Working condition diagnosis method, system and equipment of nuclear power station, medium and program product

Abstract

The invention provides a working condition diagnosis method, system and device for a nuclear power plant, a medium and a program product, and the method comprises the steps: obtaining a target parameter data set which comprises the data of a plurality of target parameters of a to-be-diagnosed nuclear power plant; and inputting the target parameter data set into a working condition diagnosis model of the nuclear power station to obtain probability distribution of the to-be-diagnosed nuclear power station in each working condition. The data of the target parameters of the nuclear power station are input into the working condition diagnosis model of the nuclear power station to diagnose the working condition of the nuclear power station, and the model outputs probability distribution of the nuclear power station in different working conditions, so that operators can combine professional knowledge of the operators to diagnose the working condition of the nuclear power station. Compared with an artificial intelligence model which only outputs one working condition, the working condition of the nuclear power station is comprehensively diagnosed, and the diagnosis result is more reliable.

Description

Technical Field

[0001] The present invention relates to the technical field of nuclear engineering, and particularly to a method, system, device, medium and program product for diagnosing the operating conditions of a nuclear power plant. Background Art

[0002] When a nuclear power plant is operating, various factors may cause the operation of the nuclear power plant to deviate from the normal operating conditions, such as human errors, mechanical and electrical defects, measurement errors, pipe ruptures, and external influences. There are more than 200 known abnormal operating conditions in a nuclear power plant. When the nuclear power plant operates abnormally, the operator needs to accurately judge the type of the current abnormal operating condition so as to be able to take corresponding measures, otherwise dangerous situations will occur.

[0003] However, when a nuclear power plant is operating, there are numerous parameter variables involved. If one wants to accurately judge the current abnormal operating condition, it is often necessary to comprehensively judge by combining the change trends of various parameters. It is difficult for humans to analyze the change situations of so many parameters in a short time, and moreover, it is difficult for the operator to be familiar with the parameter changes corresponding to each abnormal operating condition. Therefore, the reliability of pure manual judgment is relatively low.

[0004] In the prior art, although an artificial intelligence model can be trained to identify data, due to its complexity and black box characteristics, it is difficult for the operator to understand the identification process of the model. Therefore, the identification result of the model lacks convincing evidence and also has the problem of relatively low reliability. Summary of the Invention

[0005] The technical problem to be solved by the present disclosure is to overcome the defect of relatively low reliability in diagnosing the operating conditions of a nuclear power plant in the prior art, and to provide a method, system, device, medium and program product for diagnosing the operating conditions of a nuclear power plant.

[0006] The present disclosure solves the above technical problem through the following technical solutions:

[0007] The present disclosure provides a method for training a diagnosis model of the operating conditions of a nuclear power plant, including:

[0008] Obtaining a first parameter data set of the nuclear power plant and first label data corresponding to the first parameter data set, where the first parameter data set includes data of multiple target parameters of the nuclear power plant under different operating conditions, and the first label data includes the probability distribution of the nuclear power plant in each operating condition;

[0009] Using the first parameter data set as input data of a first pre-trained model, and adjusting the first pre-trained model according to the first label data to obtain the diagnosis model of the operating conditions of the nuclear power plant.

[0010] Optionally, the steps before obtaining the first parameter dataset of the nuclear power plant and the first label data corresponding to the first parameter dataset further include:

[0011] Obtain a second parameter dataset of the nuclear power plant, where the second parameter dataset includes data of multiple operating parameters of the nuclear power plant under different operating conditions;

[0012] Screen target parameters from the operating parameters based on the feature importance function of the LightGBM algorithm, where the number of the operating parameters is greater than the number of the target parameters.

[0013] Optionally, the first pre-training model is based on at least one of the following algorithms: LightGBM algorithm, XGBoost algorithm, CatBoost algorithm, HistGBM algorithm, and NGBoost algorithm.

[0014] Optionally, the first parameter dataset is obtained by a full-scope simulator of the nuclear power plant simulating the nuclear power plant under different operating conditions.

[0015] The present disclosure also provides a training method for training a detection model, including:

[0016] Obtain first original time series data, where the first original time series data is processed based on a first parameter dataset, the first parameter dataset is used to train a condition diagnosis model of a nuclear power plant, and the condition diagnosis model of the nuclear power plant is trained based on the foregoing training method of the condition diagnosis model of the nuclear power plant;

[0017] Use the original time series data as input data of a second pre-training model, the output of the second pre-training model is first reconstructed time series data, and use the first original time series data as second label data to adjust the second pre-training model to obtain a training detection model.

[0018] Optionally, the step of obtaining the first original time series data includes:

[0019] Perform normalization processing on the first parameter dataset;

[0020] Based on the sliding window technique, transform the first parameter dataset after normalization processing into first original time series data.

[0021] Optionally, the training detection model includes a GRU encoder, a GRU decoder, a VAE latent variable generator, and an attention weight calculator;

[0022] The GRU encoder is used to extract the hidden features of the time series data of the first original time series data;

[0023] The VAE latent variable generator is used to generate a first latent variable based on the hidden features;

[0024] The attention weighter is used to weight the first latent variable to obtain a second latent variable;

[0025] The GRU decoder is used to decode the second latent variable to obtain first reconstructed time series data.

[0026] The present disclosure also provides a training method for a working condition verification model, including:

[0027] Obtain a second original time series, where the second original time series data is processed based on a third parameter dataset, and the third parameter dataset includes data corresponding to a working condition of a nuclear power plant in the first parameter dataset, and the first parameter dataset is used to train a working condition diagnosis model of a nuclear power plant, and the working condition diagnosis model of the nuclear power plant is trained based on the foregoing training method for the working condition diagnosis model of the nuclear power plant;

[0028] Use the second original time series data as input data for a third pre-trained model, where the output of the third pre-trained model is second reconstructed time series data, and use the second original time series data as third label data to adjust the third pre-trained model to obtain a working condition verification model.

[0029] Optionally, the step of obtaining the second original time series includes:

[0030] Perform normalization processing on the third parameter dataset;

[0031] Based on the sliding window technique, transform the normalized third parameter dataset into second original time series data.

[0032] Optionally, the number of the working condition verification models is equal to the number of working condition types, and each working condition verification model corresponds to a different working condition.

[0033] The present disclosure also provides a working condition diagnosis method for a nuclear power plant, and the diagnosis method includes:

[0034] Obtain a target parameter dataset, where the target parameter dataset includes data of multiple target parameters of the nuclear power plant to be diagnosed;

[0035] Input the target parameter dataset into the working condition diagnosis model of the nuclear power plant to obtain the probability distribution of the nuclear power plant to be diagnosed in each working condition, and the working condition diagnosis model of the nuclear power plant is trained based on the foregoing training method for the working condition diagnosis model of the nuclear power plant.

