Nuclear energy system fault diagnosis method and system based on multi-model fusion ensemble learning

Abstract

The invention provides a nuclear energy system fault diagnosis method and system based on multi-model fusion ensemble learning, and the method comprises the following steps: S1, data processing: employing a data preprocessing module to collect the operation data of a chemical and volume control system, carrying out the cleaning, dividing and normalization of original data, and obtaining a data preprocessing module; a high-quality data set is constructed; s2, feature selection: extracting fault working condition categories and fault severity as key features through a feature selection module; s3, ensemble learning model training: constructing a Boosting and Stacking mixed framework by adopting an ensemble learning module, and training an ensemble model; and S4, model evaluation: evaluating the model performance on the test set through a comparison evaluation module. According to the method, the integrated learning model of the Boosting and Stacking mixed framework is constructed, and the hyper-parameter optimization and the adaptive mechanism are combined, so that real-time and accurate fault diagnosis is realized.

Description

Technical Field

[0001] The present invention relates to the field of nuclear power operation and maintenance support, and in particular to a nuclear energy system fault diagnosis method and system based on multi-model fusion integrated learning. Background Art

[0002] In the nuclear power sector, with the widespread adoption of nuclear energy as a clean energy source, the safety and reliability of nuclear power plants have become a global concern. As a crucial component of nuclear power plants, the Chemical and Volume Control System (CVS) is directly related to their operational efficiency and safety. Pipeline leak fault diagnosis in CVS systems remains a key challenge in nuclear power plant operations and maintenance.

[0003] However, traditional nuclear power system monitoring and diagnostic methods rely primarily on manual inspections and periodic testing. These methods are not only time-consuming and labor-intensive, but also struggle to achieve real-time monitoring and early warning of system status, making them incapable of detecting and warning of early system failures. Once an alarm condition, such as a threshold exceeding a limit, occurs, there is often insufficient time and means to conduct analysis and assessment.

[0004] Furthermore, post-fault diagnosis relies heavily on the experience of the staff, resulting in highly subjective results and placing high demands on the staff's comprehensive capabilities. The limitations of this traditional approach are particularly pronounced in the complex environment of nuclear power plants.

[0005] With advances in artificial intelligence (AI) technology, researchers have begun exploring data-driven approaches to diagnose faults and detect defects in real time. For example, for easily identifiable faults, Masoud et al. developed a new convolutional long short-term memory (CLSTM) model for fault detection and diagnosis. For fault scenarios that are difficult to classify, Saeed et al. proposed a fault diagnosis model using deep learning networks (such as LSTM and CNN). Furthermore, for unknown nuclear power faults, Li et al. proposed a new nuclear power plant operational safety margin (OSR) fault diagnosis framework based on convolutional prototype learning (CPL), which can extract discriminative fault features from raw nuclear power plant data. The introduction of these data-driven models not only improves the operating efficiency of the system, but also increases the speed and accuracy of fault response, reduces reliance on manual intervention, and thus reduces maintenance costs.

[0006] However, current mainstream data-driven models are still limited to a single model architecture, which makes it difficult to cope with the challenges posed by complex nonlinear operating conditions and sensor measurement errors in nuclear power plant operations in practical applications. This limitation of a single model significantly restricts its generalization ability.

[0007] Especially in the field of nuclear energy, which has extremely high safety requirements, it is not enough to focus only on the accuracy of model diagnosis. It is also necessary to focus on the robustness and reliability of the model.

[0008] To overcome the limitations of single models, ensemble learning frameworks have achieved significant success in various energy sectors. Their ability to synergistically integrate the strengths of multiple models to improve generalization performance has led to significant success. Ensemble models combine predictions from multiple machine learning algorithms to achieve better generalization and accuracy than a single algorithm alone. By combining the outputs of multiple models to improve overall performance, they can, to a certain extent, address the shortcomings of single models.

[0009] However, the application of ensemble learning in the nuclear energy field is still blank, and this research status is in sharp contrast to the high reliability requirements of nuclear power plant fault diagnosis.

