A method and system for concurrent fault diagnosis of a nuclear power plant based on transfer learning

Abstract

The application discloses a kind of nuclear power plant concurrent fault diagnosis method and system based on transfer learning, constructs nuclear power plant concurrent fault diagnosis model based on multilayer bidirectional long short-term memory network;Using single failure training set, the concurrent fault diagnosis model is pre-trained, and then the single failure test set is input into the trained concurrent fault diagnosis model to test the accuracy of the pre-training model;The full connection layer and classification output layer in the concurrent fault diagnosis model are reset to blank layer, and the adjusted network model is retrained using the concurrent fault training set;The concurrent fault test set is input into the trained concurrent fault diagnosis model, to realize the diagnosis of nuclear power plant sensor and actuator concurrent fault.The present application can realize the accurate online diagnosis of different types and different degrees of sensor and actuator concurrent fault during the steady-state and transient operation of nuclear power plant, and provide a reference for fault mitigation and fault-tolerant control.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent fault diagnosis of nuclear power plants, and in particular relates to a concurrent fault diagnosis method and system for nuclear power plants based on transfer learning and a bidirectional long-short term memory (BiLSTM) network. Background Art

[0002] With the rapid development of the global nuclear industry, nuclear power plants are playing an increasingly important role in both civilian and military applications. Nuclear power plants are complex and large in size, and operate in a complex and dynamic environment. To ensure safe and stable operation, numerous sensors and actuators are often deployed to monitor and control them. Because these sensors and actuators often operate in extremely harsh environments, such as high temperature and high pressure, they are prone to failure. Sensor and actuator failures can lead to unplanned shutdowns of nuclear power plants and, in severe cases, even radioactive leaks, posing a serious threat to the safe operation of nuclear power systems. Two or more simultaneous failures are called concurrent failures. Concurrent failures have more complex characteristics, more severe consequences, and are more difficult to diagnose. The numerous sensors and actuators in nuclear power plants increase the likelihood of concurrent failures. Therefore, identifying and diagnosing faults in nuclear power plants, especially concurrent failures, is crucial for the safe and stable operation of the system.

[0003] With the advent of the industrial big data era and the advancement of sensor technology, traditional fault diagnosis methods are no longer able to meet the diagnostic needs of massive amounts of data. Traditional fault diagnosis based on manual experience has very limited capabilities for abnormal operating conditions. To ensure the safety of mechanical equipment under fault conditions, personnel training can be strengthened to improve their ability to handle sudden faults. However, this fault diagnosis method requires highly rational staffing and significant manpower, and cannot guarantee high diagnostic accuracy. As can be seen, the large amount of fault data and limited staffing have gradually exposed the shortcomings of traditional manual fault diagnosis methods. Deep learning algorithms, such as recurrent neural networks and convolutional neural networks, are increasingly being used in fault diagnosis due to their powerful feature extraction and nonlinear mapping capabilities. Compared to traditional fault diagnosis methods based on expert experience, these deep learning algorithms have significantly improved diagnostic accuracy for large amounts of fault data.

[0004] Deep learning-based fault diagnosis methods extract signal features and learn different fault characteristics to fit optimal network weights for rationally classifying signal features, thereby achieving efficient and accurate end-to-end fault diagnosis for complex systems. Concurrent faults often suffer from a shortage of high-quality labeled fault data. Transfer learning methods within deep learning can effectively address the challenge of training with small amounts of data, making deep transfer learning an important approach for concurrent fault diagnosis. This intelligent diagnostic method can address the shortage of high-quality labeled data for concurrent faults in nuclear power systems, improving their safety. It can also handle large amounts of data or multiple concurrent faults quickly and effectively, significantly improving the accuracy of fault pattern recognition. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to address the deficiencies in the above-mentioned prior art and provide a method and system for diagnosing concurrent faults of nuclear power plants based on transfer learning, so as to solve the technical problems of low diagnostic efficiency and poor diagnostic effect of traditional diagnostic methods and the lack of high-quality labeled fault data for concurrent faults.

[0006] The present invention adopts the following technical solutions:

[0007] A concurrent fault diagnosis method for a nuclear power plant based on transfer learning comprises the following steps:

[0008] S1. Divide the response data of the nuclear power plant into a single fault training set and a single fault test set, as well as a concurrent fault training set and a concurrent fault test set;

[0009] S2. Preprocess the single fault training set and single fault test set, as well as the concurrent fault training set and concurrent fault test set obtained in step S1;

[0010] S3. Construct a concurrent fault diagnosis model for nuclear power plants based on a multi-layer bidirectional long short-term memory network;

[0011] S4. Pre-train the nuclear power plant concurrent fault diagnosis model constructed in step S3 using the single fault training set obtained in step S2, then input the single fault test set obtained in step S2 into the pre-trained nuclear power plant concurrent fault diagnosis model to test the diagnostic accuracy of the nuclear power plant concurrent fault diagnosis model for single faults in nuclear power plants;

[0012] S5. Adjust the nuclear power plant concurrent fault diagnosis model pre-trained in step S4, and reset the fully connected layer and output layer of the nuclear power plant concurrent fault diagnosis model to blank layers;

[0013] S6. Use the concurrent fault training set obtained in step S2 to retrain the nuclear power plant concurrent fault diagnosis model adjusted in step S5, and input the concurrent fault test set obtained in step S2 into the retrained nuclear power plant concurrent fault diagnosis model to realize sensor and actuator concurrent fault diagnosis.

