A nuclear power key indicator prediction device based on optimal deep learning

Abstract

The application discloses a nuclear power key index prediction device based on optimal deep learning, which comprises a sodium iodide spectrometer database, a data acquisition and preprocessing module, a one-dimensional CNN-LSTM cascade model module, a multi-target adaptive particle swarm optimization algorithm module and a nuclear power key index prediction and online correction module. The application realizes online prediction and parameter optimal online optimization, can robustly and agilely predict the nuclear power key index in the operation process of nuclear power facilities, does not need artificial design of prediction characteristics or rules, overcomes the defects of low precision, easy influence by human factors and easy mismatch of the prediction model of the current nuclear power key index prediction device, leading to poor prediction effect and even system collapse, and improves the prediction efficiency and accuracy of long-term online operation of the radiation environment nuclear power key index prediction.

Description

Technical Field

[0001] This invention belongs to the field of nuclear power key indicator prediction, deep learning and optimization algorithms, and particularly relates to a nuclear power key indicator prediction device based on optimal deep learning. Background Technology

[0002] In the nuclear power sector, nuclear safety is of paramount importance for nuclear power plant operation. Historically, automatic radiation environment monitoring stations have relied on manual prediction and analysis. This method is susceptible to human error, is inaccurate, and requires lengthy prediction times. Therefore, accurately predicting key nuclear power indicators is a pressing issue and challenge that urgently needs to be addressed.

[0003] In recent years, deep learning methods have been rarely used in the prediction of key nuclear power indicators. However, combining deep learning algorithms with advanced optimization algorithms can improve the accuracy, robustness, and adaptability of predictions, thereby enhancing the safety and stability of nuclear power systems. Currently available nuclear power key indicator prediction devices suffer from low prediction accuracy, susceptibility to human factors, and a tendency for prediction models to mismatch, leading to poor prediction results or even system collapse. These limitations make long-term effective online operation difficult. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing a nuclear power key indicator prediction device based on optimal deep learning.

[0005] The objective of this invention is achieved through the following technical solution: a nuclear power key indicator prediction device based on optimal deep learning, the device comprising a sodium iodide spectrometer database, a data acquisition and preprocessing module, a one-dimensional CNN-LSTM cascade model module, a multi-objective adaptive particle swarm optimization algorithm module, and a nuclear power key indicator prediction and online correction module;

[0006] The data acquisition and preprocessing module is used to process the sodium iodide spectrometer data input from the sodium iodide spectrometer database. Data preprocessing is performed to obtain preprocessed data. And upload it to the one-dimensional CNN-LSTM cascaded model module;

[0007] The one-dimensional CNN-LSTM cascaded model module is used to process the preprocessed data through a one-dimensional CNN-LSTM cascaded model. Classify the data to obtain classification vectors. ;

[0008] The multi-objective adaptive particle swarm optimization algorithm module is used to optimize the number of training iterations, learning rate, and LSTM time window size of a one-dimensional CNN-LSTM cascade model using an improved particle swarm optimization algorithm.

[0009] The nuclear power key indicator prediction and online correction module is used to update the one-dimensional CNN-LSTM cascade model online, periodically inputting the current data stream of the sodium iodide spectrometer into the training set, and using the new training set to modify the parameters in the one-dimensional CNN-LSTM cascade model.

[0010] Furthermore, the data from the sodium iodide spectrometer will be input from the sodium iodide spectrometer database. Data preprocessing is performed to obtain preprocessed data. Specifically, it includes the following sub-steps:

[0011] (a.1) Regarding the data Data cleaning involves removing duplicate and outlier information to obtain cleaned data. .

[0012] (a.2) Subsequently, the cleaned data Perform data type conversion on the cleaned data. The character-type features in the data are converted into numerical features to obtain the data. ;

[0013] (a.3) Subsequently, the data Data normalization is performed to obtain preprocessed data. The calculation formula is as follows:

[0014]

[0015] in, For data The Middle Eigenvalues ​​at; For data The smallest eigenvalue in the middle; For data The largest eigenvalue in the middle; For preprocessed data The Middle The eigenvalue at that location.