[0036] Optionally, after obtaining the target parameter dataset and before inputting the target parameter dataset into the condition diagnosis model of the nuclear power plant, the following steps are further included:

[0037] Perform normalization processing on the target parameter dataset;

[0038] Based on the sliding window technique, transform the normalized target parameter dataset into target time series data;

[0039] The step of inputting the target parameter dataset into the condition diagnosis model of the nuclear power plant includes:

[0040] Input the target time series data into the trained detection model to obtain the first target reconstructed time series data, where the trained detection model is trained based on the training method of the aforementioned trained detection model;

[0041] Calculate the first reconstruction error based on the target time series data and the first target reconstructed time series data;

[0042] In response to the first reconstruction error being less than the first preset threshold, input the target parameter dataset into the condition diagnosis model of the nuclear power plant.

[0043] Optionally, the diagnosis method further includes:

[0044] After obtaining the probability distribution of the nuclear power plant to be diagnosed in each condition, the following steps include:

[0045] Obtain the target condition of the nuclear power plant to be diagnosed based on the probability distribution;

[0046] Input the target time series data into the condition verification model corresponding to the target condition to obtain the second target reconstructed time series data, where the condition verification model is trained based on the training method of the aforementioned condition verification model;

[0047] Calculate the second reconstruction error based on the target time series data and the second target reconstructed time series data;

[0048] In response to the second reconstruction error being less than the second preset threshold, output the target condition.

[0049] Optionally, after obtaining the probability distribution of the nuclear power plant to be diagnosed in each condition, the following steps include:

[0050] Obtain the target condition based on the probability distribution;

[0051] Obtain the symptom information corresponding to the target condition from the knowledge base, where the knowledge base includes symptom information of different conditions;

[0052] Output the symptom information.

[0053] Optionally, the step after obtaining the probability distribution of the nuclear power plant to be diagnosed in each working condition includes:

[0054] Obtain a target working condition based on the probability distribution;

[0055] Obtain the symptom information corresponding to the target working condition from the knowledge base, where the knowledge base includes the symptom information of different working conditions;

[0056] Output the symptom information.

[0057] Optionally, the step after obtaining the probability distribution of the nuclear power plant to be diagnosed in each working condition includes: calculating the Shapley value of each target parameter based on the SHAP algorithm, and each Shapley value is used to characterize the contribution of the corresponding target parameter to the target working condition;

[0058] Output the Shapley value of each target parameter.

[0059] Optionally, the step after calculating the Shapley value of each target parameter based on the SHAP algorithm further includes:

[0060] Sum the absolute values of the Shapley values of each target parameter to obtain the total contribution degree;

[0061] Obtain the contribution ratio of each target parameter based on the absolute value of the Shapley value of each target parameter and the total contribution degree;

[0062] Output the contribution ratio of each target parameter.

[0063] The present disclosure also provides a working condition diagnosis system for a nuclear power plant, including:

[0064] A data set acquisition module, configured to acquire a target parameter data set, where the target parameter data set includes data of multiple target parameters of the nuclear power plant to be diagnosed;

[0065] A working condition diagnosis module, configured to input the target parameter data set into a working condition diagnosis model of the nuclear power plant to obtain the probability distribution of the nuclear power plant to be diagnosed in each working condition, and the working condition diagnosis model of the nuclear power plant is trained based on the training method of the working condition diagnosis model of the nuclear power plant described above.

[0066] Optionally, the working condition diagnosis system of the nuclear power plant further includes:

[0067] A target time series acquisition module for normalizing the target parameter dataset; transforming the normalized target parameter dataset into target time series data based on the sliding window technique;

[0068] The working condition diagnosis module is further configured to:

[0069] Input the target time series data into a trained detection model to obtain first target reconstructed time series data, where the trained detection model is trained based on the training method of the aforementioned trained detection model;

[0070] Calculate a first reconstruction error based on the target time series data and the first target reconstructed time series data;

[0071] In response to the first reconstruction error being less than a first preset threshold, input the target parameter dataset into the working condition diagnosis model of the nuclear power plant;

[0072] Optionally, the working condition diagnosis module is further configured to obtain the target working condition of the nuclear power plant to be diagnosed based on the probability distribution;

[0073] The working condition diagnosis system of the nuclear power plant further includes:

[0074] A verification module for inputting the target time series data into a working condition verification model corresponding to the target working condition to obtain second target reconstructed time series data, where the working condition verification model is trained based on the training method of the aforementioned working condition verification model;

[0075] Calculate a second reconstruction error based on the target time series data and the second target reconstructed time series data;

[0076] In response to the second reconstruction error being less than a second preset threshold, output the target working condition.

[0077] Optionally, the working condition diagnosis system of the nuclear power plant further includes:

[0078] A symptom information acquisition module for acquiring symptom information corresponding to the target working condition from a knowledge base, where the knowledge base includes symptom information of different working conditions;

[0079] Output the symptom information.

[0080] Optionally, the working condition diagnosis system of the nuclear power plant further includes:

[0081] A parameter contribution calculation module for calculating the Shapley value of each target parameter based on the SHAP algorithm, where each Shapley value is used to characterize the contribution of the corresponding target parameter to the target working condition;

[0082] Output the Shapley value of each of the target parameters.

[0083] Optionally, the parameter contribution calculation module is further configured to:

[0084] Sum the absolute values of the Shapley values of each of the target parameters to obtain the total contribution degree;

[0085] Obtain the contribution ratio of each of the target parameters based on the absolute value of the Shapley value of each of the target parameters and the total contribution degree;

[0086] Output the contribution ratio of each of the target parameters.

[0087] The present disclosure further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and configured to run on the processor. When the processor executes the computer program, the training method of the foregoing working condition diagnosis model or the training method of the training detection model or the training method of the working condition verification model or the working condition diagnosis method of the nuclear power plant is implemented.

[0088] The present disclosure further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the training method of the foregoing working condition diagnosis model or the training method of the training detection model or the training method of the working condition verification model or the working condition diagnosis method of the nuclear power plant is implemented.

[0089] The present disclosure further provides a computer program product, including a computer program. When the computer program is executed by a processor, the training method of the foregoing working condition diagnosis model or the training method of the training detection model or the training method of the working condition verification model or the working condition diagnosis method of the nuclear power plant is implemented.

