A method for nuclear reactor state synchronization based on multi-source information reverse mapping and system

Abstract

This invention discloses a digital twin architecture for a nuclear reactor, relating to a research and system for a nuclear reactor state synchronization method based on multi-source information reverse mapping. The method includes the following steps: S101, acquiring and integrating the operating condition data of the nuclear reactor to form a structured dataset for training a machine learning model; S201, training a regression prediction model that characterizes the mapping relationship between measurable and unmeasurable parameters based on the training dataset constructed in S101; S301, integrating the machine learning model trained in S201 into the reactor's online monitoring system. This invention, through a data-driven approach and utilizing readily available sensor data, achieves rapid, real-time prediction of key unmeasurable parameters. It boasts low computational cost and strong real-time performance, providing a novel and efficient technical means to ensure the safe operation of the reactor.

Description

Technical Field

[0001] This invention belongs to the field of nuclear reactor digital twin technology, specifically, it relates to a research and system for a nuclear reactor state synchronization method based on multi-source information reverse mapping. Background Technology

[0002] The safe and stable operation of nuclear reactors is the core of nuclear energy applications. To ensure safety, it is necessary to monitor the key state parameters inside the reactor in real time and accurately. Among them, the center temperature of the fuel pellets (referred to as fuel temperature) is one of the most important parameters of the reactor. It is directly related to the integrity of the fuel elements and is a key limiting condition in safety analysis. However, due to the extreme environment of high temperature, high pressure, and strong radiation inside the reactor core, it is extremely difficult to directly, continuously, and online measure the temperature inside the fuel rods using sensors. Currently, the methods for obtaining fuel temperature mainly rely on complex thermal-hydraulic calculation programs. These programs require a huge amount of computation, making it difficult to meet the needs of real-time online monitoring, and the calculation models themselves may also have certain theoretical deviations. Therefore, how to efficiently and accurately obtain key but indirectly measurable parameters such as fuel temperature is a technical problem that urgently needs to be solved in the field of reactor condition monitoring.

[0003] In view of this, the present invention is proposed. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a research and system for a nuclear reactor state synchronization method based on multi-source information reverse mapping, which solves the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the basic concept of the technical solution adopted by the present invention is as follows: A method for nuclear reactor state synchronization based on multi-source information reverse mapping includes the following steps: S101. Acquire and integrate the operating condition data of the nuclear reactor to form a structured dataset for training machine learning models; S101 includes the following steps: S1011. The reactor instrumentation system collects directly measurable parameters under different operating conditions, including core coolant outlet temperature, reactor power, coolant flow rate, and control rod position. S1012. Using a verified high-precision physical simulation calculation program, simulate and calculate the working conditions corresponding to the data collected in S1011 to generate the corresponding fuel temperature value. S1013. Associate and match the input feature data in S1011 with the output feature data in S1012 to form an original dataset containing input-output pairs, and perform data cleaning and normalization preprocessing on the original dataset. S201. Based on the training dataset constructed in S101, train a regression prediction model that can characterize the mapping relationship between measurable parameters and unmeasurable parameters. S201 includes the following steps: S2011. Based on the characteristics of the problem, select and construct a machine learning model, wherein the machine learning model is selected from the long short-term memory model, deep neural network, support vector regression or gradient boosting tree model; S2012. Divide the preprocessed dataset in S1013 into a training set, a validation set, and an independent test set. Use the training set to iteratively train the model built in S2011. Adjust the internal parameters of the model through optimization algorithms to minimize the mean square error between the predicted value and the true value. S2013. Use an independent test set to evaluate the performance of the trained model, verify its prediction accuracy and generalization ability, and ensure that it meets the engineering requirements for online monitoring. S301. Integrate the machine learning model trained in S201 into the reactor's online monitoring system; S301 includes the following steps: S3011. The machine learning model and all its parameters that have been validated in S2013 are serialized and converted into a file format that can be persistently stored to form an independent and portable model file; the parameters include the weights and biases of the neural network; the file format is selected from ONNX, HDF5 or PMML format. S3012. Load the model file generated in S3011 into the target server of the reactor online monitoring system; the online monitoring system is a DCS system or a SCADA system; in the software architecture of the monitoring system, configure a data input interface to receive real-time measurable parameters, and configure a data output interface to forward the model prediction results to the monitoring interface or alarm system. S3013. Before the model is officially put into real-time monitoring, the model integrated into the monitoring system is functionally verified by simulating input or playing back historical data. After confirming that the model can load correctly, stably receive input data, output calculation results within a specified time and the results are consistent with the offline test, the model service is activated to start processing real-time online data. S401. During reactor operation, the values ​​of unmeasurable parameters are predicted online using deployed models and real-time sensor data to achieve state synchronization. S401 includes the following steps: S4011. Obtain the current values ​​of parameters such as core coolant outlet temperature and reactor power in real time and continuously from the sensor network of the operating nuclear reactor; S4012. The real-time parameter values ​​obtained in S4011 are used as inputs and fed into the model service deployed in S301 for forward propagation calculation. S4013. The model outputs the predicted real-time fuel temperature value, which can be used for operator interface display, safety limit monitoring and alarm system, thereby completing the status synchronization.