[0016] Furthermore, the one-dimensional CNN-LSTM cascaded model module includes a one-dimensional CNN-LSTM cascaded model, the specific structure of which is as follows:

[0017] The first layer is a one-dimensional convolutional layer: the kernel size is 16×1, the stride is 2, and the number of kernels is 64.

[0018] The second layer, the max pooling layer: the pooling window size is 2;

[0019] The third layer is a one-dimensional residual layer: the first one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 1, and 64 convolutional kernels; the second one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 1, and 64 convolutional kernels.

[0020] The fourth layer is a one-dimensional residual layer: the first one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 2, and a total of 64 convolutional kernels; the second one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 1, and a total of 64 convolutional kernels.

[0021] The fifth layer is a one-dimensional residual layer: the first one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 1, and 128 convolutional kernels; the second one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 1, and 128 convolutional kernels.

[0022] The sixth layer is a one-dimensional residual layer: the first one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 2, and 128 convolutional kernels; the second one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 1, and 128 convolutional kernels.

[0023] The seventh layer, a fully connected layer, has 256 neurons, followed by a batch standard layer and a non-linear transformation layer;

[0024] The eighth layer, the LSTM layer: has 10 units and uses the tanh function as the activation function;

[0025] The ninth layer, the LSTM layer: has 10 units and uses the tanh function as the activation function;

[0026] The tenth layer, a fully connected layer: has 128 neurons, followed by a batch standard layer and a non-linear transformation layer;

[0027] The eleventh layer, the classification layer: the softmax classifier, classifies the results;

[0028] Each one-dimensional convolutional layer is followed by a non-linear transformation layer and a batch normalization layer.

[0029] Furthermore, the multi-objective adaptive particle swarm optimization algorithm module employs an improved particle swarm optimization algorithm to optimize the number of training iterations, learning rate, and LSTM time window size of the one-dimensional CNN-LSTM cascade model, specifically as follows:

[0030] (a.1) Initialize the particle swarm optimization algorithm parameters: population size P, maximum number of iterations t max ;

[0031] (a.2) Randomly initialize the particle swarm position r i and speed v i ;

[0032] (a.3) Calculate the fitness value F, and calculate the historical best value Lbest;

[0033] (a.4) According to the Pareto dominance principle, store the current non-dominated solution in the Archive set;

[0034] (a.5) Divide the target space into small regions using a grid, and use the number of particles contained in each region as the density information of the particles; the more particles a particle contains in its grid, the greater its density value, and vice versa, and use this to calculate the crowding of the Archive set.

[0035] (a.6) Select the particle with the lowest density in the Archive as the global optimum Gbest;

[0036] (a.7) Calculate the evolution factor f, which is the length of the sum of the connection vectors between the optimal particle and other particles divided by the sum of the distances between the optimal particle and other particles;

[0037] (a.8) Update the particle's velocity and position:

[0038] (twenty one)

[0039] (twenty two)

[0040] Inertia factor ; , The acceleration coefficient is typically taken as 2; , A random number between [0, 1]; For the updated particle velocity, The particle velocity before the update; For the updated particle positions, The particle positions before the update;

[0041] (a.9) Repeat steps (a.3)-(a.8) until the iteration stopping condition is met.

[0042] Furthermore, the nuclear power key indicator prediction and online correction module updates the one-dimensional CNN-LSTM cascade model online, periodically inputting the current data stream from the sodium iodide spectrometer into the training set, and using the new training set to modify the parameters in the one-dimensional CNN-LSTM cascade model, specifically as follows:

[0043] The nuclear power key indicator prediction and online correction module adopts an online correction strategy to perform real-time correction on the one-dimensional CNN-LSTM cascade model. By inputting abnormal data as new training data into the training set, the parameters in the one-dimensional CNN-LSTM cascade model are modified in real time using the new training set.

