Deep learning-driven accident parameter intelligent correction and prediction method

Abstract

The invention discloses a deep learning-driven accident parameter intelligent correction and prediction method, and the method comprises the steps: obtaining sensor data of a nuclear power plant, carrying out the screening and analysis of the sensor data, and obtaining important operation parameters and high-correlation parameters; performing data reconstruction on the high-correlation parameters through a data reconstruction model to obtain reconstructed important operation parameters; judging the important operation parameters according to the reconstructed important operation parameters, and obtaining a correction result of the important operation parameters based on a judgment result; and predicting the correction result of the important operating parameters through a data prediction model to obtain a prediction result of the important operating parameters. Wherein the data reconstruction model and the data prediction model adopt deep learning models.

Description

Technical Field

[0001] The present invention belongs to the technical field of safe operation of nuclear power plants, and in particular relates to a deep learning-driven intelligent correction and prediction method for accident parameters. Background Art

[0002] As an efficient and clean energy source, nuclear energy occupies a vital position in the global energy mix. The safe operation of nuclear power plants is not only crucial to the stability of energy supply but also directly impacts public safety and environmental protection. Nuclear power plant operations rely on precise monitoring and real-time control of key reactor parameters (such as nuclear power, core water level, pressurizer water level, and coolant temperature). However, complex operating environments (such as high temperature, high pressure, and radioactivity) can cause sensor data to be distorted or fail, thus affecting accurate assessments of the plant's operating status. Traditional data correction and prediction methods often rely on physical models and empirical formulas, but these methods have limitations when dealing with complex nonlinear relationships and dynamic changes. Consequently, accurate data correction and prediction of nuclear power plant operating data are impossible. Summary of the Invention

[0003] In order to solve the above technical problems, the present invention proposes a deep learning driven intelligent correction and prediction method for accident parameters to solve the problems existing in the above-mentioned prior art.

[0004] To achieve the above objectives, the present invention provides a deep learning-driven intelligent correction and prediction method for accident parameters, comprising:

[0005] Acquire sensor data from the nuclear power plant, screen and analyze the sensor data to obtain important operating parameters and highly relevant parameters;

[0006] Reconstructing data according to the highly correlated parameters using a data reconstruction model to obtain reconstructed important operating parameters;

[0007] Judging the important operating parameters according to the reconstructed important operating parameters, and obtaining correction results of the important operating parameters based on the judgment results;

[0008] The correction results of the important operating parameters are predicted by a data prediction model to obtain prediction results of the important operating parameters. The data reconstruction model and the data prediction model adopt a deep learning model.

[0009] Optionally, the sensor data of the nuclear power plant include: nuclear power, core water level, pressurizer water level, pressurizer pressure, coolant temperature, coolant flow, containment pressure and the total number of corresponding sensors, sensor channel number and sensor name.

[0010] Optionally, before screening and analyzing the sensor data, the following steps may be further performed:

[0011] The number of measurement values of the sensor data is obtained, and the measurement value data and the total number of sensors are judged. When the measurement value data and the total number of sensors are consistent, the sensor data is normalized. Otherwise, the invalid sensor name corresponding to the missing measurement value data is marked.

[0012] Optionally, the process of screening and analyzing the sensor data includes:

[0013] The sensor data is screened to obtain important operating parameters, and a correlation analysis is performed between the sensor data and the important operating parameters to obtain highly correlated operating parameters.

[0014] Optionally, the correlation analysis is performed using the Spearman correlation coefficient.

[0015] Optionally, before reconstructing the data according to the highly correlated parameters, the method further includes:

[0016] Training the data reconstruction model, wherein sample data is obtained, wherein the sample data includes historical real data or nuclear island operation data generated by simulation;

[0017] Dividing the sample data into a training set and a test set;

[0018] Build data reconstruction model and parameter prediction model respectively based on deep learning model;

[0019] The data reconstruction model takes highly correlated parameters as input and important operating parameters as output. The mean square error and root mean square error are used as the loss function of the deep neural network model for training based on the training set and test set.

[0020] For the parameter prediction model, the important operating parameters in the training set and the test set are time-series segmented. The time-series segmentation results under the first several time steps are used as input, and the time-series segmentation results under the last several time steps are used as output. The parameter prediction model is trained based on the time-series segmentation results.

