Nuclear chemical environment monitoring and early warning system and method based on multi-source data fusion and Bayesian maximum entropy model

Abstract

The embodiment of the invention discloses a nuclear chemical environment monitoring and early warning system and method based on multi-source data fusion and a Bayesian maximum entropy model. The system comprises a nuclear, chemical and biological environment monitoring and early warning device and an algorithm platform. The nuclear, chemical and chemical environment monitoring and early warning equipment comprises a detection assembly which is used for collecting multi-dimensional nuclear, chemical and chemical environment data in real time; the communication and positioning assembly is used for communication, positioning and time service; the three-dimensional point cloud scanning assembly is used for acquiring three-dimensional space information of a deployment environment and constructing a digital base map; the main control platform is used for data acquisition, analysis, packaging and transmission; the power supply assembly is used for providing power supply guarantee; and the algorithm platform is configured with multi-source data fusion and a Bayesian maximum entropy model, and is configured to carry out early warning on the nuclear and chemical environment crisis based on the detected multi-dimensional nuclear and chemical environment data through a Bayesian maximum entropy space-time modeling and prediction theory.

Description

Technical Field

[0001] This invention belongs to the field of chemical monitoring technology, specifically relating to a nuclear, chemical and biological environment monitoring and early warning system and method based on multi-source data fusion and Bayesian maximum entropy model. Background Technology

[0002] The current global nuclear, biological, and chemical threats are characterized by diversification and complexity. The development of nuclear energy, chemical, and biological laboratories is accompanied by multiple coupled accident risks such as leaks and explosions. However, existing monitoring technologies have limited detection dimensions, only able to identify single threat types; intelligent analysis is weak, remaining at the rudimentary stage of "threshold alarms," ​​lacking the ability to integrate multi-source data and conduct risk assessments; system architecture is fragmented, with independently deployed equipment creating data silos; and decision support is lacking, failing to integrate functions such as pollution source tracing and evacuation planning.

[0003] Patent document CN108873045A discloses a nuclear radiation monitoring system and method, which utilizes nuclear detection instruments, chemical agent detection instruments, and industrial toxic gas detection instruments for long-term monitoring of nuclear and chemical areas, and uploads the monitoring data to a platform. The platform processes the data and generates various visualization formats. Patent document CN117760489A discloses an unattended nuclear, biological, and chemical monitoring system and method, specifically consisting of multiple monitoring devices and monitoring terminals. Each monitoring device is deployed at a specific location to acquire nuclear, biological, and chemical data and transmits it through a self-organizing network. The monitoring terminal is configured with software that uses the data acquired by the monitoring devices to construct a digital twin model and deep learning algorithms for event judgment and early warning. The monitoring devices and monitoring terminals can be applied in various scenarios such as vehicles and drones.

[0004] However, existing technical solutions have the following drawbacks: poor system compatibility, making it difficult to achieve simultaneous detection and fusion analysis of multiple threats, including nuclear, biological, and chemical threats; the early warning model does not incorporate physical laws, has a low level of intelligence, and the early warning results are unreliable; the model relies on pure data-driven approaches, making it difficult to incorporate domain knowledge and physical rules, and its generalization ability is insufficient in small sample scenarios. Summary of the Invention

[0005] In view of this, on the one hand, some embodiments disclose a nuclear, chemical, and biological environment monitoring and early warning system based on multi-source data fusion and a Bayesian maximum entropy model, including nuclear, chemical, and biological environment monitoring and early warning equipment and an algorithm platform; wherein,

[0006] The nuclear, chemical, and biological environment monitoring and early warning equipment includes: a detection component for real-time acquisition of multi-dimensional nuclear, chemical, and biological environment data; a communication and positioning component for communication, positioning, and timing; a 3D point cloud scanning component for acquiring 3D spatial information of the deployment environment and constructing a digital base map; a main control platform for data acquisition, analysis, encapsulation, and transmission; and a power supply component for providing power assurance.

[0007] The algorithm platform is configured with multi-source data fusion and a Bayesian maximum entropy model. It is configured to conduct early warning of nuclear, chemical, and biological environmental crises based on multi-dimensional nuclear, chemical, and biological environmental data detected by Bayesian maximum entropy spatiotemporal modeling and prediction theory.

[0008] On the other hand, some embodiments disclose a method for monitoring and early warning of nuclear, chemical, and biological environments based on multi-source data fusion and a Bayesian maximum entropy model, implemented using the early warning system disclosed in the embodiments of this invention, specifically including:

[0009] Real-time collection of multi-dimensional nuclear, chemical, and biological environment data using nuclear, chemical, and biological environment monitoring devices;

[0010] A physical prior concentration field model describing the diffusion trend of pollutants was constructed using the Gaussian smoke and rain model.