[0044] When the one-dimensional CNN-LSTM cascaded model is working, it acquires the predicted analysis value at time t in real time. and forecast values Calculate the analysis value MI a and forecast value MI p deviation :

[0045] ;

[0046] if Greater than a positive error tolerance :

[0047] ;

[0048] Then the predicted analysis value at time t and parameter values ​​x(t) as new training data If the model parameters are not corrected, they are added to the training dataset and the prediction model parameters are retrained to perform online correction of the prediction model; otherwise, the model parameters are considered to be accurate and no online correction is required.

[0049] The beneficial effects of this invention are as follows: This invention provides a nuclear power key indicator prediction device based on optimal deep learning to predict sodium iodide spectrometer data. It mines more distinguishable subtle features at multiple scales, resulting in higher prediction accuracy. It employs a one-dimensional CNN-LSTM cascaded model to predict key indicators for industrial-scale nuclear power, while using a novel improved particle swarm optimization algorithm to select optimization parameters. This partially overcomes the drawbacks of traditional parameter tuning methods being susceptible to human factors, and can handle rare and abnormal sodium iodide spectrometer data, making the results more robust. Furthermore, it uses an online calibration method to adjust model parameters online. An online correction module collects and trains mismatched samples, and real-time online monitoring protects nuclear power plant operations, improving the prediction efficiency and accuracy of the nuclear power key indicator prediction device during long-term online operation. Attached Figure Description

[0050] Figure 1 This is a structural diagram of a nuclear power key indicator prediction device based on optimal deep learning.

[0051] In the figure, 1-Sodium iodide spectrometer database; 2-Data acquisition and preprocessing module; 3-One-dimensional CNN-LSTM cascade model module; 4-Multi-objective adaptive particle swarm optimization algorithm module; 5-Nuclear power key indicator prediction and online correction module. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention.

[0053] The technical concept of this invention is as follows: This invention is based on an online correction nuclear power key indicator prediction device, which achieves the effect of protecting nuclear power plant operation through real-time online monitoring. The nuclear power key indicator prediction device based on optimal deep learning uses raw sodium iodide spectrometer data as input, mines more distinguishable subtle features at multiple scales, and establishes a nuclear power key indicator prediction model using an optimal deep learning algorithm, resulting in higher prediction accuracy. The multi-objective adaptive particle swarm optimization algorithm used in this patent can handle rare abnormal sodium iodide spectrometer data, making the results more robust. The nuclear power key indicator prediction and online correction module collects and trains on mismatched samples, possessing advantages such as adaptability, generalization, and agility. This solves the shortcomings of traditional nuclear power key indicator prediction devices, such as low prediction accuracy and inability to predict newly emerging sodium iodide spectrometer data anomalies during nuclear power facility operation, which only address known anomalies. This invention provides a nuclear power key indicator prediction device with strong prediction capabilities, high accuracy, strong robustness, and a combination of generalization and agility. The beneficial effects of this invention are mainly reflected in the following aspects.

[0054] Example 1

[0055] like Figure 1 As shown, a nuclear power key indicator prediction device based on optimal deep learning is disclosed. The device includes a sodium iodide spectrometer database 1, a data acquisition and preprocessing module 2, a one-dimensional CNN-LSTM cascade model module 3, a multi-objective adaptive particle swarm optimization algorithm module 4, and a nuclear power key indicator prediction and online correction module 5.

[0056] The data acquisition and preprocessing module 2 is used to acquire data from the sodium iodide spectrometer database 1. Data preprocessing is performed to obtain preprocessed data. And upload it to the one-dimensional CNN-LSTM cascaded model module 3.

[0057] The data from the sodium iodide spectrometer will be input from the sodium iodide spectrometer database. Data preprocessing is performed to obtain preprocessed data. Specifically, it includes the following sub-steps:

[0058] (a.1) Regarding the data Data cleaning involves removing duplicate and outlier information to obtain cleaned data. .

[0059] (a.2) Subsequently, the cleaned data Perform data type conversion on the cleaned data. The character-type features in the data are converted into numerical features to obtain the data. ;

[0060] (a.3) Subsequently, the data Data normalization is performed to obtain preprocessed data. The calculation formula is as follows:

[0061]

[0062] in, For data The Middle Eigenvalues ​​at; For data The smallest eigenvalue in the middle; For data The largest eigenvalue in the middle; For preprocessed data The Middle The eigenvalue at that location.