[0014] Specifically, in step S1, the ratio of the single fault data set to the concurrent fault data set is 8:2, the ratio of the training set to the test set in the single fault data set is 8:2, and the ratio of the training set to the test set in the concurrent fault data set is 8:2.

[0015] Furthermore, the response data includes nuclear power, primary circuit coolant temperature and main steam pressure data of sensors and actuators at different locations, types and degrees under single and concurrent fault conditions and normal conditions of the nuclear power plant.

[0016] Specifically, in step S3, the nuclear power plant concurrent fault diagnosis model includes:

[0017] The input layer is used to receive standardized multidimensional time series training samples;

[0018] Multi-layer BiLSTM network for extracting features from multi-dimensional time series data;

[0019] The fully connected layer is used to take the spatiotemporal features obtained by the multi-layer BiLSTM network as input values ​​and output a single-dimensional time series vector that can represent the characteristics of the fault type;

[0020] The Softmax layer is used to compress all values ​​of a one-dimensional vector of any size using the Softmax function. After processing, the vector size remains unchanged, the values ​​of all elements are [0, 1], and the sum of all elements is 1;

[0021] Output layer, used to output results.

[0022] Furthermore, the gradient threshold of the multi-layer BiLSTM network is 2, and it contains 3 BiLSTM layers, 3 normalization layers, 3 nonlinear activation functions, and 3 dropout layers; the connection order is: BiLSTM layer, normalization layer, activation function, dropout layer, BiLSTM layer, normalization layer, activation function, dropout layer, BiLSTM layer, normalization layer, activation function, and dropout layer.

[0023] Furthermore, each BiLSTM layer contains 200 hidden units.

[0024] Specifically, in step S4, the learning algorithm of the pre-training process adopts sgdm, the maximum cycle is 2000, the minimum batch size is 126, the learning rate is 0.01, and the drop rate of the three dropout layers is 50%.

[0025] Specifically, in step S5, the parameter learning rate factor of the blank fully connected layer is set to 2.

[0026] Specifically, in step S6, the learning algorithm in the retraining process adopts sgdm, the maximum iteration cycle is 300, the minimum batch size is 102, the learning rate is 0.001, and the drop rate of the three dropout layers is 5%.

[0027] In a second aspect, an embodiment of the present invention provides a concurrent fault diagnosis system for nuclear power plants based on transfer learning, characterized by comprising:

[0028] A data module divides the response data of the nuclear power plant into a single fault training set and a single fault test set, as well as a concurrent fault training set and a concurrent fault test set;

[0029] A preprocessing module preprocesses the single fault training set and single fault test set, as well as the concurrent fault training set and concurrent fault test set obtained by the data module;

[0030] Construct a module to build a concurrent fault diagnosis model for nuclear power plants based on a multi-layer bidirectional long short-term memory network;

[0031] a pre-training module, which uses the single fault training set obtained by the pre-processing module to pre-train the nuclear power plant concurrent fault diagnosis model constructed by the construction module, and then inputs the single fault test set obtained by the pre-processing module into the pre-trained nuclear power plant concurrent fault diagnosis model to test the diagnostic accuracy of the nuclear power plant concurrent fault diagnosis model for the nuclear power plant single fault;

[0032] An adjustment module adjusts the nuclear power plant concurrent fault diagnosis model pre-trained by the pre-training module, and resets the fully connected layer and the output layer of the nuclear power plant concurrent fault diagnosis model to blank layers;

[0033] The diagnosis module uses the concurrent fault training set obtained by the preprocessing module to retrain the nuclear power plant concurrent fault diagnosis model adjusted by the adjustment module, and inputs the concurrent fault test set obtained by the preprocessing module into the retrained nuclear power plant concurrent fault diagnosis model to realize sensor and actuator concurrent fault diagnosis.

[0034] Compared with the prior art, the present invention has at least the following beneficial effects:

[0035] A method for concurrent fault diagnosis of nuclear power plants based on transfer learning is proposed. The acquired sample data is preprocessed through data standardization. After standardization, the influence of data units and orders of magnitude are eliminated, and only its characteristics are retained. This is conducive to improving the speed of the cross-entropy loss function when seeking the optimal solution value during gradient descent, and can also improve the accuracy of the model during training. A nuclear power plant concurrent fault diagnosis model with a multi-layer BiLSTM network is constructed. For the first time, it is proposed to apply a deep learning method based on transfer learning to the concurrent fault diagnosis of nuclear power plants. It can accurately identify the specific fault type and fault degree and has strong generalization ability.

[0036] Furthermore, response data of the main physical and thermal parameters when single faults or concurrent faults occur in sensors and actuators under steady-state conditions and various typical transient conditions are collected. About 80% of the data of single faults and concurrent faults are used as training sets, including response data of the main physical and thermal parameters when faults occur at different times / locations / types / extents. Both the training set and the test set cover various types of faults, ensuring the diversity of the training set and the comprehensiveness of the test set.