[0010] In view of this, the inventors of the present application have designed a nuclear energy system fault diagnosis method and system based on multi-model fusion integrated learning, in order to overcome the above technical problems. Summary of the Invention

[0011] The technical problem to be solved by the present invention is to overcome the defects of low efficiency, insufficient accuracy, lack of adaptability and high maintenance cost in the pipeline leakage fault diagnosis task of the CVS system in the prior art, and to provide a nuclear energy system fault diagnosis method and system based on multi-model fusion integrated learning.

[0012] The present invention solves the above technical problems through the following technical solutions:

[0013] A nuclear energy system fault diagnosis method based on multi-model fusion ensemble learning is characterized in that the nuclear energy system fault diagnosis method includes the following steps:

[0014] S1. Data processing: The data preprocessing module is used to collect the operating data of the chemical and volume control systems, clean, divide and normalize the raw data, and construct a high-quality data set;

[0015] S2. Feature selection: The fault condition category and fault severity are extracted as key features through the feature selection module;

[0016] S3. Ensemble learning model training: Use the ensemble learning module to build a hybrid framework of Boosting and Stacking to train the ensemble model;

[0017] S4. Model evaluation: Evaluate the model performance on the test set through the comparative evaluation module.

[0018] According to one embodiment of the present invention, step S1 includes:

[0019] S 11 , Clean the original fault data of the chemical and volume control systems, remove noise data and missing values, and divide the labels according to numbers to obtain the corresponding relationship between fault numbers and fault names;

[0020] S 12 , divide the processed data into training set, validation set and test set;

[0021] S 13 , normalize the numerical data to ensure that all features are on the same scale.

[0022] According to one embodiment of the present invention, the step S 12 The proportions of the training set, the validation set and the test set are 70%, 20% and 10% respectively.

[0023] According to an embodiment of the present invention, the fault condition category in step S2 includes the fault location; the fault severity is calculated by using sensor data to obtain various indicators.

[0024] According to an embodiment of the present invention, the various indicators include leakage volume and pressure change.

[0025] According to one embodiment of the present invention, step S3 includes the following steps:

[0026] S 31 ,Selection of base models: XGBoost, Random Forest and Attention-LSTM are selected as the representative models of Boosting, traditional machine learning and deep learning respectively;

[0027] S 32 ,Design of integrated framework: adopt the hybrid framework of Boosting framework and Stacking framework to train the integrated model;

[0028] S 33 ,Data partitioning: Divide the data into training set, validation set and test set. The probability distribution of the base model output is fused through soft voting to eliminate the scale differences of different model outputs and generate an integrated prediction result;

[0029] S 34 , Hyperparameter optimization: Train the XGBoost classification model, Attention-LSTM classification model and random forest classification model as base models respectively, obtain the optimal hyperparameters of different classification models, and set the base model according to the optimal hyperparameters.

[0030] S 35,Construction of metadata dataset: In the Stacking framework, the original dataset is divided into a training set and a test set, and the input data is used to construct a metadata dataset for training the meta-model;

[0031] S 36 , training ensemble learning model: the base models are trained independently on the training set, and the best model combination and parameter settings are selected through cross-validation and hyperparameter optimization. The output of each base model is processed by Softmax to generate a probability prediction value; the first-level prediction results are stacked and fused to generate a comprehensive result.

[0032] According to one embodiment of the present invention, the Boosting framework trains the base model multiple times, uses k-fold cross validation to ensure model independence, and uniformly outputs probability prediction values ​​to eliminate scale differences.

[0033] According to one embodiment of the present invention, the Stacking framework is based on the bias-variance trade-off principle, adopts logistic regression as a meta-model, and uses the probability prediction value output of the base model as the input of the meta-model. The meta-model is used to integrate the prediction results of the base model to further optimize the classification decision.

[0034] According to an embodiment of the present invention, the comparison and evaluation module in step S4 uses accuracy, F1 score and recall rate as evaluation indicators to comprehensively evaluate the classification performance of the model.

[0035] The present invention also provides a nuclear energy system fault diagnosis system based on multi-model fusion ensemble learning, which is characterized in that the nuclear energy system fault diagnosis system adopts the nuclear energy system fault diagnosis method described above, and the nuclear energy system fault diagnosis system includes:

[0036] The data preprocessing module is used to organize, divide and normalize the raw data to make the data suitable for subsequent training and testing of the base model;

[0037] Feature selection module, extracts fault condition category and fault severity as autocorrelation features, providing key information for diagnosis;

[0038] The integrated learning module builds a multi-model fusion integrated learning model, using the Boosting and Stacking hybrid framework to integrate multiple base models to achieve fault diagnosis;

[0039] Comparative evaluation module to comprehensively evaluate the performance of different models.

