A model and data double-driven intelligent peak regulation method for nuclear power plants

Abstract

The application discloses a model and data double-driven intelligent peak regulation method for a nuclear power plant, and steps are as follows: first, core modeling is performed, and core physical software is used to perform calculation to obtain core parameters; second, nuclear power plant operation data are combined with the core parameters to construct a data set, and the data set is preprocessed; third, a neural network model is trained; fourth, a rod position at a next moment is set, after calculation using the core physical software and construction of input data to be predicted, the input data to be predicted are input into the trained neural network model to predict a core axial power offset (AO) at the next moment; fifth, it is judged whether the predicted AO meets a requirement, if the requirement is met, the corresponding rod position is taken as a target rod position at the next moment, otherwise, the rod position is readjusted, and the fourth step is returned to re-predict the AO. The application can realize real-time and accurate prediction of the AO, reduce artificial decision risk, and realize intelligent peak regulation of the nuclear power plant under the premise of guaranteeing core safety.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of nuclear reactor physics and nuclear power plant operation control, and particularly relates to an intelligent peak shaving method for a nuclear power plant based on model and data double driving. BACKGROUND

[0002] With the rapid development of China's nuclear power industry, nuclear power provides more clean energy on the one hand, and on the other hand, it also brings huge peak shaving pressure to the power grid in some regions with relatively high nuclear power share and rapid growth of installed capacity. In order to ensure the stable operation of the power grid, nuclear power faces the urgent demand of participating in the peak shaving of the power grid. Moreover, at present, nuclear power units in some domestic regions have already frequently participated in peak shaving according to the demand of the power grid, and the research on the related problems of nuclear power plant peak shaving is imminent.

[0003] In the process of peak shaving of nuclear power units, the frequent changes of control rod group position and power may cause poison oscillation, resulting in local power distortion, uneven distribution of core temperature, and problems and related safety risks in reactivity control. Therefore, in the process of peak shaving, the control of the axial offset (AO) of the core and the maintenance of the power fluctuation within a certain range become the key problems in the process of peak shaving. At present, nuclear power plants generally refer to the peak shaving experience of nuclear power units in France and other countries, and adopt the constant AO control to maintain the AO within a certain range, so the calculation or prediction of the AO is crucial. However, due to the rapid change of power, frequent rod movement, and other reasons, the core condition changes dramatically, and in the traditional method, the peak shaving scheme calculated in advance by simply modeling the core is difficult to track the core condition in real time and accurately predict the AO, so the actual peak shaving process is extremely dependent on personal experience. This not only makes it difficult to meet the high-frequency and deep-amplitude peak shaving demand of pressurized water reactor nuclear power units, but also easily increases the decision risk and human risk, which is not conducive to the safety of the core. In the traditional neural network model, the main way is to use data driving alone, and the macro data directly measured as input data, but these macro data are difficult to accurately reflect the core state and accurately predict the AO.

[0004] If both model driving and data driving are used, that is, the core parameters reflecting the core state are obtained by core modeling through core physics software, and then the neural network model is used for data driving based on the core parameters and nuclear power plant operation parameters, the AO can be efficiently and accurately predicted, and the influence of the current rod position adjustment on the future AO can be accurately known, so as to intelligently assist the operator to control the AO, reduce the risk of human decision and intervention, and then realize intelligent peak shaving and ensure the safety of the core. SUMMARY

[0005] In order to solve the problems existing in the prior art, the purpose of the present application is to provide a model and data double-driven intelligent peak shaving method for nuclear power plants, which meets the peak shaving demand of nuclear power plants in China, ensures the safety of the reactor core, assists the operator in predicting the axial offset (AO) of the reactor core in real time, accurately and efficiently in the complex working conditions of the peak shaving process, reduces the risk in the peak shaving process, and can realize intelligent peak shaving of nuclear power plants under the premise of ensuring the safety of the reactor core.

[0006] In order to achieve the above purpose, the technical scheme adopted by the present application is as follows:

[0007] A model and data double-driven intelligent peak shaving method for nuclear power plants, comprising the following steps:

[0008] Step 1: Select the geometric and material parameters of the reactor core and fuel assembly of the nuclear power plant to model the reactor core, and use the reactor core physics software to simulate and calculate to obtain the reactor core parameters that need to be measured but cannot be directly measured;

[0009] Step 2: Select the nuclear power plant operation data in the historical peak shaving and the reactor core parameters obtained by the reactor core physics software simulation calculation, which together constitute the input data set required for neural network model training, and the axial offset (AO) value of the reactor core at the next time is used as the label data set; and the input data set and the label data set are preprocessed;

[0010] The combination of the reactor core data calculated by the professional reactor core physics software and the nuclear power plant operation data can greatly solve the problem of data source and improve the accuracy and robustness of the data-driven method such as neural network model.

