Training method, prediction method and system for accident prediction model of nuclear power plant

Abstract

The invention provides a training method, a prediction method and a system of a nuclear power plant accident prediction model. The training method comprises the following steps: acquiring accident condition sample data; extracting fault feature data in the accident condition sample data; acquiring a corresponding feature vector based on the fault feature data, and training an initial model by taking the feature vector as the input of the initial model; optimizing trainable parameters of the initial model by using a grey wolf algorithm to obtain target parameters; and constructing the initial model based on the target parameters to obtain a nuclear power plant accident prediction model. The hyperparameters of the LSTM network are optimized by using the grey wolf algorithm, so that the diagnosis precision and efficiency of the nuclear power plant accident prediction model obtained by training can be effectively improved.

Description

Technical Field

[0001] The present disclosure relates to the technical field of nuclear power plant data analysis, and in particular to a training method, a prediction method and a system for a nuclear power plant accident prediction model. Background Art

[0002] The safe operation of nuclear power plants is of vital importance, especially when serious accidents occur. Accurate prediction and analysis can greatly help prevent the escalation of accidents and mitigate the consequences. At present, the methods for predicting nuclear power plant failures mainly include knowledge-based expert systems, data-driven methods, and model-based methods. Among them, data-driven methods have received widespread attention due to their advantages in dealing with complex nonlinear problems.

[0003] As a deep learning model, the Long Short-Term Memory (LSTM) network can effectively capture the long-term dependencies in time series data and has broad application prospects in the field of fault prediction. However, the selection of LSTM network parameters has an important impact on its diagnostic performance. Traditional LSTM network parameter optimization methods usually rely on manual parameter adjustment, which is inefficient and difficult to obtain the global optimal solution, and it is impossible to obtain a reliable nuclear power plant accident prediction model. Summary of the invention

[0004] The technical problem to be solved by the present invention is to overcome the defect that it is difficult to obtain a reliable nuclear power plant accident prediction model through the traditional LSTM network in the prior art, and to provide a training method, prediction method, system, equipment, medium and program product for a nuclear power plant accident prediction model.

[0005] The present invention solves the above technical problems through the following technical solutions:

[0006] In a first aspect, a training method for a nuclear power plant accident prediction model is provided, the training method comprising:

[0007] Obtain accident condition sample data;

[0008] Extracting fault characteristic data from the accident condition sample data;

[0009] Acquire a corresponding feature vector based on the fault feature data, and train the initial model using the feature vector as an input of the initial model;

[0010] Optimizing the trainable parameters of the initial model using the Grey Wolf Optimization (GWO) algorithm to obtain target parameters;

[0011] The initial model is constructed based on the target parameters to obtain a nuclear power plant accident prediction model.

[0012] Preferably, the step of extracting fault feature data from the accident condition sample data comprises:

[0013] Converting the accident condition sample data into first sample matrix data, and arranging the sample matrix data in a time series to obtain second sample matrix data;

[0014] A sliding window is set, and the fault feature data is extracted from the second sample matrix data using the sliding window.

[0015] Preferably, the trainable parameters are weights and bias vectors of the initial model, and the step of optimizing the trainable parameters of the initial model using the grey wolf algorithm to obtain target parameters comprises:

[0016] Obtaining a first initial position of each wolf in the gray wolf algorithm, wherein the initial position represents an initial weight and a bias vector;

[0017] Iterate the first initial position of each wolf using the tent map to obtain a new second initial position;

[0018] The second initial position obtains a final initial position based on a normal cloud model;

[0019] Obtaining the fitness of the final initial position and the initial model;

[0020] Iterating the position of each wolf based on the fitness to obtain a final position of each wolf;

[0021] The target final position having the highest fitness with the initial model is selected, and the target parameters are obtained based on the target final position.