[0040] The present invention also provides an electronic device, which is characterized in that the electronic device includes: a processor and a memory, the memory stores programs or instructions that can be run on the processor, and the programs or instructions are executed by the processor to implement the nuclear energy system fault diagnosis method based on multi-model fusion integrated learning as described above.

[0041] The present invention also provides a readable storage medium, which is characterized in that a program or instruction is stored on the readable storage medium, and when the program or instruction is executed by the processor, the nuclear energy system fault diagnosis method based on multi-model fusion integrated learning as described above is implemented.

[0042] The positive progress effect of the present invention is:

[0043] The present invention provides a nuclear energy system fault diagnosis method and system based on multi-model fusion integrated learning, which realizes real-time and accurate fault diagnosis by constructing an integrated learning model of a hybrid framework of Boosting and Stacking, combined with hyperparameter optimization and adaptive mechanism. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The above and other features, properties and advantages of the present invention will become more apparent through the following description in conjunction with the accompanying drawings and embodiments, in which like reference numerals represent like features throughout, wherein:

[0045] Figure 1 This is a module flow diagram of the nuclear energy system fault diagnosis method based on multi-model fusion integrated learning of the present invention.

[0046] Figure 2 This is a schematic diagram of the joint model structure in the nuclear energy system fault diagnosis method based on multi-model fusion integrated learning of the present invention.

[0047] Figure 3 This is a performance comparison chart of multiple models in the nuclear energy system fault diagnosis method based on multi-model fusion integrated learning of the present invention. DETAILED DESCRIPTION

[0048] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0049] Embodiments of the present invention will now be described in detail with reference to the accompanying drawings. Reference will now be made in detail to preferred embodiments of the present invention, examples of which are illustrated in the accompanying drawings. Wherever possible, the same reference numerals will be used throughout the drawings to represent the same or similar parts.

[0050] Furthermore, although the terms used in the present invention are selected from well-known and commonly used terms, some terms mentioned in the present specification may be selected by the applicant at his or her discretion, and their detailed meanings are explained in the relevant parts of the description herein.

[0051] Furthermore, it is required that the present invention be understood not only by the actual terms used but also by the meanings lying behind each term.

[0052] like Figures 1 to 3 As shown, the present invention discloses a nuclear energy system fault diagnosis method based on multi-model fusion integrated learning, which includes the following steps:

[0053] Step S1, data processing: A data preprocessing module is used to collect the operating data of the chemical and volumetric control system (CVS) (e.g., sensor measurements such as temperature, pressure, and flow), clean, divide, and normalize the raw data, and construct a high-quality data set.

[0054] Preferably, the step S1 includes:

[0055] Step S 11 , clean the original fault data of the chemical and volume control system (CVS), remove noise data and missing values, ensure data integrity and consistency, and divide the labels according to numbers to obtain the correspondence between fault numbers and fault names (as shown in Table 1 below).

[0056] Table 1: Fault number and fault name

[0057] Fault number Fault name 0 'Leak in the upstream pipeline of the regenerative heat exchanger tube side inlet' 1 'Regenerative heat exchanger tube side leakage' 2 'Leak in the upstream pipeline of the shell side inlet of the regenerative heat exchanger' 3 'Leak in the pipeline between the regenerative heat exchanger and the downstream heat exchanger' 4 'Leakage on the side of the downflow heat exchanger tube' 5 'Desalination mixed bed inlet pipeline leaks' 6 'Leaking pipeline connecting to RNS system' 7 'Leakage of downcomer pipeline outside containment vessel' 8 'Leakage in the downcomer pipeline inside the containment vessel' 9 'V157 upstream water supply pipeline leaks' 10 'Boric acid storage tank leak' 11 'Purification circuit pipeline valve failure' 12 'Main pipe mixing valve failure at the water supply pump inlet' 13 '14A water supply pump A inlet leakage' 14 'Low flow heat exchanger cooling' 15 'Loss of cooling in the downstream heat exchanger' 16 'Faulty water supply pump outlet valve' 17 '13A water supply pump low flow heat exchanger A tube side leakage' 18 'Make-up pump PMP failure'

[0058] Step S 12 , divide the processed data into training set, validation set and test set.