[0063] The one-dimensional CNN-LSTM cascaded model module 3 is used to process the preprocessed data through the one-dimensional CNN-LSTM cascaded model. Classify the data to obtain classification vectors. .

[0064] The one-dimensional CNN-LSTM cascaded model module 3 contains a one-dimensional CNN-LSTM cascaded model, which mines more distinguishable subtle features at multiple scales, resulting in higher prediction accuracy. The specific structure of the one-dimensional CNN-LSTM cascaded model is as follows:

[0065] The first layer is a one-dimensional convolutional layer: the kernel size is 16×1, the stride is 2, and the number of kernels is 64.

[0066] The second layer, the max pooling layer: the pooling window size is 2;

[0067] The third layer is a one-dimensional residual layer: the first one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 1, and 64 convolutional kernels; the second one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 1, and 64 convolutional kernels.

[0068] The fourth layer is a one-dimensional residual layer: the first one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 2, and a total of 64 convolutional kernels; the second one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 1, and a total of 64 convolutional kernels.

[0069] The fifth layer is a one-dimensional residual layer: the first one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 1, and 128 convolutional kernels; the second one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 1, and 128 convolutional kernels.

[0070] The sixth layer is a one-dimensional residual layer: the first one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 2, and 128 convolutional kernels; the second one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 1, and 128 convolutional kernels.

[0071] The seventh layer, a fully connected layer, has 256 neurons, followed by a batch standard layer and a non-linear transformation layer;

[0072] The eighth layer, the LSTM layer: has 10 units and uses the tanh function as the activation function;

[0073] The ninth layer, the LSTM layer: has 10 units and uses the tanh function as the activation function;

[0074] The tenth layer, a fully connected layer: has 128 neurons, followed by a batch standard layer and a non-linear transformation layer;

[0075] The eleventh layer, the classification layer: the softmax classifier, classifies the results;

[0076] Each one-dimensional convolutional layer is followed by a non-linear transformation layer and a batch normalization layer.

[0077] The core algorithm of LSTM uses three control gates to control the state of each cell, thereby controlling the unit's long short-term memory. Its specific meaning is as follows:

[0078] 2.2.1) Forget Gate. The forget gate calculates the generated memory weights. Control the state of the previous moment How many are saved to the current moment? The cell state is calculated using the following formula:

[0079] ;

[0080] in, It is the weight matrix of the forget gate. It is the hidden state of the previous moment. The network input value at the current time t. To offset the forget gate, This is the Sigmoid function.

[0081] 2.2.2) Input Gate. The input gate is responsible for generating the input weights. and the current input unit status and control the state of the input unit. How many inputs are given to the current cell state? Input weights and input unit status The calculation formula is as follows:

[0082] ;

[0083] ;

[0084] in, The input gate weight matrix, For input gate bias; Given the input state weight matrix, is the input state bias, and tanh is the hyperbolic tangent function.

[0085] 2.2.3) Generate the current cell state The current state is determined by the forget gate. Previous cell state Input gate and the current input unit state The decision is made jointly, and the calculation formula is as follows:

[0086] ;

[0087] 2.2.4) Output Gate. The output gate is responsible for generating the output weights. Control the current cell state How many are hidden layer outputs at the current moment? The calculation formula is as follows:

[0088] ;

[0089] ;

[0090] in, This is the output gate weight matrix. It is the hidden layer state of the previous moment. For the current moment Network input values, The output state bias is given by tanh, which is the hyperbolic tangent function.

[0091] The nonlinear transformation layer performs a nonlinear transformation on the features using an activation function, calculated as follows:

[0092] ;

[0093] in, The features are the input.

[0094] The batch normalization layer makes the distribution more closely match the true distribution of the data, ensuring the nonlinear expressive power of the model.