[0021] Optionally, the process of obtaining the calibration results of important operating parameters includes:

[0022] The important operating parameters are judged according to the reconstructed important operating parameters, wherein when the percentage difference between the important operating parameters and the reconstructed important operating parameters is greater than the judgment threshold, the reconstructed important operating parameters are used as the correction results of the important operating parameters; otherwise, the important operating parameters and the reconstructed important operating parameters are weighted averaged to obtain the important operating parameters and the reconstructed important operating parameters;

[0023] The weights in the weighted averaging process are determined based on the test error of the data reconstruction model and the confidence of the sensor.

[0024] Optionally, the important operating parameter is the core water level.

[0025] Compared with the prior art, the present invention has the following advantages and technical effects:

[0026] By combining deep learning models with correlation analysis, it is possible to reconstruct easily distorted operating parameters with high precision, and maintain high fitting accuracy even under complex operating conditions; based on the dynamic comparison between the reconstructed value and the real-time sensor data, it is possible to judge the degree of data distortion in real time, and manage the distorted data through weighted averaging or replacement, thereby improving the reliability and availability of the data; utilizing the time series prediction capability of deep neural networks, it is possible to predict important future parameters and provide support for accident warning and emergency response of nuclear power plants; through the model training and verification process, it is possible to automatically adjust the model structure and parameters to adapt to different operating conditions and sensor characteristics, and has good generalization capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0028] Figure 1 This is a schematic diagram of the implementation process of the deep learning-driven accident parameter intelligent correction and prediction method according to an embodiment of the present invention;

[0029] Figure 2 A comparison diagram of the core water level reconstruction results based on the data reconstruction model according to an embodiment of the present invention;

[0030] Figure 3 This is a comparison diagram of the core water level after real-time distortion control and the actual water level in an embodiment of the present invention;

[0031] Figure 4 This is a comparison chart of the prediction result of the core water level after 30 seconds based on the parameter prediction model in an embodiment of the present invention and the actual water level. DETAILED DESCRIPTION

[0032] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0033] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0034] Deep learning technology demonstrates powerful capabilities in data reconstruction, anomaly detection, and predictive modeling, offering new solutions for addressing sensor data distortion in nuclear power plants. However, applying deep learning to parameter correction and prediction for nuclear power plant accidents still faces numerous challenges, such as model accuracy, reliability, and dynamic adaptability.

[0035] This paper addresses the issue of sensor data distortion in nuclear power plants by proposing a deep learning-based intelligent correction and prediction method for accident parameters. This technology uses deep neural network data reconstruction and prediction models to dynamically correct easily distorted operating parameters and predict future states. This method aims to improve the reliability and prediction accuracy of nuclear power plant operating parameters, providing strong support for their safe operation.

[0036] like Figure 1 As shown, this embodiment provides a deep learning-driven intelligent correction and prediction method for accident parameters, including:

[0037] 1. Read the sensor data of the nuclear power plant. The sensor data is the operating process parameters of the nuclear power plant reactor. The operating process parameters include nuclear power, core water level, pressurizer water level, pressurizer pressure, coolant temperature, coolant flow, containment pressure, etc., as well as the total number of corresponding sensors, sensor channel number and name of each sensor;

[0038] 2. Read the current measurement values and their location information from all sensors; check whether the number of measurement values matches the total number of sensors. If so, normalize the obtained sensor data; if not, mark the failed sensor name according to the missing value location;

[0039] 3. Using a data reconstruction model based on a deep neural network and combined with correlation analysis, we use highly correlated parameters as input to reconstruct the data of important and easily distorted operating parameters and obtain the reconstructed value of the parameter;

[0040] 4. The real-time sensor data is judged based on the parameter reconstruction value. The percentage difference σ% between the reconstructed value and the real-time sensor data is used as the judgment condition, and a judgment threshold N% is set. If the percentage difference σ% is greater than the judgment threshold N%, the sensor data is judged to be completely distorted, and the sensor data is replaced with the parameter reconstruction value as the distortion-treated data. If the percentage difference σ% is less than or equal to the judgment threshold N%, the sensor data is judged to be valid but noisy. The reconstructed value and the sensor data are weighted averaged, and the resulting data is used as the distortion-treated data. The weights of the reconstructed value and the sensor data are determined by the reconstruction model test error and the confidence level in the sensor.

[0041] 5. Utilize a data prediction model based on a deep neural network and the data obtained through data distortion control to predict important and easily distorted operating parameters. The model takes the current time and the previous n-1 time steps as input and generates predicted data m time steps later. The number of input time steps n and the output prediction time step m are determined by actual needs and model training results.