[0037] Furthermore, by obtaining experimental or simulation methods, the response data of the main physical and thermal parameters of the nuclear power plant under single / concurrent fault conditions and normal conditions of sensors and actuators at different positions / types / extents can be obtained, and the time domain characteristics of finite signals can be used to identify the time, type and extent of concurrent faults of the nuclear power plant.

[0038] Furthermore, the use of a multi-layer BiLSTM network architecture can effectively extract the spatiotemporal features of the response data of the main physical and thermal parameters when faults occur in nuclear power plant sensors and actuators, greatly improving the accuracy of the diagnosis model. Reasonable setting of the network model structure and heuristic selection of optimal hyperparameters are conducive to improving diagnostic accuracy. The fully connected layer can well match the extracted features to the fault type, thereby achieving efficient and accurate diagnosis of various fault types.

[0039] Furthermore, the multi-layer BiLSTM network module contains multiple BiLSTM network layers. Too many BiLSTM network layers can lead to model overfitting and reduce model generalization ability, while too few layers can lead to model underfitting and reduce model prediction accuracy. Setting an appropriate number of BiLSTM network layers optimizes model generalization and prediction accuracy. The module includes multiple normalization layers, nonlinear activation functions, and dropout layers. Normalization layers accelerate network training and convergence, and can control gradient explosion and prevent gradient vanishing. The nonlinear activation function performs a threshold operation on each input element, setting any value less than zero to zero, which can alleviate overfitting to a certain extent. The dropout layer randomly sets input elements to zero with a given probability, helping to prevent network overfitting.

[0040] Furthermore, the BiLSTM network layer contains multiple hidden neural units. Too many hidden neural units will lead to model overfitting and reduce the model's generalization ability. Too few hidden neural units will lead to model underfitting and reduce the model's prediction accuracy. Setting an appropriate number of hidden neural units can optimize the model's generalization ability and prediction accuracy.

[0041] Furthermore, the learning rate of the pre-training process is set to 0.01, and the maximum number of iterations is set to 2000. By setting a sufficient number of iterations, it is ensured that the loss function of the model during pre-training can reach a convergence state.

[0042] Furthermore, the parameter learning rate factor of the blank fully connected layer is set to 2 to ensure that this layer learns concurrent fault features faster than the remaining network layers. This layer can use the single fault features learned by the remaining network layers to identify concurrent fault features, thereby achieving model migration.

[0043] Furthermore, the learning rate of the retraining process is set to 0.001, and the maximum number of iterations is set to 300. During the retraining process, the concurrent fault characteristics can be learned based on the learned single fault characteristics of nuclear power plant sensors and actuators, thereby ensuring the learning speed, accuracy and generalization ability of the nuclear power plant concurrent fault diagnosis model.

[0044] It can be understood that the beneficial effects of the second to third aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here.

[0045] In summary, the present invention accurately diagnoses concurrent faults of nuclear power plant sensors and actuators by establishing a multi-layer BiLSTM network architecture. When faced with some unknown faults, the predicted result is the existing fault that is closest to its characteristics, thereby achieving the purpose of real-time and effective diagnosis of concurrent faults of nuclear power plant sensors and actuators.

[0046] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 is a logic flow chart of the present invention;

[0048] Figure 2 This is the structural diagram of the BiLSTM network;

[0049] Figure 3 This is the architecture diagram of the diagnostic model based on a three-layer BiLSTM network;

[0050] Figure 4 This is a data diagram of a single concurrent fault signal of a small pressurized water reactor sensor actuator;

[0051] Figure 5 The model training accuracy curve and model training loss curve of the single fault training set and test set during the pre-training process of the diagnosis model;

[0052] Figure 6 This is the confusion matrix result of the diagnostic model on the single fault test set after pre-training;

[0053] Figure 7 The model training accuracy curve and model training loss curve of the training set and test set for the concurrent fault diagnosis model during the retraining process;

[0054] Figure 8 This is the confusion matrix result of the diagnostic model on the concurrent fault test set after retraining;

[0055] Figure 9 Figure 1 shows the results of the diagnosis model for three types of concurrent faults, among which (a) is the real-time diagnosis result of the constant gain fault of the steam pressure sensor and the water supply valve occurring simultaneously at 50s, (b) is the real-time diagnosis result of the constant gain fault of the steam pressure sensor and the water supply valve occurring simultaneously at 120s, (c) is the real-time diagnosis result of the constant deviation fault of the steam pressure sensor and the water supply valve occurring simultaneously at 50s, and (d) is the real-time diagnosis result of the stuck fault of the steam pressure sensor and the water supply valve occurring simultaneously at 120s. DETAILED DESCRIPTION

[0056] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0057] In the description of the present invention, it is to be understood that the terms “include” and “comprise” indicate the presence of the described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0058] It should also be understood that the terms used in the present specification are only for the purpose of describing particular embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0059] It should be further understood that the term "and / or" as used in the present specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items. For example, A and / or B may represent: A alone, A and B simultaneously, or B alone. In addition, the character " / " herein generally indicates that the associated items are in an "or" relationship.

[0060] It should be understood that although the terms "first," "second," and "third" may be used to describe preset ranges in embodiments of the present invention, these preset ranges should not be limited to these terms. These terms are merely used to distinguish one preset range from another. For example, without departing from the scope of embodiments of the present invention, the first preset range may also be referred to as the second preset range, and similarly, the second preset range may also be referred to as the first preset range.