[0011] Step 3: Divide the preprocessed input data set and label data set in step 2 into a training set, a validation set and a test set, and use the training set, the validation set and the test set to train the neural network model to obtain a trained neural network model with required accuracy;

[0012] Through appropriate division of the training set, the validation set and the test set, the expression ability of the neural network model can be improved, overfitting can be prevented, and the accuracy and robustness of the neural network model can be maximized.

[0013] Step 4: In actual peak shaving, set the rod position at the next time, use the reactor core physics software to track and simulate the reactor core to obtain the reactor core parameters, then use the obtained reactor core parameters and the nuclear power plant operation data in the actual peak shaving as the input data to be predicted, preprocess the input data to be predicted, and input the trained neural network model to predict the axial offset (AO) of the reactor core at the next time in real time;

[0014] Step 5: judging whether the predicted core axial power offset AO meets the requirements, if not, re-adjusting the rod position and returning to step 4 to re-predict the core axial power offset AO; if yes, taking the corresponding rod position as the target rod position of the next moment to assist the operator in decision-making and peak shaving control, realizing intelligent peak shaving.

[0015] By presetting the control rod position and correcting, the rod position can be selected by integrating the expert experience of the operator, and the trained neural network model can be used to judge whether the predicted AO meets the requirements, thereby providing an intelligent tool for the operator to participate in peak shaving decision-making and assisting the operator in decision-making and peak shaving control.

[0016] The core parameters obtained by the core physics software in step 1 include fuel burnup and neutron poison distribution. The advantage is that the core state during peak shaving, especially the xenon oscillation, can be described through fuel burnup and neutron poison distribution.

[0017] The nuclear power plant operation data in step 2 include reactor power, rod position of each control rod group, core axial power offset AO measurement value of the ex-core detector, fuel assembly enrichment, and burnable poison quantity. The advantage is that the nuclear power plant has a large amount of nuclear power plant operation data during peak shaving, especially the AO measurement value, which can provide a large amount of data source for neural network model training and reduce the error between predicted AO and real AO.

[0018] The preprocessing in step 2 and step 4 includes:

[0019] Data set reconstruction: all input data sets and label data sets are reconstructed into the data input format required by the neural network model according to the preset order and structure;

[0020] Data set normalization: all input data sets and label data sets after reconstruction are normalized to the range of 0 to 1, reducing the influence of dimension and order of magnitude.

[0021] The neural network model in step 3 adopts a long short-term memory neural network model LSTM. The advantage of the LSTM model is that it can effectively transmit and express information in long time series without causing useful information in long time to be ignored, and can also solve the problem of gradient disappearance in conventional recurrent neural networks.

[0022] The training of the neural network model in step 3 includes forward propagation, calculation of loss function, backpropagation to solve the gradient to reduce the loss function, and use of early stopping mechanism to interrupt the training process in advance under the condition that the condition is met;

[0023] The early stopping mechanism is: when the loss function of the verification set no longer decreases for a certain number of rounds, or decreases by less than a preset value for a certain number of rounds, the training process of the neural network model is interrupted. The advantage of using the early stopping mechanism is that the feature extraction capability of the neural network model can be guaranteed, and overfitting of the neural network model can be prevented.

[0024] The loss function uses a mean square error loss function.

[0025] The requirement met in step 5 is that the error of the predicted core axial power offset AO and the target core axial power offset AO is not more than 5%. In this way, both the requirements of the nuclear power plant operation specification and the reduction of the rod position search time and the improvement of the decision-making efficiency of the operator can be met.

[0026] The present application first uses model driving to obtain relevant core parameters reflecting the core state through core modeling by core physics software, and then trains a neural network model based on the core parameters and nuclear power plant operation parameters. Finally, the trained neural network model is used for data driving to efficiently and accurately predict the core axial power offset (AO) and accurately know the impact of the current rod position adjustment on the future AO. The model and data hybrid driving nuclear power plant intelligent peak shaving method can intelligently assist the operator in AO control to reduce the risk of human decision-making and intervention, thereby realizing intelligent peak shaving and ensuring the safety of the core. Compared with the prior art, the present application has the following advantages:

[0027] 1. The present application uses model and data dual driving, innovatively adds the core parameters obtained by simulation calculation of the core physics software to the data source of the data driving of the neural network model, and solves the difficulties of single data source, few types of data, and difficulty in accurately describing the core state in the conventional data driving method.

[0028] 2. The present application uses model and data dual driving to train a set of neural network model capable of accurately predicting AO, which can efficiently and accurately predict AO.

[0029] 3. The nuclear power plant intelligent peak shaving method proposed by the present application can intelligently assist the operator in rod position adjustment and AO control, realize intelligent peak shaving, reduce the risk of human decision-making and intervention in the peak shaving process, and can effectively ensure the safety of the core. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 The present application is a method flowchart.