[0059] Preferably, the step S 12 The ratios of the training set, the validation set, and the test set are 70%, 20%, and 10%, respectively. In addition, a 5-fold cross validation is used to ensure the independence of the base model and full utilization of data.

[0060] Step S 13 , normalize the numerical data to ensure that all features are on the same scale.

[0061] This is beneficial to the convergence speed and performance improvement of the model. Here, data preprocessing is used to build a high-quality dataset suitable for the ensemble learning model, laying the foundation for subsequent feature selection and model training.

[0062] Step S2, feature selection: extract the fault condition category and fault severity as key features through the feature selection module.

[0063] Preferably, the fault condition category in step S2 includes the fault location, and the fault severity is calculated by using sensor data to obtain various indicators. Here, the various indicators include leakage and pressure change.

[0064] For example, a CVS system pipeline leakage fault involves 57 parameters. The goal of the feature selection module is to extract the most valuable parameters for fault diagnosis from the preprocessed data. Considering the particularity of the CVS system pipeline leakage fault, the feature selection module focuses on extracting parameters that are strongly correlated with the following features:

[0065] Fault condition category: including the location where the fault occurred.

[0066] Fault severity: Indicators such as leakage and pressure change calculated from sensor data reflect the severity of the fault. These features not only capture the temporal dependence of the fault but also its spatial distribution, providing rich feature information for the ensemble learning model.

[0067] Step S3: Training the ensemble learning model: Using the ensemble learning module to build a Boosting and Stacking hybrid framework, and train the ensemble model.

[0068] like Figure 2 As shown, the present invention can improve the diagnostic accuracy and robustness of the system by constructing a hybrid framework of Boosting and Stacking and integrating the advantages of multiple base models.

[0069] Preferably, step S3 includes the following steps:

[0070] Step S 31 , Selection of base models: XGBoost, Random Forest and Attention-LSTM are selected as the representative models of Boosting, traditional machine learning and deep learning respectively.

[0071] To comprehensively cover different modeling paradigms, this paper selects XGBoost (Extreme Gradient Boosting Trees), Random Forest, and Attention-LSTM as representative models for boosting, traditional machine learning, and deep learning, respectively. This design achieves feature complementarity through three mechanisms: feature importance partitioning, decision boundary optimization, and temporal dependency modeling.

[0072] Step S 32 , Design of integration framework: Use the Boosting framework and Stacking framework to form a hybrid framework to train the integration model.

[0073] Preferably, the Boosting framework trains the base model multiple times, uses k-fold cross validation to ensure model independence, and uniformly outputs probability prediction values ​​(Softmax-RF / Sigmoid-LSTM) to eliminate scale differences.

[0074] The Stacking framework is based on the bias-variance trade-off principle, adopts logistic regression as a meta-model, and uses the probability prediction value output of the base model as the input of the meta-model. The meta-model is used to integrate the prediction results of the base model to further optimize the classification decision.

[0075] Traditional k-fold cross-validation partitions data statically, which can lead to data leakage or correlation between different models during training. Therefore, the k-fold cross-validation used in this application partitions data based on timestamps, ensuring that the training sets for different models come from completely different time periods or regions.

[0076] Accordingly, the implementation of the k-fold cross validation described in this application is as follows:

[0077] 1. Sort the data by time and divide it into blocks according to time windows.

[0078] Second, assign different time periods to each base model as training sets and validation sets.

[0079] In addition, traditional k-fold partitioning is data-driven and does not consider model characteristics. This application dynamically adjusts the k-fold partitioning based on the performance of the model to obtain the optimal partitioning ratio, making the partitioning more conducive to model independence.

[0080] Accordingly, the implementation of the k-fold cross validation described in this application is as follows:

[0081] 1. After the initial partitioning, train a base model and evaluate its performance.

[0082] Second, adjust the division of subsequent models based on performance to avoid highly correlated data being assigned to different models.