[0095] The one-dimensional residual layer consists of two one-dimensional convolutional layers. Each one-dimensional convolutional layer is followed by a nonlinear transformation layer and a batch normalization layer. There is a residual connection at both ends of the two connected one-dimensional convolutional layers. Note that when the input and output feature dimensions of the two convolutional layers change, the residual connection also needs to change the feature dimension in the same way so that the output of the residual connection is added to the output of the two one-dimensional convolutional layers.

[0096] The multi-objective adaptive particle swarm optimization algorithm module 4 is used to optimize the hyperparameters of the one-dimensional CNN-LSTM cascade model, such as the number of training iterations, learning rate, and LSTM time window size, using an improved particle swarm optimization algorithm. This partially overcomes the drawbacks of traditional parameter tuning methods being susceptible to human factors and can provide more robust results for rare and anomalous sodium iodide spectrometer data. Specifically:

[0097] (a.1) Initialize the particle swarm optimization algorithm parameters: population size P, maximum number of iterations t max ;

[0098] (a.2) Randomly initialize the particle swarm position r i and speed v i ;

[0099] (a.3) Calculate the fitness value F, and calculate the historical best value Lbest;

[0100] (a.4) According to the Pareto dominance principle, store the current non-dominated solution in the Archive set;

[0101] (a.5) Divide the target space into small regions using a grid, and use the number of particles contained in each region as the particle density information. The more particles a particle contains in its grid, the higher its density value, and vice versa. Use this to calculate the crowding level of the Archive set;

[0102] (a.6) Select the particle with the lowest density in the Archive as the global optimum Gbest;

[0103] (a.7) Calculate the evolution factor f, which is the length of the sum of the connection vectors between the optimal particle and other particles divided by the sum of the distances between the optimal particle and other particles;

[0104] (a.8) Update the particle's velocity and position:

[0105] (twenty one)

[0106] (twenty two)

[0107] Inertia factor ; , The acceleration coefficient is typically taken as 2; , A random number between [0, 1]; For the updated particle velocity, The particle velocity before the update; For the updated particle positions, The position of the particle before the update.

[0108] (a.9) Repeat steps (a.3)-(a.8) until the iteration stopping condition is met.

[0109] The nuclear power key indicator prediction and online correction module 5 is used to update the one-dimensional CNN-LSTM cascade model online. It periodically inputs the current data stream from the sodium iodide spectrometer into the training set, modifies the parameters in the one-dimensional CNN-LSTM cascade model using the new training set, and adjusts the model parameters online using an online correction method. The online correction module collects and trains on mismatched samples, and through real-time online monitoring, it protects the operation of nuclear power plants and improves the prediction efficiency and accuracy of the nuclear power key indicator prediction device for long-term online operation. Specifically, this allows the one-dimensional CNN-LSTM cascade model to adapt to changes and maintain good prediction performance.

[0110] The nuclear power key indicator prediction and online correction module 5 updates the one-dimensional CNN-LSTM cascade model online, periodically inputting the current data stream from the sodium iodide spectrometer into the training set, and using the new training set to modify the parameters in the one-dimensional CNN-LSTM cascade model, specifically:

[0111] The nuclear power key indicator prediction and online correction module 5 adopts an online correction strategy to perform real-time correction on the one-dimensional CNN-LSTM cascade model. By inputting abnormal data as new training data into the training set, the parameters in the one-dimensional CNN-LSTM cascade model are modified in real time using the new training set.

[0112] When the one-dimensional CNN-LSTM cascaded model is working, it acquires the predicted analysis value at time t in real time. and forecast values Calculate the analysis value MI a and forecast value MI p deviation :

[0113] ;

[0114] if Greater than a positive error tolerance :

[0115] ;

[0116] Then the predicted analysis value at time t and parameter values ​​x(t) as new training data If the model parameters are not corrected, they are added to the training dataset and the prediction model parameters are retrained to perform online correction of the prediction model; otherwise, the model parameters are considered to be accurate and no online correction is required.