[0042] 6. Output the comparison of data before and after distortion treatment and the time and location of sensor anomaly, and output the reconstructed data for detection and comparison; output parameter prediction results.

[0043] Specifically, when the obtained sensor data is normalized, the normalization calculation formula is:

[0044]

[0045] Where x min Sensor instrument lower limit value, x max is the upper limit value of the sensor instrument, x * The data is normalized by the sensor.

[0046] Specifically, the training method of the data reconstruction model and parameter prediction model based on the deep neural network includes:

[0047] 1. Input data: import historical real data or nuclear island operation data generated by the simulator into the deep neural network;

[0048] 2. Based on engineering practice, mechanism analysis and national standards for nuclear power plant parameters and sensors, we analyze and summarize the important parameters for evaluating the operating status of nuclear power plants and the important parameters that are prone to sensor distortion;

[0049] 3. Based on correlation analysis, perform correlation analysis on important parameters of the sensor that are easily distorted, and screen out highly correlated parameters of the important parameters of the sensor that are easily distorted;

[0050] 4. Data preprocessing and data partitioning: The preprocessing method is consistent with the normalization and missing value processing method in step 2, and the data is divided into a training set and a test set, where the ratio of the training set to the test set is 8:2;

[0051] 5. Based on the deep neural network, the data reconstruction model and parameter prediction model are constructed respectively. For the data reconstruction model, the highly correlated parameters of the sensor's important parameters that are easily distorted are used as input, and the important parameters that are easily distorted are used as output. The mean square error (MSE) and root mean square error (RMSE) are selected as the loss function of the deep neural network model for training. For the parameter prediction model, the time series data is segmented, and the segmented input is {x t-n+1 ,…,x t-2 ,xt-1 ,x t}, the output is x t+m , where n represents the number of input parameter nodes used for prediction, and m represents the number of time steps for prediction. That is, n data at time t and the previous n-1 times are used to predict the data at time t+m. The selection of n and m is determined by actual needs and model training results.

[0052] 6. Verify and check the model data reconstruction model and parameter prediction model, and judge the model accuracy and generalization ability by the loss value and determination coefficient of the model on the test set. If the loss value and determination coefficient do not meet the requirements, reselect the model architecture and model structure configuration, analyze the model error, optimize the hyperparameters and adjust the model structure, and re-verify and check the model until the model meets the requirements and save the model parameters.

[0053] As some embodiments, in the correlation analysis, the correlation calculation method includes Pearson correlation coefficient calculation, Spearman correlation coefficient calculation and Kendall correlation coefficient calculation.

[0054] The above technical solution is described in detail through relevant contents, wherein the intelligent correction and prediction method of the present invention includes the following contents:

[0055] 1. Read the sensor data of the nuclear power plant. The sensor data is the operating process parameters of the nuclear power plant reactor. The operating process parameters include nuclear power, core water level, pressurizer water level, pressurizer pressure, coolant temperature, coolant flow, containment pressure, etc., as well as the total number of corresponding sensors, sensor channel number and name of each sensor;

[0056] For example, the operating process parameters read by the embodiment of the present invention include nuclear power, core water level, pressurizer water level, pressurizer pressure, coolant temperature, coolant flow rate, and containment pressure. The core water level is selected as the object of data distortion control and parameter prediction. Data distortion control and parameter prediction are performed on the core water level in the event of a large break loss of coolant accident.

[0057] 2. Read the current measurement values and their location information from all sensors; check whether the number of measurement values matches the total number of sensors. If so, normalize the obtained sensor data; if not, mark the failed sensor name according to the missing value location;

[0058] In this embodiment, when the obtained sensor data is normalized, the normalization calculation formula is:

[0059]

[0060] Where x min is the lower limit value of the sensor instrument, xmax is the upper limit value of the sensor instrument, x * In the embodiment of the present invention, the lower limit of the core water level is 0 and the upper limit is 7.85m.

[0061] 3. Using a data reconstruction model based on a deep neural network and combined with correlation analysis, we use highly correlated parameters as input to reconstruct the data of important and easily distorted operating parameters and obtain the reconstructed value of the parameter;

[0062] In this embodiment, the core water level and other parameters are analyzed for correlation. The highly correlated parameters calculated are steam generator pressure, containment pressure, pressurizer water level, reactor outlet flow rate and nuclear power. These parameters are used as input parameters of the data reconstruction model to reconstruct the core water level. 2 Under the large break loss of coolant accident, the core water level reconstruction results are as follows: Figure 2 shown by Figure 2 It can be seen that for the large break loss of coolant accident, the data reconstruction model can reconstruct the low water level data when the core is exposed well. For the complex fluctuation of the core water level during the long cooling period, although there is a certain error, the maximum error is within 10%; the determination coefficient R between the reconstructed value and the true value is 2 It reaches 0.9909, indicating that the overall fitting effect is very good.