[0083] The effectiveness of this method and the optimal performance of the integrated model can be verified through the k-fold cross validation and ablation experiments of this application.

[0084] Step S 33 Data partitioning: The data is divided into training, validation, and test sets to ensure the independence of the base models and their generalization capabilities. A soft voting mechanism is also used. The probability distribution (Softmax) output by the base models is fused through soft voting to eliminate scale differences between the outputs of different models and generate an integrated prediction result.

[0085] Step S 34, Hyperparameter optimization: Train the XGBoost classification model, Attention-LSTM classification model and Random Forest classification model as base models, hereinafter referred to as XGBoost model, LSTM model and RFC model, and obtain the optimal hyperparameters of different classification models (as shown in Table 2 below). Set the base model according to the optimal hyperparameters.

[0086] In order to keep the training and test sets randomized, the datasets are randomly permuted before each training, which will help reduce random errors and improve training efficiency.

[0087] Among them, the random forest classification model: After completing the standardization of fault data, Bayesian optimization of parameters is performed to obtain the performance impact of different tree numbers and depths on the RFC model, thereby selecting the optimal hyperparameters.

[0088] XGBoost classification model: Bayesian optimization and early stopping are used to tune parameters after normalization and feature extraction.

[0089] Attention-LSTM classification model: After standardization, the bidirectional LSTM structure is used with a multi-head attention optimizer to obtain optimal hyperparameters. The LSTM is used to capture long-term dependencies in sequence data. At the same time, the attention mechanism is introduced to weight key information in the sequence and predict the type of data fault.

[0090] Table 2: Optimal hyperparameters for different classification models

[0091]

[0092] Step S 35 ,Construction of metadata dataset: In the Stacking framework, the original dataset is divided into training set and test set, and the input data component metadata dataset is used to train the meta-model.

[0093] For example, if the input data contains 19 categories and 57 features, a 422×57-dimensional meta-dataset is constructed, retaining 10% of the important features of the original data. This meta-dataset is used to train a meta-model, which can further optimize prediction results and improve the model's robustness and accuracy.

[0094] Step S 36 , training ensemble learning model: the base models are trained independently on the training set, and the best model combination and parameter settings are selected through cross-validation and hyperparameter optimization. The output of each base model is processed by Softmax to generate a probability prediction value; the first-level prediction results are stacked and fused to generate a comprehensive result.

[0095] Furthermore, the fused feature vectors are combined into a meta-dataset. The meta-models consist of a training meta-model and a test meta-model. The meta-models are trained using the training and test inputs, respectively. This further optimizes the model's predictive capabilities and forms the final prediction value. The final integrated model is evaluated on the test set, and the fault diagnosis results are output.

[0096] Step S4, model evaluation: Evaluate the model performance on the test set through the comparative evaluation module.

[0097] Preferably, the comparison and evaluation module in step S4 uses accuracy, F1 score and recall rate as evaluation indicators to comprehensively evaluate the classification performance of the model.

[0098] The comparative evaluation module aims to comprehensively evaluate the performance of different models and ensure the advantages of the integrated model.

[0099] Comparative evaluation module: First, accuracy, F1 score, and recall are used as evaluation indicators to comprehensively evaluate the classification performance of the model.

[0100] Secondly, the single models (RF, LSTM and XGB), the ablation model (ABL model) and the integrated model are compared to verify the advantages of the integrated model.

[0101] Finally, experimental results are used to verify the improvements of the integrated model in terms of accuracy, robustness, and real-time performance.

[0102] According to the description of the above method process, the experimental results obtained by the nuclear energy system fault diagnosis method based on multi-model fusion integrated learning of the present invention are as follows Figure 3 As shown in the results, the ensemble learning model based on the hybrid framework of Boosting and Stacking proposed in the present invention has an accuracy of 99.99% (standard deviation <0.0001) in the diagnosis of pipeline leakage faults in the CVS system, which is 1.2% higher than the voting method of a single model.

[0103] Table 3: Model single prediction time

[0104] Model RFC Model XGB Model LSTM model ENS Model Single prediction time (s) 1.128386 3.349973 1.902088 0.010999

[0105] In addition, as shown in Table 3 above, the single prediction time is less than 15ms, which meets the needs of real-time monitoring.

