Fault detection method, device and equipment of nuclear power plant control system, medium and product

Abstract

The invention discloses a fault detection method, device and equipment of a nuclear power plant control system, a medium and a product, and relates to the field of intelligent monitoring of nuclear power equipment, and the method comprises the steps: obtaining the real-time operation data of the nuclear power plant control system; according to the real-time operation data, multiple models are adopted to carry out fault classification on the nuclear power plant control system, and a fault classification result and a confidence score of each model are obtained; wherein the plurality of models are respectively a random forest model, a long short-term memory network and a fuzzy inference system; each model is obtained by training a training sample set in combination with a dynamic simulation model library in advance, and samples in the training sample set comprise real samples and supplementary samples; and determining the final fault category of the nuclear power plant control system according to the fault classification result and the confidence score of each model. The fault detection accuracy of the nuclear power plant control system is improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent monitoring of nuclear power equipment, and particularly to a fault detection method, device, equipment, medium and product for a nuclear power plant control system. Background Art

[0002] In the process of fault detection of a nuclear power plant control system, related technologies have not fully considered the characteristics of high-dimensional time-series data of the nuclear power plant control system, nor solved the problem of fusion of multi-source heterogeneous data. Moreover, when dealing with the problem of data imbalance, dynamic simulation data is not combined, resulting in an increase in the false positive rate under complex working conditions and limited ability to identify hidden faults.

[0003] Therefore, there is an urgent need for a new fault detection method to improve the fault detection accuracy of the power plant control system. Summary of the Invention

[0004] The purpose of the present application is to provide a fault detection method, device, equipment, medium and product for a nuclear power plant control system, which can improve the accuracy of fault detection of the nuclear power plant control system.

[0005] To achieve the above purpose, the present application provides the following solutions: In the first aspect, the present application provides a fault detection method for a nuclear power plant control system, including: Obtaining real-time operation data of the nuclear power plant control system; According to the real-time operation data, using multiple models to classify faults of the nuclear power plant control system, and obtaining the fault classification results and confidence scores of each model; wherein, the multiple models are respectively a random forest model, a long short-term memory network and a fuzzy inference system; each model is pre-trained using a training sample set combined with a dynamic simulation model library, and the samples in the training sample set include real samples and supplementary samples; Determining the final fault category of the nuclear power plant control system according to the fault classification results and confidence scores of each model.

[0006] In the second aspect, the present application provides a fault detection device for a nuclear power plant control system, including: A multi-source data acquisition module, configured to obtain real-time operation data of the nuclear power plant control system; A multi-model analysis module, configured to classify faults of the nuclear power plant control system using multiple models according to the real-time operation data, and obtaining the fault classification results and confidence scores of each model; wherein, the multiple models are respectively a random forest model, a long short-term memory network and a fuzzy inference system; each model is pre-trained using a training sample set combined with a dynamic simulation model library, and the samples in the training sample set include real samples and supplementary samples; A fault category determination module, configured to determine the final fault category of the nuclear power plant control system according to the fault classification results and confidence scores of each model.

[0007] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the above-mentioned fault detection method for a nuclear power plant control system.

[0008] In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the above-mentioned fault detection method for a nuclear power plant control system is implemented.

[0009] In a fifth aspect, the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, the above-mentioned fault detection method for a nuclear power plant control system is implemented.

[0010] According to the specific embodiments provided by the present application, the present application has the following technical effects: The present application provides a fault detection method, device, equipment, medium and product for a nuclear power plant control system. By classifying faults in the nuclear power plant control system through multiple models, combining the advantages of the random forest model, long short-term memory network and fuzzy inference system, the problem of difficult detection of hidden faults is solved, and the comprehensiveness and accuracy of fault detection are improved. Moreover, the training sample set of the model includes real samples and supplementary samples, effectively solving the problem of data imbalance. The model is trained by combining the training sample set with a dynamic simulation model library, further improving the accuracy of fault detection in the nuclear power plant system. Description of the Drawings

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0012] Figure 1 It is an application environment diagram of a fault detection method for a nuclear power plant control system in an embodiment of the present application.

[0013] Figure 2 It is a flowchart of a fault detection method for a nuclear power plant control system provided in an embodiment of the present application.

[0014] Figure 3 It is a schematic diagram of the process of detecting faults in a nuclear power plant control system in an embodiment of the present application.

[0015] Figure 4 This is a schematic diagram of the training process of the model in an embodiment of the present application.