[0006] A nuclear reactor state synchronization system based on multi-source information reverse mapping, applying any of the above-mentioned nuclear reactor state synchronization methods, includes an offline model training module, a data acquisition module, and a state synchronization module connected in sequence. The offline model training module is used to perform steps S101 and S201, collect and generate data to build a training dataset, train and validate the machine learning model based on the dataset, and finally generate a deployable prediction model file. The offline model training module constructs a machine learning model that includes three core functions: data preprocessing, model training, and model validation. The data preprocessing function uses a normalization equation to process the collected and generated data, scaling the data to the [0,1] interval. The processing equation is as follows:

[0007] In the formula, The data is after normalization; This is the original data; , These are the minimum and maximum values ​​of the original data set, respectively. The model training function uses machine learning theory to build and train a model capable of regression prediction, adjusting network weights to minimize the mean squared error between the fuel temperature predicted by the model and the actual fuel temperature in the training data. The model validation function uses an independent test dataset to evaluate the trained deep neural network regression model. After calculating the mean squared error of the model, the model validation function can determine whether the model can be used for online deployment based on whether the mean squared error meets the preset accuracy requirements (e.g., less than 0.5). The data acquisition module is used to execute the steps of S4011, connect to the sensor system of the reactor, and acquire the measured values ​​of measurable parameters arranged at various locations in the reactor in real time. The measurable parameters acquired by the data acquisition module from the sensors include core coolant outlet temperature, reactor power, and core coolant inlet temperature. The state synchronization module is used to execute steps S4012 and S4013, and has a prediction model generated by the offline model training module. This module is used to receive real-time measurement values ​​from the data acquisition module as input, and output the predicted unmeasurable parameter values ​​through model calculation to achieve state synchronization. The operation of the state synchronization module is as follows: A set of real-time data, such as the core power value and the inlet and outlet temperature values ​​of the core coolant measured by the sensor, is substituted into the data normalization equation to obtain the normalized real-time state vector. The real-time state vector is input into the deployed machine learning model, and the normalized fuel temperature prediction value is obtained through forward propagation calculation of the model. The normalized fuel temperature prediction is substituted into the inverse function of the normalization equation for calculation to obtain the final, physically meaningful, and accurately synchronized fuel temperature value, which is then output to the monitoring system.

[0008] By adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art. Of course, any product implementing the present invention does not necessarily need to achieve all of the following advantages at the same time: This invention provides a method and system for synchronizing the state of a nuclear reactor based on multi-source information reverse mapping. By using sensors placed at various locations in the reactor to obtain the real-time state of the reactor, and on this basis, using a data-driven machine learning model, it achieves real-time and accurate synchronization of key but unmeasurable internal parameters (such as fuel temperature), providing an efficient and reliable new approach for safe monitoring and optimized operation of the reactor.

[0009] The specific embodiments of the present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0010] The accompanying drawings described below are merely some embodiments. Those skilled in the art can obtain other drawings based on these drawings without any creative effort. In the drawings: Provided that creative work is undertaken, other accompanying drawings may also be obtained based on the provided drawings.