[0022] Preferably, the calculation formula for iterating the first initial position of each wolf using Tent mapping to obtain a new second initial position is:

[0023]

[0024] Among them, X n is the first initial position, X n+1 is the second initial position;

[0025] The calculation formula for the second initial position to obtain the final initial position based on the normal cloud model is:

[0026] X i =X n+1 +He i ×N(0,1)

[0027] Among them, X iis the final initial position of the i-th wolf, He i is the super entropy, and N(0,1) is a standard normal distribution random number.

[0028] In a second aspect, a method for predicting a nuclear power plant accident is provided, the method comprising:

[0029] Obtaining operating data of the nuclear power plant;

[0030] Inputting the operation data into a nuclear power plant accident prediction model, the nuclear power plant accident prediction model outputting a fault prediction result;

[0031] Wherein, the nuclear power plant accident prediction model is obtained based on the training method of the nuclear power plant accident prediction model described in the first aspect;

[0032] The fault type and fault degree of the nuclear power plant are obtained based on the fault prediction result.

[0033] In a third aspect, a training system for a nuclear power plant accident prediction model is provided, the training system comprising a sample data acquisition module, a fault feature extraction module, an initial training module, a gray wolf optimization module and a model output module;

[0034] The sample data acquisition module is used to acquire accident condition sample data;

[0035] The fault feature extraction module is used to extract fault feature data from the accident condition sample data;

[0036] The initial training module is used to obtain a corresponding feature vector based on the fault feature data, and train the initial model using the feature vector as an input of the initial model;

[0037] The gray wolf optimization module is used to optimize the trainable parameters of the initial model using the gray wolf algorithm to obtain target parameters;

[0038] The model output module is used to construct the initial model based on the target parameters to obtain a nuclear power plant accident prediction model.

[0039] In a fourth aspect, a prediction system for nuclear power plant accidents is provided, the prediction system comprising an operation data acquisition module, an accident prediction module and a fault result output module;

[0040] The operation data acquisition module is used to acquire the operation data of the nuclear power plant;

[0041] The accident prediction module is used to input the operation data into a nuclear power plant accident prediction model, and the nuclear power plant accident prediction model outputs a fault prediction result;

[0042] Wherein, the nuclear power plant accident prediction model is obtained based on the training system of the nuclear power plant accident prediction model described in the third aspect;

[0043] The fault result output module is used to obtain the fault type and fault degree of the nuclear power plant based on the fault prediction result.

[0044] In a fifth aspect, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and used to run on the processor, wherein when the processor executes the computer program, the method described in the first aspect and the second aspect is implemented.

[0045] According to a sixth aspect, a computer storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method described in the first and second aspects is implemented.

[0046] In a seventh aspect, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the method described in the first aspect and the second aspect.

[0047] On the basis of being in accordance with the common sense in the art, the above-mentioned preferred conditions can be arbitrarily combined to obtain the preferred embodiments of the present disclosure.

[0048] The positive progress of the present disclosure is that by optimizing the hyperparameters of the LSTM network using the gray wolf algorithm, the diagnostic accuracy and efficiency of the trained nuclear power plant accident prediction model can be effectively improved. The nuclear power plant accident prediction model trained based on this method has a high accuracy, recall rate and F1 value in the prediction of nuclear power plant accidents, which are better than traditional LSTM networks and other optimization methods. It has good robustness and generalization ability, can adapt to different types of fault data, and realizes accurate diagnosis of fault prediction of severe accidents in nuclear power plants. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A flowchart of a method for training a nuclear power plant accident prediction model provided by an exemplary embodiment of the present disclosure;

[0050] Figure 2 A flowchart of a method for predicting a nuclear power plant accident provided by an exemplary embodiment of the present disclosure;

[0051] Figure 3 A schematic diagram of modules of a training system for a nuclear power plant accident prediction model provided by an exemplary embodiment of the present disclosure;

[0052] Figure 4 A schematic diagram of modules of a prediction system for nuclear power plant accidents provided by an exemplary embodiment of the present disclosure;

[0053] Figure 5A schematic diagram of the hardware structure of an electronic device provided by an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0054] The present disclosure is further described below by way of examples, but the present disclosure is not limited to the scope of the examples.