[0090] On the basis of conforming to the common knowledge in the art, the above preferred conditions can be combined arbitrarily to obtain various preferred examples of the present disclosure.

[0091] The positive and progressive effects of the present disclosure are as follows: By using the data of the target parameters of the nuclear power plant as the input data of the working condition diagnosis model of the nuclear power plant, and using the probability distribution of the nuclear power plant in each working condition as the label data to train the working condition diagnosis model, the trained working condition diagnosis model of the nuclear power plant can diagnose the working condition of the nuclear power plant according to the data of the target parameters of the nuclear power plant; and since the model outputs the probability distribution of the nuclear power plant in different working conditions, the operator can combine his professional knowledge and the probability of the nuclear power plant in different working conditions to diagnose the working condition of the nuclear power plant. Compared with the artificial intelligence model that only outputs one working condition, it has interpretability and the diagnosis result is more reliable. Description of the Drawings

[0092] Figure 1Flowchart of a method for training a working condition diagnosis model of a nuclear power plant provided by an exemplary embodiment of the present disclosure;

[0093] Figure 2 Flowchart of a method for training a detection model provided by an exemplary embodiment of the present disclosure;

[0094] Figure 3 Flowchart of a method for training a working condition verification model provided by an exemplary embodiment of the present disclosure;

[0095] Figure 4 Flowchart of a method for diagnosing the working condition of a nuclear power plant provided by an exemplary embodiment of the present disclosure;

[0096] Figure 5 Flowchart of another method for diagnosing the working condition of a nuclear power plant provided by an exemplary embodiment of the present disclosure;

[0097] Figure 6 Schematic diagram of modules of another working condition diagnosis system of a nuclear power plant provided by an exemplary embodiment of the present disclosure;

[0098] Figure 7 Schematic diagram of an electronic device provided by an exemplary embodiment of the present disclosure. Detailed implementation manners

[0099] The present disclosure will be further described below by way of examples, but the present disclosure is not limited to the scope of the described examples.

[0100] In the embodiments of the present disclosure, prefix words such as "first" and "second" are only used to distinguish different described objects, and have no limiting effect on the position, order, priority, quantity, content, etc. of the described objects. The use of ordinal numbers and other prefix words for distinguishing described objects in the embodiments of the present disclosure does not constitute a limitation on the described objects. The statements of the described objects refer to the descriptions in the context of the claims or embodiments, and should not constitute unnecessary limitations due to the use of such prefix words. In addition, in the description of this embodiment, unless otherwise specified, the meaning of "a plurality" is two or more.

[0101] Embodiment 1

[0102] Figure 1 Flowchart of a method for training a working condition diagnosis model of a nuclear power plant provided by an exemplary embodiment of the present disclosure. The training method includes:

[0103] S2. Obtain a first parameter dataset of the nuclear power plant and first label data corresponding to the first parameter dataset. The first parameter dataset includes data of multiple target parameters of the nuclear power plant under different working conditions, and the first label data includes the probability distribution of the nuclear power plant in each working condition.

[0104] S3. Use the first parameter dataset as the input data of the first pre-trained model, and adjust the first pre-trained model according to the first label data to obtain a working condition diagnosis model for the nuclear power plant.

[0105] In an alternative embodiment, the target parameters may include hundreds of parameters such as the temperature of the coolant at the outlet of the nuclear reactor, the pressure of the primary circuit, the pressure of the secondary circuit, the main steam flow rate, the control rod position, the coolant flow rate, the reactor power, the feedwater flow rate, etc. The specific number can be set by those skilled in the art in combination with the actual situation; according to the relevant national standards, the working conditions can be divided into three categories, namely: normal operation scenarios, design basis accident scenarios, and severe accident scenarios; the working conditions can also be classified into normal working conditions and various abnormal working conditions. In the present disclosure, it is more inclined to classify the working conditions into normal working conditions and various specific abnormal working conditions, so that the model can predict the specific types of the working conditions of the nuclear power plant based on the target parameters of the nuclear power plant. The abnormal working conditions may include SG (steam generator) pipe rupture (general code: SGTR), main steam pipe rupture (general code: MSLB), loss of coolant in the primary circuit (general code: LOCA), excessive pressure in the secondary circuit, feedwater pump trip, interaction between core melt and concrete (general code: MCCI), incorrect operation of the instrument control (such as incorrect set value), rupture of the direct injection pipeline of the reactor pressure vessel in the PXS (passive core cooling system) compartment, failure of the passive residual heat removal system, etc.

[0106] In an alternative embodiment, the first parameter dataset is obtained by simulating the nuclear power plant under different working conditions using a full-scope simulator of the nuclear power plant. The full-scope simulator of the nuclear power plant can use NuSIM developed by Guohua Nuclear Automation Co., Ltd. This simulator can simulate the operation scenarios of third-generation advanced passive pressurized water reactors under different working conditions, so as to obtain the data of various parameters during their operation, and further obtain the data of the target parameters. Those skilled in the art can use other types of simulators to simulate other types of nuclear power plants according to the actual situation.

[0107] In an alternative embodiment, before step S2, it further includes:

[0108] S11. Obtain a second parameter dataset of the nuclear power plant, where the second parameter dataset includes the data of multiple operating parameters of the nuclear power plant under different working conditions.

[0109] S12. Screen out the target parameters from the operating parameters based on the feature importance function of the LightGBM algorithm. The number of operating parameters is greater than the number of target parameters.

[0110] In the actual operation of nuclear power plants, the number of operating parameters involved is as high as thousands. However, not all operating parameters are associated with different operating conditions of nuclear power plants. Therefore, screening out the parameters closely related to different operating conditions as target parameters can simplify the training of the model and improve the prediction efficiency of the model.

[0111] The feature importance function of the LightGBM algorithm has the function of parameter screening. When training a condition diagnosis model of a nuclear power plant based on the LightGBM algorithm, target parameters can be screened from the operating parameters based on the feature importance function of the LightGBM algorithm, reducing the number of parameters from thousands to hundreds. The screening ratio can be set by those skilled in the art in combination with practical situations.

[0112] In an alternative embodiment, the first pre-trained model is based on at least one of the following algorithms: LightGBM algorithm, XGBoost algorithm, CatBoost algorithm, HistGBM algorithm, and NGBoost algorithm. Those skilled in the art can train a condition diagnosis model of a nuclear power plant based on each of the above algorithms according to actual needs, thus training five condition diagnosis models of nuclear power plants. Based on the diagnosis of the operating conditions of the nuclear power plant by the five models, the operator can combine their understanding of the characteristics of the above five algorithms to judge the prediction results of the models. Since the characteristics of each algorithm are different, their applicability to different scenarios is also different. By training models corresponding to different algorithms, the operator can select the prediction results of the model corresponding to the algorithm most suitable for the current scenario, further improving the interpretability and reliability of the prediction results of the model.

