Nuclear radiation indicator early warning system based on multi-algorithm fusion

Abstract

This invention discloses a nuclear radiation indicator early warning system based on multi-algorithm fusion. The system includes an environmental data acquisition module for collecting gamma-ray dose rate data, wind speed data, wind direction data, and geographic location information to construct a nuclear radiation environmental dataset; a dose rate prediction module for predicting future trends in gamma-ray dose rate data based on a sparse recurrent neural network model; an anomaly detection module for identifying abnormal mutation points and locating high-risk sensors; a diffusion region modeling module for calculating the impact area of ​​nuclear radiation diffusion; a grid risk evolution module for simulating the temporal evolution of regular grid risk states based on a multi-state cellular automaton; and a fusion judgment module for fusing output results to generate a final warning level and spatial risk distribution. This invention enables precise prediction of nuclear radiation risks and dynamic spatial early warning, significantly improving warning accuracy and response efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of nuclear radiation indicator early warning, and in particular to a nuclear radiation indicator early warning system based on multi-algorithm fusion. Background Art

[0002] In the field of nuclear radiation safety and control, accurately predicting radiation dose rate trends and promptly identifying high-risk areas have long been important research areas in environmental monitoring and emergency management. Existing radiation early warning systems primarily rely on physical sensors at fixed sites to collect gamma-ray dose rate data, supplemented by statistical analysis-based threshold judgment methods to detect abnormal fluctuations in the monitoring data. However, these methods have poor sensitivity for anomaly detection and lack in-depth modeling of time series variation patterns, making them prone to underreporting sudden high-dose events or misreporting natural fluctuations as abnormal events. Furthermore, traditional systems often use static geographic diffusion models, failing to fully account for meteorological factors, particularly the actual impact of wind speed and direction on the path and range of radiation diffusion. This results in significant deviations in spatial early warning results, making it difficult to support dynamic, real-time emergency response decisions.

[0003] In recent years, some studies have attempted to introduce machine learning models, such as support vector machines, random forests, or convolutional neural networks, for the prediction and anomaly identification of nuclear radiation dose rates, achieving some accuracy improvements. However, most of these models are black-box structures with poor interpretability. They also fail to fully utilize long-term dependency information when modeling time series, resulting in insufficient predictive stability. Furthermore, most current methods treat data anomalies as risk points for simple spatial diffusion simulations, ignoring the quantification of the degree of anomaly and the directional characteristics of propagation. Diffusion simulations carry the risk of directional and regional misjudgments. In particular, existing systems lack a unified fusion framework for collaborative judgment of multi-source models, making it difficult to effectively integrate the outputs of different algorithms, resulting in one-sided or unstable overall warning results.

[0004] To address the above shortcomings, the present invention proposes a nuclear radiation indicator early warning system based on multi-algorithm fusion. By constructing a systematic processing flow including modules such as environmental data acquisition, dose rate prediction, anomaly detection, diffusion modeling, risk evolution, and fusion judgment, the system addresses the problems of inaccurate prediction, imprecise diffusion modeling, isolated risk identification, and lack of fusion mechanisms in the existing technology. A sparse recurrent neural network combined with a time window attention mechanism is used to improve the prediction accuracy of gamma-ray dose rate time series data. Residual analysis is used to construct high-risk location data. An exponential propagation algorithm with wind direction offset correction is further used to implement non-uniform spatial diffusion modeling. A multi-state cellular automaton is used to simulate the dynamic evolution of grid risk. Finally, a historical accuracy weighting strategy is introduced to fuse the outputs of each model to generate nuclear radiation warning level results with greater timeliness and spatial accuracy, significantly improving the overall prediction accuracy and decision-making support capabilities of the system.

[0005] The introduction of this system not only overcomes the structural defects of traditional methods in data processing and model linkage, but also realizes the expansion of capabilities from single-point prediction to full-region dynamic risk distribution assessment. It has good promotion value and engineering adaptability, and is suitable for nuclear radiation safety warning tasks in nuclear power plants, nuclear facility surrounding environments and urban emergency protection scenarios.

[0006] Therefore, how to provide a nuclear radiation indicator early warning system based on multi-algorithm fusion is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0007] One purpose of the present invention is to propose a nuclear radiation indicator early warning method based on multi-algorithm fusion. The present invention fully combines sparse recurrent neural networks, anomaly detection modules, exponential propagation algorithms with wind direction offset correction, and multi-state cellular automata, and describes in detail the entire process of nuclear radiation risk from data collection, trend prediction, anomaly identification, diffusion modeling to grid evolution and fusion early warning. It has the advantages of high prediction accuracy, reasonable spatial distribution, fast response speed, and adaptability to complex meteorological conditions.

[0008] The nuclear radiation indicator early warning system based on multi-algorithm fusion according to an embodiment of the present invention includes:

[0009] Environmental data acquisition module: used to collect gamma-ray dose rate data, wind speed data, wind direction data and geographic location information, and build a nuclear radiation environment data set;

[0010] Dose rate prediction module: used to predict the future trend of gamma-ray dose rate data based on a sparse recurrent neural network model;

[0011] Anomaly detection module: used to identify abnormal mutation points and locate high-risk sensor locations;

[0012] Diffusion area modeling module: used to calculate the area affected by nuclear radiation diffusion;

[0013] Grid risk evolution module: used to divide the monitoring area into regular grids and simulate the temporal evolution of the risk state of the regular grid based on multi-state cellular automata;

[0014] Fusion judgment module: used to generate the final warning level and spatial risk distribution results.