[0016] Figure 5 This is a schematic diagram of the functional modules of a fault detection device for a nuclear power plant control system provided in another embodiment of the present application. Detailed implementation manners

[0017] Next, the technical solutions in the embodiments of the present application will be clearly and completely described with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0018] In related technologies, for example, a power generation equipment fault prediction method, system, device and medium with the Chinese invention patent publication number CN119293406A inputs features highly relevant to fault prediction into a trained random forest algorithm model and a gradient boosting machine algorithm model for power generation equipment fault prediction, and combines the two prediction results to obtain the final result. However, this method does not target the high-dimensional time-series data of nuclear power control systems and cannot handle the problem of multi-source heterogeneous data fusion.

[0019] The nuclear power valve fault diagnosis method and system with the Chinese invention patent publication number CN119295815A collects vibration signal data of different types of valves in a nuclear power plant, normalizes it, plots and saves it as a target picture, uses the normal vibration signal data picture as a training set, and the vibration signal data pictures including normal and abnormal ones as a test set, and improves the loss function of the ProGAN model based on the Wasserstein distance and introduces a threshold determination module. However, since the threshold determination module does not combine dynamic simulation data, it is difficult to accurately judge under complex working conditions and has a high misjudgment rate.

[0020] A multi-level intelligent fault diagnosis and monitoring system for a nuclear power plant digital instrument and control system with the Chinese invention patent publication number CN117687375A collects the operation status information of a distributed control system (DCS) through a data acquisition server, stores it through a data storage server, performs logical operations and threshold comparisons by a computing server, then summarizes and classifies it by a fault diagnosis unit, and finally visualizes and displays it through a terminal display unit. However, this system does not integrate a multi-model fusion algorithm and cannot effectively handle multi-device coupling faults.

[0021] The fault monitoring method, device, equipment and storage medium of electrical equipment in a nuclear power plant with the Chinese invention patent publication number CN119379248A obtain the simulation operation data and actual operation data of the electrical equipment based on a pre-established simulation model, perform fuzzy reasoning through a pre-established fuzzy expert system, and then perform defuzzification processing to obtain the fault monitoring result. However, due to the lack of introduction of a deep learning model, the recognition effect of latent faults is not good. In view of the limitations of the above prior art, the present application proposes a fault detection method, device, equipment, medium and product for a nuclear power plant control system to solve the above problems and improve the accuracy and effectiveness of fault detection in a nuclear power plant control system.

[0022] To make the above objects, features and advantages of the present application more obvious and understandable, the following further detailed description of the present application will be given in conjunction with the accompanying drawings and specific embodiments.

[0023] The fault detection method for a nuclear power plant control system provided by an embodiment of the present application can be applied to an application environment as Figure 1 shown. Among them, the terminal 102 communicates with the server 104 through a network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, integrated on the server 104, or placed on the cloud or other servers. The terminal 102 can send the real-time operation data of the nuclear power plant control system to the server 104. After receiving the real-time operation data of the nuclear power plant control system, the server 104 uses multiple models to classify the faults of the nuclear power plant control system, and determines the final fault category of the nuclear power plant control system according to the fault classification results and confidence scores of each model. The server 104 can feedback the final fault category of the nuclear power plant control system to the terminal 102. In addition, in some embodiments, the fault detection method for the nuclear power plant control system can also be implemented independently by the server 104 or the terminal 102.

[0024] Among them, the terminal 102 can be, but is not limited to, various desktop computers, laptop computers, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, etc. The portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server 104 can be implemented by an independent server or a server cluster composed of multiple servers, and can also be a cloud server.

[0025] In an exemplary embodiment, as Figure 2 and Figure 3As shown, a fault detection method for a nuclear power plant control system is provided. This method is executed by a computer device, which can be specifically executed by a computer device such as a terminal or a server alone, or jointly executed by a terminal and a server. In the embodiments of this application, taking the application of this method to Figure 1 server 104 in

[0026] as an example, the following steps 201 to 203 are included.

[0027] In a specific application example, first, the sensor data, instruction flow data, and vibration signals of the nuclear power plant control system are obtained in real time. The vibration signal is one-dimensional vibration time series data. Then, the sliding window algorithm is used to perform streaming processing on the sensor data. The Gramian Angular Field (GAF) algorithm is used to convert the vibration signal into a two-dimensional grayscale image to obtain an image-based vibration signal. The instruction flow data is the data output by the distributed control system.

[0028] Among them, the Z-Score normalization algorithm is used to normalize the real-time operation data, and the formula is: ; where is the normalized real-time operation data, is the real-time operation data before normalization, is the mean value of the real-time operation data, is the standard deviation of the real-time operation data.

[0029] For example, a set of data collected by a certain temperature sensor is [25, 28, 30, 26, 27]. First, calculate the mean value of this set of data, and the standard deviation , then the normalized data is . Through normalization processing, the data of different sensors are unified to a standard scale, eliminating the differences caused by different dimensions and value ranges between the data, making the data of different sensors comparable and facilitating subsequent model processing.