[0113] In an alternative embodiment, the first parameter dataset can be divided into a training dataset, a validation dataset, and a test dataset according to a certain ratio. For example, 70% of the data in the first parameter dataset is divided into the training dataset, 15% of the data in the first parameter dataset is divided into the validation dataset, and 15% of the data in the first parameter dataset is divided into the test dataset. After the first pre-trained model completes the adjustment steps of training, validation, and testing in sequence, it is put into use as a condition diagnosis model of a nuclear power plant.

[0114] In an alternative embodiment, when training and optimizing the condition diagnosis model of a nuclear power plant, the accuracy rate that can be output can be used as a performance evaluation index for the validation and test data. When the accuracy rate ≥ 98%, it can be put into use.

[0115] In the embodiments of the present disclosure, by using the data of the target parameters of a nuclear power plant as the input data of the working condition diagnosis model of the nuclear power plant, and using the probability distribution of the nuclear power plant in each working condition as the labeled data to train the working condition diagnosis model, the trained working condition diagnosis model of the nuclear power plant can diagnose the working condition of the nuclear power plant according to the data of the target parameters of the nuclear power plant; moreover, since the model outputs the probability distribution of the nuclear power plant in different working conditions, the operator can combine their professional knowledge and the probability of the nuclear power plant in different working conditions to diagnose the working condition of the nuclear power plant. Compared with the artificial intelligence model that only outputs one working condition, it has interpretability and the diagnosis result is more reliable. For example, if the probability of the first working condition is much higher than that of other working conditions, it indicates that the nuclear power plant is very likely to be in the first working condition currently, and the operator can make a simple confirmation; if the probabilities of the first working condition and the second working condition are close, and the probabilities of the first working condition and the second working condition are much higher than those of other working conditions, the staff needs to focus on judging whether the nuclear power plant is in the first working condition or the second working condition, rather than simply confirming that the nuclear power plant is in the working condition with the highest probability value. Therefore, the diagnosis result is also more reliable.

[0116] Embodiment 2

[0117] Figure 2 FIG. is a flowchart of a training method for training a detection model provided by an exemplary embodiment of the present disclosure. The training method includes:

[0118] S1. Obtain first original time series data, which is processed based on a first parameter dataset.

[0119] The first parameter dataset is used to train the working condition diagnosis model of the nuclear power plant, and the working condition diagnosis model of the nuclear power plant is trained based on the training method of the working condition diagnosis model of the nuclear power plant in Embodiment 1.

[0120] S2. Use the original time series data as the input data of a second pre-training model. The output of the second pre-training model is first reconstructed time series data. Use the first original time series data as the second labeled data to adjust the second pre-training model to obtain a training detection model.

[0121] In an optional implementation, step S1 includes:

[0122] S11. Perform normalization processing on the first parameter dataset.

[0123] S12. Based on the sliding window technique, transform the normalized first parameter dataset into first original time series data.

[0124] The Min-Max normalization method can be used to process the data of each target parameter in the first parameter dataset so that its value is within the range of 0 to 1. After obtaining the first parameter dataset processed by normalization, based on the sliding window technique, the two-dimensional matrix data is converted into three-dimensional time series data to obtain the first original time series data. The time step can be set by those skilled in the art according to the actual situation and can be set to five seconds.

[0125] In an alternative embodiment, the training detection model includes a GRU encoder, a GRU decoder, a VAE latent variable generator, and an attention weighter.

[0126] Among them, the GRU encoder is used to extract the hidden features of the time series data of the first original time series data; the VAE latent variable generator is used to generate the first latent variable based on the hidden features; the attention weighter is used to weight the first latent variable to obtain the second latent variable; the GRU decoder is used to decode the second latent variable to obtain the first reconstructed time series data.

[0127] Specifically, the first original time series data is:

[0128] ;

[0129] Among them, is the input variable at time step t.

[0130] The calculation formula of the GRU encoder is:

[0131] ;

[0132] ;

[0133] ;

[0134] ;

[0135] Among them, is the hidden state at time step t, GRU is the non-linear function of the standard GRU cell, is the global hidden variable; and are respectively the mean vector and variance of the first latent variable mapped to the Gaussian distribution; and are weight matrices, and are bias vectors, is the standard deviation vector of the first latent variable, is generated by the GRU encoder through applying network parameterization and represents the distribution center of the first latent variable.

[0136] The calculation formula of the VAE latent variable generator is as follows:

[0137] Sample the first latent variable through reparameterization:

[0138] ;

[0139] Among them, z represents the first latent variable obtained through sampling, which is used to represent the latent features of the encoded data. represents the random noise sampled from the standard normal distribution, which is used to introduce randomness and avoid non-differentiability in gradient propagation through the reparameterization trick. represents element-wise multiplication, the standard deviation of the first latent variable and the noise of the element-wise multiplication operation. is the standard normal distribution, I is the identity matrix, and the sampled noise is the standard normal distribution with each dimension independent.

[0140] The calculation formula of the attention weight calculator is as follows:

[0141] Perform attention weighting on the first latent variable z:

[0142] Calculate the weight score of the first latent variable : :

[0143] ;

[0144] Among them, is the weight vector of the attention mechanism, is the bias of the attention mechanism.

[0145] Use the softmax function to normalize the weights:

[0146] ;

[0147] is the attention weight of the i-th latent variable, , ensuring weight normalization.

[0148] Generate the second latent variable after attention weighting:

[0149] ;

[0150] is the second latent variable weighted by the attention mechanism and is used for subsequent decoding.

[0151] The calculation formula of the GRU decoder is as follows:

[0152] Decode the second latent variable into the initial hidden state :

[0153] ;

[0154] Generate the hidden state through a time-step-by-time-step GRU decoder:

[0155] ;

[0156] where, is the decoder hidden state at the t-th step, is the decoder input, usually the prediction value of the previous step or a specific input column.

[0157] The decoder hidden state is mapped to the reconstruction output at time step t:

[0158] ;

[0159] where, is the decoder weight matrix, is the decoder bias vector, is the reconstruction result at the t-th time step.

[0160] Through recursive operations, the decoder generates the complete first reconstructed time series:

[0161] ;

[0162] In an alternative embodiment, the trained detection model can be tested by testing the AUC value (area under the receiver operating characteristic curve) of the output data of the trained detection model. If both exceed 0.9, it indicates that the model is well-trained.