[0061] The word "if," as used herein, may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to the determination" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)," depending on the context.

[0062] The accompanying drawings illustrate various schematic diagrams of structures according to embodiments disclosed herein. These figures are not drawn to scale; for clarity, some details are exaggerated and some details may be omitted. The shapes of the various regions and layers shown in the figures, as well as their relative sizes and positional relationships, are merely exemplary and may deviate in practice due to manufacturing tolerances or technical limitations. Those skilled in the art may design regions / layers with different shapes, sizes, and relative positions as needed.

[0063] The present invention provides a nuclear power plant concurrent fault diagnosis method based on transfer learning. The method collects response data of main physical and thermal parameters such as nuclear power, primary circuit coolant temperature, and main steam pressure of sensors and actuators of different positions, types, and degrees under single / concurrent fault conditions and normal conditions of the nuclear power plant, and divides the data into a single fault training set and a test set (source data set) and a concurrent fault training set and a test set (target data set); performs preprocessing such as standardization on the single / concurrent fault training set and the test set data; constructs a nuclear power plant concurrent fault diagnosis model based on a multi-layer bidirectional long short-term memory network; pretrains the concurrent fault diagnosis model using the single fault training set, then inputs the single fault test set into the trained concurrent fault diagnosis model to test the accuracy of the pretrained model; uses a model-based transfer learning method to reset the fully connected layer and the classification output layer in the concurrent fault diagnosis model to blank layers, and retrains the adjusted network model using the concurrent fault training set; and inputs the concurrent fault test set into the trained concurrent fault diagnosis model to realize the diagnosis of concurrent faults of sensors and actuators of the nuclear power plant. The present invention can realize accurate online diagnosis of concurrent faults of different types and degrees in sensors and actuators during steady-state and transient operation of nuclear power plants, providing a reference for fault mitigation and fault-tolerant control.

[0064] See also Figure 1 The present invention provides a method for diagnosing concurrent faults in nuclear power plants based on transfer learning. The method selects numerous and fault-prone sensors and actuators in nuclear power plants as diagnostic targets, and uses a dataset of single and concurrent faults of sensors and actuators in a small pressurized water reactor nuclear steam supply system under various operating conditions as sample data. The dataset includes faults of different types and severity, such as steam pressure sensors, feedwater valves, and concurrent faults of the steam pressure sensors and feedwater valves. The method samples response data of key physical and thermal parameters, establishes a nuclear power plant concurrent fault diagnosis model based on a multi-layer BiLSTM network, and accurately diagnoses the time, location, type, and severity of concurrent faults of sensors and actuators in nuclear power plants, thereby achieving efficient and accurate diagnosis of concurrent faults in nuclear power plants. The specific steps are as follows:

[0065] S1. Collect response data of key physical and thermal parameters of nuclear power plants, such as nuclear power, primary coolant temperature, and main steam pressure, under single / concurrent fault conditions and normal conditions of sensors and actuators at different locations / types / extents. The data are then divided into single-fault training and test sets (source data sets), as well as concurrent-fault training and test sets (target data sets), which are used for training and testing in the pre-training and re-training processes, respectively.

[0066] Taking steam pressure sensors as an example, fault conditions of different types and degrees include steam pressure sensor failures occurring in various typical transient processes at high / medium / low power levels.

[0067] According to the above fault types, multiple sets of symptom data for each type of fault are collected according to different fault occurrence times, and are divided into single fault data sets and concurrent fault data sets in a ratio of approximately 8:2, including single fault training set and test set as well as concurrent fault training set and test set; on this basis, the established nuclear power plant concurrent fault diagnosis model is first pre-trained using the single fault training set, and then re-trained using the concurrent fault training set, and the concurrent fault test set is used to evaluate the diagnostic effect of the nuclear power plant concurrent fault diagnosis model.

[0068] S2. Preprocessing and standardizing the response data of the physical thermal parameters collected in step S1, wherein the preprocessing is to process the multidimensional time series signal into a format that can be accepted by the training framework used;

[0069] The standardized method is:

[0070]

[0071] Where μ is the mean, δ is the standard deviation,

[0072] After standardization, the range of the training set data is compressed very closely, which can make it converge quickly during training; after standardization, the structure and quantity of the data set remain unchanged.

[0073] S3. Construct a concurrent fault diagnosis model for nuclear power plants based on a multi-layer BiLSTM network;

[0074] A single LSTM unit includes a forget gate f t , an input gate i t and an output gate o t Each LSTM unit uses these three gate mechanisms to process the implicit information h of the previous moment. t-1 , the previous LSTM unit cell state c t-1 and the current input x t Selectively filter and determine the output of a single LSTM unit by the tanh function:

[0075] i t =σ(W i [h t-1 ,x t ]+b i )

[0076] f t =σ(W f [h t-1 ,x t ]+b f )

[0077] c t =f t c t-1 +i t tanh(W c [h t-1 ,x t ]+b c )

[0078] o t =σ(W o [h t-1 ,x t ]+b o )

[0079] h t =o t tanh(c t )

[0080] Among them, σ represents the sigmoid activation function, W i 、W f 、W o and W c Represents the weight matrices of the input gate, forget gate, output gate, and computing unit state respectively; b i 、b f 、b o and b c Represent the bias terms of the input gate, forget gate, output gate and computing unit state respectively.

