Multi-channel nuclear electronics data acquisition system based on artificial intelligence

Abstract

The present invention discloses an artificial intelligence-based multi-channel nuclear electronics data acquisition system, which relates to the technical field and includes front-end preprocessing of nuclear radiation or X-ray signals input from each detection channel; extracting features from the preprocessed signals using a deep neural network and integrating them to form a multidimensional state vector describing the channel state; inputting the multidimensional state vector into a deep reinforcement learning module to determine the sampling parameters of each channel; establishing a correlation model for the dependencies between the signals in each channel; implementing cross-compensation for channels in abnormal states by calculating and applying correction factors through a prediction method; and jointly adjusting the sampling parameters using a self-evolving multi-task optimization algorithm based on real-time feedback data to form a closed-loop data acquisition and control unit. This system can achieve automatic joint optimization of multi-task and multi-objective parameters, solving the problem that traditional methods cannot simultaneously consider real-time performance, stability, and multi-objective optimization.

Description

Technical Field

[0001] The present invention relates to the technical field of nuclear radiation or X-ray radiation measurement, and in particular to a multi-channel nuclear electronics data acquisition system based on artificial intelligence. Background Art

[0002] Existing nuclear electronics data acquisition technologies mainly rely on traditional fixed parameters and manual adjustment methods to achieve detection and data acquisition of nuclear radiation or X-ray signals. When faced with practical application scenarios such as multi-channel detection, environmental changes, and equipment aging, these methods have problems such as difficulty in flexibly adapting to sampling parameters, uneven signal processing effects, and low degree of automation. At the same time, in recent years, artificial intelligence technology, especially the application of deep learning and reinforcement learning in complex systems has continued to mature, providing a new theoretical and practical basis for accurate data acquisition and dynamic optimization in multi-task, big data environments. Therefore, the use of advanced intelligent algorithms to achieve adaptive, high-precision data acquisition and processing of multi-channel nuclear electronics systems has become a direction that urgently needs research and development. Summary of the Invention

[0003] In view of the above existing problems, the present invention is proposed.

[0004] Therefore, the present invention aims to address the problems of fixed sampling strategy, inflexible signal processing process and difficulty in handling abnormal data in existing multi-channel nuclear electronics data acquisition systems, and proposes a multi-channel nuclear electronics data acquisition method based on artificial intelligence. By introducing a deep neural network to extract features of preprocessed signals, using deep reinforcement learning to achieve adaptive allocation of dynamic sampling parameters, constructing a cross-channel correlation model to cross-compensate for abnormal data, and adopting a self-evolutionary multi-task optimization algorithm to jointly predict and globally search multiple key parameters such as sampling rate, integration time, signal amplification factor, energy consumption and hardware load, the automatic joint optimization of multi-task and multi-objective parameters is achieved, solving the problem that traditional methods cannot simultaneously take into account real-time performance, stability and multi-objective optimization.

[0005] To solve the above technical problems, the present invention provides the following technical solutions: a multi-channel nuclear electronics data acquisition system based on artificial intelligence, comprising:

[0006] Perform front-end preprocessing on the nuclear radiation or X-ray signals input from each detection channel, including noise suppression and baseline correction;

[0007] Use deep neural networks to extract signal-to-noise ratio, signal peak, and noise mean features from the front-end preprocessed signal, and integrate them to form a multidimensional state vector that describes the channel state;

[0008] The multidimensional state vector is input into the deep reinforcement learning module, and the sampling parameters of each channel are dynamically determined based on the multidimensional state vector;

[0009] Establish a correlation model of the dependency between signals in each channel to identify abnormal signals;

[0010] Implement cross compensation for channels in abnormal states, calculate and apply correction factors through predictive methods;

[0011] Based on real-time feedback data, the self-evolutionary multi-task optimization algorithm is used to jointly adjust the sampling parameters to form a closed-loop data acquisition control unit.

[0012] As a preferred solution of the artificial intelligence-based multi-channel nuclear electronics data acquisition system described in the present invention, the front-end preprocessing includes: the front-end signal processing circuit in each detection channel is integrated with a filtering and analog signal conditioning module for noise suppression and baseline drift correction of nuclear radiation or X-ray signals.

[0013] As a preferred solution of the multi-channel nuclear electronics data acquisition system based on artificial intelligence described in the present invention, the integration to form a multi-dimensional state vector describing the channel state includes: using convolutional neural network (CNN) to extract the features of the signal after front-end preprocessing, reconstructing the signal obtained after front-end preprocessing into a matrix form of a fixed size, applying a convolution layer with a 3×3 convolution kernel to perform local convolution operation on the input matrix, and using a maximum pooling layer with a window size of 2×2 to downsample the convolution output, stacking the second convolution layer and its corresponding maximum pooling layer in sequence to extract deeper features, and flattening the feature map generated by multi-layer convolution and pooling and mapping it to a fixed-dimensional feature vector through a fully connected layer.

[0014] As a preferred solution of the multi-channel nuclear electronics data acquisition system based on artificial intelligence of the present invention, the dynamic determination of the sampling parameters of each channel based on the multi-dimensional state vector includes extracting information describing the multi-channel detection state from the system input data through front-end signal processing and deep neural network, and clearly defining the action space, wherein the multi-dimensional state vector is specifically as follows:

[0015] s t =[SNR t , Peak t , NoiseMean t ,env t ]

[0016] Among them, s t Represents the multidimensional state vector of the detection channel at time t; SNR t is the signal-to-noise ratio; Peak t is the signal peak; NoiseMean t is the noise mean; envt For environmental parameters.