[0055] Prefixes such as "first" and "second" are used in the embodiments of the present disclosure only to distinguish different description objects, and have no limiting effect on the position, order, priority, quantity or content of the described objects. The use of prefixes such as ordinal numbers to distinguish description objects in the embodiments of the present disclosure does not constitute a limitation on the described objects. For the statement of the described objects, please refer to the description in the context of the claims or embodiments, and no unnecessary limitation should be constituted due to the use of such prefixes. In addition, in the description of the present embodiment, unless otherwise specified, the meaning of "plurality" is two or more.

[0056] Example 1

[0057] This embodiment provides a training method for a nuclear power plant accident prediction model, such as Figure 1 As shown, the training method comprises:

[0058] S101, obtaining accident condition sample data;

[0059] S102, extracting fault feature data from the accident condition sample data;

[0060] S103, acquiring a corresponding feature vector based on the fault feature data, and training the initial model using the feature vector as an input of the initial model;

[0061] S104, optimizing the trainable parameters of the initial model using the gray wolf algorithm to obtain target parameters;

[0062] S105. Constructing the initial model based on the target parameters to obtain a nuclear power plant accident prediction model.

[0063] In step S101, the operation data of severe accident conditions are collected, and Pearson analysis is performed on the operation data. For the operation data with an absolute value of the correlation coefficient above 0.7, it indicates that there is a very strong linear relationship between the two variables, that is, the data is considered to be strongly correlated. The data can be reduced to reduce the amount of data required for subsequent analysis.

[0064] Among them, the key parameters used to train the prediction model include reactor coolant system pressure, steam generator pressure, automatic pressure relief system level 4 valve flow, hot section rupture flow, cold section rupture flow, reactor pressure vessel water level, reactor coolant system water level, core temperature, core outlet temperature, containment pressure, containment gas temperature, hydrogen production in the pressure vessel, and other parameters screened after correlation analysis.

[0065] In step S102, the feature extraction method uses frequency domain feature extraction. Frequency domain feature extraction uses methods such as Fourier transform to convert time series data into frequency domain and extract spectrum features such as main frequency and frequency energy, so as to effectively describe the dynamic changes and periodicity of data and enhance the fault prediction capability of the model.

[0066] In step S103, the LSTM network model is used as the initial model. The feature vector contains the key information in the original data, which helps the LSTM network to more accurately capture the long-term dependencies and patterns in the sequence data. By integrating the feature data into feature vectors, the structure of the model can be simplified, the risk of overfitting can be reduced, and the interpretability and generalization ability of the model can be improved.

[0067] Among them, the structural parameters of the LSTM network are defined. The LSTM network consists of an input layer, multiple hidden layers, and an output layer. The input layer receives the data after preprocessing and feature extraction, the hidden layer uses LSTM units for calculation, and the output layer gives the diagnosis results. The structural parameters that need to be determined include the number of hidden layer units, the number of network layers, and the activation function. The number of hidden layer units determines the number of neurons in each hidden layer, the number of layers determines the depth of the network, and the activation function determines the nonlinear transformation between layers.

[0068] Secondly, initialize the weight parameters of the LSTM network. Initialization methods include random initialization and distribution-based initialization. The initialization of weight parameters has an important impact on the training process and convergence speed of the model. Reasonable initialization can avoid the problem of gradient disappearance or explosion and improve training efficiency.

[0069] Finally, select the loss function and optimization algorithm. The loss function is used to measure the prediction error of the model. Common loss functions include mean squared error (MSE) and cross entropy loss. The optimization algorithm is used to update the network weights. Common optimization algorithms include gradient descent (SGD), Adam, etc. By reasonably selecting the loss function and optimization algorithm, the training effect and diagnostic accuracy of the model can be effectively improved.

[0070] In step S104, parameter optimization is a key step to improve the performance of the LSTM network model. The hyperparameters of the long short-term memory network, such as learning rate, batch size, number of hidden units, etc., have a significant impact on the training effect of the model. Initialize the gray wolf population, set the population size and the maximum number of iterations. Each gray wolf represents a hyperparameter combination, and its position represents the specific hyperparameter value.