[0015] Optionally, modules can be connected using the following methods:

[0016] S1. Collect gamma-ray dose rate data, wind speed data, wind direction data and geographic location information within the monitoring area to construct a nuclear radiation environment data set;

[0017] S2. Using a sparse recurrent neural network model to perform trend prediction on the gamma-ray dose rate data, a nuclear radiation dose rate prediction result matrix is ​​generated;

[0018] S3. Inputting the nuclear radiation dose rate prediction result matrix and historical gamma-ray dose rate data into an anomaly detection module, identifying abnormal mutation points of the nuclear radiation dose rate in the nuclear radiation dose rate prediction result matrix, and generating high-risk location data;

[0019] S4. Inputting high-risk location data, wind speed data, and wind direction data into the improved exponential propagation algorithm to generate nuclear radiation diffusion impact area data;

[0020] S5. Divide the monitoring area into regular grids. Based on the multi-state cellular automaton model and the data of the nuclear radiation diffusion affected area and high-risk location, simulate the risk state evolution process of each regular grid and output the grid risk level results.

[0021] S6. Perform weighted fusion on the output results of the sparse recurrent neural network model, anomaly detection module, exponential propagation algorithm, and multi-state cellular automaton model, dynamically adjust the weights, and generate the final warning level and spatial risk distribution results.

[0022] Optionally, S1 specifically includes: deploying fixed sensor nodes in a set monitoring area, obtaining the geographic location information of each sensor node, collecting gamma-ray dose rate data, wind speed data and wind direction data in real time within a sampling period, constructing the gamma-ray dose rate data, wind speed data and wind direction data of each node at the same time point into a data matrix, performing time alignment, missing value filling, outlier elimination and normalization on the data matrix to obtain a standardized nuclear radiation environment data matrix, and combining them into a nuclear radiation environment data set in chronological order.

[0023] Optionally, the S2 specifically includes:

[0024] S21. Based on the nuclear radiation environment data set, construct standardized gamma-ray dose rate time series data for each sensor node, set the sensor nodes, collect standardized gamma-ray dose rate data at each sampling time point for each sensor node, arrange them in chronological order, and construct a standardized gamma-ray dose rate time series for the sensor node;

[0025] S22. Set the sliding window length and prediction step size, perform sliding sampling on the standardized gamma-ray dose rate time series of each sensor node, construct a sequence for model input, introduce a time window attention mechanism into the model input sequence, perform weighted processing on the gamma-ray dose rate values ​​at different time points in the model input sequence based on the weight coefficient of each time step, and generate an attention-weighted input sequence for model training;

[0026] S23. Based on the attention-weighted input sequence, a sparse recurrent neural network model is constructed and improved. A dynamic pruning strategy is used to control the sparsity and iteratively update the hidden state connection structure in the network to obtain a sparse connection state matrix for prediction. The model training is completed, and the trained sparse recurrent neural network model is used to predict the γ-ray dose rate time series of each sensor node to obtain the γ-ray dose rate prediction results of all sensor nodes in multiple time steps in the future. The prediction results of each sensor node at the corresponding time point are combined according to the time sequence and node number to construct a nuclear radiation dose rate prediction result matrix.

[0027] Optionally, the S3 specifically includes:

[0028] S31, based on the nuclear radiation dose rate prediction result matrix and the standardized γ-ray dose rate time series, for each sensor node , extract the time step from the nuclear radiation dose rate prediction result matrix to The prediction sequence is denoted as , extract the observation subsequence of the same time step from the standardized γ-ray dose rate time series, denoted as ,Will and Form input sample pairs;

[0029] S32: Input each input sample pair into the anomaly detection module, calculate the residual value of each time step between the predicted value of the prediction sequence and the observed value of the observation subsequence, and generate a residual sequence , where each residual value is defined as , Represents the prediction time step index;

[0030] S33, setting an anomaly detection threshold for each sensor node, and judging each time step in the residual sequence. If the residual value of a time step is greater than the corresponding anomaly detection threshold, then mark the time step as an anomaly point, record the node number and time point of the anomaly point, and form an anomaly point set;

[0031] S34. Perform spatial clustering on the outlier set, use a density clustering algorithm to aggregate outliers with close geographical locations to form a high-risk area, extract the sensor node number corresponding to the cluster center, construct a high-risk sensor node set, and output high-risk location data.

[0032] Optionally, the S4 specifically includes:

[0033] S41. Based on the high-risk location data, the set of high-risk sensor nodes is ,in, Indicates the number of high-risk nodes. Each high-risk node Including node geographic coordinates , sampling time point , γ-ray dose rate residual value , and obtain the wind speed data at the corresponding time point With wind direction data ;

[0034] S42, based on each high-risk node , construct an improved exponential propagation algorithm with wind direction deviation correction, and calculate the coordinates of any grid point in the monitoring area The diffusion impact value ;

[0035] S43. Traverse each high-risk node , calculate the diffusion impact value for all grid points within the influence range;

[0036] S44. Superimpose the diffusion impact values ​​of all high-risk nodes in the spatial dimension to obtain the superposition matrix ;

[0037] S45. Setting diffusion impact threshold , all satisfied The grid points of are determined as diffusion-affected area points, and this set is extracted to form the nuclear radiation diffusion-affected area data.