[0017] The action definition is as follows:

[0018] a t =(G t , T t , g t ) Among them, a t represents the sampling parameter combination selected by the system at time t; G t is the discrete sampling rate; T t is the discrete integration time; g t is the discrete signal amplification factor;

[0019] The states are discretized and mapped into discrete categories using clustering methods:

[0020] C(s t ) = k, k∈{1, 2, ..., K}

[0021] Among them, C(s t ) means s t The discretization result; k is the discrete category; K is the total number of categories;

[0022] The deep Q-network (DQN) is used to model the relationship between state and action, and the improved Actor-Critic structure is used to achieve strategy updates and parameter optimization. At the same time, cross-channel regularization is introduced to ensure the smoothness and consistency of the system.

[0023] The Q function is defined as follows:

[0024]

[0025] Among them, Q(s t ,a t ;θ) represents the t After taking an action, the expected value of the cumulative discounted reward is obtained; θ is the network weight parameter; γ is the discount factor, which represents the impact of future rewards on current decisions; R(s t+k ,a t+k ) is the immediate reward at the future time t+k; k1 represents the time;

[0026] Next, the Q function is used to update the network parameters as follows:

[0027] DQN updates the loss function:

[0028]

[0029] Among them, L(θ) is the loss function, which is used to measure the error between the current network prediction and the target value; r tis the actual reward obtained at the current moment; θ - Represents the parameters of the target network and plays a role in stable update; Indicates that in the next state s t+1 The maximum expected reward that can be obtained under the state; a' represents a placeholder for "all possible actions" and is used to search for the optimal action value in the next state;

[0030] Adopting an Actor-Critic structure, the strategy is refined through temporal difference updates, while regularization is introduced to ensure consistency of cross-channel decisions:

[0031] The timing difference error is as follows:

[0032] δ t =r t +γV(s t+1 θ v )-V(s t θ v )

[0033] Among them, δ t is the temporal difference error, which measures the deviation between the current state value estimate and the actual result; V(s t θ v ) is the state value function, estimated by the Critic network, θ v is the state value function parameter; V(s t θ v ) is the state value function of the next state;

[0034] The Critic network loss function is defined as follows:

[0035]

[0036] Among them, L v (θ v ) is the mean square error of the critic network, which is used to optimize the estimation of the state value function;

[0037] The Actor network update gradient is as follows:

[0038]

[0039] in, To update the policy network parameters θ a The gradient of π(a t |s t θ a ) is the strategy network in s t Select a in the state t Probability distribution of action; ||a t -at-1 || 2 Used to ensure smooth movement at consecutive moments; Used to constrain the consistency between sampling parameters of different detection channels; λ and λ a is the regularization coefficient, balancing the importance of temporal smoothness and cross-channel consistency; is the gradient operator, which is a vector operation for finding the partial derivative of the network parameter vector;

[0040] In order to ensure that the system can take into account data quality, energy consumption, and cross-channel consistency, a mechanism is introduced in which the auxiliary network dynamically generates the reward weights. The details are as follows:

[0041] The generation of dynamic weights is as follows:

[0042] w u =f u (s t , a t ;φ), and

[0043] Among them, w u represents the dynamic weight of the u-th reward, which is determined by the auxiliary network f u (s t , a t ; φ) calculation, where φ is the parameter of the auxiliary network; normalization condition Ensure that the sum of all weights is 1;

[0044] The comprehensive reward function is as follows:

[0045]

[0046] Among them, R(s t , a t ) is the system in s t Take a t The comprehensive reward obtained by the action; R1(s t , a t ) is the reward corresponding to the data quality; R2(s t , a t ) is the energy consumption penalty term; R3(s t , a t ) is the cross-channel consistency related reward.

[0047] As a preferred solution of the multi-channel nuclear electronics data acquisition system based on artificial intelligence of the present invention, the dynamic determination of the sampling parameters of each channel based on the multi-dimensional state vector includes using a self-attention mechanism to achieve effective fusion of cross-channel information in order to fully utilize the spatiotemporal dependencies in the multi-channel detection information, thereby providing more comprehensive input features for the decision module;

[0048] The query, key, and value mappings are:

[0049]

[0050] in, represents the state vector of the jth detection channel at time t; W Q , W K , W V They are used to generate query Q j , key K j and the value V j The linear transformation matrix of ; N is the total number of detection channels; the attention weight is calculated as follows:

[0051]

[0052] Among them, α ij is the attention weight from channel i to channel j; d k K j The dimension is used for normalization; l is the sum index used to traverse all channels; K l Representation refers to the key of each traversed channel l;

[0053] The fusion state vector is defined as follows:

[0054]

[0055] in, is the fused state vector, which represents the comprehensive state of the i-th channel after considering all channel information.