[0030] In the actual operation scenario of a nuclear power plant, there are various types of sensors in the nuclear power plant, which monitor different operation parameters respectively. For example, temperature sensors are used to measure the working temperature of equipment in real time, pressure sensors are responsible for monitoring the pressure changes in pipelines, and flow sensors record the flow rate of media, etc. The sampling frequencies of these sensors vary according to monitoring requirements. The temperature sensor may collect data every 1 second, that is, the sampling frequency is 1 Hz; the pressure sensor is set to a sampling frequency of 5 Hz in order to more accurately capture pressure fluctuations.

[0031] In this application, for sensor data with different sampling frequencies, a sliding window algorithm is used for streaming processing. The window length is dynamically adjusted according to the sampling frequency of the sensor. Taking a temperature sensor as an example, if its sampling frequency is 1 Hz, considering the requirements of data integrity and trend analysis, the window length may be set to 10 sampling points, which means the window duration is 10 seconds; when the sampling frequency becomes 2 Hz, to ensure data processing efficiency and accuracy, the window length is correspondingly adjusted to 5 sampling points, that is, 2.5 seconds. At the same time, to ensure the continuity and accuracy of data processing, the window overlap rate is not less than 50%, thereby avoiding the loss of key information during data processing and ensuring the reliability of subsequent analysis results.

[0032] The process of the vibration signal changing over time is like a curve fluctuating on a one-dimensional number axis. The GAF algorithm unfolds this curve and plots it on a two-dimensional plane, using grayscale values to represent the magnitude of the vibration amplitude. Thus, features originally hidden in the one-dimensional data, such as the periodic pattern of vibration and sudden abnormal vibration points, become more intuitive in the two-dimensional grayscale image and are more easily captured and analyzed by subsequent models, providing richer and more intuitive information for fault detection.

[0033] Step 202: According to the real-time operation data, use multiple models to classify the faults of the nuclear power plant control system, and obtain the fault classification results and confidence scores of each model.

[0034] Among them, the multiple models are respectively a random forest model, a long short-term memory network, and a fuzzy inference system.

[0035] In this application, the random forest model, the long short-term memory network, and the fuzzy inference system are run in parallel. Each model analyzes the data according to its own algorithm and outputs the fault classification result and the confidence score. By integrating multiple models, the problems of data imbalance and hidden fault detection are solved, and the comprehensiveness and accuracy of fault detection are improved.

[0036] (1) The random forest model first uses the Gini coefficient feature selection algorithm to screen out highly relevant features, and then conducts fault classification based on the highly relevant features. Specifically, the random forest model conducts classification based on the numerous decision trees it constructs. Each decision tree judges the data separately, and finally the results of all decision trees are combined to obtain the fault classification result and the confidence score.

[0037] In the vast ocean of data in the nuclear power plant control system, there are a large number of various features, such as various sensor measurement values, equipment operating status parameters, environmental parameters, etc. The Gini coefficient feature selection algorithm can accurately select features with relatively high correlation with fault detection from numerous features. For example, when detecting a specific equipment fault, after being screened by this algorithm, several key features such as temperature, pressure, and vibration frequency may be determined, while some features with less impact on fault detection are excluded. This not only reduces the computational amount of random forest model training but also improves the accuracy of feature selection, enabling the random forest model to focus more on key information and enhancing the detection efficiency and accuracy.

[0038] (2) The long short-term memory network has the ability to remember and learn temporal data and uses the time attention mechanism algorithm to capture implicit features in temporal data. When processing the temporal data of the nuclear power plant control system, the time attention weights in the long short-term memory network are calculated using the Softmax function: ; where is t the attention score at time is t the time attention weight at time T and

[0039] is the total number of time steps. Taking the monitoring of the core temperature change of the nuclear reactor as an example, the core temperature remains relatively stable during normal operation, but there may be some slight fluctuations in the early stage of some potential faults. Through the time attention mechanism, the long short-term memory network can pay attention to these fluctuation points that account for a small proportion in the overall data but are crucial, thereby discovering early signs of implicit faults and enhancing the detection ability of implicit faults.

[0040] (3) The fuzzy inference system conducts inferences based on the defined deviation membership rules and converts the deviation degrees of various parameters into fault judgment bases. The fuzzy inference system uses the Gaussian membership function algorithm to define the deviation membership rules and conducts fault classification based on the deviation membership rules.

[0041] In the fault detection of the nuclear power plant control system, the deviation judgment of various operating parameters is a key link in detection. The Gaussian membership function can more accurately describe the membership relationship of parameter deviations at different levels. For example, for pressure deviation, through the Gaussian membership function, it is possible to determine the degree to which different pressure deviation values belong to different states such as normal, slightly abnormal, and severely abnormal. Assuming the normal pressure range is [0.8 MPa, 1.2 MPa], when the pressure deviation is 0.1 MPa, through the calculation of the Gaussian membership function, it can be judged that the degree to which it belongs to the slightly abnormal state is 0.6 (the specific value is obtained according to the function calculation). Such a description is more accurate and detailed, improving the logic and accuracy of detection.

