Nuclear power safety radiation dose rate predictor based on attention decomposition mechanism

Abstract

The application discloses a nuclear power safety radiation dose rate predictor based on an attention decomposition mechanism, which comprises a database, a host computer and a sensor, and the database, the host computer and the sensor are sequentially connected. The database contains the radiation dose rate data of the nuclear power safety historical monitoring. The host computer comprises a supervised training module, a data prediction module and a prediction display module, the supervised training module uses the nuclear power safety historical monitoring data to train a deep prediction network model and transmits the data into the data prediction module. The sensor acquires the nuclear power safety monitoring data in real time and transmits the data into the data prediction module of the host computer for prediction and estimation, and the prediction result is displayed on the prediction display module of the host computer. The application innovatively uses the attention decomposition mechanism, obtains the seasonal and trend characteristics of the data, better grasps the inherent characteristics and rules of the data, and realizes the nuclear power safety radiation dose rate predictor based on the attention decomposition mechanism with high accuracy and online prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, and particularly relates to a nuclear power safety radiation dose rate predictor based on an attention decomposition mechanism. BACKGROUND

[0002] Nuclear power safety radiation dose rate prediction is a key technology widely used in nuclear energy industry, nuclear accident emergency management, environmental protection and other fields. Accurate prediction of nuclear power safety radiation dose rate is of great significance, which can be used to protect people's health and environmental safety, and provide decision support. However, due to the complexity and uncertainty of nuclear power safety, the prediction of nuclear radiation dose rate faces many challenges. First, nuclear power safety is affected by many factors, such as nuclear accidents, weather conditions, environmental factors, etc. Changes in these factors will lead to instability and volatility of radiation dose rate data. Second, the monitoring data of nuclear power safety may be incomplete and inaccurate, which may be caused by limitations of monitoring equipment, difficulties of data collection and errors in monitoring data processing. Incomplete and inaccurate data will affect the establishment of prediction model and the accuracy of prediction results.

[0003] Currently, there have been some researches on nuclear power safety radiation dose rate prediction, mainly including statistical and model-based methods, machine learning and artificial intelligence-based methods, and time series analysis-based methods. However, these methods have some common limitations in nuclear power safety radiation dose rate prediction: on the one hand, they cannot effectively handle the seasonal and trend characteristics of the data, limiting the accuracy and stability of the prediction model; on the other hand, they lack understanding of the complex characteristics and rules of nuclear power safety data and are cumbersome to process, resulting in the need to improve the reliability and real-time performance of the prediction model. Therefore, it is urgent to develop a new type of nuclear power safety radiation dose rate predictor to overcome the shortcomings of existing methods and improve the accuracy and real-time performance of prediction. The instrument should have the ability to effectively process and analyze nuclear power safety data, be able to make full use of historical monitoring data and real-time monitoring data, extract key features and make accurate predictions. At the same time, the instrument should have good adaptability and scalability to cope with different environmental conditions and monitoring needs, providing reliable radiation dose rate prediction services for the nuclear energy industry and environmental protection. The innovation of the present application lies in the introduction of the attention decomposition mechanism, which can better capture the seasonal and trend characteristics of the data, improve the accuracy and real-time performance of the predictor, and provide a new solution for nuclear power safety radiation dose rate prediction. SUMMARY

[0004] In order to overcome the deficiencies of the existing nuclear power safety radiation dose rate predictor in poor accuracy and poor real-time performance, the present application aims to provide a nuclear power safety radiation dose rate predictor based on attention decomposition mechanism, which can make full use of the data rules of historical monitoring data and real-time monitoring data, efficiently extract potential features in the data, and realize reliable, accurate and timely prediction results.

[0005] The technical scheme adopted by the present application to solve its technical problems is: a nuclear power safety radiation dose rate predictor based on an attention decomposition mechanism, comprising a database, a host computer and a sensor, which are connected in sequence; the database is used to store the historical monitoring radiation dose rate data of nuclear power safety, and the data stored in the database is used by the host computer; the host computer comprises a supervision training module, a data prediction module and a prediction display module connected in sequence; the supervision training module of the host computer uses the historical monitoring radiation dose rate data of nuclear power safety to train a deep prediction network model, and then uploads the trained deep prediction network model to the data prediction module of the host computer; the sensor acquires real-time nuclear environment monitoring data and uploads it to the data prediction module of the host computer, and the data prediction module predicts and estimates the real-time nuclear environment monitoring data; finally, the prediction result is displayed on the prediction display module of the host computer.