[0056] As a preferred solution of the multi-channel nuclear electronics data acquisition system based on artificial intelligence of the present invention, wherein: the correlation model of the dependency relationship between the signals of each channel includes performing statistical correlation analysis on the state vector of each detection channel, calculating the statistical correlation r between the i-th channel and the j-th channel ij , the formula is:

[0057]

[0058] in, Represents the state vector of the i-th channel at time t; represents the state vector of the jth channel at time t; μ i Represents the state vector The mean of μ j Represents the state vector The mean of i Represents the state vector The standard deviation of j Represents the state vector The standard deviation of ;∈ is a small positive constant to avoid the denominator being zero;

[0059] In order to adapt to the graph neural network processing, a nonlinear mapping F is used to transform Convert to a low-dimensional embedding vector h i :

[0060]

[0061] Among them, F:R d →R d′ is the designed nonlinear mapping function, which is composed of fully connected layers or convolutional layers; d is the dimension of the original state vector, d' is the dimension after node embedding; h i is the embedded representation of the i-th channel after mapping;

[0062] In obtaining the node embedding h i After that, the unnormalized attention score e is calculated through the graph attention mechanism ij , the formula is: e ij =LeakyReLU(a T [h i ||h j ])

[0063] Among them, [h i ||h j ] means embedding the node into h i With h j Perform splicing; a∈R 2d' is the parameter vector to be trained; a T [h i ||h j ] represents the linear combination of the concatenated vectors to obtain a scalar; LeakyReLU(·) is a linear rectification activation function with a negative slope, and the processed e ij represents the initial attention score of i to j;

[0064] Normalize the unnormalized attention scores through the Softmax operation and calculate the attention weights between each node

[0065]

[0066] in, represents the normalized attention weight from i to j; exp(e ij ) will score e ij Mapping to the positive real space; e iqIn the graph attention mechanism, the unnormalized attention score calculated from channel i to channel q is represented; the denominator normalizes all attention scores under channel i to ensure

[0067] In order to simultaneously utilize statistical correlation and attention scores obtained by graph neural networks, a weighted average method is used to generate the final cross-channel dependency matrix A, which is formulated as follows:

[0068]

[0069] Among them, A ij represents the comprehensive dependency between the i-th channel and the j-th channel; β is the weight parameter, and its value range is [0,1]; r ij It is the value calculated by statistical correlation; is the normalized attention weight calculated by the graph neural network.

[0070] As a preferred embodiment of the multi-channel nuclear electronics data acquisition system based on artificial intelligence of the present invention, the calculation and application of the correction factor by the prediction method includes:

[0071] The cross-channel correlation matrix is ​​used to perform traditional regression prediction on the abnormal channel to obtain the predicted value based on the remaining normal channel signals. The formula is:

[0072]

[0073] in, It represents the prediction signal of abnormal channel o obtained by statistical regression method at time t; represents the actual collected signal of the jth detection channel at time t; N is the total number of detection channels, where abnormal channels are excluded when summing; A oj represents the comprehensive dependency between the oth channel and the jth channel;

[0074] On the basis of traditional regression prediction, a neural network model is designed to input the signals of all channels except the abnormal channel, and predict the nonlinear residual that the regression model cannot capture. The formula is:

[0075]

[0076] in, represents the residual value predicted by the neural network at time t, corresponding to the nonlinear prediction part of the abnormal channel; N(·; Θ) represents the neural network that constitutes this prediction model, and its parameters are represented by Θ. The model can adopt a multilayer perceptron or convolutional neural network structure; represents the actual signal set of all channels except the abnormal channel at time t;

[0077] The traditional regression prediction error and the residual error of the neural network prediction are weighted and fused proportionally to calculate the comprehensive compensation factor of the abnormal channel. The formula is:

[0078]

[0079] in, It represents the comprehensive compensation factor of the abnormal channel at time t; is a signal predicted based on statistical regression; represents the actual collected signal of abnormal channel o at time t; represents the regression prediction error; is the residual error predicted by the neural network; γ1 is a weight factor between 0 and 1, which is used to balance the contributions of the two parts;

[0080] Apply the comprehensive compensation factor to the original signal of the abnormal channel to complete the data correction. The formula is:

[0081]

[0082] in, It represents the signal of abnormal channel o after correction at time t.

[0083] As a preferred solution of the multi-channel nuclear electronics data acquisition system based on artificial intelligence of the present invention, wherein: the self-evolutionary multi-task optimization algorithm includes: input data x t , composed of a multi-task deep neural network f(x t ;θ) outputs predictions of multiple target parameters Construct a joint loss function L(θ) to evaluate the network output error and perform a global optimal search for the network parameter θ; use the particle swarm optimization algorithm to update the speed and parameters, forming a self-evolving parameter adjustment mechanism;

[0084] The parameter update formula directly uses the updated velocity vector to correct the network parameters to form a new parameter set θ t+1 ; This process is repeated in each iteration until the joint loss function L(θ) meets the predetermined convergence condition.

[0085] A computer device includes a memory and a processor, wherein the memory stores a computer program and the processor implements the steps of a multi-channel nuclear electronics data acquisition system based on artificial intelligence when executing the computer program.

[0086] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a multi-channel nuclear electronics data acquisition system based on artificial intelligence.

[0087] Beneficial effects of the present invention: Through the technical solution proposed in the present invention, the system can realize automation, intelligence and closed-loop dynamic optimization in multi-channel nuclear electronics data acquisition. Each detection channel can adjust the sampling parameters and data processing strategies in real time according to the actual signal status and environmental changes. At the same time, the abnormal data can be compensated and corrected through the cross-channel correlation model, and the self-evolutionary multi-task optimization algorithm is used to realize the joint adjustment of multiple key target parameters, ensuring that various technical indicators reach the best ratio, and providing effective technical support for the precise measurement, automatic control and long-term operation of nuclear radiation and X-ray data acquisition systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0089] Figure 1 A flowchart of a multi-channel nuclear electronics data acquisition system based on artificial intelligence is provided for one embodiment of the present invention. DETAILED DESCRIPTION

[0090] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0091] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0092] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0093] The present invention is described in detail with reference to schematic diagrams. For ease of illustration, cross-sectional views of device structures may be partially enlarged and not to scale when describing embodiments of the present invention. Furthermore, the schematic diagrams are merely illustrative and should not limit the scope of the present invention. Furthermore, in actual production, the three-dimensional dimensions of length, width, and depth should be included.