[0106] In this application, the base model and meta-model can be replaced by other machine learning algorithm models, as long as the model has classification function. The above embodiments are only examples, and other machine learning algorithm models are also within the scope of protection of this application.

[0107] like Figures 1 to 3As shown, the present invention also provides a nuclear energy system fault diagnosis system based on multi-model fusion integrated learning, which adopts the nuclear energy system fault diagnosis method as described above. The nuclear energy system fault diagnosis system includes: a data preprocessing module for organizing, dividing and normalizing the original data so that the data is suitable for subsequent training and testing of the base model. A feature selection module extracts the fault condition category and fault severity as autocorrelation features to provide key information for diagnosis. An integrated learning module constructs a multi-model fusion integrated learning model, and uses the Boosting and Stacking hybrid framework to integrate the advantages of multiple base models to achieve efficient and accurate fault diagnosis. A comparative evaluation module comprehensively evaluates the performance of different models, for example, using accuracy, F1 score and recall as evaluation indicators.

[0108] The present invention also provides an electronic device comprising: a processor and a memory, wherein the memory stores programs or instructions that can be run on the processor, and the programs or instructions are executed by the processor to implement the nuclear energy system fault diagnosis method based on multi-model fusion integrated learning as described above.

[0109] The present invention also provides a readable storage medium, which stores a program or instruction. When the program or instruction is executed by a processor, the nuclear energy system fault diagnosis method based on multi-model fusion integrated learning as described above is implemented.

[0110] As described above, the nuclear energy system fault diagnosis method based on multi-model fusion ensemble learning of the present invention has the following characteristics:

[0111] First, this paper adopts a two-stage integrated architecture. First, it builds XGBoost, random forest, and Attention-LSTM base models through the bagging framework to achieve feature importance partitioning, decision boundary optimization, and temporal dependency modeling. Second, based on the stacking strategy, it uses a logistic regression meta-model for bias-variance optimization and fuses the base model outputs through a soft voting mechanism.

[0112] This two-stage integrated architecture leverages the diversity of base models, optimizing feature representation, decision boundaries, and temporal dependency modeling, significantly improving the accuracy and robustness of fault diagnosis. The introduction of a soft voting mechanism and meta-model further optimizes classification decisions, ensuring the model's efficiency and accuracy.

[0113] Second, this invention uses a multi-model fusion strategy, combined with the characteristics of the base model, XGBoost, and random forest to achieve feature importance partitioning and decision boundary optimization, enhancing the processing capabilities of high-dimensional data. Attention-LSTM specifically models temporal dependencies and captures the dynamic characteristics of fault data.

[0114] 3. The present invention proposes to use a soft voting mechanism to fuse the probability prediction values ​​of multiple base models, and obtain a fault diagnosis model with stronger real-time performance and higher accuracy after training based on nuclear power system operating data.

[0115] The nuclear energy system fault diagnosis method based on multi-model fusion integrated learning of the present invention has the following advantages:

[0116] First, this paper proposes a fault diagnosis model for nuclear CVS systems based on a hybrid stacking and boosting framework, employing ensemble learning. This hybrid strategy combines the strengths of multiple base models, significantly improving diagnostic accuracy and robustness. This ensemble learning framework overcomes the limitations of single models in complex tasks.

[0117] 2. The present invention deeply stacks and fuses three heterogeneous models, namely XGBoost, random forest and LSTM, for nuclear power CVS fault data modeling, and conducts 100 repeated experiments to verify the stability of the standard deviation <0.0001.

[0118] 3. The present invention uses weighted average ablation experiments to prove that the ensemble learning model improves the accuracy by an average of 1.2% compared with the single model voting, which is significantly better than the existing single model and ablation model.

[0119] Fourth, the model's single prediction time is less than 15ms, meeting the real-time monitoring requirements of nuclear power plants. This high efficiency and real-time performance ensures timely detection and resolution of faults, improving the overall response speed of the system.

[0120] Therefore, the present invention achieves comprehensive utilization of high-dimensional data and effective capture of fault information by integrating the advantages of multiple models.

[0121] In summary, the nuclear energy system fault diagnosis method and system of the present invention, based on multi-model fusion integrated learning, realizes real-time and accurate fault diagnosis by constructing an integrated learning model of the Boosting and Stacking hybrid framework, combining hyperparameter optimization and adaptive mechanism.