[0163] In an alternative embodiment, the output of the trained detection model can be connected to a reconstruction error calculation module. The reconstruction error calculation module is used to calculate the reconstruction error of the trained detection model. The reconstruction error is the square of the difference between the input and output of the trained detection model. Since the first original time series is processed based on the first parameter dataset of the working condition diagnosis model of the training nuclear power plant, therefore, if the data subsequently input to the trained detection model is relatively close to the first parameter dataset, the reconstruction error is usually small. A first preset threshold can be set. By comparing the reconstruction error and the first preset threshold, if the reconstruction error is less than the first preset threshold, it indicates that they are relatively close, and the reliability of diagnosing this data using the working condition diagnosis model of the nuclear power plant is relatively high, otherwise the reliability is relatively low.

[0164] In the embodiments of the present disclosure, by training a training detection model, after the working condition diagnosis model of the nuclear power plant is put on line, the data to be input into the working condition diagnosis model of the nuclear power plant is input into the training detection model, and the reconstruction error is calculated based on the input and output of the training detection model. If the reconstruction error is small, it indicates that the input data is relatively close to the data used to train the working condition diagnosis model of the nuclear power plant, and the reliability of the output result of the working condition diagnosis model of the nuclear power plant is stronger. If the reconstruction error is large, it indicates that the difference between the two is large, and the output result of the working condition diagnosis model of the nuclear power plant is weak. Furthermore, the interpretability and reliability of the output of the working condition diagnosis model of the nuclear power plant are enhanced; and, a new GRU-VAE-Attention model is proposed. By introducing the variational inference method of VAE, the first latent variable can be generated by normal distribution sampling, enhancing the generalization ability of the model; using the GRU decoder to decode and generate the original time series can optimize the reconstruction error; by adding an attention mechanism to weight-process the first latent variable, the sensitivity of the model to important features is enhanced, and the detection accuracy and reliability of the model for complex time series data are further improved.

[0165] Embodiment 3

[0166] Figure 3 The flowchart of a training method for a working condition verification model provided by an exemplary embodiment of the present disclosure. The training method includes:

[0167] S1. Obtain a second original time series, and the second original time series data is processed based on a third parameter dataset, and the third parameter dataset includes the data corresponding to a working condition of a nuclear power plant in the first parameter dataset.

[0168] The first parameter dataset is used to train the working condition diagnosis model of the nuclear power plant, and the working condition diagnosis model of the nuclear power plant is trained based on the training method of the working condition diagnosis model in Embodiment 1.

[0169] S2. Use the second original time series data as the input data of the third pre-trained model, and the output of the third pre-trained model is the second reconstructed time series data. Use the second original time series data as the third label data to adjust the third pre-trained model to obtain the working condition verification model.

[0170] In an alternative implementation, the number of working condition verification models is equal to the number of working condition types, and the working conditions corresponding to each working condition verification model are different. For example, if there are two hundred different working conditions, a working condition verification model is trained for each working condition.

[0171] In an alternative implementation, step S1 includes:

[0172] S11. Normalize the third parameter dataset;

[0173] S12. Transform the third parameter dataset after normalization processing into the second original time series data based on the sliding window technique;

[0174] The Min-Max normalization method can be used to process the data of each target parameter in the third parameter dataset so that its value is within the range of 0 to 1. After obtaining the third parameter dataset after normalization processing, based on the sliding window technique, the two-dimensional matrix data is converted into three-dimensional time series data to obtain the second original time series data. The time step can be set by those skilled in the art according to the actual situation, and the time step can be set to five seconds.

[0175] In an alternative embodiment, the condition verification model can also adopt the GRU-VAE-ATTENTION structure of the training detection model. Since the condition diagnosis model is trained based on the data of the target parameters of one condition, it can verify whether a specific condition is accurate; while the training detection model is trained based on the data of the target parameters under different conditions. Therefore, it can only detect whether the data input into the condition diagnosis model of the nuclear power plant is close to the data used to train the condition diagnosis model of the nuclear power plant, and cannot play the role of verifying specific conditions.

[0176] In an alternative embodiment, the condition verification model can be tested by testing the AUC value (area under the receiver operating characteristic curve) of the output data of the condition verification model. If both exceed 0.9, it indicates that the model is well-trained.

[0177] In the embodiments of the present disclosure, multiple condition verification models are trained, and each model matches one condition. When the probability distribution is output with the condition diagnosis model of the nuclear power plant, several condition verification models corresponding to the several conditions with the highest probability can be matched according to the highest probabilities. The data input into the condition diagnosis model of the nuclear power plant is respectively input into these several condition verification models. The reconstruction error is calculated based on the input and output of these several condition verification models. Based on the reconstruction error, these several conditions can be verified. If the corresponding reconstruction error is small, it indicates that the probability of the nuclear power plant being in the corresponding condition is high. If the corresponding reconstruction error is large, it indicates that the probability of the nuclear power plant being in the corresponding condition is small, further enhancing the interpretability and reliability of the condition diagnosis.

[0178] Embodiment 4

[0179] Figure 4 It is a flowchart of a method for diagnosing the conditions of a nuclear power plant provided by an exemplary embodiment of the present disclosure. The training method includes:

[0180] S1. Obtain a target parameter dataset, where the target parameter dataset includes the data of multiple target parameters of the nuclear power plant to be diagnosed.

[0181] S3. Input the target parameter dataset into the operating condition diagnosis model of the nuclear power plant to obtain the probability distribution of the nuclear power plant to be diagnosed in each operating condition.

[0182] The operating condition diagnosis model of the nuclear power plant is trained based on the training method of the operating condition diagnosis model in Embodiment 1.

[0183] In an alternative embodiment, referring to Figure 5 , the steps after step S1 and before step S3 further include:

[0184] S21. Perform normalization processing on the target parameter dataset.

[0185] S22. Based on the sliding window technique, transform the normalized target parameter dataset into target time series data.

[0186] Step S3 specifically includes:

[0187] S31. Input the target time series data into the training and detection model to obtain the first target reconstructed time series data.

[0188] The training and detection model is trained based on the training method of the training and detection model in Embodiment 2.

[0189] S32. Calculate the first reconstruction error based on the target time series data and the first target reconstructed time series data.

[0190] S33. Determine whether the first reconstruction error is the first preset threshold. If so, execute step S34. If not, notify the operator to intervene and handle.

[0191] S34. In response to the first reconstruction error being less than the first preset threshold, input the target parameter dataset into the operating condition diagnosis model of the nuclear power plant.