[0011] Figure 1 This is a flowchart of a nuclear reactor state synchronization method based on multi-source information reverse mapping according to the present invention; Figure 2 This is a flowchart of the S101 method in this invention; Figure 3 This is a flowchart of the S201 method in this invention; Figure 4 This is a flowchart of the S301 method in this invention; Figure 5This is a flowchart of the S401 method in this invention; Figure 6 This is a block diagram of a nuclear reactor state synchronization system based on multi-source information reverse mapping according to the present invention. Figure 7 This is a structural diagram of the machine learning algorithm used for state synchronization in this invention; Figure 8 The performance indicators of the integrated pressurized water reactor model after using the state synchronization method of this invention.

[0012] It should be noted that these accompanying drawings and textual descriptions are not intended to limit the scope of the invention in any way, but rather to illustrate the concept of the invention to those skilled in the art by referring to specific embodiments. Detailed Implementation

[0013] The invention will now be described in further detail with reference to the accompanying drawings.

[0014] Please see Figure 1-8 As shown, this embodiment provides a nuclear reactor state synchronization method based on multi-source information reverse mapping, including the following steps: S101. Acquire and integrate the operating condition data of the nuclear reactor to form a structured dataset for training machine learning models; S201. Based on the training dataset constructed in S101, train a regression prediction model that can characterize the mapping relationship between measurable parameters and unmeasurable parameters. S301. Integrate the machine learning model trained in S201 into the reactor's online monitoring system; S401. During reactor operation, the values ​​of unmeasurable parameters are predicted online using deployed models and real-time sensor data to achieve state synchronization.

[0015] In one specific embodiment, refer to Figure 2 As shown, S101 includes the following steps: S1011. Collect directly measurable parameters under different operating conditions through the reactor instrumentation system, such as core coolant outlet temperature, reactor power, coolant flow rate, and control rod position. S1012. Using a verified high-precision physical simulation calculation program, simulate and calculate the working conditions corresponding to the data collected in S1011 to generate the corresponding fuel temperature value. S1013. Associate and match the input feature data in S1011 with the output feature data in S1012 to form an original dataset containing input-output pairs, and perform preprocessing operations such as data cleaning and normalization.

[0016] In one specific embodiment, the content of S1013 is as follows: To eliminate the impact of different physical dimensions on model training efficiency and stability, and to ensure that all input features are of the same order of magnitude, the Min-Max Normalization method is used to process the dataset constructed in S1013. The processing equation is as follows:

[0017] In the formula, The normalized data has a range of [0, 1]. These are the original sensor measurements or simulation calculations; and These are the minimum and maximum values ​​of the feature in the entire dataset, respectively. All input features (such as core power, coolant temperature, etc.) and output features (fuel temperature) are processed by this equation.

[0018] In one specific embodiment, refer to Figure 3 As shown, S201 includes the following steps: S2011. Based on the characteristics of the problem, select and construct a suitable machine learning model, such as Long Short-Term Memory (LSTM), Deep Neural Network (DNN), Support Vector Regression (SVR), or Gradient Boosting Tree (GBT) model. S2012. Divide the preprocessed dataset in S1013 into a training set and a validation set. Use the training set to iteratively train the model built in S2011. Continuously adjust the internal parameters of the model through optimization algorithms (such as gradient descent) to minimize the error between the predicted value and the true value. S2013. Use an independent test set to evaluate the performance of the trained model, verify its prediction accuracy and generalization ability, and ensure that it meets the engineering requirements for online monitoring.

[0019] In one specific embodiment, the content of S2011 is as follows: A Long Short-Term Memory (LSTM) network regression model is constructed using machine learning theory. The model consists of an input layer, one or more LSTM layers (as hidden layers), optionally multiple fully connected hidden layers (Dense layers), and an output layer. The number of neurons in the input layer corresponds to the number of measurable input features acquired in S1011; the output layer has one neuron, corresponding to the fuel temperature generated in S1012.