[0122] For those skilled in the art, the above invention disclosure is intended only as an example and does not constitute a limitation of this application. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements, and amendments to this application. Such modifications, improvements, and amendments are suggested in this application and remain within the spirit and scope of the exemplary embodiments of this application.

[0123] At the same time, this application uses specific terms to describe the embodiments of this application. For example, "one embodiment," "an embodiment," and / or "some embodiments" refer to a certain feature, structure, or characteristic related to at least one embodiment of this application. Therefore, it should be emphasized and noted that "one embodiment," "an embodiment," or "an alternative embodiment" mentioned twice or multiple times in different locations in this specification does not necessarily refer to the same embodiment. In addition, certain features, structures, or characteristics in one or more embodiments of this application may be appropriately combined.

[0124] Some aspects of the present application can be performed entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The above hardware or software can be referred to as "data blocks", "modules", "engines", "units", "components" or "systems". The processor can be one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DAPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors or combinations thereof. In addition, various aspects of the present application may be expressed as computer products located in one or more computer-readable media, which include computer-readable program code. For example, computer-readable media may include, but are not limited to, magnetic storage devices (e.g., hard disks, floppy disks, tapes...), optical disks (e.g., compact disks CDs, digital versatile disks DVDs...), smart cards, and flash memory devices (e.g., cards, sticks, key drives...).

[0125] A computer-readable medium may include a propagated data signal embodying computer program code, for example, in baseband or as part of a carrier wave. The propagated signal may be in a variety of forms, including electromagnetic, optical, etc., or a suitable combination thereof. A computer-readable medium may be any computer-readable medium other than a computer-readable storage medium that can be connected to an instruction execution system, apparatus, or device to communicate, propagate, or transmit the program for use. The program code on the computer-readable medium may be transmitted via any suitable medium, including radio, cable, fiber optic cable, radio frequency signal, or similar medium, or any combination of the above.

[0126] Similarly, it should be noted that, in order to simplify the presentation of the present disclosure and facilitate understanding of one or more embodiments of the invention, the foregoing description of the embodiments of the present disclosure sometimes combines various features into a single embodiment, figure, or description thereof. However, this disclosure method does not imply that the subject matter of the present disclosure requires more features than those recited in the claims. In fact, embodiments may feature fewer than all the features of a single disclosed embodiment. Some embodiments use numbers to describe the quantity of components or attributes. It should be understood that such numbers used in the embodiment descriptions are, in some instances, modified by the qualifiers "about," "approximately," or "substantially." Unless otherwise indicated, "about," "approximately," or "substantially" indicate that the number can vary by ±20%. Accordingly, in some embodiments, the numerical parameters used in the specification and claims are approximate and may vary depending on the desired characteristics of individual embodiments. In some embodiments, numerical parameters should be considered to the specified number of significant digits and adopt the usual method of retaining significant digits. Although the numerical ranges and parameters used to identify the breadth of the ranges in some embodiments of the present disclosure are approximate, in specific embodiments, such numerical values ​​are set as accurately as practicable.

[0127] Although specific embodiments of the present invention have been described above, those skilled in the art will appreciate that these are merely illustrative and that the scope of the present invention is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, and such changes and modifications are intended to fall within the scope of the present invention.

Claims

1. A nuclear energy system fault diagnosis method based on multi-model fusion ensemble learning, characterized in that: The nuclear energy system fault diagnosis method comprises the following steps: S1. Data processing: The data preprocessing module is used to collect the operating data of the chemical and volume control systems, clean, divide and normalize the raw data, and construct a high-quality data set; S2. Feature selection: The fault condition category and fault severity are extracted as key features through the feature selection module; S3. Ensemble learning model training: Use the ensemble learning module to build a hybrid framework of Boosting and Stacking to train the ensemble model; S4. Model evaluation: Evaluate the model performance on the test set through the comparative evaluation module.

2. The nuclear energy system fault diagnosis method based on multi-model fusion ensemble learning according to claim 1 is characterized in that: The step S1 includes: S 11 , Clean the original fault data of the chemical and volume control systems, remove noise data and missing values, and divide the labels according to numbers to obtain the corresponding relationship between fault numbers and fault names; S 12 , divide the processed data into training set, validation set and test set; S 13 , normalize the numerical data to ensure that all features are on the same scale.