[0192] In an alternative embodiment, the steps after step S3 include:

[0193] S41. Obtain the target operating condition of the nuclear power plant to be diagnosed based on the probability distribution.

[0194] The operating condition with the highest probability in the probability distribution can be used as the target operating condition. The target operating condition can be one or multiple with the highest probability.

[0195] S42. Input the target time series data into the operating condition verification model corresponding to the target operating condition to obtain the second target reconstructed time series data. The operating condition verification model is trained based on the training method of the operating condition verification model in Embodiment 3.

[0196] S43. Calculate the second reconstruction error based on the target time series data and the second target reconstructed time series data.

[0197] S44. Output the target operating condition in response to the second reconstruction error being less than the second preset threshold.

[0198] Through re-verifying with the operating condition verification model, the reliability of the target operating condition can be further improved.

[0199] In an alternative embodiment, the steps after step S3 include:

[0200] S51. Obtain the target operating condition based on the probability distribution.

[0201] S52. Obtain the symptom information corresponding to the target operating condition from the knowledge base, where the knowledge base includes symptom information for different operating conditions.

[0202] S53. Output the symptom information.

[0203] The operator can verify whether the output of the model is correct based on the symptom information. It is also possible to automatically determine by the system which symptoms in the symptom information are met by the nuclear power plant, and mark the met symptoms in red and display them to the staff.

[0204] In an alternative embodiment, the steps after step S3 include:

[0205] S61. Calculate the Shapley value of each target parameter based on the SHAP algorithm, and each Shapley value is used to characterize the contribution of the corresponding target parameter to the target operating condition.

[0206] S63. Output the Shapley value of each target parameter.

[0207] The Shapley value includes positive and negative values. A positive value indicates a positive correlation, and a negative value indicates a negative correlation. A positive correlation can be displayed in red, and a negative correlation can be displayed in blue.

[0208] In an alternative embodiment, the steps after step S61 include:

[0209] S621. Sum the absolute values of the Shapley values of each target parameter to obtain the total contribution degree.

[0210] S622. Obtain the contribution ratio of each target parameter based on the absolute value of the Shapley value of each target parameter and the total contribution degree.

[0211] S623. Output the contribution ratio of each target parameter.

[0212] The target parameters with a contribution ratio greater than 10% can be defined as the main evidence, and the target parameters with a contribution ratio of 1% - 10% can be defined as the secondary evidence.

[0213] In an alternative embodiment, after step S44, it includes:

[0214] S45. If the target working condition is a normal working condition, control the indicator light to display green; if the target working condition is an abnormal working condition, control the indicator light to display red.

[0215] In this embodiment, the colors used to prompt the staff can be set according to the actual situation.

[0216] In an alternative embodiment, all information such as the data input into each model, the output data of each model, and the diagnostic results can be summarized to generate a log. Through the log, the consistency of the model diagnostic results over time can be verified, which helps to judge the reliability of the AI. At the same time, it can be used to verify the necessity of the working condition diagnosis function.

[0217] In the embodiment of the present disclosure, the working condition of the nuclear power plant is diagnosed by inputting the data of the target parameters of the nuclear power plant into the working condition diagnosis model of the nuclear power plant. Since the model outputs the probability distribution of the nuclear power plant in different working conditions, the operator can combine their professional knowledge and the probability of the nuclear power plant in different working conditions to comprehensively diagnose the working condition of the nuclear power plant. Compared with the artificial intelligence model that only outputs one working condition, it has interpretability and the diagnostic results are more reliable.

[0218] Embodiment 5

[0219] Corresponding to the foregoing embodiment of the working condition diagnosis method of the nuclear power plant, the present disclosure also provides an embodiment of the working condition diagnosis system of the nuclear power plant.

[0220] Figure 6 The module schematic diagram of a working condition diagnosis system of a nuclear power plant provided for an exemplary embodiment of the present disclosure. The system includes:

[0221] A data set acquisition module 1, configured to acquire a target parameter data set, where the target parameter data set includes the data of multiple target parameters of the nuclear power plant to be diagnosed.

[0222] A working condition diagnosis module 2, configured to input the target parameter data set into the working condition diagnosis model of the nuclear power plant to obtain the probability distribution of the nuclear power plant to be diagnosed in each working condition.

[0223] The working condition diagnosis model of the nuclear power plant is trained based on the training method of the working condition diagnosis model of the nuclear power plant in Embodiment 1.

[0224] In an alternative embodiment, the diagnosis method further includes:

[0225] A target time series acquisition module, configured to perform normalization processing on the target parameter data set; based on the sliding window technology, convert the normalized target parameter data set into target time series data.

[0226] The operating condition diagnosis module is further configured to: input the target time series data into the trained detection model to obtain the first target reconstructed time series data, where the trained detection model is trained based on the training method of the trained detection model in Embodiment 2; calculate the first reconstruction error based on the target time series data and the first target reconstructed time series data; compare the first reconstruction error with the first preset threshold, if the first reconstruction error is less than the first preset threshold, input the target parameter data set into the operating condition diagnosis model of the nuclear power plant, and if the first reconstruction error is greater than the first preset threshold, notify the operator to intervene for processing.

[0227] In an alternative embodiment, the operating condition diagnosis module is further configured to obtain the target operating condition of the nuclear power plant to be diagnosed based on the probability distribution.

[0228] The operating condition with the highest probability in the probability distribution can be used as the target operating condition, and the target operating condition can be one or multiple with the highest probability.

[0229] The operating condition diagnosis system of the nuclear power plant further includes:

[0230] A verification module, configured to input the target time series data into the operating condition verification model corresponding to the target operating condition to obtain the second target reconstructed time series data, where the operating condition verification model is trained based on the training method of the operating condition verification model in Embodiment 3; calculate the second reconstruction error based on the target time series data and the second target reconstructed time series data; and output the target operating condition in response to the second reconstruction error being less than the second preset threshold.

[0231] Through the re-verification of the operating condition verification model, the reliability of the target operating condition can be further improved.

[0232] In an alternative embodiment, the operating condition diagnosis system of the nuclear power plant further includes:

[0233] A symptom information acquisition module, configured to obtain the target operating condition based on the probability distribution; acquire the symptom information corresponding to the target operating condition from the knowledge base, where the knowledge base includes symptom information of different operating conditions; and output the symptom information.

[0234] The operator can verify whether the output of the model is correct based on the symptom information. It is also possible to automatically determine by the system which symptoms of the symptom information are met by the nuclear power plant, and mark the met symptoms in red and display them to the staff.