[0020] During training, the training set data preprocessed by S1013 is input into the network. Using the backpropagation algorithm and the Adam optimizer, the connection weights and biases between neurons in the network are iteratively updated with the goal of minimizing the mean squared error (MSE). The formula for calculating the mean squared error is as follows:

[0021] In the formula, The number of samples in the training set; For the first The actual fuel temperature value of each sample; For the model to the first The fuel temperature value predicted for each sample.

[0022] In one specific embodiment, the content of S2013 is as follows: To quantitatively evaluate the performance of the trained model and test its generalization ability to new data, mean squared error (MSE) is used as the key evaluation metric. The model is considered qualified only when the MSE reaches a preset standard (e.g., below 0.5) on an independent test set.

[0023] The reactor was simulated using a high-precision thermal-hydraulic analysis program, and the resulting operational data was used to train the machine learning model. The training set included data under steady-state operation, stepped load reduction, and typical accident conditions (such as pump power outages) to ensure that the model could handle a variety of operating conditions.

[0024] In one specific embodiment, refer to Figure 4 As shown, S301 includes the following steps: S3011. Serialize the machine learning model and all its parameters (such as the weights and biases of the neural network) that have been validated in S2013, converting them from objects in memory into a file format that can be persistently stored (such as ONNX, HDF5 or PMML format) to form an independent and portable model file. S3012. Load the model file generated in S3011 into the target server of the reactor online monitoring system (such as DCS or SCADA system). In the software architecture of the monitoring system, configure the data input interface to receive real-time measurable parameters from the data acquisition module, and configure the data output interface to forward the model's prediction results to the monitoring interface or alarm system; S3013. Before the model is officially put into real-time monitoring, the model integrated into the monitoring system is subjected to final functional verification through simulated input or historical data playback. After confirming that the model can load correctly, stably receive input data, and provide calculation results within the specified time, and that the results are consistent with the offline test, the model service is officially activated to begin processing real-time online data.

[0025] In one specific embodiment, refer to Figure 5 As shown, S401 includes the following steps: S4011 The monitoring system reads the values ​​of five measurable parameters, including reactor power and coolant temperature, from the DCS system in real time at a frequency of 1Hz. S4012. After the vector formed by these 5 parameter values ​​is normalized in the same way as during training, it is input into the model service deployed in S301. S4013: The model service returns the predicted fuel temperature value (after inverse normalization) within milliseconds. This value is immediately updated on the operator monitoring screen in the main control room and compared in real time with the safety limit. If the predicted temperature exceeds the warning line, an audible and visual alarm is triggered.

[0026] Reference Figure 6 As shown, the present invention also discloses a nuclear reactor state synchronization system based on multi-source information reverse mapping, which applies a nuclear reactor state synchronization method based on multi-source information reverse mapping, including an offline model training module, a data acquisition module and a state synchronization module connected in sequence. The offline model training module is used to perform steps S101 and S201, collect and generate data to build a training dataset, train and validate machine learning models based on the dataset, and finally generate a deployable prediction model file. The data acquisition module is used to execute the steps of S4011. It is connected to the reactor's sensor system to acquire the measured values ​​of measurable parameters located at various points in the reactor in real time. The state synchronization module is used to execute steps S4012 and S4013. It has a built-in prediction model generated by the offline model training module. This module receives real-time measurement values ​​from the data acquisition module as input, calculates and outputs predicted unmeasurable parameter values ​​through the model, and realizes state synchronization.

[0027] In one specific embodiment, the machine learning model constructed by the offline model training module includes three core functions: data preprocessing, model training, and model validation. Among them, the data preprocessing function uses data normalization theory to process the collected and generated data, scaling the data to the [0,1] interval. The processing equation is as follows:

[0028] In the formula, The data is after normalization; This is the original data; , These are the minimum and maximum values ​​of the original data set, respectively. The model training function uses machine learning theory to build and train a model capable of regression prediction, adjusting network weights to minimize the mean squared error (MSE) between the fuel temperature predicted by the model and the actual fuel temperature in the training data. The model validation function uses an independent test dataset to evaluate the trained deep neural network regression model. After calculating the model's mean squared error (MSE), the model validation function can determine whether the model can be used for online deployment based on whether the MSE meets the preset accuracy requirements (e.g., less than 0.5).