[0117] The nuclear power key indicator prediction and online correction module 5 is used to update the one-dimensional CNN-LSTM cascade model online. It periodically inputs the current data stream from the sodium iodide spectrometer into the training set, modifies the parameters in the one-dimensional CNN-LSTM cascade model using the new training set, and adjusts the model parameters online using an online correction method. The online correction module collects and trains on mismatched samples, and through real-time online monitoring, it protects the operation of nuclear power plants and improves the prediction efficiency and accuracy of the nuclear power key indicator prediction device for long-term online operation. Specifically, this allows the one-dimensional CNN-LSTM cascade model to adapt to changes and maintain good prediction performance.

[0118] The nuclear power key indicator prediction and online correction module 5 updates the one-dimensional CNN-LSTM cascade model online, periodically inputting the current data stream from the sodium iodide spectrometer into the training set, and using the new training set to modify the parameters in the one-dimensional CNN-LSTM cascade model, specifically:

[0119] The nuclear power key indicator prediction and online correction module 5 adopts an online correction strategy to perform real-time correction on the one-dimensional CNN-LSTM cascade model. By inputting abnormal data as new training data into the training set, the parameters in the one-dimensional CNN-LSTM cascade model are modified in real time using the new training set.

[0120] When the one-dimensional CNN-LSTM cascaded model is working, it acquires the predicted analysis value at time t in real time. and forecast values Calculate the analysis value MIa and forecast value MI p deviation :

[0121] ;

[0122] if Greater than a positive error tolerance :

[0123] ;

[0124] Then the predicted analysis value at time t and parameter values ​​x(t) as new training data If the model parameters are not corrected, they are added to the training dataset and the prediction model parameters are retrained to perform online correction of the prediction model; otherwise, the model parameters are considered to be accurate and no online correction is required.

[0125] The proposed multi-objective adaptive particle swarm optimization module 4 is used to obtain the optimal weight of the new sample data x(t) in the training set to determine the magnitude of its influence on the training model results. This includes the hyperparameters of the CNN-LSTM prediction module, such as the number of training iterations, learning rate, and LSTM time window size. This allows for the rapid correction of the mismatched nuclear power key indicator prediction model, resulting in better prediction performance for nuclear power key indicators.

[0126] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A nuclear power key indicator prediction device based on optimal deep learning, characterized in that, The device includes a sodium iodide spectrometer database, a data acquisition and preprocessing module, a one-dimensional CNN-LSTM cascaded model module, a multi-objective adaptive particle swarm optimization algorithm module, and a nuclear power key indicator prediction and online correction module. The data acquisition and preprocessing module is used to process the sodium iodide spectrometer data input from the sodium iodide spectrometer database. Data preprocessing is performed to obtain preprocessed data. And upload it to the one-dimensional CNN-LSTM cascaded model module; The one-dimensional CNN-LSTM cascaded model module is used to process the preprocessed data through a one-dimensional CNN-LSTM cascaded model. Classify the data to obtain classification vectors. ; The multi-objective adaptive particle swarm optimization algorithm module is used to optimize the training iterations, learning rate, and LSTM time window size of the one-dimensional CNN-LSTM cascade model using an improved particle swarm optimization algorithm, specifically: (a.1) Initialize the particle swarm optimization algorithm parameters: population size P, maximum number of iterations t max ; (a.2) Randomly initialize the particle swarm position r i and speed v i ; (a.3) Calculate the fitness value F, and calculate the historical best value Lbest; (a.4) According to the Pareto dominance principle, store the current non-dominated solution in the Archive set; (a.5) Divide the target space into small regions using a grid, and use the number of particles contained in each region as the density information of the particles; the more particles a particle contains in its grid, the greater its density value, and vice versa, and use this to calculate the crowding of the Archive set. (a.6) Select the particle with the lowest density in the Archive as the global optimum Gbest; (a.7) Calculate the evolution factor f, which is the length of the sum of the connection vectors between the optimal particle and other particles divided by the sum of the distances between the optimal particle and other particles; (a.8) Update the particle's velocity and position: (21) (22) Inertia factor ; , The acceleration coefficient is typically taken as 2; , A random number between [0, 1]; For the updated particle velocity, The particle velocity before the update; For the updated particle positions, The particle positions before the update; (a.9) Repeat steps (a.3)-(a.8) until the iteration stopping condition is met; The nuclear power key indicator prediction and online correction module is used to update the one-dimensional CNN-LSTM cascade model online, periodically inputting the current data stream of the sodium iodide spectrometer into the training set, and using the new training set to modify the parameters in the one-dimensional CNN-LSTM cascade model.