[0042] Each of the above models is pre-trained using a training sample set in combination with a dynamic simulation model library. The samples in the training sample set include real samples and supplementary samples.

[0043] In a specific application example, such as Figure 4 shown, the training process of multiple models includes the following steps 301 to 305.

[0044] Step 301, obtain the historical operation data of the nuclear power plant control system and the corresponding fault category labels to obtain real samples.

[0045] Step 302, use the ProGAN data augmentation model to generate supplementary samples corresponding to different fault categories. Specifically, when using the ProGAN data augmentation model to generate supplementary samples corresponding to different fault categories, the distribution consistency between the supplementary samples and the real data is constrained by the Kullback-Leibler divergence algorithm (KLDivergence), and the constraint condition is: ; where is the KL divergence, is the distribution of the real data, is the distribution of the supplementary samples.

[0046] In practical applications, the ProGAN data augmentation model optimizes the generative adversarial network based on the Wasserstein distance and the gradient penalty algorithm to generate supplementary samples with a distribution consistent with the real fault data, and verifies the effectiveness of the generated data through a dynamic simulation model library.

[0047] Among them, the generator adopts a residual neural network structure, which can effectively avoid the problem of gradient disappearance, enabling the generator to better learn data features during the training process, thereby improving the quality of the generated samples. The detector introduces the spectral normalization algorithm, enhancing the stability and detection ability of the detector, enabling it to more accurately distinguish between real data and generated data.

[0048] The effectiveness verification of the generated data calculates the temporal similarity between the simulation operation data and the generated data through the dynamic time warping algorithm, and the similarity threshold is set to 0.85. If the calculated similarity is greater than 0.85, the generated data is considered valid.

[0049] The supplementary samples generated by the ProGAN data augmentation model need to meet the following 3 conditions.

[0050] (1) The Wasserstein distance between the supplementary samples and the distribution of the real fault data is less than a preset threshold.

[0051] (2) After the supplementary samples are verified by the dynamic simulation model, the consistency with the physical mechanism of the target fault exceeds 90%.

[0052] (3) The supplementary samples need to pass the physical consistency verification algorithm, and the verification formula is: ; where is the simulation model function, which is used to simulate the output result according to the input data, is the threshold for judging the consistency between the generated data and the simulation model output, , N is the number of samples, is the i th input data of the generated sample, is the i th actual output data of the generated sample.

[0053] For example, when detecting a certain type of fault, the data of this type of fault accounts for a relatively small proportion in the training sample set, perhaps only 10% of the total data volume, resulting in insufficient learning of this type of fault during the training of the model. This application generates supplementary samples through the ProGAN data enhancement model to make them similar to the distribution of real fault data, increasing the proportion of this type of fault data in the training sample set to 30% or even higher, thereby enhancing the detection ability of the model for this type of fault.

[0054] Through the above strict condition restrictions, it is ensured that the quality of the supplementary samples is reliable, which can effectively solve the data imbalance problem and provide richer and more representative data for model training.

[0055] Step 303: According to the real samples and the supplementary samples, use each model to perform fault classification respectively, and obtain the preliminary prediction category and confidence score of each model for each sample.

[0056] Step 304: Use the dynamic simulation model library to perform dynamic simulation on the nuclear power plant control system to generate simulation operation data corresponding to different fault categories.

[0057] The dynamic simulation model library is based on the equipment benchmark parameter library of the full-scope simulator of the nuclear power plant to generate simulation operation data under normal conditions and typical fault states. In the actual modeling process, the improved Runge-Kutta algorithm (R-K method) is used to solve the equipment dynamics differential equations. Taking the cooling pump system in the nuclear power plant as an example, its dynamics differential equations describe the dynamic relationships among parameters such as the pump speed, flow rate, and pressure, and these parameters affect and restrict each other. By using the improved Runge-Kutta algorithm to solve these equations, the operating states of the cooling pump under different working conditions can be simulated, and the corresponding simulation data of the operating parameters can be obtained, such as the change curves of the pump flow rate and pressure under normal operating conditions, and the abnormal fluctuations of these parameters under specific fault conditions.

[0058] The improved Runge-Kutta algorithm performs adaptive step-size control, optimizes the step-size adjustment strategy, and optimizes the system matrix based on the Runge-Kutta algorithm.

[0059] In adaptive step-size control, the initial step size is changed from the traditional fixed step size to dynamic adjustment. For example, the initial step size is set to 0.1 s, the maximum step size is set to 1 s, and the minimum step size is set to 0.01 s to avoid calculation divergence or insufficient accuracy. The relative tolerance value of the local truncation error is set to , ensuring the simulation accuracy.