[0038] Optionally, the S5 specifically includes:

[0039] S51. Divide the monitoring area into regular grids according to the geographical boundary information of the monitoring area, and define the regular grid set as , where each grid cell Corresponding to the unique center coordinate ;

[0040] S52, based on the superposition matrix , for each grid cell The center coordinates of The corresponding diffusion impact value Risk levels are divided and the diffusion impact threshold is , the gradient interval is , construct the initial risk state vector ,in:

[0041] like , then suppose , indicating a low-risk state;

[0042] like , then suppose , indicating a medium-risk state;

[0043] like , then suppose , indicating a high-risk state;

[0044] S53, based on the initial state vector , establish a multi-state cellular automaton model, for each evolution round , update the state value according to the following state update rules:

[0045] If the current status , and the number of adjacent cells with a state value of 2 is not less than the threshold , then update to ;

[0046] If the current status , and the number of adjacent cells with a state value of 2 is not less than the threshold , then update to ;

[0047] If the current status , then remains unchanged;

[0048] Other situations remain the same, i.e. ;

[0049] in, 、 The adjacency state trigger threshold is preset as a constant parameter;

[0050] S54. Set the maximum number of evolution rounds , for the state vector conduct Round iteration to obtain the final state vector ,in Represents a grid cell The final risk level;

[0051] S55, the final state vector With each grid coordinate Bind and output grid risk level results including geographic spatial distribution.

[0052] Optionally, the S6 specifically includes: collecting the nuclear radiation dose rate prediction result matrix, high-risk location data, nuclear radiation diffusion impact area data and grid risk level results, summarizing the output results, dynamically assigning weights, and weighted fusion of each output result to obtain a comprehensive risk score for each grid unit, dividing the comprehensive score into low warning level, medium warning level and high warning level, combining geographic location information, and outputting the final warning level and spatial risk distribution results of the nuclear radiation indicator for warning display and response deployment.

[0053] The beneficial effects of the present invention are:

[0054] First, the present invention introduces a sparse recurrent neural network model and combines it with a time window attention mechanism to achieve accurate prediction of γ-ray dose rate time series data, overcoming the problem of insufficient time-dependent modeling in traditional methods and effectively improving the timeliness and accuracy of the prediction.

[0055] Secondly, the present invention identifies high-risk locations by constructing an anomaly detection module and directional models the nuclear radiation diffusion path in combination with an exponential propagation algorithm corrected for wind direction offset. This method is more in line with the actual environmental diffusion process than the existing static diffusion method, and significantly improves the reliability of spatial warning results.

[0056] In addition, the present invention is based on the grid risk evolution simulation implemented by multi-state cellular automata, combined with a multi-model fusion judgment mechanism, and dynamically weighted by comprehensively considering the output results of each model and historical accuracy, which effectively improves the comprehensiveness and stability of nuclear radiation warning level judgment, and has stronger practicality and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0058] Figure 1 This is a flow chart of the method of the nuclear radiation indicator early warning system based on multi-algorithm fusion proposed by the present invention;

[0059] Figure 2 This is a schematic diagram of the module structure of the nuclear radiation indicator early warning system based on multi-algorithm fusion proposed by the present invention. DETAILED DESCRIPTION

[0060] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0061] refer to Figure 1 and Figure 2 , a nuclear radiation indicator early warning system based on multi-algorithm fusion, including:

[0062] Environmental data acquisition module: used to collect gamma-ray dose rate data, wind speed data, wind direction data and geographic location information, and build a nuclear radiation environment data set;

[0063] Dose rate prediction module: used to predict the future trend of gamma-ray dose rate data based on a sparse recurrent neural network model;

[0064] Anomaly detection module: used to identify abnormal mutation points and locate high-risk sensor locations;

[0065] Diffusion area modeling module: used to calculate the area affected by nuclear radiation diffusion;

[0066] Grid risk evolution module: used to divide the monitoring area into regular grids and simulate the temporal evolution of the risk state of the regular grid based on multi-state cellular automata;

[0067] Fusion judgment module: used to generate the final warning level and spatial risk distribution results.

[0068] This invention implements data perception, intelligent modeling, and collaborative analysis for the entire nuclear radiation early warning process by constructing six functional modules: an environmental data acquisition module, a dose rate prediction module, an anomaly detection module, a diffusion zone modeling module, a grid risk evolution module, and a fusion judgment module. The closed-loop data flow and complementary functional logic between these modules effectively address the issues of information isolation, incomplete modeling, and delayed response in existing technologies. The system boasts a clear structure, well-defined division of labor, high processing efficiency, and strong scalability, significantly enhancing the intelligence and practical performance of nuclear radiation monitoring and early warning.