[0094] In the description of the present invention, it should be noted that the terms "upper, lower, inner, and outer" and other references to orientations or positional relationships are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate and simplify the description of the present invention. They are not intended to indicate or imply that the devices or components referred to must have, be constructed, or operate in a specific orientation, and therefore should not be construed as limitations on the present invention. Furthermore, the terms "first, second, or third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0095] In this disclosure, unless otherwise specified or limited, the terms "mounted," "connected," and "connected" should be interpreted broadly. For example, they may refer to fixed, removable, or integral connections. They may also refer to mechanical, electrical, or direct connections, indirect connections through an intermediary, or internal communication between two components. Those skilled in the art will understand the specific meanings of these terms in this disclosure.

[0096] Example 1, with reference to Figure 1 , which is the first embodiment of the present invention, provides a multi-channel nuclear electronics data acquisition system based on artificial intelligence, comprising:

[0097] S1: Perform front-end preprocessing on the nuclear radiation or X-ray signals input by each detection channel, including noise suppression and baseline correction.

[0098] The front-end preprocessing includes that the front-end signal processing circuit in each detection channel is integrated with a filtering and analog signal conditioning module for performing noise suppression and baseline drift correction on nuclear radiation or X-ray signals.

[0099] S2: Use a deep neural network to extract the signal-to-noise ratio, signal peak, and noise mean features of the signal after front-end preprocessing, and integrate them to form a multidimensional state vector that describes the channel state.

[0100] The integration to form a multidimensional state vector describing the channel state includes extracting features from the pre-processed signal using a convolutional neural network (CNN), reconstructing the pre-processed signal into a fixed-size matrix, applying a convolutional layer with a 3×3 kernel to the input matrix, and downsampling the convolution output using a maximum pooling layer with a window size of 2×2. A second convolutional layer and its corresponding maximum pooling layer are then stacked to extract deeper features. The feature maps generated by the multiple convolutions and pooling layers are then flattened and mapped into a fixed-dimensional feature vector through a fully connected layer. This feature vector forms the main part of the state vector for use by subsequent modules.

[0101] S3: Input the multidimensional state vector into the deep reinforcement learning module and dynamically determine the sampling parameters of each channel based on the multidimensional state vector.

[0102] The dynamic determination of sampling parameters for each channel based on the multi-dimensional state vector includes extracting information describing the multi-channel detection state from system input data through front-end signal processing and deep neural networks, and clearly defining the action space, wherein the multi-dimensional state vector is specifically as follows:

[0103] s t =[SNR t , Peak t , NoiseMean t ,env t ]

[0104] Among them, s t Represents the multidimensional state vector of the detection channel at time t; SNR t is the signal-to-noise ratio; Peak t is the signal peak; NoiseMean t is the noise mean; env t are environmental parameters (such as temperature, voltage, etc.);

[0105] The multidimensional state vector converts the original preprocessed signal into a numerical description that can be input to the subsequent decision-making module, providing basic information for reinforcement learning decision-making.

[0106] The action definition is as follows:

[0107] a t =(G t , T t , g t )

[0108] Among them, a t represents the sampling parameter combination selected by the system at time t; G t is the discrete sampling rate; T t is the discrete integration time; gt is the discrete signal amplification factor;

[0109] In order to reduce the complexity of the continuous state space, the state is discretized and mapped into discrete categories through clustering method:

[0110] C(s t ) = k, k∈{1, 2, ..., K}

[0111] Among them, C(s t ) means s t The discretization result; k is the discrete category; K is the total number of categories;

[0112] The deep Q-network (DQN) is used to model the relationship between state and action, and the improved Actor-Critic structure is used to achieve strategy updates and parameter optimization. At the same time, cross-channel regularization is introduced to ensure the smoothness and consistency of the system.

[0113] The Q function is defined as follows:

[0114]

[0115] Among them, Q(s t , a t ;θ) represents the t After taking an action, the expected value of the cumulative discounted reward is obtained; θ is the network weight parameter; γ is the discount factor, which represents the impact of future rewards on current decisions; R(s t+k , a t+k ) is the immediate reward at the future time t+k; k1 represents the time;

[0116] By defining the Q function, we can establish a value mapping between states and actions, which can provide a basis for action selection. Next, we use the Q function to update the network parameters as follows:

[0117] DQN updates the loss function:

[0118]

[0119] Among them, L(θ) is the loss function, which is used to measure the error between the current network prediction and the target value; r t is the actual reward obtained at the current moment; θ - Represents the parameters of the target network and plays a role in stable update; Indicates that in the next state s t+1 The maximum expected reward that can be obtained under the state; a' represents a placeholder for "all possible actions" and is used to search for the optimal action value in the next state;

[0120] The DQN part provides state-based value estimation for the system. It adopts an Actor-Critic structure, refines the strategy through temporal difference updates, and introduces regularization to ensure consistency of cross-channel decisions:

[0121] The timing difference error is as follows:

[0122] δ t =r t +γV(s t+1 θ v )-V(s t θ v )

[0123] Among them, δ t is the temporal difference error, which measures the deviation between the current state value estimate and the actual result; V(s t θ v ) is the state value function, estimated by the Critic network, θ v is the state value function parameter; V(s t+1 θ v ) is the state value function of the next state;