[0071] Secondly, in each iteration, the positions of each wolf are updated. The leader wolf guides the search direction, the follower wolf assists in the guidance, and the explorer wolf searches around. The quality of the current solution is evaluated according to the loss function value of the model, and the position of the gray wolf is adjusted.

[0072] Finally, the algorithm continues to iterate until the stopping condition is reached, such as the maximum number of iterations or convergence criteria, and finally determines the optimal hyperparameter combination.

[0073] By using the Gray Wolf Algorithm to optimize the trainable parameters of the initial model, more accurate target parameters can be obtained, thereby improving the performance of the model. It helps to find the optimal parameter combination so that the model performs better in prediction or classification tasks; and the Gray Wolf Algorithm can automatically search for the optimal parameters, reducing the need for human intervention and saving time and energy.

[0074] In step S105, the optimal parameters obtained by using the Grey Wolf algorithm are adjusted by the back propagation algorithm to significantly improve the prediction performance of the model. The model can more accurately predict the occurrence of nuclear power plant accidents and provide strong support for prevention and response.

[0075] In this scheme, by using the Grey Wolf algorithm to optimize the hyperparameters of the LSTM network, the diagnostic accuracy and efficiency of the trained nuclear power plant accident prediction model can be effectively improved.

[0076] As an implementable manner, step S102 includes:

[0077] Converting the accident condition sample data into first sample matrix data, and arranging the sample matrix data in a time series to obtain second sample matrix data;

[0078] A sliding window is set, and the fault feature data is extracted from the second sample matrix data using the sliding window.

[0079] In this solution, the accident condition sample data is organized into first sample matrix data in matrix form; the first sample matrix data is output to an EXCEL table; wherein the first column of the table is a time series, and the operating parameters of other columns correspond one-to-one to the time series, to obtain second sample matrix data.

[0080] Use methods such as sliding windows to capture dynamic changes in time series data. Specifically, feature extraction includes the following steps:

[0081] First, set the sliding window size and sliding step size. The sliding window is used to divide the continuous time series data into multiple overlapping or non-overlapping segments, each of which contains a fixed length of time series data. The sliding step size determines the distance the window moves. The smaller the step size, the more features are extracted, but the more computation is required.

[0082] Secondly, feature extraction is performed on the data in each sliding window. The feature extraction method uses frequency domain feature extraction. Frequency domain feature extraction uses methods such as Fourier transform to convert time series data into frequency domain and extract spectrum features such as main frequency and frequency energy as fault feature data. These features can effectively describe the dynamic changes and periodicity of data and enhance the fault prediction ability of the model.

[0083] As an implementable manner, the trainable parameters are weights and bias vectors of the initial model, and the step of optimizing the trainable parameters of the initial model using the gray wolf algorithm to obtain target parameters includes:

[0084] Obtaining a first initial position of each wolf in the gray wolf algorithm, wherein the initial position represents an initial weight and an initial bias vector;

[0085] Iterate the first initial position of each wolf using Tent mapping to obtain a new second initial position;

[0086] The second initial position obtains a final initial position based on a normal cloud model;

[0087] Performing a preset number of iterations based on the final initial position to obtain the final position of each wolf;

[0088] The final position of each wolf is decoded to obtain the target parameters of the initial model.

[0089] In this scheme, the wolf pack is first initialized by setting parameters: setting the number of gray wolves (N), the maximum number of iterations (T), and the structure of LSTM, such as the number of layers and the number of neurons in each layer.

[0090] Generate a random initial population X0 of size N, where N is the population size, and the position of each individual is randomly distributed in the search space. Then use the Tent mapping formula to iterate each initial position X0 to generate a new second initial position X n+1 .

