A machine learning based neutron noise anomaly detection method

Abstract

The present application relates to the field of intelligent nuclear power, and more particularly to a neutron noise anomaly detection method based on machine learning. The method comprises the following steps: step 1, collecting neutron noise data in a nuclear power plant and performing data preprocessing; step 2, constructing an LSTM model and training it; step 3, using the trained LSTM model to extract features; step 4, using the isolated forest algorithm to construct an isolated forest model, and inputting the feature group extracted in the above steps to identify abnormal data. The present application can accurately detect abnormal signals in neutron noise, and timely evaluate and monitor the state of the nuclear reactor, thereby improving the safety and reliability of the nuclear power plant.

Description

Technical Field

[0001] This invention relates to the field of smart nuclear power, and in particular to a neutron noise anomaly detection method based on machine learning. Background Technology

[0002] During the operation of nuclear power plants, monitoring neutron noise is crucial for assessing the state of the nuclear reactor. Monitoring neutron noise enables effective measurement of the neutron flux density in the reactor core, thereby detecting early faults in the reactor's primary circuit and ensuring safe reactor operation. However, because neutron noise signals also contain anomalous signals, these anomalous signals can lead to incorrect assessments of the reactor's state. Therefore, detecting anomalous neutron noise signals is essential.

[0003] Currently, machine learning-based anomaly detection methods have been widely applied in various fields, including signal processing, image processing, and speech recognition. However, research on neutron noise anomaly detection is extremely limited worldwide. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a neutron noise anomaly detection method based on machine learning, which can accurately detect abnormal signals in neutron noise and timely assess and monitor the state of nuclear reactors, thereby improving the safety and reliability of nuclear power plants.

[0005] This invention provides a neutron noise anomaly detection method based on machine learning, comprising:

[0006] Step 1: Collect neutron noise data from the nuclear power plant and perform data preprocessing;

[0007] Step 2: Build and train the LSTM model;

[0008] Step 3: Use the trained LSTM model to extract features;

[0009] Step 4: Construct an isolated forest model using the isolated forest algorithm, taking the feature groups extracted in the above steps as input, and identify abnormal data.

[0010] In a specific embodiment of the present invention, step 1, the data preprocessing includes:

[0011] The collected neutron noise data from nuclear power plants was subjected to denoising, filtering, and dimensionality reduction processing.

[0012] Perform normalization or standardization operations to ensure that it is suitable for the data input requirements of subsequent LSTM models.

[0013] In a specific embodiment of the present invention, step 2 specifically includes:

[0014] Construct the LSTM model network structure, and set the input layer, intermediate hidden layer and output layer;

[0015] The preprocessed neutron noise data is input into the LSTM model and trained in a time series manner to learn the time dependence and patterns in the neutron noise data.

[0016] The LSTM model parameters are adjusted using the backpropagation algorithm and gradient descent method to minimize the prediction error;

[0017] Save the trained LSTM model.

[0018] In a specific embodiment of the present invention, step 3 specifically comprises:

[0019] The trained LSTM model is used to predict the newly collected data, and the error between the actual value and the predicted value is calculated. The features with errors are then extracted.

[0020] In one specific embodiment of the present invention, the feature includes the amount of error or the rate of change.

[0021] In a specific embodiment of the present invention, step 4 specifically comprises:

[0022] The error data of the extracted features are input into the Isolation Forest algorithm to build an Isolation Forest model, and the data is separated by decision trees;

[0023] Based on the output of the isolated forest, determine whether the data points are abnormal, obtain the score of the abnormal data, and determine whether there are any abnormalities based on the set threshold.

[0024] In one specific embodiment of the present invention, the preprocessed neutron noise data includes normal operation data and known fault data.

[0025] In one specific embodiment of the present invention, the result output of abnormal data is also included;

[0026] Specifically, this involves visualizing abnormal data, generating anomaly detection reports, and summarizing the frequency, type, and potential impact of abnormal data.

[0027] In one specific embodiment of the present invention, the performance of the LSTM model is monitored in real time and the model is automatically adjusted.

[0028] In one specific embodiment of the present invention, the method further includes real-time monitoring of the performance of the isolated forest model and automatic adjustment of the model.

[0029] Compared with existing technologies, the neutron noise anomaly detection method based on machine learning in this invention utilizes machine learning techniques for feature extraction and anomaly detection of neutron noise. This enables accurate and rapid detection of neutron noise anomalies, thereby improving the safety and reliability of nuclear power plants and preventing accidents caused by neutron noise anomalies. Furthermore, this method employs an adaptive learning strategy, which automatically adapts to changes in data, improving detection accuracy and robustness. Detailed Implementation

[0030] To further understand the present invention, embodiments of the present invention are described below in conjunction with examples. However, it should be understood that these descriptions are only for further illustrating the features and advantages of the present invention, and not for limiting the present invention.

[0031] An embodiment of the present invention discloses a neutron noise anomaly detection method based on machine learning, comprising:

[0032] Step 1: Collect neutron noise data from the nuclear power plant and perform data preprocessing;

[0033] Specifically:

[0034] Neutron noise data from nuclear power plants is collected using sensors or data loggers.

[0035] To ensure the accuracy and completeness of the data, gaps in the sensor layout are eliminated during the preprocessing process, and measurement errors are minimized as much as possible.

[0036] The preprocessing steps include denoising, noise reduction, and filtering operations to improve data quality. Preferably, wavelet transform is used for denoising.

[0037] The preprocessed neutron noise data will be normalized or standardized to ensure that it is suitable for the data input requirements of the subsequent LSTM model. The normalization process is preferably maximum value normalization.