[0124] The Critic network loss function is defined as follows:

[0125]

[0126] Among them, L v (θ v ) is the mean square error of the critic network, which is used to optimize the estimation of the state value function;

[0127] The Actor network update gradient (including cross-channel regularization) is as follows:

[0128]

[0129] in, To update the policy network parameters θ a The gradient of π(a t |s t θ a ) is the strategy network in s t Select a in the state t Probability distribution of action; ||a t -a t-1 || 2 Used to ensure smooth movement at consecutive moments; Used to constrain the consistency between sampling parameters of different detection channels; λ and λ a is the regularization coefficient, balancing the importance of temporal smoothness and cross-channel consistency; is the gradient operator, which is a vector operation for finding the partial derivative of the network parameter vector;

[0130] The Actor-Critic structure completes the iterative update of the strategy through the above formula, and after introducing regularization constraints, it can further coordinate the decisions between channels.

[0131] In order to ensure that the system can take into account data quality, energy consumption, and cross-channel consistency, a mechanism is introduced in which the auxiliary network dynamically generates the reward weights. The details are as follows:

[0132] The generation of dynamic weights is as follows:

[0133] w u =f u (s t , a t ;φ), and

[0134] Among them, w u represents the dynamic weight of the u-th reward, which is determined by the auxiliary network f u (s t , a t ; φ) calculation, where φ is the parameter of the auxiliary network; normalization condition Ensure that the sum of all weights is 1;

[0135] The comprehensive reward function is as follows:

[0136]

[0137] Among them, R(s t , a t ) is the system in s t Take a t The comprehensive reward obtained by the action; R1(s t , a t ) is the reward corresponding to the data quality; R2(s t ,a t ) is the energy consumption penalty term; R3(s t ,a t ) is the cross-channel consistency related reward.

[0138] Through the dynamic weight generation mechanism, the reward function can be autonomously adjusted according to the actual state and action, thereby more flexibly guiding the update of reinforcement learning.

[0139] In order to fully utilize the spatiotemporal dependencies in multi-channel detection information, a self-attention mechanism is used to achieve effective fusion of cross-channel information, thereby providing more comprehensive input features for the decision module;

[0140] The query, key, and value mappings are:

[0141]

[0142] in, represents the state vector of the jth detection channel at time t; W Q ,W K ,W V They are used to generate query Q j , key K j and the value V j The linear transformation matrix; N is the total number of detection channels;

[0143] These mappings help capture the similarities and interaction information between different channels.

[0144] The attention weight is calculated as follows:

[0145]

[0146] Among them, α ij is the attention weight from channel i to channel j; d k K j The dimension is used for normalization; l is the sum index used to traverse all channels; K l Representation refers to the key of each traversed channel l;

[0147] The fusion state vector is defined as follows:

[0148]

[0149] in, is the fused state vector, which represents the comprehensive state of the i-th channel after considering all channel information.

[0150] S4: Establish a correlation model of the dependency relationship between the signals of each channel to identify abnormal signals.

[0151] The correlation model of the dependency relationship between the signals of each channel is established by performing statistical correlation analysis on the state vector of each detection channel and calculating the statistical correlation r between the i-th channel and the j-th channel. ij , the formula is:

[0152]

[0153] in, Represents the state vector of the i-th channel at time t; represents the jth pass at time t

[0154] The state vector of the Tao; μ i Represents the state vector The mean of μ j Represents the state vector The mean of i Represents the state vector The standard deviation of j Represents the state vector The standard deviation of ;∈ is a small positive constant to avoid the denominator being zero;

[0155] This formula uses the idea of ​​Pearson correlation coefficient to express the similarity of the two channel state vectors with the value r ij It represents the basic statistical information for the subsequent introduction of attention weights extracted by graph neural networks.

[0156] In order to adapt to the graph neural network processing, a nonlinear mapping F is used to transform Convert to a low-dimensional embedding vector h i :

[0157]

[0158] Among them, F:R d →R d′ is the designed nonlinear mapping function, which is composed of fully connected layers or convolutional layers; d is the dimension of the original state vector, d' is the dimension after node embedding; h i is the embedded representation of the i-th channel after mapping;

[0159] The mapped node embedding h i It will serve as the input for calculating the similarity between nodes in the graph attention mechanism, laying the foundation for further constructing the attention score of the graph neural network.

[0160] In obtaining the node embedding h i After that, the unnormalized attention score e is calculated through the graph attention mechanism ij , the formula is:

[0161] e ij =LeakyReLU(a T [h i ||h j ])

[0162] Among them, [h i ||h j ] means embedding the node into h i With h j Perform splicing; a∈R 2d' is the parameter vector to be trained; a T [h i ||h j ] represents the linear combination of the concatenated vectors to obtain a scalar; LeakyReLU(·) is a linear rectification activation function with a negative slope, and the processed e ijrepresents the initial attention score of i to j;

[0163] Normalize the unnormalized attention scores through the Softmax operation and calculate the attention weights between each node

[0164] in, represents the normalized attention weight from i to j; exp(e ij ) will score e ij Mapping to the positive real space; e iq In the graph attention mechanism, the unnormalized attention score calculated from channel i to channel q is represented; the denominator normalizes all attention scores under channel i to ensure

[0165] Attention weights obtained by normalization It reflects the correlation information between channels obtained through graph neural network calculation. The next step will be combined with statistical correlation to generate the final cross-channel dependency matrix.