[0063] 4. The real-time sensor data is judged based on the parameter reconstruction value. The percentage difference σ% between the reconstructed value and the real-time sensor data is used as the judgment condition, and a judgment threshold N% is set. If the percentage difference σ% is greater than the judgment threshold N%, the sensor data is judged to be completely distorted, and the sensor data is replaced with the parameter reconstruction value as the distortion-treated data. If the percentage difference σ% is less than the judgment threshold N%, the sensor data is judged to be valid but noisy. The reconstructed value and the sensor data are weighted averaged, and the resulting data is used as the distortion-treated data. The weights of the reconstructed value and the sensor data are determined by the reconstruction model test error and the confidence level in the sensor.

[0064] In this embodiment, the judgment threshold is set to 20%. When the percentage difference between the reconstructed value and the real-time sensor data is greater than the threshold, the reconstructed value is used as the governance value; when the percentage difference between the reconstructed value and the real-time sensor data is less than the threshold, the reconstructed value and the sensor data are weighted averaged as the governance value, and the weight of the reconstructed value is determined by the coefficient R. 2 The square of the given (ie 0.9909 2 ), the weight of the reconstruction value is set to 0.982, and the weight of the sensor data is set to 0.018 (obtained by 1-reconstruction value weight). The obtained governance results are as follows Figure 3As shown, the overall data management is relatively ideal, among which there are some individual data with large errors, but they are concentrated in the long-term cooling period, the core water level fluctuates greatly, and the fitting is difficult.

[0065] 5. Utilize a data prediction model based on a deep neural network and the data obtained through data distortion control to predict important and easily distorted operating parameters. The model takes the current time and the previous n-1 time steps as input and generates predicted data m time steps later. The number of input time steps n and the output prediction time step m are determined by actual needs and model training results.

[0066] In this embodiment, the input time step is selected as 20 steps, and the output prediction time step is selected as 60 steps, that is, the core water level at the current time and the previous 20 time steps is used to predict the core water level 60 time steps later; since the time step of the used data is 0.5s and the accident time is 500s, the core water level data at 500s and the previous 10s is used to predict the core water level 30s later as the starting point, and then recursively predicts it. The prediction results are as follows: Figure 4 As shown in the figure, the core water level prediction effect is relatively good during the core exposure stage at the early stage of a large break loss of coolant accident, but there are errors in the subsequent re-flooding and long-term cooling stages. Especially in the long-term cooling stage, it is difficult to predict too detailed fluctuations, and the core water level can only be predicted to be at a certain level. However, overall, the expected effect is achieved.

[0067] 6. Output the comparison of data before and after distortion treatment and the time and location of sensor anomaly, and output the reconstructed data for detection and comparison; output parameter prediction results.

[0068] In this embodiment, if Figure 1 As shown, the training method of the data reconstruction model and parameter prediction model based on the deep neural network includes:

[0069] 1. Input data: import historical real data or nuclear island operation data generated by the simulator into the deep neural network;

[0070] For example, the training data of the embodiment of the present invention is based on RELAP5 to generate large break loss of coolant accident data, and 0.1-0.5m 2 A total of 5 sets of breach area data were used as training samples;

[0071] 2. Based on engineering practice, mechanism analysis and national standards for nuclear power plant parameters and sensors, we analyze and summarize the important parameters for evaluating the operating status of nuclear power plants and the important parameters that are prone to sensor distortion;

[0072] For example,

[0073] The specific steps for determining the corresponding important parameters through mechanism analysis are as follows:

[0074] Select the research object and accident type: Select a reactor in the historical data as the research object, identify the specific accident type, and analyze the state of the reactor / nuclear power plant over time after the accident.

[0075] Analyze the accident process and phenomena: Based on the above-mentioned time evolution status, different stages are divided, and each stage of a specific accident type (blowdown stage, refilling stage, reflooding stage, and long-term core cooling stage) is analyzed in detail to clarify the physical phenomena and change laws of each stage.

