A deep learning-based transient calculation method for pressurized water reactors

Abstract

The application discloses a kind of based on deep learning's pressurized water reactor transient calculation method, first establish the deep learning network model of single time point reactor core physical field prediction with pressurized water reactor temperature and power distribution etc. History data of core physical field is input, the physical field prediction of transient condition changes with time;Single time point reactor core physical field prediction deep learning network model features are: using the core physical field data of several historical time points before the predicted time point and the core rod position of the predicted time point as input parameters, the core physical field of the predicted time point is obtained using deep learning network model.The physical field prediction of core physical field changes with time: the predicted physical field and the physical field of earlier historical time point are jointly used as input, and the physical field of next time point is predicted.The method overcomes the limitation that the previous deep learning network model can only predict the physical field of steady state condition, and realizes the prediction of the physical field of transient condition changes with time.

Description

Technical Field

[0001] This invention relates to the fields of nuclear reactor core design and safety technology, specifically to a deep learning-based transient calculation method for pressurized water reactors. Background Technology

[0002] Traditional pressurized water reactor transient calculation methods often require several seconds for transient prediction, making it impossible to predict the transient physical field of the reactor core in real time under rapid power change conditions and thus failing to guarantee the safe operation of the reactor. In contrast, deep learning-based pressurized water reactor transient calculation can complete the prediction of a point in time in just a few milliseconds, achieving real-time prediction of the transient physical field of the reactor core. Summary of the Invention

[0003] To overcome the problems existing in the prior art, the purpose of this invention is to propose a deep learning-based transient calculation method for pressurized water reactors. The time interval of this invention is not affected by the implicit differential discrete approximation error of traditional methods, thus enabling the prediction of physical fields over a large time interval. Combined with the high efficiency of deep learning algorithms, it can achieve ultra-real-time prediction of the transient physical fields of the reactor core under rapid power change conditions.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] A deep learning-based transient calculation method for pressurized water reactors (PWRs) is proposed to predict the time-varying core physical field distribution under transient operating conditions. The method includes the following steps:

[0006] Step 1: Read the layout information of the pressurized water reactor core to be simulated;

[0007] Step 2: Model and calculate the pressurized water reactor core fuel management calculation program based on the core layout information read in Step 1, output the core physical field changing over time under various transient conditions, and create a dataset for training the deep learning network; the dataset includes the core temperature distribution, core power distribution, and control rod position parameters at each time point;

[0008] Step 3: For pressurized water reactors, establish a deep learning network model for predicting the core physics field at a single time point, and train it using the dataset generated in Step 2; the deep learning network model includes:

[0009] (a) Input layer: includes the control rod position at the current time point to be predicted, as well as the control rod positions at multiple previous time points, core temperature distribution, core power distribution, total core power, and time interval;

[0010] (b) Intermediate layer: including 5 to 8 hidden layers;

[0011] (c) Output layer: Outputs the core physical field, which includes the temperature distribution, power distribution and total power of the pressurized water reactor core;

[0012] Step 4: Based on the deep learning network model for predicting the core physical field at a single time point established in Step 3, predict the core physical field of transient operating conditions that change over time. The specific process is as follows: using the core physical fields at multiple time points before the current time point as input parameters, the deep learning network model is used to predict the core physical field at the current time point; then, using the core physical field at the current time point and the core physical fields at multiple time points before the current time point as input, the core physical field at the next time point is predicted; and so on, to achieve the prediction of the core physical field at subsequent time points.

[0013] Preferably, the previous multiple time points are the previous 8 time points.

[0014] Compared with the prior art, the present invention has the following outstanding advantages:

[0015] Because this invention uses a deep learning model to replace the calculation process, the time interval is not affected by the implicit differential discrete approximation error of the traditional method, thus enabling the prediction of physical fields over a large time interval. Combined with the high efficiency of the deep learning algorithm, it can achieve the prediction of ultra-real-time transient physical fields of the reactor core under rapid power change conditions.

[0016] The method of this invention overcomes the limitation of previous deep learning network models that can only predict the physical field under steady-state conditions, and realizes the prediction of the physical field that changes over time under transient conditions. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method of the present invention.