[0069] In this embodiment, the modules are connected through the following methods:

[0070] S1. Collect gamma-ray dose rate data, wind speed data, wind direction data and geographic location information within the monitoring area to construct a nuclear radiation environment data set;

[0071] S2. Using a sparse recurrent neural network model to perform trend prediction on the gamma-ray dose rate data, a nuclear radiation dose rate prediction result matrix is ​​generated;

[0072] S3. Inputting the nuclear radiation dose rate prediction result matrix and historical gamma-ray dose rate data into an anomaly detection module, identifying abnormal mutation points of the nuclear radiation dose rate in the nuclear radiation dose rate prediction result matrix, and generating high-risk location data;

[0073] S4. Inputting high-risk location data, wind speed data, and wind direction data into the improved exponential propagation algorithm to generate nuclear radiation diffusion impact area data;

[0074] S5. Divide the monitoring area into regular grids. Based on the multi-state cellular automaton model and the data of the nuclear radiation diffusion affected area and high-risk location, simulate the risk state evolution process of each regular grid and output the grid risk level results.

[0075] S6. Perform weighted fusion on the output results of the sparse recurrent neural network model, anomaly detection module, exponential propagation algorithm, and multi-state cellular automaton model, dynamically adjust the weights, and generate the final warning level and spatial risk distribution results.

[0076] The nuclear radiation indicator early warning method proposed in this paper, based on multi-algorithm fusion, constructs a full-process technical system from multi-source data acquisition, time series prediction, anomaly detection, diffusion modeling, risk evolution, to fusion judgment, achieving an organic connection between the algorithm functions of each stage. Compared with existing nuclear radiation early warning methods that focus on single-point processing and lack model coordination, this method improves data processing accuracy, diffusion modeling authenticity, and the reliability of early warning results through modular design and dynamic fusion strategies. It has the advantages of clear structure, accurate prediction, and adaptability to complex environments, providing systematic technical support for the construction of an efficient and intelligent nuclear radiation early warning system.

[0077] In this embodiment, S1 specifically includes: deploying fixed sensor nodes in the set monitoring area, obtaining the geographical location information of each sensor node, and collecting gamma-ray dose rate data, wind speed data and wind direction data in real time during the sampling period. The gamma-ray dose rate data is measured in real time by the radiation detector and recorded at a frequency of seconds. The wind speed data is obtained by an ultrasonic anemometer to obtain the wind speed size. The wind direction data is measured by a wind vane or a built-in wind direction sensor to measure the current wind direction angle. The three types of data are collected synchronously and stored with timestamps. The gamma-ray dose rate data, wind speed data and wind direction data of each node at the same time point are constructed into a data matrix. After the data are aligned by timestamp, interpolation is used to fill in missing values, and out-of-limit abnormal points are eliminated. Finally, the data are normalized to a unified scale by column to obtain a standardized nuclear radiation environment data matrix, which is combined into a nuclear radiation environment data set in chronological order.

[0078] This invention deploys fixed sensor nodes within the monitoring area to collect information on gamma-ray dose rate, wind speed, wind direction, and geographic location, constructing a structured, standardized nuclear radiation environment dataset. This ensures the continuity, integrity, and analyzability of data input. Compared to traditional nuclear radiation monitoring, which relies solely on a single physical quantity or non-standardized data, this invention achieves the simultaneous collection and standardized preprocessing of multi-source monitoring data, providing a unified input format for subsequent prediction and analysis, and improving the data quality and scalability of the overall system.

[0079] In this embodiment, S2 specifically includes:

[0080] S21. Based on the nuclear radiation environment data set, construct standardized gamma-ray dose rate time series data for each sensor node, set the sensor nodes, collect standardized gamma-ray dose rate data at each sampling time point for each sensor node, arrange them in chronological order, and construct a standardized gamma-ray dose rate time series for the sensor node;

[0081] S22. Set the sliding window length and prediction step size, perform sliding sampling on the standardized gamma-ray dose rate time series of each sensor node, construct a sequence for model input, introduce a time window attention mechanism into the model input sequence, perform weighted processing on the gamma-ray dose rate values ​​at different time points in the model input sequence based on the weight coefficient of each time step, and generate an attention-weighted input sequence for model training;

[0082] S23. Based on the attention-weighted input sequence, a sparse recurrent neural network model is constructed. By introducing a connection sparsification mechanism, redundant connections between neurons are reduced, reducing computational complexity while retaining the time series modeling capability. Improvements are made, and a dynamic pruning strategy is used to control the sparsity and iteratively update the hidden state connection structure in the network. Specifically, during each round of training, the importance of each connection between hidden states is evaluated based on the amplitude and gradient sensitivity of the connection weight, and connections with small weights or low contributions are dynamically pruned to improve network sparsity. At the same time, a recovery mechanism is retained for some pruned connections. In subsequent iterations, the connection structure is dynamically adjusted in combination with network error feedback to achieve the coexistence of adding and removing connections. This controls the parameter scale while maintaining the expressiveness of the model, improves training efficiency and generalization capability, and obtains a sparse connection state matrix for prediction. After model training is completed, the trained sparse recurrent neural network model is used to predict the γ-ray dose rate time series of each sensor node, obtaining the γ-ray dose rate prediction results for all sensor nodes in the future multiple time steps. The prediction results corresponding to each sensor node at the corresponding time point are combined according to the chronological order and node number to construct the nuclear radiation dose rate prediction result matrix.

[0083] This paper uses a sparse recurrent neural network model with an attention mechanism to predict gamma-ray dose rates. This effectively captures long-term dependencies in time series and reduces redundant model connections through dynamic pruning, improving computational efficiency and generalization. Compared to traditional LSTM or black-box neural network prediction methods, this paper significantly reduces model complexity while maintaining prediction accuracy and enhances responsiveness to unusual trends, effectively supporting early detection of sudden nuclear radiation events.