[0235] In an alternative embodiment, the operating condition diagnosis system of the nuclear power plant further includes:

[0236] A parameter contribution calculation module, configured to calculate the Shapley value of each target parameter based on the SHAP algorithm, where each Shapley value is used to characterize the contribution of the corresponding target parameter to the target operating condition; and output the Shapley value of each target parameter.

[0237] The Shapley values include positive and negative values. A positive value indicates a positive correlation, and a negative value indicates a negative correlation. A positive correlation can be displayed in red, and a negative correlation can be displayed in blue.

[0238] In an alternative embodiment, the parameter contribution calculation module is further configured to: sum the absolute values of the Shapley values of each target parameter to obtain the total contribution degree; obtain the contribution ratio of each target parameter based on the absolute value of the Shapley value of each target parameter and the total contribution degree; and output the contribution ratio of each target parameter.

[0239] The target parameters with a contribution ratio greater than 10% can be defined as main evidence, and the target parameters with a contribution ratio of 1% - 10% can be defined as secondary evidence.

[0240] In an alternative embodiment, the condition diagnosis system of the nuclear power plant further includes:

[0241] An indicator light control module, configured to control the indicator light to display green when the target condition is a normal condition; and control the indicator light to display red when the target condition is an abnormal condition.

[0242] The colors used to prompt the staff involved in this embodiment can be set according to the actual situation.

[0243] In an alternative embodiment, all information such as the data input to each model, the output data of each model, and the diagnosis results can be summarized to generate a log. The log can be used to verify the consistency of the model diagnosis results over time, which helps to judge the reliability of the AI and can also be used to verify the necessity of the condition diagnosis function.

[0244] In the embodiment of the present disclosure, the condition of the nuclear power plant is diagnosed by inputting the data of the target parameters of the nuclear power plant into the condition diagnosis model of the nuclear power plant. Since the model outputs the probability distribution of the nuclear power plant in different conditions, the operator can combine their professional knowledge and the probability of the nuclear power plant in different conditions to comprehensively diagnose the condition of the nuclear power plant. Compared with the artificial intelligence model that only outputs one condition, it has interpretability and the diagnosis results are more reliable.

[0245] 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. The units described as separate components may or may not be physically separated. The components as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the present disclosure solution.

[0246] Example 6

[0247] Figure 7 As shown in the structural schematic diagram of an electronic device according to an exemplary embodiment of the present disclosure, the electronic device includes a memory, a processor, and a computer program stored in the memory and configured to run on the processor. When the processor executes the computer program, it implements the training method of the working condition diagnosis model or the training method of the training detection model or the training method of the working condition verification model or the working condition diagnosis method of a nuclear power plant described in any of the above embodiments. Figure 7 The electronic device 90 shown is merely an example and should not impose any limitation on the functions and usage scope of the embodiments of the present disclosure.

[0248] As Figure 7 shown, the electronic device 90 may be presented 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 of the above-mentioned processors 91, at least one of the above-mentioned memories 92, and a bus 93 connecting different system components (including the memory 92 and the processor 91).

[0249] The bus 93 includes a data bus, an address bus, and a control bus.

[0250] The memory 92 may include 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.

[0251] The memory 92 may further include a program tool 925 (or utility tool) having a set (at least one) of program modules 924. Such program modules 924 include, but are not limited to: an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include the implementation of a network environment.

[0252] 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 working condition diagnosis model or the training method of the training detection model or the training method of the working condition verification model or the working condition diagnosis method of a nuclear power plant provided in any of the above embodiments.

[0253] The electronic device 90 may also communicate with one or more external devices 94 (such as a keyboard, a pointing device, etc.). Such communication may be performed through an input / output (I / O) interface 95. And, the electronic device 90 may also communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through a network adapter 96. As Figure 7As shown, network adapter 96 communicates with other modules of electronic device 90 via bus 93. It should be understood that although Figure 7 not shown in Figure 7 , other hardware and / or software modules can be used in conjunction with electronic device 90, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (redundant array of independent disks) systems, tape drives, and data backup storage systems, etc.

[0254] 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 / modules. Conversely, the features and functions of one unit / modules described above can be further divided and embodied by multiple units / modules.

[0255] Embodiment 7

[0256] The embodiments of the present disclosure also provide a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the training method of the working condition diagnosis model or the training method of the training detection model or the training method of the working condition verification model or the working condition diagnosis method of the nuclear power plant provided in any one of the above embodiments.

[0257] Among them, the readable storage medium can more specifically include but not limited to: portable disks, hard disks, random access memories, read-only memories, erasable programmable read-only memories, optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0258] Embodiment 8

[0259] The embodiments of the present disclosure also provide a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the training method of the working condition diagnosis model or the training method of the training detection model or the training method of the working condition verification model or the working condition diagnosis method of the nuclear power plant described in any one of the above.

[0260] 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 entirely on the user device, partially on the user device, executed as an independent software package, partially on the user device and partially on a remote device, or entirely on a remote device.

[0261] Although the specific embodiments of the present disclosure have been 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. Without departing from the principle and essence of the present disclosure, those skilled in the art can make various changes or modifications to these embodiments, but these changes and modifications all fall within the protection scope of the present disclosure.

Claims

1. A training method for a working condition diagnosis model of a nuclear power plant, characterized in that Including: Obtain a first parameter dataset of the nuclear power plant and first label data corresponding to the first parameter dataset, where the first parameter dataset includes data of multiple target parameters of the nuclear power plant under different working conditions, and the first label data includes the probability distribution of the nuclear power plant in each working condition; Use the first parameter dataset as input data for a first pre-trained model, and adjust the first pre-trained model according to the first label data to obtain a working condition diagnosis model of the nuclear power plant.

2. The training method of the operating condition diagnosis model of a nuclear power plant according to claim 1, characterized in that The steps before obtaining the first parameter dataset of the nuclear power plant and the first label data corresponding to the first parameter dataset further include: Obtain a second parameter dataset of the nuclear power plant, where the second parameter dataset includes data of multiple operating parameters of the nuclear power plant under different working conditions; Screen out target parameters from the operating parameters based on the feature importance function of the LightGBM algorithm, and the number of the operating parameters is greater than the number of the target parameters.

3. The training method of the operating condition diagnosis model of a nuclear power plant according to claim 1, characterized in that The first pre-trained model is based on at least one of the following algorithms: LightGBM algorithm, XGBoost algorithm, CatBoost algorithm, HistGBM algorithm, and NGBoost algorithm; And / or The first parameter dataset is obtained by a full-scope simulator of the nuclear power plant simulating the nuclear power plant under different working conditions.