[0076] Identify important parameters and easily distorted parameters: Identify significant parameters in the evolution of physical phenomena and change laws: Important parameters: parameters that directly or indirectly affect the core state and radioactive leakage during the accident process; Easy to distort parameters: important parameters that easily lead to inaccurate or distorted readings of measuring equipment or detectors due to their drastic changes or fluctuations during the measurement or monitoring process.

[0077] The embodiment of the present invention uses a reactor in historical data as an actual engineering object to analyze the process of loss of coolant accident (LOCA). The large break loss of coolant accident is divided into the blowdown stage, the refilling stage, the reflooding stage, and the long-term cooling stage. During the accident process, the core state and the fuel element temperature are strongly coupled. The rapid loss of coolant in the blowdown stage causes the core water level to drop sharply, and the fuel elements heat up rapidly due to the loss of cooling; in the refilling stage, the coolant intermittently contacts the core, causing violent temperature fluctuations; in the reflooding stage, the water level continues to rise to restore the cooling capacity, and the fuel temperature gradually drops. Since the fuel element cladding temperature directly reflects the integrity of the core (a zirconium-water reaction will occur if it exceeds 1204°C), but it cannot be monitored in real time, the core water level becomes the core indicator for judging the coolant coverage status. This parameter indirectly reflects the core residual heat removal efficiency by characterizing the degree of coolant coverage of the fuel. Its violent fluctuation characteristics (such as the ±1m / s water level fluctuation in the refilling stage) can easily lead to measurement distortion;

[0078] Analysis shows that the core water level parameter has a direct impact on the fuel element temperature, and the fuel element temperature reflects whether the fuel element is damaged, which is crucial for operators to judge. However, the fuel element temperature cannot be directly measured, so the core water level data is very important for accident judgment. During the loss of coolant accident process, this parameter fluctuates greatly and is a parameter that is easily distorted. Combined with the national standard GB / T 13627-2010, specific criteria for determining nuclear power plant accident monitoring variables are given. Important operating process parameters are selected, including nuclear power, core water level, pressurizer water level, pressurizer pressure, coolant temperature, coolant flow rate, and containment pressure. The core water level is selected as the object of data distortion control and parameter prediction.

[0079] 3. Based on the correlation analysis, the important parameters of the sensor that are easily distorted are analyzed to screen out the parameters with high correlation with the important parameters that are easily distorted. In this embodiment, the Spearman correlation coefficient is selected as the correlation calculation method, and its expression is:

[0080]

[0081] Where n represents the number of data, d i Indicates the difference between two data orders.

[0082] d i = rg(X i )-rg(Y i )

[0083] rg(X i ) means that in the sequence {X1,X2,…,X n} in ascending order.

[0084] The Spearman correlation coefficient also ranges from [-1,1]. The closer the value is to ±1, the stronger the monotonic correlation between the two sequences.

[0085] For example, the calculation results of the correlation between the core water level and other parameters in the embodiment of the present invention are shown in Table 1:

[0086] Table 1

[0087]

[0088]

[0089] 4. Data preprocessing and data partitioning: The preprocessing method is the same as step 2, and the data is divided into a training set and a test set. In this embodiment, the ratio of the training set to the test set is 8:2.

[0090] 5. Based on the deep neural network, the data reconstruction model and parameter prediction model are constructed respectively. For the data reconstruction model, the highly correlated parameters of the sensor's important parameters that are easily distorted are used as input, and the important parameters that are easily distorted are used as output. The mean square error (MSE) and root mean square error (RMSE) are selected as the loss function of the deep neural network model for training. For the parameter prediction model, the time series data is segmented, and the segmented input is {x t-n+1 ,…,x t-2 ,x t-1 ,x t}, the output is x t+m , where n represents the number of input parameter nodes used for prediction, and m represents the number of time steps for prediction. That is, n data at time t and the previous n-1 times are used to predict the data at time t+m. The selection of n and m is determined by actual needs and model training results.

[0091] For example, in the embodiment of the present invention, the data reconstruction model is constructed using a BP neural network, which includes three hidden layers. The activation function uses the ReLU function, and the optimizer uses the Adam method to accelerate convergence and improve training efficiency. Each time a hidden layer is passed, a function transformation is performed:

[0092] y=ReLU(Wx+b)

[0093] Where W and b represent the hidden layer weight matrix and bias vector respectively, and the ReLU function is defined as:

[0094] ReLU(x)=max(0,x)

[0095] In the embodiment of the present invention, the parameter prediction model is constructed using an LSTM neural network, which includes three hidden layers and one fully connected layer. The optimizer uses the Adam method to accelerate convergence and improve training efficiency. The mathematical expression of LSTM can be simplified as follows:

[0096] h t =LSTM(x t ,h t-1 )

[0097] y=W·h T +b

[0098] Among them, h T represents the hidden state at the t-th time step, W and b are the weight and bias of the fully connected layer, respectively.