[0081] See also Figure 2 The BiLSTM network consists of two LSTM recurrent layers that transmit information in opposite directions. The first layer transmits information in chronological order, while the second layer transmits information in reverse chronological order. Compared to LSTM, the BiLSTM has a more complex structure and parameters, and can not only utilize "past" information but also capture "future" information.

[0082] See also Figure 3 The architecture of the nuclear power plant concurrent fault diagnosis model of the multi-layer BiLSTM network includes an input layer, a multi-BiLSTM network layer, a fully connected layer, a Softmax layer and an output layer; the gradient threshold is set to 2 in the multi-layer network model, and the remaining initial weights are randomly generated and continuously optimized during the training process.

[0083] The structure of the nuclear power plant concurrent fault diagnosis model based on the multi-layer BiLSTM network is as follows:

[0084] Input layer: receives multidimensional time series training samples with a 12-dimensional sequence size of 3001 after normalization;

[0085] Multi-BiLSTM network layer: Contains three BiLSTM layers, three normalization layers, three nonlinear activation functions, and three dropout layers. They are connected in the following order: BiLSTM layer, normalization layer, nonlinear activation function, dropout layer, BiLSTM layer, normalization layer, nonlinear activation function, dropout layer, BiLSTM layer, normalization layer, nonlinear activation function, dropout layer, BiLSTM layer, normalization layer, nonlinear activation function, and dropout layer.

[0086] Each BiLSTM layer contains 200 hidden units, which are used to extract the hidden temporal features in the input signal.

[0087] Fully connected layer: takes the spatiotemporal features obtained by the feature merging layer as input, and outputs a single-dimensional vector that can represent the characteristics of the fault type;

[0088] Softmax layer: Use the Softmax function to compress all values ​​of a one-dimensional vector of any size. After processing, the size of the vector remains unchanged, the values ​​of all elements are compressed to between [0, 1], and the sum of all elements is 1.

[0089] S4. Pre-train the nuclear power plant concurrent fault diagnosis model constructed in step S3 using the single fault training set obtained in step S2, then input the single fault test set obtained in step S2 into the pre-trained concurrent fault diagnosis model to test the diagnostic accuracy of the pre-trained model for the nuclear power plant single fault;

[0090] The single fault training set is input into the nuclear power plant concurrent fault diagnosis model based on the multi-layer BiLSTM network for pre-training. Since deep neural networks need to continuously extract features from a large number of samples and optimize network weight parameters to achieve ideal classification results, the learning algorithm of the pre-training process adopts SGDM. During the network training process, the maximum iteration cycle is set to 2000, the learning rate is 0.01, and the dropout rate of the three-layer dropout layer is 50%.

[0091] S5. Adjust the pre-trained network obtained in step S4 and reset the fully connected layer and output layer of the pre-trained model to blank layers;

[0092] Set the learning rate factor of the blank fully connected layer parameters to 2 to ensure that the learning speed of this network layer is faster than that of the other layers.

[0093] S6. Use the concurrent fault training set obtained in step S2 to retrain the nuclear power plant concurrent fault diagnosis model adjusted in step S5, and input the concurrent fault test set obtained in step S2 into the retrained nuclear power plant concurrent fault diagnosis model to realize sensor and actuator concurrent fault diagnosis.

[0094] The concurrent fault training set is input into the nuclear power plant concurrent fault diagnosis model of the multi-layer BiLSTM network for retraining. The learning algorithm of the retraining process adopts SGDM. During the training process, the maximum iteration cycle is set to 300, the learning rate is 0.001, and the dropout rate of the three dropout layers is 5%.

[0095] In another embodiment of the present invention, a nuclear power plant concurrent fault diagnosis system based on transfer learning is provided. The system can be used to implement the above-mentioned nuclear power plant concurrent fault diagnosis method based on transfer learning. Specifically, the nuclear power plant concurrent fault diagnosis system based on transfer learning includes a data module, a preprocessing module, a construction module, a pretraining module, an adjustment module and a diagnosis module.

[0096] The data module divides the response data of the nuclear power plant into a single fault training set and a single fault test set, as well as a concurrent fault training set and a concurrent fault test set;

[0097] A preprocessing module preprocesses the single fault training set and single fault test set, as well as the concurrent fault training set and concurrent fault test set obtained by the data module;

[0098] Construct a module to build a concurrent fault diagnosis model for nuclear power plants based on a multi-layer bidirectional long short-term memory network;

[0099] a pre-training module, which uses the single fault training set obtained by the pre-processing module to pre-train the nuclear power plant concurrent fault diagnosis model constructed by the construction module, and then inputs the single fault test set obtained by the pre-processing module into the pre-trained nuclear power plant concurrent fault diagnosis model to test the diagnostic accuracy of the nuclear power plant concurrent fault diagnosis model for the nuclear power plant single fault;

[0100] An adjustment module adjusts the nuclear power plant concurrent fault diagnosis model pre-trained by the pre-training module, and resets the fully connected layer and the output layer of the nuclear power plant concurrent fault diagnosis model to blank layers;

[0101] The diagnosis module uses the concurrent fault training set obtained by the preprocessing module to retrain the nuclear power plant concurrent fault diagnosis model adjusted by the adjustment module, and inputs the concurrent fault test set obtained by the preprocessing module into the retrained nuclear power plant concurrent fault diagnosis model to realize sensor and actuator concurrent fault diagnosis.