[0084] In this embodiment, S3 specifically includes:

[0085] S31, based on the nuclear radiation dose rate prediction result matrix and the standardized γ-ray dose rate time series, for each sensor node , extract the time step from the nuclear radiation dose rate prediction result matrix to The prediction sequence is denoted as , extract the observation subsequence of the same time step from the standardized γ-ray dose rate time series, denoted as ,Will and Form input sample pairs;

[0086] S32: Input each input sample pair into the anomaly detection module, calculate the residual value of each time step between the predicted value of the prediction sequence and the observed value of the observation subsequence, and generate a residual sequence , where each residual value is defined as , Represents the prediction time step index;

[0087] S33, setting an anomaly detection threshold for each sensor node, and judging each time step in the residual sequence. If the residual value of a time step is greater than the corresponding anomaly detection threshold, then mark the time step as an anomaly point, record the node number and time point of the anomaly point, and form an anomaly point set;

[0088] S34. Perform spatial clustering on the outlier set, use a density clustering algorithm to aggregate outliers with close geographical locations to form a high-risk area, extract the sensor node number corresponding to the cluster center, construct a high-risk sensor node set, and output high-risk location data.

[0089] This invention utilizes a residual analysis mechanism to compare predictions with historical observations point by point, generating an anomaly score sequence. This, combined with spatial clustering methods to identify high-risk nodes, effectively improves the sensitivity and accuracy of anomaly detection. Compared to existing anomaly detection methods that rely on fixed thresholds or single-point judgments, this invention comprehensively considers temporal deviations and spatial distribution characteristics, accurately identifying areas with sudden or localized dose rate anomalies, and enhancing the accuracy and stability of early warnings.

[0090] In this embodiment, the S4 specifically includes:

[0091] S41. Based on the high-risk location data, the set of high-risk sensor nodes is ,in, Indicates the number of high-risk nodes. Each high-risk node Including node geographic coordinates , sampling time point , γ-ray dose rate residual value , and obtain the wind speed data at the corresponding time point With wind direction data ;

[0092] S42, based on each high-risk node , construct an improved exponential propagation algorithm with wind direction deviation correction, and calculate the coordinates of any grid point in the monitoring area The diffusion impact value , defined as:

[0093] ;

[0094] in, Indicates that the grid point and node The Euclidean distance, Indicates the direction angle of the grid point relative to the node, Indicates that the node is at the time point Wind direction data, is the wind direction deviation penalty coefficient, is the diffusion scale parameter, affected by wind speed control, represents the exponential mapping function;

[0095] The diffusion impact value formula dynamically modifies the radiation impact intensity based on the degree of angular offset between wind direction and the diffusion path of the radiation source. By introducing the wind direction offset angle, the exponential attenuation coefficient is adjusted, so that the diffusion path is enhanced in the downwind direction and attenuated in the upwind direction. This more realistically simulates the asymmetric propagation behavior of nuclear radiation under different meteorological conditions and improves the spatial accuracy and physical rationality of diffusion area modeling.

[0096] S43. Traverse each high-risk node , calculate the diffusion impact value for all grid points within the influence range;

[0097] S44. Superimpose the diffusion impact values ​​of all high-risk nodes in the spatial dimension to obtain the superposition matrix ,Right now:

[0098] ;

[0099] S45. Setting diffusion impact threshold , all satisfied The grid points of are determined as diffusion-affected area points, and this set is extracted to form the nuclear radiation diffusion-affected area data.

[0100] The proposed wind-direction-biased exponential propagation algorithm comprehensively considers radiation source intensity, distance attenuation, and wind direction deviation angles. Through asymmetric diffusion modeling, it more realistically reproduces the actual diffusion path of nuclear radiation under complex meteorological conditions. Compared to traditional isotropic diffusion models or static diffusion maps, this method improves the spatial accuracy and directional representation of diffusion area simulations, facilitating more accurate identification of potentially affected areas and guiding response deployment during nuclear accident emergencies.

[0101] In this embodiment, the S5 specifically includes:

[0102] S51. Divide the monitoring area into regular grids according to the geographical boundary information of the monitoring area, and define the regular grid set as , where each grid cell Corresponding to the unique center coordinate ;

[0103] S52, based on the superposition matrix , for each grid cell The center coordinates of The corresponding diffusion impact value Risk levels are divided and the diffusion impact threshold is , the gradient interval is , construct the initial risk state vector ,in:

[0104] like , then suppose , indicating a low-risk state;

[0105] like , then suppose , indicating a medium-risk state;

[0106] like , then suppose , indicating a high-risk state;

[0107] S53, based on the initial state vector , establish a multi-state cellular automaton model, for each evolution round , update the state value according to the following state update rules:

[0108] If the current status , and the number of adjacent cells with a state value of 2 is not less than the threshold , then update to ;

[0109] If the current status , and the number of adjacent cells with a state value of 2 is not less than the threshold , then update to ;

[0110] If the current status , then remains unchanged;

[0111] Other situations remain the same, i.e. ;

[0112] in, 、 The adjacency state trigger threshold is preset as a constant parameter;

[0113] S54. Set the maximum number of evolution rounds , for the state vector conduct Round iteration to obtain the final state vector ,in Represents a grid cell The final risk level;

[0114] S55, the final state vector With each grid coordinate Bind and output grid risk level results including geographic spatial distribution.