[0029] In one specific embodiment, the sensor measurements acquired by the data acquisition module include: core coolant outlet temperature, reactor power, core coolant inlet temperature, etc.

[0030] In one specific embodiment, the state synchronization module normalizes the values ​​of multiple measurable parameters collected in real time by the data acquisition module and then inputs them into the deployed machine learning model to finally obtain the accurate synchronization values ​​of the unmeasurable parameters; the process is as follows: A set of real-time data, such as the core power value and the inlet and outlet temperature values ​​of the core coolant measured by the sensor, is substituted into the data normalization equation to obtain the normalized real-time state vector. The real-time state vector is input into the deployed deep neural network regression model, and a normalized fuel temperature prediction value is obtained through forward propagation calculation of the model. The normalized fuel temperature prediction is substituted into the inverse function of the normalization equation for calculation, resulting in the final, physically meaningful, and precisely synchronized fuel temperature value, which is then output to the monitoring system.

[0031] The LSTM structure used in this invention is as follows: Figure 7 As shown; The performance parameters of the integrated pressurized water reactor after using the state synchronization system of this invention are as follows: Figure 8 As shown; Figure 8 -(a) is a scatter plot of predicted values ​​versus actual values ​​for the stepped power reduction training set; Figure 8 -(b) is a scatter plot of predicted values ​​versus actual values ​​for the stepped power reduction test set; Figure 8 -(c) is a scatter plot of predicted and actual values ​​for the training set of pump power failure; Figure 8-(d) is a scatter plot of predicted and actual values ​​for the pump power failure test set.

[0032] The results above demonstrate that this method, using a data-driven neural network regression model, can accurately predict key unmeasurable parameters (such as fuel temperature) in real time based on real-time measurable parameters of the reactor (such as power and temperature). This solves the problems of large computational load, poor real-time performance, and infeasibility of direct measurement in traditional physical models, ensuring that operators have a precise grasp of the core state of the reactor, enhancing reactor safety and reliability, and providing a new solution for nuclear reactor condition monitoring, fault diagnosis, and the construction of digital twin systems.

[0033] It should be noted that all electrical devices involved in this application can be powered by batteries or external power sources.

[0034] This invention is not limited to the embodiments described above. Anyone should understand that structural changes made under the guidance of this invention, and any technical solutions that are the same as or similar to this invention, fall within the protection scope of this invention. Technical aspects, shapes, and structures not described in detail in this invention are all publicly known technologies.

Claims

1. A method for nuclear reactor state synchronization based on multi-source information reverse mapping, characterized in that, Includes the following steps: S101. Acquire and integrate the operating condition data of the nuclear reactor to form a structured dataset for training machine learning models; S201. Based on the training dataset constructed in S101, train a regression prediction model that can characterize the mapping relationship between measurable parameters and unmeasurable parameters. S301. Integrate the machine learning model trained in S201 into the reactor's online monitoring system; S401. During reactor operation, the values ​​of unmeasurable parameters are predicted online using deployed models and real-time sensor data to achieve state synchronization.

2. The method of claim 1, wherein, S101 includes the following steps: S1011. The reactor instrumentation system collects directly measurable parameters under different operating conditions, including core coolant outlet temperature, reactor power, coolant flow rate, and control rod position. S1012. Using a verified high-precision physical simulation calculation program, simulate and calculate the working conditions corresponding to the data collected in S1011 to generate the corresponding fuel temperature value. S1013. Associate and match the input feature data in S1011 with the output feature data in S1012 to form an original dataset containing input-output pairs, and perform data cleaning and normalization preprocessing operations on the original dataset.

3. The method of claim 1, wherein, S201 includes the following steps: S2011. Based on the characteristics of the problem, select and construct a machine learning model, wherein the machine learning model is selected from the long short-term memory model, deep neural network, support vector regression or gradient boosting tree model; S2012. Divide the preprocessed dataset in S1013 into a training set, a validation set, and an independent test set. Use the training set to iteratively train the model built in S2011. Adjust the internal parameters of the model through optimization algorithms to minimize the mean square error between the predicted value and the true value. S2013. Use an independent test set to evaluate the performance of the trained model, verify its prediction accuracy and generalization ability, and ensure that it meets the engineering requirements for online monitoring.