[0102] In another embodiment of the present invention, a terminal device is provided, which includes a processor and a memory, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of the concurrent fault diagnosis method of nuclear power plants based on transfer learning, including:

[0103] The response data of the nuclear power plant is divided into a single fault training set and a single fault test set, as well as a concurrent fault training set and a concurrent fault test set; the single fault training set and the single fault test set, as well as the concurrent fault training set and the concurrent fault test set are preprocessed; a nuclear power plant concurrent fault diagnosis model based on a multi-layer bidirectional long short-term memory network is constructed; the nuclear power plant concurrent fault diagnosis model is pretrained using the single fault training set, and the single fault test set is input into the pretrained nuclear power plant concurrent fault diagnosis model to test the diagnostic accuracy of the nuclear power plant concurrent fault diagnosis model for single faults of the nuclear power plant; the pretrained nuclear power plant concurrent fault diagnosis model is adjusted, and the fully connected layer and the output layer of the nuclear power plant concurrent fault diagnosis model are reset to blank layers; the adjusted nuclear power plant concurrent fault diagnosis model is retrained using the concurrent fault training set, and the concurrent fault test set is input into the retrained nuclear power plant concurrent fault diagnosis model to realize sensor and actuator concurrent fault diagnosis.

[0104] In another embodiment of the present invention, the present invention further provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a terminal device for storing programs and data. It can be understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and, of course, the extended storage medium supported by the terminal device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space. These instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory (Non-Volatile Memory), such as at least one disk memory.

[0105] The processor may load and execute one or more instructions stored in a computer-readable storage medium to implement the corresponding steps of the method for concurrent fault diagnosis of nuclear power plants based on transfer learning in the above embodiment. The processor may load and execute the following steps:

[0106] The response data of the nuclear power plant is divided into a single fault training set and a single fault test set, as well as a concurrent fault training set and a concurrent fault test set; the single fault training set and the single fault test set, as well as the concurrent fault training set and the concurrent fault test set are preprocessed; a nuclear power plant concurrent fault diagnosis model based on a multi-layer bidirectional long short-term memory network is constructed; the nuclear power plant concurrent fault diagnosis model is pretrained using the single fault training set, and the single fault test set is input into the pretrained nuclear power plant concurrent fault diagnosis model to test the diagnostic accuracy of the nuclear power plant concurrent fault diagnosis model for single faults of the nuclear power plant; the pretrained nuclear power plant concurrent fault diagnosis model is adjusted, and the fully connected layer and the output layer of the nuclear power plant concurrent fault diagnosis model are reset to blank layers; the adjusted nuclear power plant concurrent fault diagnosis model is retrained using the concurrent fault training set, and the concurrent fault test set is input into the retrained nuclear power plant concurrent fault diagnosis model to realize sensor and actuator concurrent fault diagnosis.

[0107] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0108] Example

[0109] Relying on a simulation platform for a small pressurized water reactor nuclear steam supply system, we generated datasets for single and concurrent faults of steam pressure sensors and feedwater valves under various simulated operating conditions. This included response data for key physical and thermal parameters, such as nuclear power, primary coolant temperature, and main steam pressure, under steady-state conditions at high, medium, and low power levels, and various typical transient conditions. We also collected response data for key physical and thermal parameters, such as nuclear power, primary coolant temperature, and main steam pressure, under normal operating conditions, under single and concurrent fault conditions of sensors and actuators at different locations, types, and degrees of failure in nuclear power plants. The data was then divided into training and test sets for single faults, as well as training and test sets for concurrent faults.

[0110] See also Figure 4 , which are sample data collected by the present invention on the nuclear steam supply system simulation platform, including steam pressure sensor constant gain fault (fault degree 0.7 times), steam pressure sensor constant deviation fault (fault deviation 0.3MPa), steam pressure sensor stuck fault, feed water valve constant gain fault (fault degree 0.7 times), feed water valve constant deviation fault (fault deviation -8%), feed water valve stuck fault, steam pressure sensor constant gain fault (fault degree 1.3 times), steam pressure sensor constant deviation fault (fault deviation 0.5MPa), ... 0.5MPa), feed water valve constant gain fault (fault degree 0.7 times), feed water valve constant deviation fault (fault deviation 0.5MPa), feed water valve constant gain fault (fault degree 0.7 times), feed water valve constant deviation fault (fault deviation 0.5MPa), feed water valve constant gain fault (fault degree 0.7 times), feed water valve constant deviation fault (fault deviation 0.5MPa), feed water valve constant gain fault (fault degree 0.7 times), feed water valve constant deviation fault (fault deviation 0.5MPa), feed water valve constant gain fault (fault degree 0. The data collected included 14 fault types, including 10 single faults and 3 concurrent faults, as well as data from normal operating conditions. The data included data from various operating conditions and multiple fault times for each fault type, except for the stuck fault. For the stuck fault, data was collected under various transient conditions and multiple fault times. To test the real-time diagnostic effectiveness of the concurrent fault diagnosis model for nuclear power plants, data from various fault types at different fault times were collected as a test set.