[0099] In the embodiment of the present invention, the mean squared error (MSE) is used as the loss function, and the loss function is defined as:

[0100]

[0101] Among them, y i is the true value, is the predicted value of the model, and n is the number of samples.

[0102] 6. Verify and check the model data reconstruction model and parameter prediction model, and judge the model accuracy and generalization ability by the loss value and determination coefficient of the model on the test set. If the loss value and determination coefficient do not meet the requirements, reselect the model architecture and model structure configuration, analyze the model error, optimize the hyperparameters and adjust the model structure, and re-verify and check the model until the model meets the requirements and save the model parameters.

[0103] For example, in the embodiment of the present invention, the test set is input into the trained deep neural network model for testing, and the model output and the true value are evaluated. The evaluation criteria include the root mean square error RMSE, the determination coefficient R 2 In this embodiment, the evaluation criteria expressions are:

[0104]

[0105] Among them, n is the number of test points, y i represents the true value at the i-th sample, is the model prediction value at this point, is the mean of the true values. 2 RMSE is used to evaluate the prediction accuracy of the model, R 2 The closer the value is to 1 or the smaller the RMSE value is, the higher the accuracy of the model prediction is.

[0106] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A deep learning driven intelligent correction and prediction method for accident parameters, characterized in that: include: Acquire sensor data from the nuclear power plant, screen and analyze the sensor data to obtain important operating parameters and highly relevant parameters; Reconstructing data according to the highly correlated parameters using a data reconstruction model to obtain reconstructed important operating parameters; Judging the important operating parameters according to the reconstructed important operating parameters, and obtaining correction results of the important operating parameters based on the judgment results; The correction results of the important operating parameters are predicted by a data prediction model to obtain prediction results of the important operating parameters. The data reconstruction model and the data prediction model adopt a deep learning model.

2. The method according to claim 1, characterized in that The sensor data of the nuclear power plant include: nuclear power, core water level, pressurizer water level, pressurizer pressure, coolant temperature, coolant flow, containment pressure and the total number of corresponding sensors, sensor channel numbers and sensor names.

3. The method according to claim 2, characterized in that Before screening and analyzing the sensor data, the method further includes: The number of measurement values of the sensor data is obtained, and the measurement value data and the total number of sensors are judged. When the measurement value data and the total number of sensors are consistent, the sensor data is normalized. Otherwise, the invalid sensor name corresponding to the missing measurement value data is marked.

4. The method according to claim 1, wherein The process of screening and analyzing the sensor data includes: The sensor data is screened to obtain important operating parameters, and a correlation analysis is performed between the sensor data and the important operating parameters to obtain highly correlated operating parameters.

5. The method according to claim 1, wherein The correlation analysis was performed using the Spearman correlation coefficient.

6. The method according to claim 1, characterized in that Before reconstructing the data according to the highly correlated parameters, the following steps are further included: Training the data reconstruction model, wherein sample data is obtained, wherein the sample data includes historical real data or nuclear island operation data generated by simulation; Dividing the sample data into a training set and a test set; Build data reconstruction model and parameter prediction model respectively based on deep learning model; The data reconstruction model takes highly correlated parameters as input and important operating parameters as output. The mean square error and root mean square error are used as the loss function of the deep neural network model for training based on the training set and test set. For the parameter prediction model, the important operating parameters in the training set and the test set are time-series segmented. The time-series segmentation results under the first several time steps are used as input, and the time-series segmentation results under the last several time steps are used as output. The parameter prediction model is trained based on the time-series segmentation results.

7. The method according to claim 1, characterized in that The process of obtaining the calibration results of important operating parameters includes: The important operating parameters are judged according to the reconstructed important operating parameters, wherein when the percentage difference between the important operating parameters and the reconstructed important operating parameters is greater than the judgment threshold, the reconstructed important operating parameters are used as the correction results of the important operating parameters; otherwise, the important operating parameters and the reconstructed important operating parameters are weighted averaged to obtain the important operating parameters and the reconstructed important operating parameters; The weights in the weighted averaging process are determined based on the test error of the data reconstruction model and the confidence of the sensor.

8. The method according to claim 1, characterized in that The important operating parameter is the core water level.

Images