[0091] As an achievable method, the calculation formula for iterating the first initial position of each wolf using Tent mapping to obtain a new second initial position is:

[0092]

[0093] Among them, X n is the first initial position, X n+1 is the second initial position;

[0094] The normal cloud model is introduced to further increase the diversity of the population. For each position generated by tent mapping, a standard normal distribution random number N(0,1) is generated. The final initial position X of each wolf is calculated using the normal cloud model formula. i .

[0095] As an achievable method, the calculation formula for obtaining the final initial position based on the normal cloud model from the second initial position is:

[0096] X i =X n+1 +He i ×N(0,1)

[0097] Among them, X i is the final initial position of the i-th wolf, He i is the super entropy, and N(0,1) is a standard normal distribution random number.

[0098] Then set the control parameters to linearly decrease the number of iterations from 2 to 0;

[0099] The final initial position of each gray wolf is X i Decoded into the weights and biases of the LSTM model;

[0100] Use the feature vector to input the LSTM model for training and adjust the model parameters through the back-propagation algorithm;

[0101] Calculate the loss function of the LSTM model, such as the mean square error (MSE), and use the mean square error to represent the fitness of each wolf position in the gray wolf algorithm;

[0102] The position of each wolf is updated based on the fitness, and the wolf α with the best fitness, the wolf β with the second best fitness, and the wolf δ with the third best fitness are determined.

[0103] Calculate the distance between the gray wolf and the three wolves α, β, and δ, such as:

[0104] Distance D: represents the distance between the current position of the gray wolf and the positions of α, β, and δ wolves. C is a coefficient vector representing the search range.

[0105] D α =|C1X α -X|;

[0106] D β =|C2Xβ -X|;

[0107] D α =|C3X δ -X|;

[0108] Among them, X α , X β , X δ Respectively represent the positions of α, β, δ wolves, C1, C2, C3 represent random function vectors, and X represents the position of the gray wolf.

[0109] Update the position of the gray wolf according to the positions of the three wolves α, β, and δ, such as:

[0110] X1=X α -A1D α ;

[0111] X2=X β -A2D β ;

[0112] X3=X δ -A3D δ ;

[0113] Among them, X1 represents the new position of the α wolf, X2 represents the new position of the β wolf, X3 represents the new position of the δ wolf, and A1, A2 and A3 are coefficient vectors.

[0114] Combine the three updated positions to get the latest position of the gray wolf.

[0115]

[0116] Then, the coefficient vectors A and C are calculated as follows:

[0117] A = 2ar1-a;

[0118] C = 2cr2-c;

[0119] Where r1, r2 are random vectors in [0,1].

[0120] Recalculate the fitness, retrain the LSTM model, decode the new gray wolf position into the weight and bias vector corresponding to the LSTM, retrain the LSTM model and calculate the fitness.

[0121] And based on the newly calculated fitness, the positions of α, β, and δ wolves are re-determined.

[0122] Repeat the iteration of the positions of the α, β, δ wolves, calculate the fitness, and determine the position of the gray wolf until the maximum number of iterations T or the fitness value converges.

[0123] Finally, the optimal LSTM target weight and target bias vector are obtained as target parameters, and the nuclear power plant accident prediction model is constructed with the target parameters.

[0124] The training method of the nuclear power plant accident prediction model provided in this embodiment optimizes the trainable parameters of the initial model by using the gray wolf algorithm to obtain more accurate target parameters, thereby improving the performance of the model. It helps to find the optimal parameter combination so that the model performs better in prediction or classification tasks; and the gray wolf algorithm can automatically search for the optimal parameters, reducing the need for human intervention and saving time and energy.

[0125] Example 2

[0126] This embodiment provides a method for predicting nuclear power plant accidents. Figure 2 As shown, the prediction method includes:

[0127] S201, obtaining the operation data of the nuclear power plant;

[0128] S202, inputting the operation data into a nuclear power plant accident prediction model, and the nuclear power plant accident prediction model outputs a fault prediction result;

[0129] Wherein, the nuclear power plant accident prediction model is obtained based on the training method of the nuclear power plant accident prediction model described in Example 1;

[0130] S203: Obtain the fault type and fault degree of the nuclear power plant based on the fault prediction result.