[0060] In the step-size adjustment strategy, the step-size increase factor is set to 1.5 and the decrease factor is set to 0.5, and the step size is dynamically scaled according to the error estimate.

[0061] In the coefficient matrix optimization, the coefficients of the classical fourth-order R-K method are adjusted. For example, an implicit R-K method applicable to stiff equations is introduced to improve the numerical stability of the dynamic equations of nuclear equipment.

[0062] The above implicit R-K method performs adaptive optimization for stiff problems, dynamically adjusts the step-size strategy, and improves the calculation efficiency.

[0063] Specifically, the dynamic differential equations of the nuclear power plant control system usually contain coupled processes with fast and slow changes (such as heat conduction and mechanical vibration). Such problems have stiff characteristics. The implicit R-K method enhances stability through the following two methods: (1) Reconstruct the coefficient matrix using the Butcher table of the implicit R-K method (such as the RadauIIA or Gauss-Legendre method), and its coefficient matrix contains non-zero diagonal elements to enhance numerical stability. (2) Introduce the Newton-Raphson iteration method into the implicit equation, and combine the Jacobian matrix to update the iteration solution strategy to ensure convergence even at a large step size.

[0064] In the dynamic step-size adjustment strategy, the step size is dynamically adjusted through local truncation error estimation (such as the embedded implicit R-K method). For example: the relative tolerance value of the local truncation error is set to , the absolute error tolerance . The step-size scaling factor is adjusted according to the error feedback. For example, when the error exceeds the threshold, the step size is reduced to 0.5 times the original value, and vice versa, it is expanded to 1.5 times.

[0065] In improving the calculation efficiency, for the sparsity of the dynamic equations of nuclear power equipment, a sparse Jacobian matrix storage and a fast linear algebra solver (such as the Krylov subspace method) are used to reduce the calculation cost of implicit iteration. The iterative solution of the implicit equation is executed in parallel on a multi-core processor to shorten the single-step simulation time.

[0066] In a specific application example, the dynamic simulation model library includes: a device simulation sub-model based on physical mechanisms and a deviation threshold calibration module driven by fuzzy inference.

[0067] (1) The device simulation sub-model based on physical mechanisms is used to simulate the dynamic response of the nuclear power plant control system. The device simulation sub-model based on physical mechanisms is carefully constructed according to the physical principles and working characteristics of each device in the nuclear power plant. Taking the steam generator as an example, this device simulation sub-model fully considers the complex heat exchange process in the steam generator, including how heat is transferred from high-temperature media to low-temperature media, and the flow characteristics of the fluid in the pipeline, such as flow velocity, pressure distribution, etc. By accurately simulating these physical processes, it can accurately present the operating state of the steam generator under different working conditions, providing a reliable reference basis for fault detection.

[0068] (2) The deviation threshold calibration module driven by fuzzy inference is used to compare the simulation data with the actual operation data in real time and output the dynamic warning threshold.

[0069] In the deviation threshold calibration module: The fuzzy inference rule base adopts IF-THEN type fuzzy rules. For example: "IF the pressure deviation > 0.2MPa AND the temperature change rate < 5℃ / min THEN the warning level = level III". This rule is formulated by nuclear power plant operation experts through long-term accumulation of practical experience and in-depth research on the characteristics of equipment failures.

[0070] The dynamic threshold update is achieved through the exponentially weighted moving average algorithm, and the value range of the smoothing factor α is [0.2, 0.5]. If α is taken as 0.3, when the deviation between the actual operation data and the simulation operation data changes, the exponentially weighted moving average algorithm will calculate according to a certain weight based on the current deviation and historical deviation data. Specifically, it will pay more attention to the recent deviation data while taking into account the influence of historical data, so as to obtain a more reasonable dynamic warning threshold, greatly improving the timeliness and accuracy of fault warning.

[0071] Step 305: According to the simulation operation data corresponding to different fault categories, the preliminary prediction category and confidence score of each model for each sample, respectively correct the parameters of each model to obtain a trained model.

[0072] Specifically, input the preliminary prediction category and confidence score of each model for each sample into the dynamic simulation model library, compare the deviation between the sample corresponding to the preliminary prediction category and the simulation operation data corresponding to the same fault category, trigger a trend warning and correct misjudged cases.

[0073] In a specific application example, the dynamic simulation comparison process in step 305 is as follows.

[0074] (1)Calculate the deviation membership degree of the simulation operation data and the actual operation data (samples) based on the fuzzy expert system. The triangle membership function algorithm is used for the calculation of the deviation membership degree: ; where is the value of the triangle membership function, that is, the deviation membership degree, is the input parameter of the triangle membership function, c is the membership degree center, d is the width parameter.