[0011] The Bayesian maximum entropy model is used for data fusion and uncertainty quantification to obtain the posterior probability distribution;

[0012] Based on the posterior probability distribution, a dynamic risk map is synthesized and multi-level early warnings are triggered.

[0013] An expert feedback mechanism is introduced to continuously optimize model parameters and early warning rules.

[0014] The nuclear, chemical, and biological (NCB) environmental monitoring and early warning system and method disclosed in this invention, based on multi-source data fusion and a Bayesian maximum entropy model, integrates integrated environmental detection technology, nuclear detection technology, chemical detection technology, industrial toxic gas detection technology, and biological warfare agent detection technology, and can meet the needs of various application scenarios. The system features high integration and ease of use, as well as a high degree of intelligence. Based on a constructed physical prior concentration field model of pollutant diffusion trends, it utilizes a Bayesian maximum entropy model for data fusion and uncertainty quantification to obtain a posterior probability distribution, synthesizes a dynamic risk map and a multi-level early warning triggering mechanism, and combines the expert experience, relevant standards, and processing methods accumulated over many years of development in nuclear and chemical detection with real-time detection data to provide scientific and reasonable early warning and handling opinions for NCB crises. It has promising application prospects in the field of NCB environmental monitoring and early warning. Attached Figure Description

[0015] Figure 1 Example 1: Schematic diagram of the nuclear, chemical, and biological environment monitoring and early warning system based on multi-source data fusion and Bayesian maximum entropy model;

[0016] Figure 2 Example 1: Schematic diagram of the composition of an unattended nuclear, biological, and chemical environment monitoring and early warning device;

[0017] Figure 3 Example 1: Schematic diagram of the nuclear, chemical, and biological environment monitoring and early warning system based on multi-source data fusion and Bayesian maximum entropy model;

[0018] Figure 4 Example 2: Flowchart of a nuclear, chemical, and biological environment monitoring and early warning method based on multi-source data fusion and Bayesian maximum entropy model. Detailed Implementation

[0019] The term "embodiment" used herein, as an example, is not necessarily to be construed as superior to or better than other embodiments. Performance testing in these embodiments of the invention, unless otherwise specified, employs conventional testing methods in the art. It should be understood that the terminology used in these embodiments is merely for describing particular implementations and is not intended to limit the scope of the disclosure of these embodiments.

[0020] Unless otherwise stated, the technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the embodiments of this invention pertain; other experimental methods and technical means not specifically noted in the embodiments of this invention refer to experimental methods and technical means commonly used by one of ordinary skill in the art.

[0021] The terms “basic” and “approximately” as used herein are used to describe small fluctuations. For example, they can mean less than or equal to ±5%, such as less than or equal to ±2%, such as less than or equal to ±1%, such as less than or equal to ±0.5%, such as less than or equal to ±0.2%, such as less than or equal to ±0.1%, such as less than or equal to ±0.05%. Numerical data presented or expressed in range format herein are used for convenience and brevity only, and should therefore be interpreted flexibly to include not only the explicitly listed values ​​that define the range, but also all independent values ​​or subranges contained within that range. For example, a numerical range of “1–5%” should be interpreted to include not only the explicitly listed values ​​from 1% to 5%, but also the independent values ​​and subranges within the indicated range. Thus, this numerical range includes independent values ​​such as 2%, 3.5%, and 4%, and subranges such as 1%–3%, 2%–4%, and 3%–5%, etc. This principle also applies to ranges that list only one value. Furthermore, this interpretation applies regardless of the width of the range or the characteristics described.

[0022] In this document, including in the claims, conjunctions such as "comprising," "including," "with," "having," "containing," "involving," and "accommodating" are understood to be open-ended, meaning "including but not limited to." Only the conjunctions "consisting of" and "composed of" are closed conjunctions.

[0023] To better illustrate the content of this invention, numerous specific details are provided in the following detailed embodiments. Those skilled in the art should understand that the invention can be practiced even without certain specific details. In the embodiments, some methods, means, instruments, and devices well-known to those skilled in the art are not described in detail, in order to highlight the main points of the invention.

[0024] Without conflict, the technical features disclosed in the embodiments of the present invention can be combined arbitrarily, and the resulting technical solution belongs to the content disclosed in the embodiments of the present invention.

[0025] In some implementations, the nuclear, chemical, and biological environment monitoring and early warning system based on multi-source data fusion and a Bayesian maximum entropy model includes nuclear, chemical, and biological environment monitoring and early warning equipment and an algorithm platform; wherein,

[0026] The nuclear, chemical, and biological environment monitoring and early warning equipment includes: a detection component for real-time acquisition of multi-dimensional nuclear, chemical, and biological environment data; a communication and positioning component for communication, positioning, and timing; a 3D point cloud scanning component for acquiring 3D spatial information of the deployment environment and constructing a digital base map; a main control platform for data acquisition, analysis, encapsulation, and transmission; and a power supply component for providing power assurance.