[0115] This paper constructs a multi-state cellular automaton model to iteratively simulate the evolution of risk states in grid cells. Combining initial diffusion impact values ​​with neighborhood state propagation rules, this approach dynamically models the evolution of risk over time. Compared to existing approaches that only statically assess current risk, this paper captures the process of risk spreading from a point to a surface, improving the consistency, scalability, and predictability of nuclear radiation risk area identification and providing effective support for continuous spatial early warning.

[0116] In this embodiment, the S6 specifically includes: collecting the nuclear radiation dose rate prediction result matrix, high-risk location data, nuclear radiation diffusion impact area data and grid risk level results, summarizing the output results, dynamically assigning weights, and weighted fusion of each output result to obtain a comprehensive risk score for each grid unit, and dividing the comprehensive score into low warning level, medium warning level and high warning level, specifically including: when the fusion comprehensive score is less than or equal to the first threshold, it is judged to be a low warning level, indicating that there is no significant abnormality in the area and the risk is controllable; when the fusion comprehensive score is greater than the first threshold and less than or equal to the second threshold, it is judged to be a medium warning level, indicating that there is a potential risk in the area and needs to be paid more attention; when the fusion comprehensive score is greater than the second threshold, it is judged to be a high warning level, indicating that the area There may be significant nuclear radiation anomalies, which require timely response and intervention. Combined with geographic location information, the final warning level and spatial risk distribution results of nuclear radiation indicators are output for warning display and response deployment. The specific process is: the system maps the fusion scoring results to the grid map of the monitoring area, and uses different colors to identify the three warning levels of low, medium and high, to intuitively present the risk status of each area. Users can view the geographic location, risk level and associated model scoring data of each risk point through the interface; at the same time, the system automatically pushes high warning areas to the emergency management platform, and links relevant units to implement response mechanisms, such as initiating on-site inspections, adjusting monitoring frequencies or issuing personnel evacuation warnings. The entire process realizes closed-loop control from risk identification to emergency response, improving the timeliness and pertinence of nuclear radiation incident handling.

[0117] This invention constructs a fusion judgment algorithm that dynamically weights and fuses dose rate prediction results, anomaly detection data, diffusion area information, and grid risk levels. Weights are automatically adjusted based on historical prediction accuracy, avoiding model bias and the instability of a single algorithm. Compared to traditional multi-model combinations that use static weights or manually assign weights based on experience, this invention improves the scientific nature and robustness of the warning level output, significantly enhancing the overall warning decision-making capabilities and practicality of the system.

[0118] Example 1:

[0119] To verify the feasibility of this invention, it was applied to a radiation safety monitoring scenario surrounding a nuclear facility. This area is equipped with multiple fixed monitoring nodes. The existing radiation early warning mechanism primarily relies on a single-point judgment model with set thresholds. This lacks the ability to predict data trends and a mechanism for linking multiple sources of information. This leads to delayed recognition, high false alarm rates, and insufficient spatial response capabilities during emergencies. To address these issues, the multi-algorithm fusion-based nuclear radiation indicator early warning system proposed in this invention was deployed to comprehensively model and assess nuclear radiation risks in real time.

[0120] The system first deployed 60 fixed sensor nodes within the monitoring area. Each node collected data such as gamma-ray dose rate, wind speed, and wind direction every five minutes. Edge computing nodes normalized the data, time-aligned it, filled in missing values, and removed anomalies, creating a standardized nuclear radiation environment dataset. Within the experimental area, the dose rate at one node gradually increased from a normal 0.12 μSv / h to 0.46 μSv / h over four consecutive hours, then fluctuated and reached 0.59 μSv / h. Based on this, the system used a sparse recurrent neural network model for time series forecasting, successfully predicting two time steps (10 minutes) in advance that the dose rate would exceed 0.50 μSv / h.

[0121] By comparing historical data with forecast results using a residual anomaly detection model, the system accurately identified abnormal mutation points and located specific node numbers. Within the same time window, the system identified four adjacent nodes with significant abnormal residuals (mean residual reached 0.27 μSv / h, standard deviation 0.05), and the spatial clustering algorithm identified this as a high-risk cluster area. Subsequently, the location of the area's central node, wind speed (1.9 m / s), and wind direction (southeastward) were input into an exponential propagation model with wind direction offset correction. The system calculated that under current meteorological conditions, the radiation impact area expands in an elliptical shape in a northeasterly direction, with a maximum diffusion distance of approximately 980 meters and 143 affected grid cells.

[0122] During the risk evolution process, the system assigned initial states to all grid cells based on grid division and initial diffusion impact values. The system then used a multi-state cellular automaton model to evolve the risk state, setting adjacent trigger thresholds and simulating the expansion of risk from the center to the periphery. After three rounds of evolution, the high-risk area expanded outward by approximately 160 meters, and the trend of risk level changes within the grid cells was highly consistent with the actual radiation values.

[0123] The fusion judgment module comprehensively analyzes the dose rate prediction output, anomaly detection results, diffusion impact matrix, and grid evolution status, and automatically assigns weights based on historical model accuracy to generate a fusion score. Using thresholds of 0.35 and 0.55 as the warning level, the system ultimately classified 43 grids as high warning, 82 as medium warning, and the remainder as low warning. Comparing this result with subsequent manual inspection data, it was found that the overlap rate between high warning areas and measured exceedance areas reached 92%, and the prediction accuracy rate for medium warning areas reached 87%, effectively reducing false positives and missed alerts.