2. The nuclear power key indicator prediction device based on optimal deep learning according to claim 1, characterized in that, The data from the sodium iodide spectrometer will be input from the sodium iodide spectrometer database. Data preprocessing is performed to obtain preprocessed data. Specifically, it includes the following sub-steps: (a.1) Regarding the data Data cleaning involves removing duplicate and outlier information to obtain cleaned data. ; (a.2) Subsequently, the cleaned data Perform data type conversion on the cleaned data. The character-type features in the data are converted into numerical features to obtain the data. ; (a.3) Subsequently, the data Data normalization is performed to obtain preprocessed data. The calculation formula is as follows: in, For data The Middle Eigenvalues ​​at; For data The smallest eigenvalue in the middle; For data The largest eigenvalue in the middle; For preprocessed data The Middle The eigenvalue at that location.

3. The nuclear power key indicator prediction device based on optimal deep learning according to claim 1, characterized in that, The one-dimensional CNN-LSTM cascaded model module includes a one-dimensional CNN-LSTM cascaded model, and the specific structure of the one-dimensional CNN-LSTM cascaded model is as follows: The first layer is a one-dimensional convolutional layer: the kernel size is 16×1, the stride is 2, and the number of kernels is 64. The second layer, the max pooling layer: the pooling window size is 2; The third layer is a one-dimensional residual layer: the first one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 1, and 64 convolutional kernels; the second one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 1, and 64 convolutional kernels. The fourth layer is a one-dimensional residual layer: the first one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 2, and a total of 64 convolutional kernels; the second one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 1, and a total of 64 convolutional kernels. The fifth layer is a one-dimensional residual layer: the first one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 1, and 128 convolutional kernels; the second one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 1, and 128 convolutional kernels. The sixth layer is a one-dimensional residual layer: the first one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 2, and 128 convolutional kernels; the second one-dimensional convolutional layer has a 16×1 convolutional kernel, a stride of 1, and 128 convolutional kernels. The seventh layer, a fully connected layer, has 256 neurons, followed by a batch standard layer and a non-linear transformation layer; The eighth layer, the LSTM layer: has 10 units and uses the tanh function as the activation function; The ninth layer, the LSTM layer: has 10 units and uses the tanh function as the activation function; The tenth layer, a fully connected layer: has 128 neurons, followed by a batch standard layer and a non-linear transformation layer; The eleventh layer, the classification layer: the softmax classifier, classifies the results; Each one-dimensional convolutional layer is followed by a non-linear transformation layer and a batch normalization layer.

4. The nuclear power key indicator prediction device based on optimal deep learning according to claim 1, characterized in that, The nuclear power key indicator prediction and online correction module updates the one-dimensional CNN-LSTM cascade model online, periodically inputting the current data stream from the sodium iodide spectrometer into the training set, and using the new training set to modify the parameters in the one-dimensional CNN-LSTM cascade model, specifically: The nuclear power key indicator prediction and online correction module adopts an online correction strategy to perform real-time correction on the one-dimensional CNN-LSTM cascade model. By inputting abnormal data as new training data into the training set, the parameters in the one-dimensional CNN-LSTM cascade model are modified in real time using the new training set. When the one-dimensional CNN-LSTM cascaded model is working, it acquires the predicted analysis value at time t in real time. and forecast values Calculate the analysis value MI a and forecast value MI p deviation : ； if Greater than a positive error tolerance : ； Then the predicted analysis value at time t and parameter values ​​x(t) as new training data If the model parameters are not corrected, they are added to the training dataset and the prediction model parameters are retrained to perform online correction of the prediction model; otherwise, the model parameters are considered to be accurate and no online correction is required.

Images