[0027] The algorithm platform is configured with multi-source data fusion and a Bayesian maximum entropy model. It is configured to conduct early warning of nuclear, chemical, and biological environmental crises based on multi-dimensional nuclear, chemical, and biological environmental data detected by Bayesian maximum entropy spatiotemporal modeling and prediction theory.

[0028] Some embodiments disclose a method for monitoring and early warning of nuclear, chemical, and biological environments based on multi-source data fusion and a Bayesian maximum entropy model, implemented using the early warning system disclosed in the embodiments of this invention, specifically including:

[0029] Real-time collection of multi-dimensional nuclear, chemical, and biological environment data using nuclear, chemical, and biological environment monitoring devices;

[0030] A physical prior concentration field model describing the diffusion trend of pollutants was constructed using the Gaussian smoke and rain model.

[0031] Data fusion and uncertainty quantification are performed using the Bayesian maximum entropy model to obtain the posterior probability distribution.

[0032] Based on the posterior probability distribution, a dynamic risk map is synthesized and multi-level early warnings are triggered.

[0033] An expert feedback mechanism is introduced to continuously optimize model parameters and early warning rules.

[0034] In some embodiments, the detection component includes:

[0035] The environmental detection module is used to detect environmental data. Generally, the environmental detection module adopts multi-dimensional integrated sensing technology, specifically including: temperature and humidity detection based on semiconductor technology, which transmits the data to the main control platform after calibration, correlation and data fitting smoothing caused by changes in ambient temperature and humidity; wind speed and direction using ultrasonic sensors (wear-free, fast response, and high accuracy), which use the speed difference of propagation with / against the wind to convert into time difference (linearly related to wind speed) to achieve wind speed measurement, and invert the wind direction through the characteristics of propagation time change; optical rainfall detection based on the principle of particle forward scattering, which consists of an optical path composed of a near-infrared modulated beam emitter and a high-sensitivity receiver. Raindrops passing through the sampling area cause light signal attenuation. The microprocessor analyzes the time-domain and frequency-domain characteristics of the attenuation parameters, and combines the Mie scattering theory algorithm to invert the raindrop spectrum, calculate the instantaneous rainfall intensity and cumulative amount, which has the advantages of not requiring physical water collection and high response sensitivity; air pressure detection adopts silicon piezoresistive technology, which utilizes the piezoresistive effect of single-crystal silicon. Atmospheric pressure causes deformation of the silicon thin-film Wheatstone bridge, resulting in a change in bridge resistance, and outputs an electrical signal proportional to the pressure;

[0036] The nuclear detection module is used to detect gamma rays. For example, the nuclear detection needle uses a sodium iodide scintillator detector to detect gamma rays. The gamma rays interact with the scintillator to excite atomic transitions and release photons. These photons are collected by a photomultiplier tube, converted into electrons, and amplified into a current signal. The pulse acquisition is calibrated by a multichannel analyzer to determine the energy of the rays and the type of radioactive material.

[0037] The chemical agent and industrial gas detection module is used to detect the types of chemical agents and industrial gases. For example, chemical agent detection is based on ion mobility spectrometry (IMS): after the agent is vaporized, it is ionized by an ion source (discharge / chemical / radioactive). The ions form a characteristic migration time spectrum due to differences in size and charge in an electric field, which is compared with a reference standard to identify the type of agent.

[0038] Industrial toxic gas detection employs a sensor array composed of multiple types of sensors with varying selectivity and sensitivity. The target gas triggers a characteristic response pattern (response curve shape, amplitude, and time), and the gas type and concentration are determined by matching the data processing system with a database.

[0039] The biological agent detection module is used to detect the type of biological agent. For example, the detection of biological warfare agent aerosols combines laser-induced fluorescence (LIF) and ion mobility spectrometry (IMS): In LIF technology, a laser irradiates aerosol particles, and fluorescent substances (NADH, riboflavin, etc.) within the bioactive particles absorb photons and release characteristic fluorescence. A photoelectric sensor collects the signal (to determine the presence or absence of active particles and to quantify the concentration by intensity); combined with IMS, the agent type can be identified.