[0018] Figure 2 It is a deep learning network model for predicting physical fields at a single time point. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0020] This invention proposes a deep learning-based transient calculation method for pressurized water reactors, the flowchart of which is shown below. Figure 1 As shown, a deep learning network model is first established to predict the core physics field at a single time point, using historical core physics field data such as temperature and power distribution of the pressurized water reactor as input. Based on this, the physics field prediction for transient operating conditions over time is performed, such as... Figure 2As shown. The basic characteristics of the deep learning network model for predicting the core physics field at a single time point are: using core physics field data from several historical time points prior to the time point to be predicted and the core rod positions at the time point to be predicted as input parameters, the deep learning network model obtains the core physics field at the time point to be predicted. The prediction process for the core physics field that changes over time is as follows: using the predicted physics field and the physics field from even earlier historical time points as input, the physics field at the next time point is predicted. The specific steps are as follows:

[0021] Step 1: Read the layout information of the pressurized water reactor core to be simulated;

[0022] Step 2: Using the core layout information obtained in Step 1, modeling and calculations are performed using the NECP-Bamboo program from the NECP Laboratory of Xi'an Jiaotong University. The output shows the core physical field changing over time under various transient conditions, creating a dataset for training the deep learning network. The dataset includes parameters such as core temperature distribution, core power distribution, and control rod position at each time point.

[0023] Step 3: For pressurized water reactors, establish a deep learning network model for predicting the core physics field at a single time point, and train it using the dataset generated in Step 2. The deep learning network model includes:

[0024] (a) Input layer: includes the control rod position at the current time point to be predicted, as well as the control rod positions, core temperature distribution, core power distribution, total core power and time intervals at the previous 8 time points;

[0025] (b) Intermediate layer: including 5 to 8 hidden layers;

[0026] (c) Output layer: Outputs the core physical field, which includes the temperature distribution, power distribution and total power of the pressurized water reactor core;

[0027] Step 4: Based on the deep learning network model for predicting the core physical field at a single time point established in Step 3, predict the core physical field of transient operating conditions over time. Specifically, using the core physical fields of the previous 8 time points as input parameters, the deep learning network model predicts the core physical field at the current time point. Then, using the core physical field at the current time point and the core physical fields of the previous 7 time points as input, the core physical field at the next time point is predicted. This process is repeated to predict the core physical field at subsequent time points.

Claims

1. A deep learning-based transient calculation method for pressurized water reactors, characterized in that: Predicting the time-varying core physical field distribution under transient operating conditions of a pressurized water reactor involves the following steps: Step 1: Read the layout information of the pressurized water reactor core to be simulated; Step 2: Use the pressurized water reactor core fuel management calculation program to model and calculate the core layout information read in Step 1, output the core physical field changing over time under various transient conditions, and create a dataset for training the deep learning network; the dataset includes the core temperature distribution, core power distribution and control rod position parameters at each time point; Step 3: For pressurized water reactors, establish a deep learning network model for predicting the core physics field at a single time point, and train it using the dataset generated in Step 2; the deep learning network model includes: (a) Input layer: includes the control rod position at the current time point to be predicted, as well as the control rod positions at multiple previous time points, core temperature distribution, core power distribution, total core power, and time interval; (b) Intermediate layer: including 5 to 8 hidden layers; (c) Output layer: Outputs the core physical field, which includes the temperature distribution, power distribution and total core power of the pressurized water reactor core; Step 4: Based on the deep learning network model for predicting the core physical field at a single time point established in Step 3, predict the core physical field of transient operating conditions that change over time. The specific process is as follows: using the core physical fields at multiple time points before the current time point as input parameters, the deep learning network model is used to predict the core physical field at the current time point; then, using the core physical field at the current time point and the core physical fields at multiple time points before the current time point as input, the core physical field at the next time point is predicted; and so on, to achieve the prediction of the core physical field at subsequent time points.

2. The transient calculation method for pressurized water reactors based on deep learning according to claim 1, characterized in that: The previous multiple time points refer to the previous 8 time points.

Images