[0038] Step 2: Build and train the LSTM model;

[0039] Specifically:

[0040] Construct the LSTM model network structure, and set the input layer, intermediate hidden layer and output layer;

[0041] The preprocessed neutron noise data is input into the LSTM model and trained in a time series manner to learn the time dependence and patterns in the neutron noise data.

[0042] The preprocessed neutron noise data includes normal operation data and known fault data;

[0043] The LSTM model parameters are adjusted using the backpropagation algorithm and gradient descent method to minimize the prediction error;

[0044] Save the trained LSTM model.

[0045] Monitor the performance of the LSTM model in real time and automatically adjust the model.

[0046] In this modeling step, the long short-term memory (LSTM) property of the LSTM network is used to analyze the time series data to capture the temporal dependencies and patterns within the data. During training, the LSTM model continuously adjusts its weight parameters to minimize prediction errors, ensuring it can accurately capture the dynamic changes in neutron noise data.

[0047] Step 3: Use the trained LSTM model to extract features;

[0048] Specifically:

[0049] The trained LSTM model is used to predict the newly collected data, and the error between the actual value and the predicted value is calculated. The features with errors are then extracted.

[0050] The features include the amount of error or the rate of change.

[0051] The features will be used for anomaly detection.

[0052] Step 4: Construct an isolated forest model using the isolated forest algorithm, taking the feature groups extracted in the above steps as input, and identify abnormal data.

[0053] The Isolation Forest algorithm is used to identify anomalies in neutron noise data. The Isolation Forest algorithm is suitable for multi-dimensional feature data, and therefore can effectively handle features extracted by the LSTM model.

[0054] Step 4 specifically involves:

[0055] The error data of the extracted features are input into the Isolation Forest algorithm to build an Isolation Forest model, and the data is separated by decision trees;

[0056] Based on the output of the isolated forest, determine whether the data points are abnormal, obtain the score of the abnormal data, and determine whether there are any abnormalities based on the set threshold.

[0057] The threshold is adaptively determined based on the isolated forest algorithm.

[0058] Monitor the performance of the isolated forest model in real time and automatically adjust the model.

[0059] In a specific embodiment, the parameters of the isolated forest algorithm are set as follows:

[0060] (1) The sample size for each tree is 256;

[0061] (2) Randomly select a subsample;

[0062] (3) Randomly selected features.

[0063] The output anomaly detection result is a binary classification result, where 1 indicates anomaly and 0 indicates normal.

[0064] In this step, instead of using the traditional Support Vector Machine (SVM), unsupervised anomaly detection is achieved through an isolated forest to improve the accuracy and adaptability of the detection.

[0065] This invention also includes the output of results for abnormal data;

[0066] Specifically, this involves visualizing abnormal data, generating anomaly detection reports, and summarizing the frequency, type, and potential impact of abnormal data to support decision-making.

[0067] The visualization includes the time point of the anomaly, its severity, etc., so that nuclear power plant operators can quickly understand the abnormal situation.

[0068] The above description of the embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. It should be noted that those skilled in the art can make several improvements and modifications to the present invention without departing from the principles of the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

[0069] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A neutron noise anomaly detection method based on machine learning, characterized in that, include: Step 1: Collect neutron noise data from the nuclear power plant and perform data preprocessing; Step 2: Build and train the LSTM model; Step 3: Use the trained LSTM model to extract features; Step 4: Construct an isolated forest model using the isolated forest algorithm, taking the feature groups extracted in the above steps as input, and identify abnormal data.

2. The neutron noise anomaly detection method based on machine learning according to claim 1, characterized in that, In step 1, the data preprocessing includes: The collected neutron noise data from nuclear power plants was subjected to denoising, filtering, and dimensionality reduction processing. Perform normalization or standardization operations to ensure that it is suitable for the data input requirements of subsequent LSTM models.

3. The neutron noise anomaly detection method based on machine learning according to claim 1, characterized in that, Step 2 specifically includes: Construct the LSTM model network structure, and set the input layer, intermediate hidden layer and output layer; The preprocessed neutron noise data is input into the LSTM model and trained in a time series manner to learn the time dependence and patterns in the neutron noise data. The LSTM model parameters are adjusted using the backpropagation algorithm and gradient descent method to minimize the prediction error; Save the trained LSTM model.

4. The neutron noise anomaly detection method based on machine learning according to claim 1, characterized in that, Step 3 specifically involves: The trained LSTM model is used to predict the newly collected data, and the error between the actual value and the predicted value is calculated. The features with errors are then extracted.

5. The neutron noise anomaly detection method based on machine learning according to claim 4, characterized in that, The features include the amount of error or the rate of change.

6. The neutron noise anomaly detection method based on machine learning according to claim 4, characterized in that, Step 4 specifically involves: The error data of the extracted features are input into the Isolation Forest algorithm to build an Isolation Forest model, and the data is separated by decision trees; Based on the output of the isolated forest, determine whether the data points are abnormal, obtain the score of the abnormal data, and determine whether there are any abnormalities based on the set threshold.

7. The neutron noise anomaly detection method based on machine learning according to claim 3, characterized in that, The preprocessed neutron noise data includes normal operation data and known fault data.

8. The neutron noise anomaly detection method based on machine learning according to claim 1, characterized in that, It also includes the output of results for abnormal data; Specifically, this involves visualizing abnormal data, generating anomaly detection reports, and summarizing the frequency, type, and potential impact of abnormal data.

9. The neutron noise anomaly detection method based on machine learning according to claim 1, characterized in that, It also includes real-time monitoring of the LSTM model's performance and automatic model adjustments.

10. The neutron noise anomaly detection method based on machine learning according to claim 1, characterized in that, It also includes real-time monitoring of the performance of the isolated forest model and automatic adjustment of the model.