[0124] In this implementation scenario, this invention demonstrates outstanding trend prediction capabilities, anomaly response speed, and spatial diffusion simulation, resolving issues such as delayed response, crude spatial warnings, and simplified risk diffusion models in traditional monitoring models. The system overall outperforms existing solutions in data processing efficiency, risk location accuracy, and comprehensive warning capabilities, demonstrating strong adaptability and engineering practicality.

[0125] Table 1: Comparison of nuclear radiation multi-model fusion warning results and actual observation data

[0126]

[0127] The aforementioned "Nuclear Radiation Multi-Model Fusion Warning Results and Actual Observation Data Comparison Table" demonstrates the system's ability to predict gamma-ray dose rates, detect anomalies, model diffusion impacts, and compare risk evolution results with actual observations in different grid cells. The table shows that the system's warning levels are highly consistent with actual observations in most grid cells. For example, grid cells such as G001, G017, and G082 were all assigned a high warning level by the system, consistent with their observed dose rates exceeding normal levels, validating the system's accuracy in identifying high risks.

[0128] In medium-risk grids such as G005, G029, and G101, the predicted values, residual values, and diffusion impact values ​​were at intermediate levels, and the system identified them as medium warning levels, which was also highly consistent with the actual situation. This shows that the fusion judgment algorithm's weighted processing of multi-model results has strong stability and hierarchical recognition capabilities. In low-risk areas such as G045 and G120, the system's scores were low in all aspects, and the final judgment was a low warning level, with no excessive warnings, demonstrating the system's good false alarm control capabilities.

[0129] From the data in the comprehensive analysis table, it can be seen that the multi-algorithm fusion mechanism described in the present invention has good risk identification ability in complex spaces, and the fusion results can effectively fit the actual monitoring situation, with high engineering practicality and promotion value.

[0130] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. The nuclear radiation indicator early warning system based on multi-algorithm fusion is characterized by: include: Environmental data acquisition module: used to collect gamma-ray dose rate data, wind speed data, wind direction data and geographic location information, and build a nuclear radiation environment data set; Dose rate prediction module: used to predict the future trend of gamma-ray dose rate data based on a sparse recurrent neural network model; Anomaly detection module: used to identify abnormal mutation points and locate high-risk sensor locations; Diffusion area modeling module: used to calculate the area affected by nuclear radiation diffusion; Grid risk evolution module: used to divide the monitoring area into regular grids and simulate the temporal evolution of the risk state of the regular grid based on multi-state cellular automata; Fusion judgment module: used to generate the final warning level and spatial risk distribution results; The modules are implemented as follows: S1. Collect gamma-ray dose rate data, wind speed data, wind direction data and geographic location information within the monitoring area to construct a nuclear radiation environment data set; S2. Using a sparse recurrent neural network model to perform trend prediction on the gamma-ray dose rate data, a nuclear radiation dose rate prediction result matrix is ​​generated; S3. Compare the nuclear radiation dose rate prediction result matrix with the historical gamma-ray dose rate data. For each sensor node, calculate the residual value between the predicted value of the prediction sequence and the observed value of the observation subsequence at each time step to generate a residual sequence. Based on the residual sequence, identify abnormal mutation points and perform spatial clustering on the abnormal point set to construct a high-risk sensor node set containing the geographic coordinates, sampling time points and gamma-ray dose rate residual values ​​of each high-risk node, and generate high-risk location data. S4. Inputting the high-risk location data, wind speed data, and wind direction data into an improved exponential propagation algorithm with a wind direction offset correction, calculating the diffusion impact value of each high-risk node on any grid point in the monitoring area, and superimposing the diffusion impact values ​​of all high-risk nodes in the spatial dimension to obtain a superposition matrix M(x, y), thereby generating nuclear radiation diffusion impact area data; S5. Divide the monitoring area into regular grids and classify the risk level of the superposition matrix of each grid unit to construct the initial state vector , based on the initial state vector , establish a multi-state cellular automaton model. Based on the multi-state cellular automaton model, according to the nuclear radiation diffusion affected area data and high-risk location data, simulate the risk state evolution process of each regular grid unit and output the grid risk level result; S6. Perform weighted fusion on the output results of the sparse recurrent neural network model, anomaly detection module, exponential propagation algorithm, and multi-state cellular automaton model, dynamically adjust the weights, and generate the final warning level and spatial risk distribution results.

2. The nuclear radiation indicator early warning system based on multi-algorithm fusion according to claim 1 is characterized in that: The S1 specifically includes: deploying fixed sensor nodes in the set monitoring area, obtaining the geographical location information of each sensor node, collecting gamma-ray dose rate data, wind speed data and wind direction data in real time during the sampling period, constructing the gamma-ray dose rate data, wind speed data and wind direction data of each node at the same time point into a data matrix, performing time alignment, missing value filling, outlier removal and normalization on the data matrix to obtain a standardized nuclear radiation environment data matrix, and combining them into a nuclear radiation environment data set in chronological order.