[0075] (2)When the deviation membership degree exceeds the dynamic threshold, trigger a multi-level warning signal. The rules for triggering the multi-level warning signal are as follows.

[0076] Level I warning: The deviation membership degree > 0.6 and the duration < 10s, indicating that a slight abnormal situation may have occurred and requires close attention.

[0077] Level II warning: The deviation membership degree > 0.8 or the Level I warning lasts for more than 30s, indicating that the abnormal situation is relatively serious and may have a greater impact on the equipment operation.

[0078] Level III warning: The deviation membership degree > 0.95 or two or more subsystems trigger the Level II warning simultaneously, which means that a serious fault has occurred and immediate measures must be taken.

[0079] (3)Reverse-correct the parameters of the model according to the warning signal. For example, if the Level II warning is triggered, the feature selection range of the random forest model may be readjusted, adding some relevant features that were not considered before, or adjusting the time step of the long short-term memory network, or directly marking it as a misjudgment case (such as excluding classification results with low confidence), so as to more accurately capture the fault trend and improve the accuracy of fault detection.

[0080] Finally, output the adjusted confidence score (dynamically correct the confidence of each model according to the simulation deviation value, such as reducing the weight of the misjudgment model) and the filtered classification result (excluding the cases marked as misjudgments due to the mismatch of simulation comparison).

[0081] Step 203: Determine the final fault category of the nuclear power plant control system according to the fault classification results and confidence scores of each model.

[0082] In a specific application example, based on the fault classification results and confidence scores of each model, the final fault category of the nuclear power plant control system is determined based on the confidence dynamic weighted voting mechanism (such as the D-S evidence theory). The fault location information and disposal suggestions can also be further generated. This application provides a reference basis through the dynamic simulation model library, combines the dynamic simulation data and the confidence dynamic weighted voting mechanism, and improves the adaptive ability and the accuracy of fault detection.

[0083] In the confidence dynamic weighted voting mechanism, the confidence scores of the models are fused through the D-S evidence theory algorithm and satisfy: ; where is the basic probability assignment function, is the belief degree of proposition A , A is the proposition, B is A subset of.

[0084] The confidence dynamic weighted voting mechanism adopts the following rules.

[0085] (1) The initial value of the model weight is assigned according to its accuracy rate in historical failure cases.

[0086] (2) The confidence of real-time data is jointly calculated through the dynamic simulation deviation value and the model output entropy value.

[0087] (3) The final failure determination result needs to satisfy that the confidence scores of at least two models exceed 80%. For example, if the confidence scores of both the random forest model and the long short-term memory network exceed 80%, and the final result calculated according to the weighted voting mechanism points to a specific failure category, such as "a certain valve seal failure", then the failure category is determined as this category, and the corresponding location information, such as "the valve located at pipeline X", and the disposal suggestion, such as "immediately stop the operation of the relevant equipment and replace the valve seal", are output.

[0088] Specifically, the confidence dynamic weighted voting mechanism first assigns the initial weight according to the historical accuracy rate output by each model. The initial weight calculation formula is: ; where K is the total number of models, K = 3, is the initial weight of model j , j = 1, 2, 3. Model 1 represents the random forest model, model 2 represents the long short-term memory network, and model 3 represents the fuzzy inference system. Suppose that within a certain period of time, the accuracy rate of the random forest model in historical failure detection is 70%, the accuracy rate of the long short-term memory network is 60%, and the accuracy rate of the fuzzy inference system is 50%. Then the initial weight of the random forest model .

[0089] Then, combined with the real-time operation data and the dynamic simulation deviation value, the weight allocation is adjusted in real time. The real-time weight adjustment calculation formula is: ; where is the weight of model j after real-time adjustment. If the dynamic simulation deviation value of a certain equipment under the current working condition is 30% of the maximum allowable deviation, and the initial weight of the random forest model is 0.39, then the weight after real-time adjustment .

[0090] In another exemplary embodiment, the fault detection method for the nuclear power plant control system further includes: using the real-time operation data of the nuclear power plant control system and the corresponding final fault categories to perform online learning and optimization on each model, and the process includes the following three points.

[0091] (1) Store the real-time operation data confirmed to be a fault in the historical database, and record in detail information such as the accurate fault detection time, fault type, relevant equipment operation data, and current working conditions each time.

[0092] (2) Update the parameters of the random forest model and the long short-term memory network based on the historical database using the incremental learning algorithm.

[0093] The random forest model uses the incremental Gini coefficient update algorithm, and the formula is: , is the updated Gini coefficient, is the Gini coefficient of the old data, is the sample number of the old data, is the sample number of the newly added data, is the Gini coefficient of the newly added data. When new fault case data is added, update the Gini coefficient of the random forest model according to this algorithm, and adjust the model structure so that the model can learn new fault features.