[0166] In order to simultaneously utilize statistical correlation and attention scores obtained by graph neural networks, a weighted average method is used to generate the final cross-channel dependency matrix A, which is formulated as follows:

[0167]

[0168] Among them, A ij represents the comprehensive dependency between the i-th channel and the j-th channel; β is the weight parameter, and its value range is [0,1]; r ij It is the value calculated by statistical correlation; is the normalized attention weight calculated by the graph neural network.

[0169] By A ij The matrix A∈R N×N It is the cross-channel dependency matrix used in subsequent systems. The cross-channel dependency matrix provides an information expression of the mutual dependence between multiple channels and is the final output of the entire cross-channel correlation modeling process.

[0170] S5: Implement cross compensation for channels in abnormal states and calculate and apply correction factors through prediction methods.

[0171] The calculation and application of the correction factor by the prediction method includes,

[0172] The cross-channel correlation matrix is ​​used to perform traditional regression prediction on the abnormal channel to obtain the predicted value based on the remaining normal channel signals. The formula is:

[0173]

[0174] in, It represents the prediction signal of abnormal channel o obtained by statistical regression method at time t; represents the actual collected signal of the jth detection channel at time t; N is the total number of detection channels, where abnormal channels are excluded when summing; A oj represents the comprehensive dependency between the oth channel and the jth channel;

[0175] The abnormal channel is preliminarily predicted using the signals of other channels through the linear regression method, providing a benchmark prediction component for the subsequent introduction of neural network for nonlinear compensation.

[0176] On the basis of traditional regression prediction, a neural network model is designed to input the signals of all channels except the abnormal channel, and predict the nonlinear residual that the regression model cannot capture. The formula is:

[0177]

[0178] in, represents the residual value predicted by the neural network at time t, corresponding to the nonlinear prediction part of the abnormal channel; N(·; Θ) represents the neural network that constitutes this prediction model, and its parameters are represented by Θ. The model can adopt a multilayer perceptron or convolutional neural network structure; represents the actual signal set of all channels except the abnormal channel at time t;

[0179] The traditional regression prediction error and the residual error of the neural network prediction are weighted and fused proportionally to calculate the comprehensive compensation factor of the abnormal channel. The formula is:

[0180]

[0181] in, It represents the comprehensive compensation factor of the abnormal channel at time t; is a signal predicted based on statistical regression; represents the actual collected signal of abnormal channel o at time t; represents the regression prediction error (i.e., the deviation between the predicted signal and the actual signal); is the residual predicted by the neural network; γ1 is a weight factor between 0 and 1, which is used to balance the contributions of the two parts; its value can be determined by the validation set according to the actual situation or adjusted using an adaptive strategy.

[0182] By integrating the linear prediction error and the nonlinear residual, a comprehensive compensation factor is formed. This factor reflects the relative role of each method in the correction of abnormal data and provides a direct basis for the final data correction.

[0183] Apply the comprehensive compensation factor to the original signal of the abnormal channel to complete the data correction. The formula is:

[0184]

[0185] in, It represents the signal of abnormal channel o after correction at time t.

[0186] S6: Based on real-time feedback data, the self-evolutionary multi-task optimization algorithm is used to jointly adjust the sampling parameters to form a closed-loop data acquisition control unit.

[0187] The self-evolutionary multi-task optimization algorithm includes: input data x t , composed of a multi-task deep neural network f(x t ;θ) outputs predictions of multiple target parameters Construct a joint loss function L(θ) to evaluate the network output error and perform a global optimal search for the network parameter θ; use the particle swarm optimization algorithm to update the speed and parameters, forming a self-evolving parameter adjustment mechanism;

[0188] The parameter update formula directly uses the updated velocity vector to correct the network parameters to form a new parameter set θ t+1 ; This process is repeated in each iteration until the joint loss function L(θ) meets the predetermined convergence condition.

[0189] The entire process is completed sequentially, achieving the joint prediction and optimization of multiple objectives such as sampling rate, integration time, signal amplification factor, energy consumption, and hardware load. The innovation of this model lies in combining multi-task deep prediction with the global search algorithm PSO, providing an adaptive, globally optimal, self-evolving multi-objective optimization solution.

[0190] Example 2

[0191] The second embodiment of the present invention is different from the previous embodiment in that:

[0192] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0193] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0194] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0195] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0196] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0197] Obviously, those skilled in the art may make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is intended to include these modifications and variations.

