter matrix.

[0075] Optionally, the target fault diagnosis dataset includes a training set, a validation set, and a test set. The fault diagnosis model is trained based on the target fault diagnosis dataset, including: training the fault diagnosis model using the training set and evaluating the model using the validation set during training to calculate the model accuracy; selecting the fault diagnosis model with the highest accuracy as the target fault diagnosis model; evaluating the performance of the target fault diagnosis model using the test set to calculate the target model accuracy; and if the target model accuracy is not less than a preset threshold, then the target fault diagnosis model is considered the trained fault diagnosis model.

[0076] Optionally, the device further includes: if the accuracy of the target model is less than a preset threshold, updating the target fault diagnosis dataset and retraining the fault diagnosis model based on the updated target fault diagnosis dataset until the accuracy of the target model is not less than the preset threshold.

[0077] Optionally, fault diagnosis of the reactor system is performed based on the trained fault diagnosis model, including: obtaining the current system operating parameters of the reactor system in the current time window, and determining the current system operating parameter matrix based on the current system operating parameters; inputting the current system operating parameter matrix into the fault diagnosis model so that the fault diagnosis model outputs fault diagnosis information; wherein, the fault diagnosis information includes fault probability distribution information.

[0078] The reactor system fault diagnosis device based on sample enhancement provided in this embodiment of the invention has the same technical features as the reactor system fault diagnosis method based on sample enhancement provided in the above embodiment, so it can also solve the same technical problems and achieve the same technical effects.

[0079] This invention also provides an electronic device, including a processor and a memory. The memory stores machine-executable instructions that can be executed by the processor. The processor executes the machine-executable instructions to implement the above-described sample-enhanced reactor system fault diagnosis method.

[0080] See Figure 8 As shown, the electronic device includes a processor 100 and a memory 101. The memory 101 stores machine-executable instructions that can be executed by the processor 100. The processor 100 executes the machine-executable instructions to implement the above-described sample-enhanced reactor system fault diagnosis method.

[0081] Furthermore, Figure 8 The electronic device shown also includes a bus 102 and a communication interface 103, with the processor 100, the communication interface 103 and the memory 101 connected via the bus 102.

[0082] The memory 101 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 103 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc. The bus 102 may be an ISA (Industrial Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Enhanced Industry Standard Architecture) bus, etc. These buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 8 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0083] Processor 100 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 100 or by instructions in software form. Processor 100 may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software module can reside in a readily available storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory 101, and the processor 100 reads the information from memory 101 and, in conjunction with its hardware, completes the steps of the method described in the foregoing embodiments.

[0084] This embodiment also provides a computer-readable storage medium storing a computer program, which is executed by a processor to perform the above-described sample-enhanced reactor system fault diagnosis method.

[0085] The computer program product of the reactor system fault diagnosis method, apparatus and electronic device based on sample enhancement provided in the embodiments of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation, please refer to the method embodiments, which will not be repeated here.

[0086] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and apparatus described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0087] Furthermore, in the description of the embodiments of the present invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.

[0088] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part 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 several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0089] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0090] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A sample augmentation based reactor system fault diagnosis method, characterized by, The method includes: An initial fault diagnosis dataset for the reactor system under various fault scenarios is constructed; wherein, the initial fault diagnosis dataset includes multiple fault diagnosis samples, and each fault diagnosis sample includes: a system operating parameter matrix, expert fault annotation text information, and fault labels; Obtain target expert fault annotation text information corresponding to multiple rare faults, and input the target expert fault annotation text information into a pre-trained generative network model so that the generative network model outputs the target system operating parameter matrix corresponding to the rare fault; wherein, the rare fault is a fault not included in the multiple fault scenarios. Multiple rare fault diagnosis samples are generated based on the target expert fault annotation text information, the target system operating parameter matrix and the corresponding target fault labels, and the target fault diagnosis dataset is determined based on the multiple rare fault diagnosis samples and the initial fault diagnosis dataset. The fault diagnosis model is trained based on the target fault diagnosis dataset, and the fault diagnosis of the reactor system is performed based on the trained fault diagnosis model.