[0094] The parameters of the long short-term memory network are updated using the Elastic Weight Consolidation (EWC) algorithm to prevent catastrophic forgetting, ensure that the model does not forget the important knowledge learned before while learning new data, and enable the model to continuously adapt to new fault cases. The update formula is , where, is the Fisher information matrix, which is used to measure the importance of the parameters, is the total loss function, is the loss function on the new data, is the hyperparameter, which is used to control the degree of retention of the old knowledge, is the current parameter value, is the old parameter value.

[0095] (3) According to the cross-unit migration requirements, adjust the benchmark parameter thresholds of the dynamic simulation model library. For example, the equipment parameters of different nuclear power units may be different. By adjusting the benchmark parameter thresholds, the dynamic simulation model library can be adapted to the characteristics of different nuclear power units.

[0096] In addition, a transfer learning algorithm can also be used to adapt to the control characteristic differences of different nuclear power units, so that the fault detection method for the nuclear power plant control system provided by this application can be flexibly applied in different nuclear power units and improve the adaptive ability.

[0097] In summary, compared with the prior art, the present application has at least the following beneficial effects.

[0098] (1) By collecting multiple data in real time and performing streaming processing through a sliding window algorithm, high-dimensional time series data can be efficiently processed, with a window overlap rate of no less than 50%, ensuring the continuity and accuracy of data processing. The Gram angular field transform algorithm is used to convert one-dimensional time series vibration signals into two-dimensional grayscale images, which helps to mine potential features in the data and provide richer information for subsequent model analysis. (2) The ProGAN data enhancement model, random forest classification model, long short-term memory network and fuzzy inference system are integrated. The ProGAN data enhancement model optimizes the generative adversarial network based on the Wasserstein distance and gradient penalty algorithm, which can generate supplementary samples consistent with the distribution of real fault data, effectively solving the problem of data imbalance. The validity of the generated data is verified by the dynamic simulation model library, ensuring the data quality; the random forest model selects highly relevant features through the Gini coefficient feature selection algorithm, improving the accuracy of feature selection; the long short-term memory network uses the time attention mechanism algorithm to capture the trend of hidden faults and enhance the detection ability of hidden faults; the fuzzy inference system defines the deviation membership rule through the Gaussian membership function algorithm, improving the logic and accuracy of fault detection. The models work together to significantly improve the comprehensiveness and accuracy of fault detection in nuclear power plant control systems. (3) The dynamic simulation model library is based on the equipment benchmark parameter library of the full-range simulator of the nuclear power plant, which can generate dynamic simulation data under normal and typical fault conditions, providing a reliable reference for fault detection. Its deviation threshold calibration module compares the simulation and actual operation data in real time, outputs the dynamic warning threshold, and improves the timeliness and accuracy of fault warning. The output results of multiple models are further integrated through the confidence dynamic weighted voting mechanism, and the fault type is determined and located in combination with the simulation operation data comparison. The initial weight is assigned according to the historical accuracy of each model, and the weight distribution can be adjusted in real time in combination with the current operating conditions and dynamic simulation deviation. At the same time, the transfer learning algorithm is used to adapt to the control characteristics of different nuclear power units, which greatly improves the adaptive ability and detection accuracy of the fault detection system.

[0099] (4) The entire fault detection method has clear and comprehensive steps, from acquiring real-time operating data and standardizing it, to using the ProGAN data enhancement model to generate supplementary samples to build a balanced training sample set, to running multiple models in parallel to output preliminary fault classification results, and then correcting misjudgments through dynamic simulation model library comparison, and finally outputting the final result based on the confidence dynamic weighted voting mechanism. Each step is closely linked to form a complete and efficient fault detection process. In addition, the online learning optimization step can continuously update the model parameters and adjust the baseline parameter thresholds, so that the fault detection method can continuously adapt to new fault cases and different unit requirements, further improving the performance and reliability of the fault detection method.

[0100] Based on the same inventive concept, the embodiment of the present application also provides a device for implementing the above-mentioned method for detecting faults in a nuclear power plant control system. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in the embodiments of the one or more devices for detecting faults in a nuclear power plant control system provided below can refer to the limitations of the method for detecting faults in a nuclear power plant control system in the above text, and will not be repeated here.

[0101] In an exemplary embodiment, Figure 5 As shown, a fault detection device for a nuclear power plant control system is provided, comprising: a multi-source data acquisition module 501 , a multi-model analysis module 502 and a fault category determination module 503 .

[0102] The multi-source data acquisition module 501 is used to obtain real-time operation data of the nuclear power plant control system.