[0031] Figure 2 The present application is a relative power and R rod rod position variation diagram with time for a certain M310 unit in February 2023.

[0032] Figure 3The figure shows the comparison between the predicted AO and the actual AO during the peak regulation process of a certain M310 unit in February 2023.

[0033] Figure 4 The figure shows the actual R rod position change over time during the peak regulation process of a certain M310 unit in February 2023.

[0034] Figure 5 The figure shows the target R rod position change over time during the peak regulation process of a certain M310 unit in February 2023. DETAILED DESCRIPTION

[0035] In order to make the principles, methods, etc. of the present application more easily understood, the specific implementation cases of the present application are further described in detail below in conjunction with the accompanying drawings. It must be emphasized that the specific implementation described here is only used to illustrate the present application in specific cases, and is not intended to limit the present application.

[0036] As shown in Figure 1 The present application is a model and data dual-driven intelligent peak regulation method for nuclear power plants, which comprises the following steps:

[0037] Step 1: Select the geometric and material parameters of the reactor core and fuel assemblies of the nuclear power plant for core modeling, use the core physics software Bamboo-C for simulation calculation, and obtain the required but not directly measurable core parameters, including fuel burnup and neutron poison distribution;

[0038] Step 2: Select the relative power of the reactor, the position of each control rod group, the core axial power offset AO measurement value of the reactor external detector, the fuel assembly enrichment, and the number of burnable poisons, etc. Nuclear power plant operation data in historical peak regulation, and the core parameters obtained by the core physics software Bamboo-C simulation calculation together constitute the input data set required for neural network model training, and the continuous data is sampled according to a 4-minute time interval, each peak regulation is divided into different number of samples; the core axial power offset AO value at the next time is used as the label data set; the input data set and the label data set are preprocessed: all input data sets and label data sets are arranged in order and subjected to sliding window operation, and processed into corresponding input and output samples of the neural network model. The reconstructed data is subjected to "Min-Max normalization", and is uniformly converted to the range of 0 to 1;

[0039] The conventional data-driven method relies on data itself to a high degree, especially in the field of nuclear energy application. Although it can achieve certain effects, it is difficult to overcome the disadvantages of single data source, insufficient data types, and low data quality. In the field of nuclear energy application, the cost of obtaining core experimental data is very high. Due to the reasons of detection method and measurement technology, the types of measurement data during the operation of a nuclear power plant alone are few, and it is difficult to accurately depict the core state during the peak regulation operation of the nuclear power plant, which is a rapid change process of the core. The core data calculated by the professional core physics software and the operation data of the nuclear power plant are combined in the present application, which can greatly solve the problem of data source and improve the accuracy and robustness of the data-driven method such as neural network model.

[0040] Step 3: The preprocessed input data set and label data set in step 2 are divided into a training set, a validation set and a test set, and the neural network model is trained using the training set, the validation set and the test set. In this example, the neural network model uses a long short-term memory neural network model LSTM, which includes forward propagation, calculation of a loss function, and reverse gradient calculation to reduce the loss function in the model training. The loss function uses a mean squared error loss function. An early stopping mechanism is used, and when the loss function of the validation set does not decrease within 50 rounds, the training process is interrupted in advance.

[0041] Step 4: In actual peak regulation, set the rod position at the next time, use the core physics software to track and simulate the core to obtain the core parameters, then use the core parameters and the nuclear power plant operation data in the actual peak regulation as the input data to be predicted, and use the same preprocessing method as in step 2 to preprocess the input data to be predicted, and input the trained neural network model to predict the core axial power offset AO at the next time in real time. The single prediction time is not more than 1 second.

[0042] Step 5: Determine whether the core axial power offset AO predicted in step 4 meets the error of not more than 5% with the target core axial power offset AO. If it does not meet the requirement, the rod position is adjusted and returned to step 4 to predict the core axial power offset AO again. If it meets the requirement, the corresponding rod position is used as the target rod position for the next moment to assist the operator in decision-making and peak regulation control, realizing intelligent peak regulation.

[0043] In this example, a certain period of peak regulation in February 2023 is taken as an example, and the real AO is taken as the target AO. Under the premise that the error between the predicted AO and the target AO is not more than 1%, the rod position is adjusted to obtain the required target rod position. Then, the adjusted target R rod position is compared with the actual R rod position to verify the feasibility of the intelligent peak regulation method described in the present application.