[0131] In this scheme, the fault type and fault degree are identified based on the fault prediction results. For example, the output fault type is the rupture of the direct injection pipeline of the passive core cooling system compartment reactor pressure vessel, and the passive residual heat removal system fails, and the output mitigation measures are the automatic pressure relief system, core water tank, and injection tank. The prediction model can quickly identify the fault type when the nuclear power plant simulator is inserted into a severe accident, and timely provide mitigation measures to the operator, which can effectively improve the operator's skill level.

[0132] The nuclear power plant accident prediction method provided in this embodiment is based on a nuclear power plant accident prediction model. In the prediction of nuclear power plant accidents, it has high accuracy, recall rate and F1 value, which are superior to traditional LSTM networks and other optimization methods. It has good robustness and generalization ability, can adapt to different types of fault data, and realize accurate diagnosis of fault prediction of serious accidents in nuclear power plants.

[0133] Example 3

[0134] This embodiment provides a training system 100 for a nuclear power plant accident prediction model. Figure 3As shown, the training system includes a sample data acquisition module 101, a fault feature extraction module 102, an initial training module 103, a gray wolf optimization module 104 and a model output module 105;

[0135] The sample data acquisition module 101 is used to acquire accident condition sample data;

[0136] The fault feature extraction module 102 is used to extract fault feature data from the accident condition sample data;

[0137] The initial training module 103 is used to obtain a corresponding feature vector based on the fault feature data, and train the initial model using the feature vector as an input of the initial model;

[0138] The gray wolf optimization module 104 is used to optimize the trainable parameters of the initial model using the gray wolf algorithm to obtain target parameters;

[0139] The model output module 105 is used to construct the initial model based on the target parameters to obtain a nuclear power plant accident prediction model.

[0140] As an implementable manner, the fault feature extraction module 102 includes a matrix conversion unit and a sliding window unit;

[0141] The matrix conversion unit is used to convert the accident condition sample data into first sample matrix data, and arrange the sample matrix data in a time series to obtain second sample matrix data;

[0142] The sliding window unit is used to set a sliding window, and to extract the fault feature data from the second sample matrix data using the sliding window.

[0143] As an implementable manner, the trainable parameters are the weights and bias vectors of the initial model, and the gray wolf optimization module 104 includes an initialization unit, a mapping unit, a normal cloud unit, a fitness calculation unit, an iteration unit, and a selection unit;

[0144] The initialization unit is used to obtain a first initial position of each wolf in the gray wolf algorithm, wherein the initial position represents an initial weight and an initial bias vector;

[0145] The mapping unit is used to iterate the first initial position of each wolf using Tent mapping to obtain a new second initial position;

[0146] The normal cloud unit is used for obtaining the final initial position of the second initial position based on the normal cloud model;

[0147] The fitness calculation unit is used to obtain the fitness of the final initial position and the initial model;

[0148] The iteration unit is used to iterate the position of each wolf based on the fitness to obtain the final position of each wolf;

[0149] The selection unit is used to select the target final position with the highest fitness with the initial model, and obtain the target parameter based on the target final position.

[0150] As an achievable method, the calculation formula for iterating the first initial position of each wolf using Tent mapping to obtain a new second initial position is:

[0151]

[0152] Among them, X n is the first initial position, X n+1 is the second initial position;

[0153] The calculation formula for the second initial position to obtain the final initial position based on the normal cloud model is:

[0154] X i =X n+1 +He i ×N(0,1)

[0155] Among them, X i is the final initial position of the ith one, EX i is the second initial position, He i is the super entropy, and N(0,1) is a standard normal distribution random number.

[0156] As for the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The system embodiment described above is only illustrative, wherein the units described as separate components may or may not be physically separated, and the components as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the disclosed solution.

[0157] The training system of the nuclear power plant accident prediction model provided in this embodiment optimizes the trainable parameters of the initial model by using the gray wolf algorithm to obtain more accurate target parameters, thereby improving the performance of the model. It helps to find the optimal parameter combination so that the model performs better in prediction or classification tasks; and the gray wolf algorithm can automatically search for the optimal parameters, reducing the need for human intervention and saving time and energy.