[0103] The multi-model analysis module 502 is used to classify the faults of the nuclear power plant control system using multiple models according to the real-time operation data, and obtain the fault classification results and confidence scores of each model. Among them, the multiple models are random forest models, long short-term memory networks and fuzzy inference systems. Each model is pre-trained using a training sample set combined with a dynamic simulation model library, and the samples in the training sample set include real samples and supplementary samples.

[0104] The multi-model analysis module 502 integrates the ProGAN data enhancement model, the random forest model, the long short-term memory network and the fuzzy reasoning system. Each model cooperates with each other to jointly improve the fault detection capability.

[0105] The fault category determination module 503 is used to determine the final fault category of the nuclear power plant control system according to the fault classification results and confidence scores of each model.

[0106] In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0107] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0108] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0109] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0110] Those of ordinary skill in the art can understand that all or part of the processes of implementing the above method embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above method embodiments. Among them, any reference to a memory, database, or other medium used in the various embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0111] In each of the embodiments provided in this application, the database involved may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., and is not limited thereto. In each of the embodiments provided in this application, the processor may be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., and is not limited thereto.

[0112] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0113] In this article, specific examples are used to elaborate on the principle and implementation manner of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, according to the idea of this application, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. A fault detection method for a nuclear power plant control system, characterized in that, Including: Obtain the real-time operation data of the nuclear power plant control system; According to the real-time operation data, use multiple models to classify the faults of the nuclear power plant control system, and obtain the fault classification results and confidence scores of each model; among them, the multiple models are a random forest model, a long short-term memory network, and a fuzzy inference system respectively; each model is pre-trained using a training sample set combined with a dynamic simulation model library, and the samples in the training sample set include real samples and supplementary samples; Determine the final fault category of the nuclear power plant control system according to the fault classification results and confidence scores of each model.

2. The fault detection method of the nuclear power plant control system according to claim 1, wherein The real-time operation data includes sensor data, instruction flow data, and visualized vibration signals; Obtaining the real-time operation data of the nuclear power plant control system specifically includes: Obtain the sensor data, instruction flow data, and vibration signals of the nuclear power plant control system in real time; Perform streaming processing on the sensor data using a sliding window algorithm; Convert the vibration signal into a two-dimensional grayscale image using the Gramian angular field transformation algorithm to obtain a visualized vibration signal.

3. The fault detection method of the nuclear power plant control system according to claim 1, characterized in that, The training process of multiple models includes: Obtain the historical operation data of the nuclear power plant control system and the corresponding fault category labels to obtain real samples; Use the ProGAN data augmentation model to generate supplementary samples corresponding to different fault categories; According to the real samples and the supplementary samples, use each model to classify faults respectively, and obtain the preliminary prediction category and confidence score of each model for each sample; Perform dynamic simulation on the nuclear power plant control system using a dynamic simulation model library to generate simulation operation data corresponding to different fault categories; According to the simulation operation data corresponding to different fault categories, the preliminary prediction category and confidence score of each model for each sample, correct the parameters of each model respectively to obtain a trained model.

4. The fault detection method of the nuclear power plant control system according to claim 3, wherein, When using the ProGAN data augmentation model to generate supplementary samples corresponding to different fault categories, the distribution consistency between the supplementary samples and the real data is constrained by the Kullback-Leibler divergence algorithm.

5. The fault detection method for a nuclear power plant control system according to claim 3, characterized in that, When using the fuzzy inference system for fault classification, the fuzzy inference system defines the deviation membership rule using the Gaussian membership function algorithm and performs fault classification based on the deviation membership rule.

6. The fault detection method of the nuclear power plant control system according to claim 1, characterized in that Determining the final fault category of the nuclear power plant control system according to the fault classification results and confidence scores of each model specifically includes: Based on the confidence dynamic weighted voting mechanism, determine the final fault category of the nuclear power plant control system according to the fault classification results and confidence scores of each model.

7. A fault detection device for a nuclear power plant control system, which is applied to the fault detection method of the nuclear power plant control system described in any one of claims 1-6, and is characterized in that, Including: A multi-source data acquisition module for obtaining the real-time operation data of the nuclear power plant control system; A multi-model analysis module for classifying the faults of the nuclear power plant control system using multiple models according to the real-time operation data, and obtaining the fault classification results and confidence scores of each model; among them, the multiple models are a random forest model, a long short-term memory network, and a fuzzy inference system respectively; each model is pre-trained using a training sample set combined with a dynamic simulation model library, and the samples in the training sample set include real samples and supplementary samples; A fault category determination module, configured to determine the final fault category of the nuclear power plant control system according to the fault classification results and confidence scores of each model.

8. A computer device, comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the fault detection method of the nuclear power plant control system according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the fault detection method of the nuclear power plant control system according to any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the fault detection method of the nuclear power plant control system according to any one of claims 1-6.

Images