[0006] Further, given the historical monitoring radiation dose rate data of nuclear power safety, referred to as historical monitoring data X t =[x t+1 ,x t+2 ,...,x t+T ], where x t+1 represents the historical monitoring data at the t+1 time step, T represents the length of the historical monitoring data time window, where D represents the dimension of the historical monitoring data; is the radiation dose rate at the t+T+1 to t+T+H time step, H represents the length of the prediction data time window; the target is to learn a function such that Further, the supervision training module comprises a deep prediction network embedding layer module, a deep prediction network encoder module, a deep prediction network decoder module and a deep prediction network training module:

[0007] 1) the deep prediction network embedding layer module, which encodes and processes the historical monitoring radiation dose rate data of nuclear power safety, encodes the data using multiple embedding forms, better grasps the inherent characteristics and rules of the data, and improves the prediction accuracy. It contains position embedding, label embedding and time embedding:

[0008] X em=Position(X in )+Token(X in )+Temporal(X in (1)

[0009] Among them, X in =X t This represents the input radiation dose rate data from historical nuclear power plant safety monitoring. Position(·), Token(·), and Temporal(·) represent the position embedding, token embedding, and time embedding operations, respectively. X em This represents the output of the embedding layer module of the deep prediction network.

[0010] 2) The deep prediction network encoder module efficiently extracts data features, further improving prediction accuracy. The deep prediction network encoder module encodes the output of the deep prediction network embedding layer module for feature extraction. The specific process is as follows:

[0011] First, the multi-head attention representation of the sequence is:

[0012] MultiHead(Q,K,V)=Concat(h1,h2,…,h n W (2)

[0013] h i =Attention(Q) i ,K i V i (3)

[0014]

[0015] Where MultiHead(·) represents multi-head attention operation, Concat(·) represents concatenation operation, Attention(·) represents attention operation, SoftMax(·) represents standardization operation on attention scores, Q, K, and V represent query, key, and value matrices, respectively, and h i Let W represent the i-th attention head, 1≤i≤n, where n represents the total number of attention heads, W represents the mapping matrix, and Q represents the mapping matrix. i K i and V i Let d represent the matrix of keys and values ​​for the i-th query. k Q represents i The first dimension of the matrix. Multi-head attention extraction is performed on the output of the deep prediction network embedding layer module:

[0016] X a =MultiHead(X em (5)

[0017] Where X a This represents the encoded output after multi-head attention.

[0018] Next, the output of the multi-head attention layer is decomposed, and the module that decomposes the multi-head attention layer output is defined as the decomposer. The decomposer pads the input before the feature extraction step, and then uses a one-dimensional average pooling layer to traverse the entire data time series and find its seasonal and trend features. Different sized convolutional kernels extract seasonal and trend features of different lengths, better grasping the inherent characteristics and patterns of the data and improving prediction accuracy. The final seasonal and trend terms are obtained:

[0019]

[0020] Where Decomposer(·) represents the decomposer operation. and These represent the seasonal and trend features of the data in the encoder, respectively.

[0021] Then, convolutional layers are used to learn seasonal and trend features from the encoder, respectively. The output of the convolutional layers is then passed through a decomposer to further reveal the temporal correlation of the data.

[0022]

[0023] in, and These represent the seasonal and trend features of the data in the encoder after feature extraction, respectively.

[0024] Finally, the output of the depth prediction network encoder module is obtained from the feedforward layer:

[0025]

[0026] Where LayerNorm(·) represents the layer normalization operation, Linear(·) represents a multilayer perceptron, and X en This represents the output of the depth prediction network encoder module.

[0027] 3) A deep prediction network decoder module, which decodes the output of the deep prediction network encoder module to obtain the final prediction output. The specific process is as follows:

[0028] First, the raw data, i.e., the detection data, goes directly into the decoder without embedding processing. The first module of the decoder is the decomposer:

[0029]

[0030] in, and X represents the seasonal and trend features of the data in the decoder, respectively. de =X t This represents the input data to the decoder, i.e., the raw data.

[0031] Next, convolutional layers are used to learn the seasonal and trend features of the decoder, respectively, to obtain the seasonal and trend features in the decoder after feature extraction. and

[0032]

[0033] Finally, the output of the deep network encoder module is summed with the seasonal and trend features in the feature-extracted decoder and then fed into the feedforward layer to obtain the decoder output:

[0034]

[0035] in, It is the radiation dose rate predicted by the network from the (t+T+1)th to the (t+T+H)th time step. α represents the output X of the depth prediction network encoder module. en The output percentage factor.

[0036] 4) Deep prediction network training module, which is used to train a deep prediction network model. The training objective is:

[0037]

[0038] in, This represents the loss function, and the model is trained using backpropagation during the training process.

[0039] 5) Upload the trained deep prediction network model to the data prediction module.