4. A training method for training a detection model, characterized in that, Including: Obtain first original time series data, which is processed based on a first parameter dataset used to train a working condition diagnosis model of a nuclear power plant, and the working condition diagnosis model of the nuclear power plant is trained based on the training method of the working condition diagnosis model of the nuclear power plant described in any one of claims 1-3; Use the original time series data as input data for a second pre-trained model, the output of the second pre-trained model is first reconstructed time series data, and use the first original time series data as second label data to adjust the second pre-trained model to obtain a training detection model.

5. The training method of the training detection model according to claim 4, characterized in that, The training detection model includes a GRU encoder, a GRU decoder, a VAE latent variable generator, and an attention weight calculator; The GRU encoder is used to extract the hidden features of the time series data of the first original time series data; The VAE latent variable generator is used to generate a first latent variable based on the hidden features; The attention weight calculator is used to weight the first latent variable to obtain a second latent variable; The GRU decoder is used to decode the second latent variable to obtain first reconstructed time series data; And / or The step of obtaining the first original time series data includes: Perform normalization processing on the first parameter dataset; Based on the sliding window technique, transform the first parameter dataset after normalization processing into first original time series data.

6. A training method for a working condition verification model, characterized in that Including: Obtain a second original time series, where the second original time series data is processed based on a third parameter dataset, and the third parameter dataset includes data in the first parameter dataset corresponding to the working conditions of a nuclear power plant. The first parameter dataset is used to train a working condition diagnosis model of the nuclear power plant, and the working condition diagnosis model of the nuclear power plant is trained based on the training method of the working condition diagnosis model of the nuclear power plant as described in any one of claims 1-3; Use the second original time series data as the input data of a third pre-trained model, and the output of the third pre-trained model is the second reconstructed time series data. Use the second original time series data as the third label data to adjust the third pre-trained model to obtain a working condition verification model.

7. The training method of the operating condition verification model according to claim 6, characterized in that The step of obtaining the second original time series includes: Perform normalization processing on the third parameter dataset; Based on the sliding window technique, transform the normalized third parameter dataset into the second original time series data; and / or, The number of the working condition verification models is equal to the number of working condition types, and the working conditions corresponding to each working condition verification model are different.

8. A method for diagnosing the operating conditions of a nuclear power plant, characterized in that, The diagnosis method includes: Obtain a target parameter dataset, where the target parameter dataset includes data of multiple target parameters of the nuclear power plant to be diagnosed; Input the target parameter dataset into the working condition diagnosis model of the nuclear power plant to obtain the probability distribution of the nuclear power plant to be diagnosed in each working condition. The working condition diagnosis model of the nuclear power plant is trained based on the training method of the working condition diagnosis model of the nuclear power plant as described in any one of claims 1-3.

9. The method for diagnosing the operating condition of a nuclear power plant according to claim 8, wherein, After the step of obtaining the target parameter dataset and before the step of inputting the target parameter dataset into the working condition diagnosis model of the nuclear power plant, the steps further include: Perform normalization processing on the target parameter dataset; Based on the sliding window technique, transform the normalized target parameter dataset into target time series data; The step of inputting the target parameter dataset into the working condition diagnosis model of the nuclear power plant includes: Input the target time series data into a trained detection model to obtain first target reconstructed time series data. The trained detection model is trained based on the training method of the trained detection model as described in claim 4 or 5; Calculate a first reconstruction error based on the target time series data and the first target reconstructed time series data; In response to the first reconstruction error being less than a first preset threshold, input the target parameter dataset into the working condition diagnosis model of the nuclear power plant; and / or, After the step of obtaining the probability distribution of the nuclear power plant to be diagnosed in each working condition, the steps include: Obtain the target working condition of the nuclear power plant to be diagnosed based on the probability distribution; Input the target time series data into the working condition verification model corresponding to the target working condition to obtain second target reconstructed time series data. The working condition verification model is trained based on the training method of the working condition verification model as described in claim 6 or 7; Calculate a second reconstruction error based on the target time series data and the second target reconstructed time series data; In response to the second reconstruction error being less than a second preset threshold, output the target working condition.

10. The method for diagnosing the operating condition of a nuclear power plant according to claim 8 or 9, characterized in that, The steps after obtaining the probability distribution of the nuclear power plant to be diagnosed in each working condition include: Obtaining a target working condition based on the probability distribution; Obtaining symptom information corresponding to the target working condition from a knowledge base, where the knowledge base includes symptom information for different working conditions; Outputting the symptom information; And / or, Calculating the Shapley value of each target parameter based on the SHAP algorithm, where each Shapley value is used to characterize the contribution of the corresponding target parameter to the target working condition; Outputting the Shapley value of each target parameter.

11. The method for diagnosing the operating condition of a nuclear power plant according to claim 10, characterized in that, The steps after calculating the Shapley value of each target parameter based on the SHAP algorithm further include: Summing the absolute values of the Shapley values of each target parameter to obtain a total contribution degree; Obtaining the contribution ratio of each target parameter based on the absolute value of the Shapley value of each target parameter and the total contribution degree; Outputting the contribution ratio of each target parameter.

12. A condition diagnosis system for a nuclear power plant, characterized in that, Including: A data set acquisition module for acquiring a target parameter data set, where the target parameter data set includes data of multiple target parameters of the nuclear power plant to be diagnosed; A working condition diagnosis module for inputting the target parameter data set into a working condition diagnosis model of the nuclear power plant to obtain the probability distribution of the nuclear power plant to be diagnosed in each working condition, and the working condition diagnosis model of the nuclear power plant is trained based on the training method of the working condition diagnosis model of the nuclear power plant according to any one of claims 1-3.

13. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and configured to run on the processor, characterized in that, When the processor executes the computer program, it implements the training method of the working condition diagnosis model according to any one of claims 1 to 3, or the training method of the training detection model according to claim 4 or 5, or the training method of the working condition verification model according to claim 6 or 7, or the working condition diagnosis method of the nuclear power plant according to any one of claims 8 to 11.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the training method of the working condition diagnosis model according to any one of claims 1 to 3, or the training method of the training detection model according to claim 4 or 5, or the training method of the working condition verification model according to claim 6 or 7, or the working condition diagnosis method of the nuclear power plant according to any one of claims 8 to 11.

15. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the training method of the working condition diagnosis model according to any one of claims 1 to 3, or the training method of the training detection model according to claim 4 or 5, or the training method of the working condition verification model according to claim 6 or 7, or the working condition diagnosis method of the nuclear power plant according to any one of claims 8 to 11.

Images