[0111] Based on the aforementioned fault types, 3001 data points were collected for each data set, with a signal acquisition frequency of 10 Hz. The ratio of the single-fault dataset to the concurrent-fault dataset was approximately 8:2, and the ratio of the training set to the test set for both the single-fault and concurrent-fault datasets was approximately 8:2. The established nuclear power plant concurrent-fault diagnosis model was pre-trained using the single-fault training set, re-trained using the concurrent-fault training set, and evaluated using the concurrent-fault test set, resulting in more accurate fault diagnosis.

[0112] After standardization, the range of the data is compressed very closely, which can make it converge quickly during network training; after standardization, the structure and quantity of the data set remain unchanged.

[0113] The single fault test set was used to verify the pre-training effect of the nuclear power plant concurrent fault diagnosis model of the multi-layer BiLSTM network. The concurrent fault test set was used to verify the retraining and diagnosis effect of the nuclear power plant concurrent fault diagnosis model of the multi-layer BiLSTM network. The confusion matrix was used to evaluate the diagnosis performance.

[0114] The diagnostic accuracy of the nuclear power plant concurrent fault diagnosis model is evaluated using a confusion matrix. The diagnostic accuracy is equal to the number of correct predictions divided by the sum of the number of correct predictions and the number of misdiagnoses. The rows and columns of the confusion matrix represent the fault types and data predicted by the nuclear power plant single / concurrent fault diagnosis model and the actual fault types, respectively. The numbers in the confusion matrix record the number of correct predictions, the number of misdiagnoses, and the categories of misdiagnoses.

[0115] See also Figure 5 The following curves show the change in single-fault diagnosis accuracy and cross-entropy loss during the pre-training process of the nuclear power plant concurrent fault diagnosis model. The closer the accuracy is to 100% and the closer the loss is to 0, the better the classification effect. As can be seen from the figure, as the number of iterations increases, the model's prediction accuracy on the training and test sets increases, while the model's loss on the training and test sets decreases. The pre-training single-fault diagnosis accuracy on the test set is 98.06%.

[0116] See also Figure 6 The diagonal line shows the amount of correctly classified data, which allows for a visual assessment of the model's diagnostic performance on the test data. As can be seen from the figure, the constructed BiLSTM-based fault diagnosis model can accurately predict the majority of data in the single-fault test dataset and accurately diagnose single-fault sensors and actuators in nuclear power plants.

[0117] See also Figure 7The graph below shows the change in concurrent fault diagnosis accuracy and cross-entropy loss during the retraining process of the nuclear power plant concurrent fault diagnosis model. The closer the accuracy is to 100% and the closer the loss is to 0, the better the classification effect. As can be seen from the figure, as the number of iterations increases, the model's prediction accuracy on the training and test sets increases, while the model's loss on the training and test sets decreases. The retrained concurrent fault diagnosis accuracy on the test set is 92.75%.

[0118] See also Figure 8 , where the diagonal line shows the amount of correctly classified data, providing a visual assessment of the model's diagnostic performance on the test data. As can be seen from the figure, the constructed BiLSTM-based fault diagnosis model can accurately predict the majority of data in the concurrent fault test dataset and accurately diagnose concurrent faults in nuclear power plant sensors and actuators.

[0119] See also Figure 9 The dark curve represents the real-time diagnosis result of the diagnosis model for concurrent faults of nuclear power plant sensors and actuators, and the light curve represents the actual occurrence curve of concurrent faults of nuclear power plant sensors and actuators. It can be seen from the figure that the nuclear power plant concurrent fault diagnosis model can accurately diagnose the fault time and fault type of the three types of concurrent faults.

[0120] In summary, the present invention provides a method and system for diagnosing concurrent faults of nuclear power plants based on transfer learning. Taking the steam pressure sensor and feedwater valve of a small pressurized water reactor as verification examples, the method completes the sampling of training data by relying on the single / concurrent fault data samples of nuclear power plant sensors and actuators generated by its nuclear steam supply system simulation platform, and constructs a fault diagnosis model based on a multi-layer BiLSTM network. Through pre-training, the model can effectively learn and extract the data features of single faults of nuclear power plant sensors and actuators, and through retraining, the model is applied to extract and identify the data features of concurrent faults of nuclear power plant sensors and actuators, thereby realizing accurate prediction of concurrent faults of nuclear power plant sensors and actuators. It verifies that the method of the present invention can achieve the purpose of accurately diagnosing concurrent faults of nuclear power plant sensors and actuators of different time / locations / types / degrees.

[0121] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0122] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0123] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0124] In the embodiments provided by the present invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are merely illustrative. For example, the division of the modules or units is merely a logical functional division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interface, indirect coupling or communication connection of devices or units, and can be electrical, mechanical, or other forms.

[0125] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0126] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0127] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of the above-mentioned various method embodiments. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0128] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0129] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0130] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0131] The above content is only for explaining the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the claims of the present invention.