[0040] Furthermore, the sensor acquires radiation dose rate data from nuclear environment monitoring in real time and uploads it to the data prediction module. The data prediction module receives the trained deep prediction network model from the supervised training module and uses the model to predict and estimate the real-time monitoring data. Finally, the prediction results are displayed in the prediction display module.

[0041] The beneficial effects of this invention are as follows: Addressing the problem of nuclear power plant safety radiation dose rate prediction, this invention uses various embedding methods to encode data, employs an attention-based encoder and decoder to efficiently extract data features and perform predictions, and utilizes an attention decomposition mechanism to obtain the seasonal and trend characteristics of the data, thus better grasping the inherent characteristics and patterns of the data, ultimately achieving real-time intelligent prediction of nuclear power plant safety. It has the following advantages: 1. High prediction accuracy and strong stability; 2. Strong real-time performance and high reliability. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a schematic diagram of the hardware structure of the instrument proposed in this invention;

[0044] Figure 2 This is a schematic diagram of the functional modules of the host computer proposed in this invention. Detailed Implementation

[0045] The present invention will now be described in detail with reference to the accompanying drawings. Unless otherwise specified, the features of the following embodiments and implementations can be combined with each other.

[0046] refer to Figure 1 , Figure 2 A nuclear power plant safety radiation dose rate predictor based on an attention decomposition mechanism includes a database 1, a host computer 2, and sensors 3, which are connected sequentially. The database 1 stores radiation dose rate data from historical nuclear power plant safety monitoring, and the data stored in the database 1 is used by the host computer 2. The host computer 2 includes a supervised training module 4, a data prediction module 5, and a prediction display module 6, which are connected sequentially. The supervised training module 4 of the host computer 2 uses historical nuclear power plant safety monitoring data to train a deep prediction network model, and then uploads the trained deep prediction network model to the data prediction module 5 of the host computer. The sensors 3 acquire nuclear environment monitoring data in real time and upload it to the data prediction module 5 of the host computer 2. The data prediction module 5 uses the trained deep prediction network model to predict and estimate the real-time monitoring data. Finally, the prediction results are displayed on the prediction display module 6 of the host computer 2.

[0047] Specifically, database 1 is responsible for storing a large amount of historical radiation dose rate data from nuclear power plant safety monitoring. Given historical monitoring data X... t =[x t+1 ,xt+2 ,…,x t+T ], where x t+1 This represents the historical monitoring data at time step t+1, where T represents the length of the historical monitoring data time window. Where D represents the dimension of historical monitoring data; It is the radiation dose rate from the (t+T+1)th time step to the (t+T+H)th time step. H represents the length of the time window for the prediction data; the goal is to learn a function. Make

[0048] Specifically, the supervised training module 4 includes a deep prediction network embedding layer module 7, a deep prediction network encoder module 8, a deep prediction network decoder module 9, and a deep prediction network training module 10.

[0049] 1) Deep Prediction Network Embedding Layer Module 7: This module performs embedding and encoding processing on the raw data from nuclear power safety monitoring. It uses multiple embedding methods to encode the data, better grasping its inherent characteristics and patterns, and improving prediction accuracy. These include positional embedding, label embedding, and temporal embedding. Positional embedding uses sine and cosine functions to encode the input order of the data; label embedding is a one-dimensional convolution operation to expand the feature dimension of the data; and temporal embedding is responsible for further extracting the temporal relationships between the data.

[0050] X em =Position(X in )+Token(X in )+Temporal(X in (1)

[0051] Among them, X in =X t This represents the raw data from nuclear power safety monitoring. Position(·), Token(·), and Temporal(·) represent the position embedding, token embedding, and time embedding operations, respectively. X em This represents the output of the embedding layer module 7 of the deep prediction network.

[0052] 2) The deep prediction network encoder module 8 efficiently extracts data features, further improving prediction accuracy. The deep prediction network encoder module 8 encodes the output of the deep prediction network embedding layer module 7 for feature extraction. The specific process is as follows:

[0053] First, the multi-head attention representation of the sequence is:

[0054] MultiHead(Q,K,V)=Concat(h1,h2,…,h nW (2)

[0055] h i =Attention(Q) i ,K i V i (3)

[0056]

[0057] Where MultiHead(·) represents multi-head attention operation, Concat(·) represents concatenation operation, Attention(·) represents attention operation, SoftMax(·) represents standardization operation on attention scores, Q, K, and V represent query, key, and value matrices, respectively, and h i Let represent the i-th attention head, n represent the total number of attention heads, W represent the mapping matrix, and Q represent the... i K i and V i Let d represent the matrix of keys and values ​​for the i-th query. k Q represents i The first dimension of the matrix. Multi-head attention extraction is performed on the output of the embedding layer module 7 of the deep prediction network:

[0058] X a =MultiHead(X em (5)

[0059] Where X a This represents the encoded output after multi-head attention.