3. The nuclear energy system fault diagnosis method based on multi-model fusion ensemble learning according to claim 2 is characterized in that: The step S 12 The proportions of the training set, the validation set and the test set are 70%, 20% and 10% respectively.

4. The nuclear energy system fault diagnosis method based on multi-model fusion ensemble learning according to claim 1 is characterized in that: In step S2, the fault condition category includes the fault location; the fault severity is obtained by calculating various indicators using sensor data.

5. The nuclear energy system fault diagnosis method based on multi-model fusion ensemble learning according to claim 4 is characterized in that: The various indicators include leakage volume and pressure change.

6. The nuclear energy system fault diagnosis method based on multi-model fusion ensemble learning according to claim 1 is characterized in that: The step S3 comprises the following steps: S 31 ,Selection of base models: XGBoost, Random Forest and Attention-LSTM are selected as the representative models of Boosting, traditional machine learning and deep learning respectively; S 32 ,Design of integrated framework: adopt the hybrid framework of Boosting framework and Stacking framework to train the integrated model; S 33 ,Data partitioning: Divide the data into training set, validation set and test set. The probability distribution of the base model output is fused through soft voting to eliminate the scale differences of different model outputs and generate an integrated prediction result; S 34 , Hyperparameter optimization: Train the XGBoost classification model, Attention-LSTM classification model and Random Forest classification model as base models respectively, obtain the optimal hyperparameters of different classification models, and set the base model according to the optimal hyperparameters; S 35 ,Construction of metadata dataset: In the Stacking framework, the original dataset is divided into a training set and a test set, and the input data is used to construct a metadata dataset for training the meta-model; S 36 , training ensemble learning model: the base model is trained independently on the training set, and the best model combination and parameter setting are selected through cross-validation and hyperparameter optimization. The output of each base model is processed by Softmax to generate a probability prediction value; the first-level prediction result is stacked and fused to generate a comprehensive result.

7. The nuclear energy system fault diagnosis method based on multi-model fusion ensemble learning according to claim 6 is characterized in that: The Boosting framework trains the base model multiple times, uses k-fold cross validation to ensure model independence, and uniformly outputs probability prediction values ​​to eliminate scale differences.

8. The nuclear energy system fault diagnosis method based on multi-model fusion ensemble learning according to claim 6 is characterized in that: The Stacking framework is based on the bias-variance trade-off principle, adopts logistic regression as a meta-model, and uses the probability prediction value output of the base model as the input of the meta-model. The meta-model is used to integrate the prediction results of the base model to further optimize the classification decision.

9. The nuclear energy system fault diagnosis method based on multi-model fusion ensemble learning according to claim 6, characterized in that: In step S4, the comparison and evaluation module uses accuracy, F1 score and recall rate as evaluation indicators to comprehensively evaluate the classification performance of the model.

10. A nuclear energy system fault diagnosis system based on multi-model fusion integrated learning, characterized in that: The nuclear energy system fault diagnosis system adopts the nuclear energy system fault diagnosis method according to any one of claims 1 to 9, and the nuclear energy system fault diagnosis system includes: The data preprocessing module is used to organize, divide and normalize the raw data to make the data suitable for subsequent training and testing of the base model; Feature selection module, extracts fault condition category and fault severity as autocorrelation features, providing key information for diagnosis; The integrated learning module builds a multi-model fusion integrated learning model, using the Boosting and Stacking hybrid framework to integrate multiple base models to achieve fault diagnosis; Comparative evaluation module to comprehensively evaluate the performance of different models.

11. An electronic device, characterized in that: The electronic device includes: a processor and a memory, the memory stores programs or instructions that can be run on the processor, and the programs or instructions are executed by the processor to implement the nuclear energy system fault diagnosis method based on multi-model fusion integrated learning as described in any one of claims 1-9.

12. A readable storage medium, characterized in that: The readable storage medium stores programs or instructions, and when the programs or instructions are executed by the processor, the nuclear energy system fault diagnosis method based on multi-model fusion integrated learning as described in any one of claims 1 to 9 is implemented.

Images