[0158] Example 4

[0159] This embodiment provides a prediction system 200 for nuclear power plant accidents, such as Figure 4 As shown, the prediction system includes an operation data acquisition module 201, an accident prediction module 202 and a fault result output module 203;

[0160] The operation data acquisition module 201 is used to acquire the operation data of the nuclear power plant;

[0161] The accident prediction module 202 is used to input the operation data into a nuclear power plant accident prediction model, and the nuclear power plant accident prediction model outputs a fault prediction result;

[0162] Wherein, the nuclear power plant accident prediction model is obtained based on the training system of the nuclear power plant accident prediction model described in Example 3;

[0163] The fault result output module 203 is used to obtain the fault type and fault degree of the nuclear power plant based on the fault prediction result.

[0164] The nuclear power plant accident prediction system provided in this embodiment is based on a nuclear power plant accident prediction model. In the prediction of nuclear power plant accidents, it has high accuracy, recall rate and F1 value, which are superior to traditional LSTM networks and other optimization methods. It has good robustness and generalization ability, can adapt to different types of fault data, and realize accurate diagnosis of fault prediction of serious accidents in nuclear power plants.

[0165] Example 5

[0166] Figure 5 This is a schematic diagram of the structure of an electronic device shown in an example embodiment of the present disclosure, the electronic device includes a memory, a processor, and a computer program stored in the memory and used to run on the processor, and when the processor executes the computer program, it implements the training method of the nuclear power plant accident prediction model or the prediction method of nuclear power plant accidents described in any of the above embodiments. Figure 5 The electronic device 90 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0167] like Figure 5 As shown, the electronic device 90 may be in the form of a general-purpose computing device, for example, it may be a server device. The components of the electronic device 90 may include, but are not limited to: at least one processor 91, at least one memory 92, and a bus 93 connecting different system components (including the memory 92 and the processor 91).

[0168] The bus 93 includes a data bus, an address bus, and a control bus.

[0169] The memory 92 may include a volatile memory, such as a random access memory (RAM) 921 and / or a cache memory 922 , and may further include a read-only memory (ROM) 923 .

[0170] The memory 92 may also include a program tool 925 (or utility) having a set (at least one) of program modules 924, such program modules 924 including but not limited to: an operating system, one or more application programs, other program modules and program data, each of which or some combination may include an implementation of a network environment.

[0171] The processor 91 executes various functional applications and data processing by running the computer program stored in the memory 92, such as the training method of the nuclear power plant accident prediction model or the prediction method of the nuclear power plant accident provided in any of the above embodiments.

[0172] The electronic device 90 may also communicate with one or more external devices 94 (e.g., keyboards, pointing devices, etc.). Such communication may be performed via an input / output (I / O) interface 95. Furthermore, the electronic device 90 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 96. As shown, the network adapter 96 communicates with other modules of the electronic device 90 via a bus 93. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 90, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems, etc.

[0173] It should be noted that although several units / modules or sub-units / modules of the electronic device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided into multiple units / modules to be embodied.

[0174] Example 6

[0175] The embodiments of the present disclosure also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the training method of a nuclear power plant accident prediction model or the prediction method of a nuclear power plant accident provided in any of the above embodiments.

[0176] The readable storage medium may include but is not limited to: a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical storage device, a magnetic storage device or any suitable combination of the above.

[0177] Example 5

[0178] The embodiments of the present disclosure also provide a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned methods for training a nuclear power plant accident prediction model or a method for predicting a nuclear power plant accident.

[0179] Among them, the program code for executing the computer program product of the present disclosure can be written in any combination of one or more programming languages, and the program code can be executed completely on the user device, partially on the user device, as an independent software package, partially on the user device and partially on a remote device, or completely on the remote device.