Claims

1. An artificial intelligence-based multi-channel nuclear electronics data acquisition system, characterized by: include, Perform front-end preprocessing on the nuclear radiation or X-ray signals input from each detection channel, including noise suppression and baseline correction; Use deep neural networks to extract signal-to-noise ratio, signal peak, and noise mean features from the front-end preprocessed signal, and integrate them to form a multidimensional state vector that describes the channel state; The multidimensional state vector is as follows: s t =[SNR t ,Peak t ,NoiseMean t ,env t ]; Among them, s t Represents the multidimensional state vector of the detection channel at time t; SNR t is the signal-to-noise ratio; Peak t is the signal peak; NoiseMean t is the noise mean; env t is the environmental parameter; The multi-dimensional state vector is input into the deep reinforcement learning module, and the sampling parameters of each channel are dynamically determined according to the multi-dimensional state vector; starting from the system input data, through front-end signal processing and deep neural network, the information describing the multi-channel detection state is extracted and the action space is clearly defined. The action definition is as follows: a t =(G t ,T t ,g t ); Among them, a t represents the sampling parameter combination selected by the system at time t; G t is the discrete sampling rate; T t is the discrete integration time; g t is the discrete signal amplification factor; Discretize the state into discrete categories through clustering method; The deep Q-network (DQN) is used to model the relationship between state and action, and the improved Actor-Critic structure is used to achieve strategy updates and parameter optimization. At the same time, cross-channel regularization is introduced to ensure the smoothness and consistency of the system. Adopting an Actor-Critic structure, the strategy is refined through temporal difference updates, while regularization is introduced to ensure consistency of cross-channel decisions: In order to ensure that the system can take into account data quality, energy consumption, and cross-channel consistency, a mechanism is introduced in which the auxiliary network dynamically generates the reward weights. The details are as follows: The generation of dynamic weights is as follows: w u = f u (s t , a t ; φ), and ; Among them, w u represents the dynamic weight of the u-th reward, which is determined by the auxiliary network f u (s t , a t ; φ) calculation, where φ is the parameter of the auxiliary network; normalization condition Ensure that the sum of all weights is 1; The comprehensive reward function is as follows: ； Among them, R(s t ,a t ) is the system in s t Take a t The comprehensive reward obtained by the action; R1(s t ,a t ) is the reward corresponding to the data quality; R2(s t , a t ) is the energy consumption penalty term; R3(s t , a t ) is the cross-channel consistency related reward; In order to fully utilize the spatiotemporal dependencies in multi-channel detection information, a self-attention mechanism is used to achieve effective fusion of cross-channel information, thereby providing more comprehensive input features for the decision module; Establish a correlation model of the dependency between signals in each channel to identify abnormal signals; Implement cross compensation for channels in abnormal states, calculate and apply correction factors through predictive methods; Based on real-time feedback data, the self-evolutionary multi-task optimization algorithm is used to jointly adjust the sampling parameters to form a closed-loop data acquisition control unit.

2. The multi-channel nuclear electronics data acquisition system based on artificial intelligence according to claim 1, characterized in that: The front-end preprocessing includes that the front-end signal processing circuit in each detection channel is integrated with a filtering and analog signal conditioning module for performing noise suppression and baseline drift correction on nuclear radiation or X-ray signals.

3. The multi-channel nuclear electronics data acquisition system based on artificial intelligence according to claim 2, characterized in that: The integration to form a multi-dimensional state vector describing the channel state includes: using a convolutional neural network (CNN) to extract features from the signal after front-end preprocessing, reconstructing the signal obtained after front-end preprocessing into a matrix form of a fixed size, applying a convolution layer with a 3×3 convolution kernel to perform a local convolution operation on the input matrix, and using a maximum pooling layer with a window size of 2×2 to downsample the convolution output, stacking the second convolution layer and its corresponding maximum pooling layer in sequence to extract deeper features, and flattening the feature map generated by multiple layers of convolution and pooling and mapping it to a fixed-dimensional feature vector through a fully connected layer.

4. The artificial intelligence-based multi-channel nuclear electronics data acquisition system according to claim 3, wherein: The dynamically determining the sampling parameters of each channel according to the multidimensional state vector includes discretizing the state into discrete categories by a clustering method: ； in, express The discretization result of ; is a discrete category; is the total number of categories; The deep Q-network (DQN) is used to model the relationship between state and action, and the improved Actor-Critic structure is used to achieve strategy updates and parameter optimization. At the same time, cross-channel regularization is introduced to ensure the smoothness and consistency of the system. The Q function is defined as follows: ； Where Q(s t , a t ;θ) represents the t After taking an action in the state, the expected value of the cumulative discounted reward is obtained; θ is the network weight parameter; γ is the discount factor, which represents the impact of future rewards on the current decision; R(s t+k , a t+k ) is the immediate reward at the future time t+k; k1 represents the time; Next, the Q function is used to update the network parameters as follows: DQN updates the loss function: ； Among them, L(θ) is the loss function, which is used to measure the error between the current network prediction and the target value; r t is the actual reward obtained at the current moment; θ - Represents the parameters of the target network and plays a role in stable update; maxQ(s t+1 , a′;θ - ) means in the next state s t+1 The maximum expected reward that can be obtained under the state; a' represents a placeholder for "all possible actions" and is used to search for the optimal action value in the next state; Adopting an Actor-Critic structure, the strategy is refined through temporal difference updates, while regularization is introduced to ensure consistency of cross-channel decisions: The timing difference error is as follows: δ t =r t +γV(s t+1 ;θ v )-V(s t ;θ v ); Among them, δ t is the temporal difference error, which measures the deviation between the current state value estimate and the actual result; V(s t θ v ) is the state value function, estimated by the Critic network, θ v is the state value function parameter; V(s t+1 θ v ) is the state value function of the next state; The Critic network loss function is defined as follows: ； Among them, L v (θ v ) is the mean square error of the critic network, which is used to optimize the estimation of the state value function; The Actor network update gradient is as follows: ； in, To update the policy network parameters gradient; For strategic networks Select in status Probability distribution of actions; Used to ensure smooth movement at consecutive moments; Used to constrain the consistency between sampling parameters of different detection channels; and is the regularization coefficient, balancing the importance of temporal smoothness and cross-channel consistency; ∇ is the gradient operator, which is a vector operation that finds the partial derivative of the network parameter vector.