4. The method of claim 1, wherein, S301 includes the following steps: S3011. The machine learning model and all its parameters that have been validated in S2013 are serialized and converted into a file format that can be persistently stored to form an independent and portable model file; the parameters include the weights and biases of the neural network; the file format is selected from ONNX, HDF5 or PMML format. S3012. Load the model file generated in S3011 into the target server of the reactor online monitoring system; the online monitoring system is a DCS system or a SCADA system; in the software architecture of the monitoring system, configure a data input interface to receive real-time measurable parameters, and configure a data output interface to forward the model prediction results to the monitoring interface or alarm system. S3013. Before the model is officially put into real-time monitoring, the model integrated into the monitoring system is functionally verified by simulating input or playing back historical data. After confirming that the model can load correctly, stably receive input data, output calculation results within a specified time and the results are consistent with the offline test, the model service is activated to start processing real-time online data.

5. The method of claim 1, wherein, S401 includes the following steps: S4011. Obtain the current values ​​of parameters such as core coolant outlet temperature and reactor power in real time and continuously from the sensor network of the operating nuclear reactor; S4012. The real-time parameter values ​​obtained in S4011 are used as inputs and fed into the model service deployed in S301 for forward propagation calculation. S4013. The model outputs the predicted real-time fuel temperature value, which can be used for operator interface display, safety limit monitoring and alarm system, thereby completing the status synchronization.

6. A nuclear reactor state synchronization system based on multi-source information reverse mapping, characterized in that, The nuclear reactor state synchronization method according to any one of claims 1-5 includes an offline model training module, a data acquisition module and a state synchronization module connected in sequence. The offline model training module is used to perform steps S101 and S201, collect and generate data to build a training dataset, train and validate the machine learning model based on the dataset, and finally generate a deployable prediction model file. The data acquisition module is used to execute the steps of S4011, connect to the sensor system of the reactor, and acquire the measured values ​​of measurable parameters arranged at various locations in the reactor in real time. The state synchronization module is used to execute steps S4012 and S4013, and has a prediction model generated by the offline model training module. This module is used to receive real-time measurement values ​​from the data acquisition module as input, and output the predicted unmeasurable parameter values ​​through model calculation to achieve state synchronization.

7. A nuclear reactor state synchronization system based on multi-source information reverse mapping according to claim 6, characterized in that, The machine learning model built by the offline model training module includes three core functions: data preprocessing, model training, and model validation.

8. The nuclear reactor core state synchronization system based on multi-source information reverse mapping of claim 7, wherein, The data preprocessing function uses a normalization equation to process the collected and generated data; the data is scaled to the [0,1] interval, and the processing equation is as follows: wherein, is the normalized data; is the original data; , are the minimum and maximum values of the original data set, respectively; the model training function uses machine learning theory to construct and train a model capable of making regression predictions, adjusting network weights to minimize the mean squared error of the model's predicted fuel temperature and the true fuel temperature in the training data; The model validation function uses an independent test dataset to evaluate the trained deep neural network regression model. After calculating the mean squared error of the model, the model validation function can determine whether the model can be used for online deployment based on whether the mean squared error meets the preset accuracy requirements (such as less than 0.5).

9. The nuclear reactor core state synchronization system based on multi-source information reverse mapping of claim 6, wherein, The measurable parameters acquired by the data acquisition module from the sensors include core coolant outlet temperature, reactor power, and core coolant inlet temperature.

10. The nuclear reactor core state synchronization system based on multi-source information reverse mapping of claim 6, wherein, The working process of the state synchronization module is as follows: A set of real-time data, such as the core power value and the inlet and outlet temperature values ​​of the core coolant measured by the sensor, is substituted into the data normalization equation to obtain the normalized real-time state vector. The real-time state vector is input into the deployed machine learning model, and the normalized fuel temperature prediction value is obtained through forward propagation calculation of the model. The normalized fuel temperature prediction is substituted into the inverse function of the normalization equation for calculation to obtain the final, physically meaningful, and accurately synchronized fuel temperature value, which is then output to the monitoring system.

Images