[0060] Next, the output of the multi-head attention layer is decomposed to obtain the seasonal and trend features of the input data, effectively predicting nuclear radiation data. Nuclear radiation data itself has complex patterns and is difficult to predict. Dividing the data into two parts allows for a better understanding of its inherent characteristics and patterns. The module that decomposes the output of the multi-head attention layer is defined as a decomposer. The decomposer performs a padding operation on the input before the feature extraction step, allowing the input dimension to be the same as the output dimension. To better utilize the information at both ends of the time series data, the data at both ends is first copied. Then, a one-dimensional average pooling layer is used to traverse the entire time series data and find its seasonal and trend features. Convolutional kernels of different sizes extract seasonal and trend features of different lengths, better grasping the inherent characteristics and patterns of the data and improving prediction accuracy. The final seasonal and trend terms are obtained:

[0061]

[0062] Where Decomposer(·) represents the decomposer operation. and These represent the seasonal and trend features of the data in the encoder, respectively.

[0063] Then, convolutional layers are used to learn seasonal and trend features from the encoder, respectively. The output of the convolutional layers is then passed through a decomposer to further reveal the temporal correlation of the data.

[0064]

[0065] in, and These represent the seasonal and trend features of the data in the encoder after feature extraction, respectively.

[0066] Finally, the output of the depth prediction network encoder module 8 is obtained from the feedforward layer:

[0067]

[0068] Where LayerNorm(·) represents the layer normalization operation, Linear(·) represents a multilayer perceptron, and X en This indicates the output of depth prediction network encoder module 8.

[0069] 3) Depth prediction network decoder module 9, which decodes the output of the depth prediction network encoder module 8 to obtain the final prediction output. The specific process is as follows:

[0070] First, to preserve the initial features of the original input data, the raw input data goes directly into the decoder without embedding. The first module of the decoder is the decomposer:

[0071]

[0072] in, and X represents the seasonal and trend features of the data in the decoder, respectively. de =X t This represents the input data to the decoder, i.e., the raw data.

[0073] Next, convolutional layers are used to learn the seasonal and trend features of the decoder, respectively, to obtain the seasonal and trend features in the decoder after feature extraction. and

[0074]

[0075] Finally, the output of the deep network encoder module 8 is summed with the seasonal and trend features in the feature-extracted decoder and then input into the feedforward layer to obtain the decoder output:

[0076]

[0077] in, It is the radiation dose rate predicted by the network from the (t+T+1)th to the (t+T+H)th time step. α represents the output X of the depth prediction network encoder module. en The output percentage factor.

[0078] 4) Deep prediction network training module 10, which is used to train the deep prediction network defined in steps 1), 2), and 3), with the training objective being:

[0079]

[0080] in, This represents the loss function, and the model is trained using backpropagation during the training process.

[0081] 5) Upload the trained deep prediction network model to the data prediction module 5.

[0082] Specifically, the sensor 3 acquires radiation dose rate data from nuclear environment monitoring in real time and uploads it to the data prediction module 5. The data prediction module 5 receives the trained deep prediction network model from the supervised training module 4 and uses the model to predict and estimate the real-time monitoring data. Finally, the prediction results are displayed in the prediction display module 6.