5. The multi-channel nuclear electronics data acquisition system based on artificial intelligence according to claim 4, characterized in that: In order to make full use of the spatiotemporal dependencies in multi-channel detection information, a self-attention mechanism is used to achieve effective fusion of cross-channel information, thereby providing more comprehensive input features for the decision module; The query, key, and value mappings are: ； in, represents the state vector of the jth detection channel at time t; W Q , W K , W V They are used to generate query Q j , key K j and the value V j The linear transformation matrix; N is the total number of detection channels; The attention weight is calculated as follows: ； Among them, α ij is the attention weight from channel i to channel j; d k K j The dimension is used for normalization; l is the sum index used to traverse all channels; K l Representation refers to the key of each traversed channel l; The fusion state vector is defined as follows: ； in, is the fused state vector, which represents the comprehensive state of the i-th channel after considering all channel information.

6. The artificial intelligence-based multi-channel nuclear electronics data acquisition system according to claim 5, characterized in that: The correlation model of the dependency relationship between the signals of each channel is established by performing statistical correlation analysis on the state vector of each detection channel and calculating the statistical correlation r between the i-th channel and the j-th channel. ij , the formula is: ； in, Indicates time Next The state vector of each channel; Indicates time Next The state vector of each channel; Represents the state vector The mean of Represents the state vector The mean of Represents the state vector The standard deviation of Represents the state vector The standard deviation of is a small positive constant to avoid the denominator being zero; In order to adapt to the graph neural network processing, a nonlinear mapping F is used to transform Convert to a low-dimensional embedding vector h i : ； Among them, F:R d →R d' is the designed nonlinear mapping function, which is composed of fully connected layers or convolutional layers; d is the dimension of the original state vector, d' is the dimension after node embedding; h i is the embedded representation of the i-th channel after mapping; In obtaining the node embedding h i After that, the unnormalized attention score e is calculated through the graph attention mechanism ij , the formula is: have been ij =LeakyReLU(a T [h i ||h j ]) Among them, [h i ‖h j ] means embedding the node into h i With h j Perform splicing; a∈R 2d' is the parameter vector to be trained; a T [h i ‖h j ] represents the linear combination of the concatenated vectors to obtain a scalar; LeakyReLU(·) is a linear rectification activation function with a negative slope, and the processed e ij represents the initial attention score of i to j; Normalize the unnormalized attention scores through the Softmax operation and calculate the attention weights between each node : ； in, represents the normalized attention weight from i to j; exp(e ij ) will score e ij Mapping to the positive real space; e iq In the graph attention mechanism, the unnormalized attention score calculated from channel i to channel q is represented; the denominator normalizes all attention scores under channel i to ensure ; In order to simultaneously utilize statistical correlation and attention scores obtained by graph neural networks, a weighted average method is used to generate the final cross-channel dependency matrix A, which is formulated as follows: ； Among them, A ij represents the comprehensive dependency between the i-th channel and the j-th channel; β is the weight parameter, and its value range is [0,1]; r ij It is the value calculated by statistical correlation; is the normalized attention weight calculated by the graph neural network.

7. The artificial intelligence-based multi-channel nuclear electronics data acquisition system according to claim 6, characterized in that: The calculation and application of the correction factor by the prediction method includes, The cross-channel correlation matrix is ​​used to perform traditional regression prediction on the abnormal channel to obtain the predicted value based on the remaining normal channel signals. The formula is: ； in, It represents the prediction signal of abnormal channel o obtained by statistical regression method at time t; represents the actual collected signal of the jth detection channel at time t; N is the total number of detection channels, where abnormal channels are excluded when summing; A oj represents the comprehensive dependency between the oth channel and the jth channel; On the basis of traditional regression prediction, a neural network model is designed to input the signals of all channels except the abnormal channel, and predict the nonlinear residual that the regression model cannot capture. The formula is: ； in, represents the residual value predicted by the neural network at time t, corresponding to the nonlinear prediction part of the abnormal channel; N(·; Θ) represents the neural network that constitutes this prediction model, and its parameters are represented by Θ. The model can adopt a multilayer perceptron or convolutional neural network structure; represents the actual signal set of all channels except the abnormal channel at time t; The traditional regression prediction error and the residual error of the neural network prediction are weighted and fused proportionally to calculate the comprehensive compensation factor of the abnormal channel. The formula is: ； in, Indicates time Comprehensive compensation factor of the lower abnormal channel; is a signal predicted based on statistical regression; Indicates time Lower abnormal channel The actual collected signal; represents the regression prediction error; is the residual error predicted by the neural network; is a weight factor between 0 and 1, used to balance the contributions of the two parts; Apply the comprehensive compensation factor to the original signal of the abnormal channel to complete the data correction. The formula is: ； in, Indicates time The abnormal channel after correction Signal.

8. The multi-channel nuclear electronics data acquisition system based on artificial intelligence according to claim 7, characterized in that: The self-evolutionary multi-task optimization algorithm includes: input data x t , composed of a multi-task deep neural network f(x t ;θ) outputs predictions of multiple target parameters Construct a joint loss function L(θ) to evaluate the network output error and perform a global optimal search for the network parameter θ; use the particle swarm optimization algorithm to update the speed and parameters, forming a self-evolving parameter adjustment mechanism; The parameter update formula directly uses the updated velocity vector to correct the network parameters to form a new parameter set θ t+1 ; This process is repeated in each iteration until the joint loss function L(θ) meets the predetermined convergence condition.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the artificial intelligence-based multi-channel nuclear electronics data acquisition system according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the multi-channel nuclear electronics data acquisition system based on artificial intelligence according to any one of claims 1 to 8 are implemented.

Images