2. The method of claim 1, wherein, The construction of the initial fault diagnosis dataset for the reactor system under various fault scenarios includes: The system operating parameters of the reactor system under various failure scenarios are obtained, and the corresponding system operating parameter matrix, the failure label, and the expert failure annotation text information are determined based on the system operating parameters; wherein, the expert failure annotation text information includes the failure component and the change information of key system parameters; The fault diagnosis samples are generated based on the system operating parameter matrix, the expert fault annotation text information, and the fault labels, and the initial fault diagnosis dataset is constructed based on multiple fault diagnosis samples.

3. The method of claim 1, wherein, The generative network model includes an encoding layer, a mapping layer, and a decoding layer; the step of inputting the target expert fault annotation text information into the pre-trained generative network model, so that the generative network model outputs the target system operating parameter matrix corresponding to the rare fault, includes: The target expert fault annotation text information is input into the encoding layer so that the encoding layer outputs the corresponding target semantic information. The target semantic information is input into the mapping layer so that the mapping layer outputs the corresponding target latent semantic feature information. The target latent semantic feature information is input into the decoding layer so that the decoding layer outputs the corresponding target system operating parameter matrix.

4. The method of claim 1, wherein, The target fault diagnosis dataset includes a training set, a validation set, and a test set. Training the fault diagnosis model based on the target fault diagnosis dataset includes: The fault diagnosis model is trained based on the training set, and the fault diagnosis model is evaluated using the validation set during the training process to calculate the model accuracy; and the fault diagnosis model with the highest model accuracy is selected as the target fault diagnosis model. The target fault diagnosis model is evaluated for performance based on the test set, and the accuracy of the target model is calculated. If the accuracy of the target model is not less than a preset threshold, then the target fault diagnosis model is used as the trained fault diagnosis model.

5. The method of claim 4, wherein, The method further includes: If the accuracy of the target model is less than the preset threshold, the target fault diagnosis dataset is updated, and the fault diagnosis model is retrained based on the updated target fault diagnosis dataset until the accuracy of the target model is not less than the preset threshold.

6. The method of claim 1, wherein, The step of performing fault diagnosis on the reactor system based on the trained fault diagnosis model includes: Obtain the current system operating parameters of the reactor system in the current time window, and determine the current system operating parameter matrix based on the current system operating parameters; The current system operating parameter matrix is ​​input into the fault diagnosis model so that the fault diagnosis model outputs fault diagnosis information; wherein, the fault diagnosis information includes fault probability distribution information.

7. A sample-based enhancement-based reactor system fault diagnosis apparatus characterized by, The device includes: The construction module is used to construct an initial fault diagnosis dataset for the reactor system under various fault scenarios; wherein, the initial fault diagnosis dataset includes multiple fault diagnosis samples, and each fault diagnosis sample includes: a system operating parameter matrix, expert fault annotation text information, and fault labels; The acquisition module is used to acquire target expert fault annotation text information corresponding to multiple rare faults, and input the target expert fault annotation text information into a pre-trained generative network model so that the generative network model outputs the target system operating parameter matrix corresponding to the rare fault; wherein, the rare fault is a fault not included in the multiple fault scenarios. The generation module is used to generate multiple scarce fault diagnosis samples based on the target expert fault annotation text information, the target system operating parameter matrix and the corresponding target fault labels, and to determine the target fault diagnosis dataset based on the multiple scarce fault diagnosis samples and the initial fault diagnosis dataset. The training module is used to train the fault diagnosis model based on the target fault diagnosis dataset, and to perform fault diagnosis on the reactor system based on the trained fault diagnosis model.

8. The apparatus of claim 7, wherein, The step of performing fault diagnosis on the reactor system based on the trained fault diagnosis model includes: Obtain the current system operating parameters of the reactor system in the current time window, and determine the current system operating parameter matrix based on the current system operating parameters; The current system operating parameter matrix is ​​input into the fault diagnosis model so that the fault diagnosis model outputs fault diagnosis information; wherein, the fault diagnosis information includes fault probability distribution information.

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

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, performs the steps of the method described in any one of claims 1-6.

Images