[0180] Although the specific embodiments of the present disclosure are described above, those skilled in the art should understand that this is only an example, and the protection scope of the present disclosure 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 disclosure, but these changes and modifications all fall within the protection scope of the present disclosure.

Claims

1. A training method for a nuclear power plant accident prediction model, characterized in that: The training method comprises: Obtain accident condition sample data; Extracting fault characteristic data from the accident condition sample data; Acquire a corresponding feature vector based on the fault feature data, and train the initial model using the feature vector as an input of the initial model; Optimizing the trainable parameters of the initial model using the Grey Wolf Algorithm to obtain target parameters; The initial model is constructed based on the target parameters to obtain a nuclear power plant accident prediction model.

2. The training method of the nuclear power plant accident prediction model according to claim 1, characterized in that: The step of extracting fault feature data from the accident condition sample data comprises: Converting the accident condition sample data into first sample matrix data, and arranging the sample matrix data in a time series to obtain second sample matrix data; A sliding window is set, and the fault feature data is extracted from the second sample matrix data using the sliding window.

3. The training method of the nuclear power plant accident prediction model according to claim 1, characterized in that: The trainable parameters are weights and bias vectors of the initial model, and the step of optimizing the trainable parameters of the initial model using the gray wolf algorithm to obtain target parameters includes: Obtaining a first initial position of each wolf in the gray wolf algorithm, wherein the initial position represents an initial weight and an initial bias vector; Iterate the first initial position of each wolf using Tent mapping to obtain a new second initial position; The second initial position obtains a final initial position based on a normal cloud model; Obtaining the fitness of the final initial position and the initial model; Iterating the position of each wolf based on the fitness to obtain a final position of each wolf; The target final position having the highest fitness with the initial model is selected, and the target parameters are obtained based on the target final position.

4. The training method of the nuclear power plant accident prediction model according to claim 3, characterized in that: The calculation formula for iterating the first initial position of each wolf using Tent mapping to obtain a new second initial position is: Among them, X n is the first initial position, X n+1 is the second initial position; The calculation formula for the second initial position to obtain the final initial position based on the normal cloud model is: X i =X n+1 +He i ×N(0,1) Among them, X i is the final initial position of the i-th wolf, He i is the super entropy, and N(0,1) is a standard normal distribution random number.

5. A method for predicting nuclear power plant accidents, characterized in that: The prediction method comprises: Obtaining operating data of the nuclear power plant; Inputting the operation data into a nuclear power plant accident prediction model, the nuclear power plant accident prediction model outputting a fault prediction result; Wherein, the nuclear power plant accident prediction model is obtained based on the training method of the nuclear power plant accident prediction model according to any one of claims 1 to 4; The fault type and fault degree of the nuclear power plant are obtained based on the fault prediction result.

6. A training system for a nuclear power plant accident prediction model, characterized in that: The training system includes a sample data acquisition module, a fault feature extraction module, an initial training module, a gray wolf optimization module and a model output module; The sample data acquisition module is used to acquire accident condition sample data; The fault feature extraction module is used to extract fault feature data from the accident condition sample data; The initial training module is used to obtain a corresponding feature vector based on the fault feature data, and train the initial model using the feature vector as an input of the initial model; The gray wolf optimization module is used to optimize the trainable parameters of the initial model using the gray wolf algorithm to obtain target parameters; The model output module is used to construct the initial model based on the target parameters to obtain a nuclear power plant accident prediction model.

7. A prediction system for nuclear power plant accidents, characterized in that: The prediction system includes an operation data acquisition module, an accident prediction module and a fault result output module; The operation data acquisition module is used to acquire the operation data of the nuclear power plant; The accident prediction module is used to input the operation data into a nuclear power plant accident prediction model, and the nuclear power plant accident prediction model outputs a fault prediction result; Wherein, the nuclear power plant accident prediction model is obtained based on the training system of the nuclear power plant accident prediction model according to claim 6; The fault result output module is used to obtain the fault type and fault degree of the nuclear power plant based on the fault prediction result.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and used to run on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 5 is implemented.

9. A computer storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images