[0044] wherein Figure 2Fig. 1 is a schematic diagram of the relative power and the R-rod position varying with time during a certain period of peak regulation in February 2023 for a certain M310 unit, Figure 3 Fig. 2 is a schematic diagram of the comparison between the predicted AO and the actual AO during the peak regulation, Figure 2 and Figure 3 there are 4 data points between each two data points in Figs. 1 and 2, that is, the interval between adjacent two data points in the figures is 20 minutes. Figure 2 and Figure 3 It can be seen from Figs. 1 and 2 that the relative power and the rod position change constantly during the peak regulation, but the predicted AO value of the application is in good agreement with the actual AO value, and the error is small. Figure 4 and Figure 5 Figs. 3 and 4 are a schematic diagram of the actual R-rod position varying with time and a schematic diagram of the target R-rod position varying with time during a certain period of peak regulation in February 2023 for a certain M310 unit, Figure 4 and Figure 5 there are 4 data points between each two data points in Figs. 3 and 4, that is, the interval between adjacent two data points in the figures is 20 minutes. Figure 4 and Figure 5 It can be seen from Figs. 3 and 4 that the target R-rod position is in good agreement with the actual R-rod position, which proves that the intelligent peak regulation method of the nuclear power plant based on the model and data double driving of the application can efficiently and accurately predict the axial offset AO of the core, can intelligently assist the operator to adjust the rod position and control the axial offset AO of the core, reduce the risk of human decision and intervention, and then realize intelligent peak regulation and ensure the safety of the core.

Claims

1. A smart peak-shaving method for nuclear power plants based on a model- and data-driven approach, characterized in that: Includes the following steps: Step 1: Select the geometric and material parameters of the nuclear power plant reactor core and fuel assemblies to perform core modeling, and use core physics software to perform simulation calculations to obtain the core parameters that are needed but cannot be directly measured; Step 2: Select the nuclear power plant operation data from historical peak shaving and the core parameters obtained from core physics simulation calculations. The two together constitute the input dataset required for training the neural network model. Use the core axial power offset (AO) value at the next time step as the label dataset. Preprocess the input dataset and the label dataset. Step 3: Divide the preprocessed input dataset and label dataset from Step 2 into training set, validation set and test set, and use the training set, validation set and test set to train the neural network model to obtain a trained neural network model with the required accuracy. Step 4: In actual peak shaving, set the rod position for the next moment. After using core physics software to track and simulate the core to obtain core parameters, use the obtained core parameters and the nuclear power plant operation data in actual peak shaving as the input data to be predicted. After preprocessing the input data to be predicted, input it into the trained neural network model to predict the core axial power offset AO for the next moment in real time. Step 5: Determine whether the predicted core axial power offset AO meets the requirements. If it does not meet the requirements, readjust the rod position and return to Step 4 to re-predict the core axial power offset AO. If it meets the requirements, use the corresponding rod position as the target rod position for the next moment to assist the operator in decision-making and peak-shaving control, and realize intelligent peak-shaving. The nuclear power plant operating data mentioned in step 2 includes reactor power, rod position of each control rod group, core axial power offset (AO) measurement value of the external detector, fuel assembly enrichment and quantity of combustible poisons; The preprocessing described in steps 2 and 4 includes: Dataset Reconstruction: Reconstruct all input datasets and label datasets into the data input format required by the neural network model according to a preset order and structure; Dataset normalization: Normalize all the reconstructed input and label datasets to the range of 0 to 1, reducing the influence of units and orders of magnitude.

2. The intelligent peak-shaving method for nuclear power plants based on model and data dual-drive as described in claim 1, characterized in that: The core parameters obtained from the simulation calculations performed by the core physics software in step 1 include fuel burnup and neutron poison distribution.

3. The intelligent peak-shaving method for nuclear power plants based on model and data dual-drive as described in claim 1, characterized in that: The time interval between adjacent moments shall not exceed 60 minutes.

4. The intelligent peak-shaving method for nuclear power plants based on model and data dual-drive as described in claim 1, characterized in that: The neural network model described in step 3 is the Long Short-Term Memory (LSTM) neural network model.

5. The intelligent peak-shaving method for nuclear power plants based on model and data dual-drive as described in claim 1, characterized in that: Step 3 describes training the neural network model, which includes forward propagation, calculating the loss function, backpropagating the gradient to reduce the loss function, and using an early stopping mechanism to interrupt the training process in advance when the conditions are met. The early stopping mechanism is as follows: when the loss function of the validation set no longer decreases after a certain number of rounds, or when the decrease is less than a preset value after a certain number of rounds, the training process of the neural network model is interrupted.

6. The intelligent peak-shaving method for nuclear power plants based on model and data dual-drive as described in claim 5, characterized in that: The loss function used is the mean squared error loss function.

7. The intelligent peak-shaving method for nuclear power plants based on model and data dual-drive as described in claim 1, characterized in that: The requirement mentioned in step 5 means that the error between the predicted core axial power offset AO and the target core axial power offset AO does not exceed 5%.

Images