Claims

1. A concurrent fault diagnosis method for nuclear power plants based on transfer learning, characterized in that: The following steps are involved: S1. Divide the response data of the nuclear power plant into a single fault training set and a single fault test set, as well as a concurrent fault training set and a concurrent fault test set; S2. Preprocess the single fault training set and single fault test set, as well as the concurrent fault training set and concurrent fault test set obtained in step S1; S3. Construct a concurrent fault diagnosis model for nuclear power plants based on a multi-layer bidirectional long short-term memory network; S4. Pre-train the nuclear power plant concurrent fault diagnosis model constructed in step S3 using the single fault training set obtained in step S2, then input the single fault test set obtained in step S2 into the pre-trained nuclear power plant concurrent fault diagnosis model to test the diagnostic accuracy of the nuclear power plant concurrent fault diagnosis model for single faults in nuclear power plants; S5. Adjust the nuclear power plant concurrent fault diagnosis model pre-trained in step S4, and reset the fully connected layer and output layer of the nuclear power plant concurrent fault diagnosis model to blank layers; S6. Use the concurrent fault training set obtained in step S2 to retrain the nuclear power plant concurrent fault diagnosis model adjusted in step S5, and input the concurrent fault test set obtained in step S2 into the retrained nuclear power plant concurrent fault diagnosis model to realize sensor and actuator concurrent fault diagnosis.

2. The method for concurrent fault diagnosis of nuclear power plants based on transfer learning according to claim 1, characterized in that: In step S1, the ratio of the single fault data set to the concurrent fault data set is 8:2, the ratio of the training set to the test set in the single fault data set is 8:2, and the ratio of the training set to the test set in the concurrent fault data set is 8:

2.

3. The method for concurrent fault diagnosis of nuclear power plants based on transfer learning according to claim 2, characterized in that: The response data include nuclear power, primary circuit coolant temperature and main steam pressure data of sensors and actuators at different locations, types and degrees under single and concurrent fault conditions and normal conditions of nuclear power plants.

4. The method for concurrent fault diagnosis of nuclear power plants based on transfer learning according to claim 1, characterized in that: In step S3, the nuclear power plant concurrent fault diagnosis model includes: The input layer is used to receive standardized multidimensional time series training samples; Multi-layer BiLSTM network for extracting features from multi-dimensional time series data; The fully connected layer is used to take the spatiotemporal features obtained by the multi-layer BiLSTM network as input values ​​and output a single-dimensional time series vector that can represent the characteristics of the fault type; The Softmax layer is used to compress all values ​​of a one-dimensional vector of any size using the Softmax function. After processing, the vector size remains unchanged, the values ​​of all elements are [0, 1], and the sum of all elements is 1; Output layer, used to output results.

5. The method for concurrent fault diagnosis of nuclear power plants based on transfer learning according to claim 4, characterized in that: The multi-layer BiLSTM network has a gradient threshold of 2 and consists of three BiLSTM layers, three normalization layers, three nonlinear activation functions, and three dropout layers. The connection order is: BiLSTM layer, normalization layer, activation function, dropout layer, BiLSTM layer, normalization layer, activation function, dropout layer, BiLSTM layer, normalization layer, activation function, and dropout layer.

6. The method for concurrent fault diagnosis of nuclear power plants based on transfer learning according to claim 5, characterized in that: Each BiLSTM layer contains 200 hidden units.

7. The method for concurrent fault diagnosis of nuclear power plants based on transfer learning according to claim 1, characterized in that: In step S4, the learning algorithm of the pre-training process adopts sgdm, the maximum cycle is 2000, the minimum batch size is 126, the learning rate is 0.01, and the drop rate of the three dropout layers is 50%.

8. The method for concurrent fault diagnosis of nuclear power plants based on transfer learning according to claim 1, characterized in that: In step S5, the parameter learning rate factor of the blank fully connected layer is set to 2.

9. The method for concurrent fault diagnosis of nuclear power plants based on transfer learning according to claim 1, characterized in that: In step S6, the learning algorithm used in the retraining process is SGDM, the maximum iteration cycle is 300, the minimum batch size is 102, the learning rate is 0.001, and the drop rate of the three dropout layers is 5%.

10. A concurrent fault diagnosis system for nuclear power plants based on transfer learning, characterized in that: include: A data module divides the response data of the nuclear power plant into a single fault training set and a single fault test set, as well as a concurrent fault training set and a concurrent fault test set; A preprocessing module preprocesses the single fault training set and single fault test set, as well as the concurrent fault training set and concurrent fault test set obtained by the data module; Construct a module to build a concurrent fault diagnosis model for nuclear power plants based on a multi-layer bidirectional long short-term memory network; a pre-training module, which uses the single fault training set obtained by the pre-processing module to pre-train the nuclear power plant concurrent fault diagnosis model constructed by the construction module, and then inputs the single fault test set obtained by the pre-processing module into the pre-trained nuclear power plant concurrent fault diagnosis model to test the diagnostic accuracy of the nuclear power plant concurrent fault diagnosis model for the nuclear power plant single fault; An adjustment module adjusts the nuclear power plant concurrent fault diagnosis model pre-trained by the pre-training module, and resets the fully connected layer and the output layer of the nuclear power plant concurrent fault diagnosis model to blank layers; The diagnosis module uses the concurrent fault training set obtained by the preprocessing module to retrain the nuclear power plant concurrent fault diagnosis model adjusted by the adjustment module, and inputs the concurrent fault test set obtained by the preprocessing module into the retrained nuclear power plant concurrent fault diagnosis model to realize sensor and actuator concurrent fault diagnosis.

Images