[0083] The above embodiments are used to explain and illustrate the present invention, but not to limit the present invention. Any modifications and changes made to the present invention within the spirit and scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A nuclear power plant safety radiation dose rate predictor based on an attention decomposition mechanism, characterized in that: The system comprises a database, a host computer, and sensors, which are connected sequentially. The database stores radiation dose rate data from historical nuclear power plant safety monitoring, and the data stored in the database is used by the host computer. The host computer includes a supervised training module, a data prediction module, and a prediction display module, all connected sequentially. The supervised training module uses the radiation dose rate data from historical nuclear power plant safety monitoring to train a deep prediction network model, and then uploads the trained deep prediction network model to the data prediction module of the host computer. The sensors acquire real-time nuclear environment monitoring data and upload it to the data prediction module of the host computer, which then performs prediction and estimation on the real-time nuclear environment monitoring data. Finally, the prediction results are displayed on the prediction display module of the host computer. The supervised training module includes a deep prediction network embedding layer module, a deep prediction network encoder module, a deep prediction network decoder module, and a deep prediction network training module. 1) A deep prediction network embedding layer module, which performs embedding encoding processing on radiation dose rate data from historical nuclear power safety monitoring, including location embedding, label embedding, and time embedding: X em =Position(X in )+Token(X in )+Temporal(X in ) (1) Among them, X in =X t This represents the input radiation dose rate data from historical nuclear power plant safety monitoring. Position(·), Token(·), and Temporal(·) represent the position embedding, token embedding, and time embedding operations, respectively. X em This represents the output of the embedding layer module in the deep prediction network; 2) A deep prediction network encoder module, which encodes the output of the deep prediction network embedding layer module for feature extraction; the specific process is as follows: First, multi-head attention is extracted from the output of the deep prediction network embedding layer module: X a =MultiHead(X em ) (5) Where X a This represents the encoded output after multi-head attention; Next, the output of the multi-head attention layer is decomposed, and the module that decomposes the output of the multi-head attention layer is defined as the decomposer. The decomposer performs padding operations on the input before the feature extraction step, and then passes it through a one-dimensional average pooling layer. The average pooling layer is used to traverse the entire data time series and find its seasonal and trend features. Convolutional kernels of different sizes extract seasonal and trend features of different lengths; thus, the final seasonal and trend terms are obtained. Where Decomposer(·) represents the decomposer operation. and These represent the seasonal and trend features of the data in the encoder, respectively. Then, convolutional layers are used to learn seasonal and trend features from the encoder, respectively; the output of the convolutional layers is then passed through a decomposer to further reveal the temporal correlation of the data. in, and These represent the seasonal and trend features of the data in the encoder after feature extraction, respectively; Finally, the output of the depth prediction network encoder module is obtained from the feedforward layer: Where LayerNorm(·) represents the layer normalization operation, Linear(·) represents a multilayer perceptron, and X en This represents the output of the depth prediction network encoder module; 3) A deep prediction network decoder module, which decodes the output of the deep prediction network encoder module to obtain the final prediction output; the specific process is as follows: First, the raw data, i.e., the detection data, goes directly into the decoder without embedding processing; the first module of the decoder is the decomposer: in, and X represents the seasonal and trend features of the data in the decoder, respectively. de =X t This represents the input data to the decoder, i.e., the raw data; Next, convolutional layers are used to learn the seasonal and trend features of the decoder, respectively, to obtain the seasonal and trend features in the decoder after feature extraction. and Finally, the output of the deep prediction network encoder module is summed with the seasonal and trend features in the feature-extracted decoder and then fed into the feedforward layer to obtain the decoder output: in, It is the radiation dose rate predicted by the network from the (t+T+1)th to the (t+T+H)th time step. α represents the output X of the depth prediction network encoder module. en ; output proportion factor; 4) Deep prediction network training module, which is used to train a deep prediction network model. The training objective is: in, This represents the loss function, and the model is trained using backpropagation during the training process. 5) Upload the trained deep prediction network model to the data prediction module.

2. The nuclear power plant safety radiation dose rate predictor based on attention decomposition mechanism as described in claim 1, characterized in that, Given radiation dose rate data from historical nuclear power plant safety monitoring, referred to as historical monitoring data X t =[x t+1 ,x t+2 ,...,x t+T ], where x t+1 This represents the historical monitoring data at time step t+1, where T represents the length of the historical monitoring data time window. Where D represents the dimension of historical monitoring data; It is the radiation dose rate from the (t+T+1)th time step to the (t+T+H)th time step. H represents the length of the forecast data time window; The goal is to learn a function Make 3. A nuclear power plant safety radiation dose rate predictor based on an attention decomposition mechanism as described in claim 1, characterized in that, Bullish attention is represented as: MultiHead(Q,K,V)=Concat(h1,h2,…,h n )W (2) h i =Attention(Q i ,K i ,V i ) (3) Where MultiHead(·) represents multi-head attention operation, Concat(·) represents concatenation operation, Attention(·) represents attention operation, SoftMax(·) represents standardization operation on attention scores, Q, K, and V represent query, key, and value matrices, respectively, and h i Let W represent the i-th attention head, 1≤i≤n, where n represents the total number of attention heads, W represents the mapping matrix, and Q represents the mapping matrix. i K i and V i Let d represent the matrix of keys and values ​​for the i-th query. k Q represents i The first dimension of the matrix.

4. A nuclear power plant safety radiation dose rate predictor based on an attention decomposition mechanism as described in claim 1, characterized in that, The sensor acquires radiation dose rate data from nuclear environment monitoring in real time and uploads it to the data prediction module. The data prediction module receives the trained deep prediction network model from the supervised training module and uses the model to predict and estimate the real-time monitoring data. Finally, the prediction results are displayed in the prediction display module.

Images