3. The nuclear radiation indicator early warning system based on multi-algorithm fusion according to claim 2 is characterized in that: The S2 specifically includes: S21. Based on the nuclear radiation environment data set, construct standardized gamma-ray dose rate time series data for each sensor node, set the sensor nodes, collect standardized gamma-ray dose rate data at each sampling time point for each sensor node, arrange them in chronological order, and construct a standardized gamma-ray dose rate time series for the sensor node; S22. Set the sliding window length and prediction step size, perform sliding sampling on the standardized gamma-ray dose rate time series of each sensor node, construct a sequence for model input, introduce a time window attention mechanism into the model input sequence, perform weighted processing on the gamma-ray dose rate values ​​at different time points in the model input sequence based on the weight coefficient of each time step, and generate an attention-weighted input sequence for model training; S23. Based on the attention-weighted input sequence, a sparse recurrent neural network model is constructed and improved. A dynamic pruning strategy is used to control the sparsity and iteratively update the hidden state connection structure in the network to obtain a sparse connection state matrix for prediction. The model training is completed, and the trained sparse recurrent neural network model is used to predict the γ-ray dose rate time series of each sensor node to obtain the γ-ray dose rate prediction results of all sensor nodes in multiple time steps in the future. The prediction results of each sensor node at the corresponding time point are combined according to the time sequence and node number to construct a nuclear radiation dose rate prediction result matrix.

4. The nuclear radiation indicator early warning system based on multi-algorithm fusion according to claim 3 is characterized in that: The S3 specifically includes: S31, based on the nuclear radiation dose rate prediction result matrix and the standardized γ-ray dose rate time series, for each sensor node , extract the time step from the nuclear radiation dose rate prediction result matrix to The prediction sequence is denoted as , extract the observation subsequence of the same time step from the standardized γ-ray dose rate time series, denoted as ,Will and Form input sample pairs; S32, input each input sample into the anomaly detection module, calculate the residual value of each time step between the predicted value of the prediction sequence and the observed value of the observation subsequence, and generate the residual sequence { } , where each residual value is defined as , Represents the prediction time step index; S33, setting an anomaly detection threshold for each sensor node, and judging each time step in the residual sequence. If the residual value of a time step is greater than the corresponding anomaly detection threshold, then mark the time step as an anomaly point, record the node number and time point of the anomaly point, and form an anomaly point set; S34. Perform spatial clustering on the outlier set, use a density clustering algorithm to aggregate outliers with close geographical locations to form a high-risk area, extract the sensor node number corresponding to the cluster center, construct a high-risk sensor node set, and output high-risk location data.

5. The nuclear radiation indicator early warning system based on multi-algorithm fusion according to claim 4 is characterized in that: The S4 specifically includes: S41. Based on the high-risk location data, the set of high-risk sensor nodes is ,in, Indicates the number of high-risk nodes. Each high-risk node Including node geographic coordinates , sampling time point , γ-ray dose rate residual value , and obtain the wind speed data at the corresponding time point With wind direction data ; S42, based on each high-risk node , construct an improved exponential propagation algorithm with wind direction deviation correction, and calculate the coordinates of any grid point in the monitoring area The diffusion impact value ; S43. Traverse each high-risk node , calculate the diffusion impact value for all grid points within the influence range; S44. Superimpose the diffusion impact values ​​of all high-risk nodes in the spatial dimension to obtain the superposition matrix ; S45. Setting diffusion impact threshold , all satisfied The grid points are determined as diffusion-affected area points, and the set of all diffusion-affected area points is extracted to form the nuclear radiation diffusion-affected area data.

6. The nuclear radiation indicator early warning system based on multi-algorithm fusion according to claim 5 is characterized in that: The S5 specifically includes: S51. Divide the monitoring area into regular grids according to the geographical boundary information of the monitoring area, and define the regular grid set as , where each grid cell Corresponding to the unique center coordinate ; S52, based on the superposition matrix , for each grid cell The center coordinates of The corresponding superposition matrix Risk levels are divided and the diffusion impact threshold is , the gradient interval is , construct the initial state vector ,in: like , then suppose , indicating a low-risk state; like , then suppose , indicating a medium-risk state; like , then suppose , indicating a high-risk state; S53, based on the initial state vector , establish a multi-state cellular automaton model, for each evolution round , update the state value according to the following state update rules: If the current status , and the number of adjacent cells with a state value of 2 is not less than the threshold , then update to ; If the current status , and the number of adjacent cells with a state value of 2 is not less than the threshold , then update to ; If the current status , then remains unchanged; Other situations remain the same, i.e. ; in, 、 The adjacency state trigger threshold is preset as a constant parameter; S54. Set the maximum number of evolution rounds , for the state vector conduct Round iteration to obtain the final state vector ,in Represents a grid cell The final risk level; S55, the final state vector With each grid coordinate Bind and output grid risk level results including geographic spatial distribution.

7. The nuclear radiation indicator early warning system based on multi-algorithm fusion according to claim 6 is characterized in that: The S6 specifically includes: collecting the nuclear radiation dose rate prediction result matrix, high-risk location data, nuclear radiation diffusion impact area data and grid risk level results, unifying the output results, dynamically assigning weights, and weighted fusion of each output result to obtain a comprehensive risk score for each grid unit, dividing the comprehensive score into low warning level, medium warning level and high warning level, combining geographic location information, and outputting the final warning level and spatial risk distribution results of the nuclear radiation indicator for warning display and response deployment.

Images