[0040] Some embodiments disclose a nuclear, chemical, and biological environment monitoring and early warning system based on multi-source data fusion and a Bayesian maximum entropy model. The communication and positioning components include: a wireless communication module; integrated wireless communication technology supporting a decentralized self-organizing network architecture; a satellite navigation module; and the provision of positioning information and precise time synchronization. For example, the wireless communication and positioning system integrates mainstream wireless communication technologies such as 4G / 5G, LoRa, and BeiDou short message service, supporting a decentralized self-organizing network architecture; wherein, the BeiDou satellite navigation module, while realizing communication functions, can simultaneously provide high-precision positioning and precise time synchronization services.

[0041] In some embodiments, a 3D point cloud scanning device is a device that acquires 3D spatial information of an object's surface by actively emitting lasers or light waves and receiving their reflected signals. Its core principle is ranging and positioning: using technologies such as LiDAR or structured light, the distance from the scanner to each point on the object's surface is accurately measured. Then, combined with a built-in precision angle sensor or camera, the precise 3D coordinates (X, Y, Z) of each point are calculated. Thus, a massive set of points constitutes a "point cloud," digitally reproducing the true shape of the object.

[0042] In some embodiments, the power supply component is designed with low power consumption and long battery life in tandem, employing a dual strategy of "large-capacity energy storage + dynamic power consumption control" to ensure continuous and reliable operation in unattended scenarios. The energy storage system of the power supply component uses high-energy-density 21700 lithium battery cells, achieving large-capacity power supply through multiple series and parallel connections. The specific number of series and parallel connections is configured according to power requirements, ensuring that the device has a battery life of ≥24 hours under full-power operation (e.g., including full load of sensors, continuous operation of data processing and communication modules), meeting the needs of high-intensity monitoring. The voltage components feature a low-power design and an integrated intelligent sleep-wake mechanism. Through an embedded microcontroller (MCU), the device's operating mode is dynamically managed. During non-monitoring periods, it automatically switches to a low-power silent mode (the core module is in hibernation, with only the wake-up trigger circuit and real-time clock running, and the standby current controlled at the microampere level). In this mode, the device can continuously operate silently for ≥7 days. When the wake-up condition is triggered (timed wake-up or external event wake-up), the system quickly recovers to full-power operation and maintains a full-power battery life of ≥24 hours after wake-up, achieving energy optimization allocation of "long-cycle silence - instantaneous high-efficiency response".

[0043] In some embodiments, the main control platform employs a high-performance microcontroller unit (MCU) to coordinate the entire detection process. It acquires raw data from the front-end detector in real time via serial communication protocols such as UART and RS-485, using either polling or interrupt methods. The embedded firmware program within the MCU parses, verifies, and preprocesses the acquired data, and encapsulates and frames the data according to predefined communication protocols (such as Modbus, custom binary frames, etc.). Finally, the encapsulated data packets are uploaded to a remote monitoring terminal or cloud platform through a communication interface driven by the MCU (such as 4G, LoRa, BeiDou short message service, or Wi-Fi module), achieving remote, real-time, and reliable transmission of monitoring data. The entire system possesses local processing, protocol conversion, and status management capabilities.

[0044] Typically, the mounting and fixing accessories adopt a "quick-release" adapter design, enabling unattended nuclear, biological, and chemical environment monitoring and early warning equipment to be used in various scenarios such as drones, unmanned vehicles, and tripods.

[0045] Typically, depending on the specific application scenario, multiple nuclear, chemical, and biological (NCB) environmental monitoring and early warning devices can be deployed using suitable transportation methods such as drones, unmanned vehicles, or personnel-carried equipment. Each NCB device functions as a single node, with wireless communication distances between nodes supporting ≥3 kilometers. Up to 16 devices can be used simultaneously. During deployment, various network topologies, such as star and linear topologies, can be configured via terminal software according to actual application needs. After deployment, the devices are automatically networked upon one-click startup, and the shortest path algorithm is used to achieve efficient data transmission between nodes. Ultimately, the monitoring data is uniformly uploaded to the terminal platform, enabling real-time monitoring and early warning in unattended environments.

[0046] Multi-source data fusion and the Bayesian maximum entropy model are implemented based on Bayesian maximum entropy spatiotemporal modeling and prediction theory (BME). Bayesian maximum entropy spatiotemporal modeling and prediction theory (BME) is a spatiotemporal geostatistical method whose core advantage lies in its ability to seamlessly integrate hard data (precise measurements) and soft data (uncertain, qualitative information), while strictly adhering to the laws of spatiotemporal conservation. Compared to traditional geostatistical methods, its greatest feature is its ability to integrate data information from different sources and formats to achieve accurate prediction and mapping effects. Through Bayesian maximum entropy spatiotemporal modeling and prediction theory (BME), information such as temperature, humidity, wind speed and direction, and latitude and longitude collected by detectors is combined with industry physical laws, industry standards, and expert experience to estimate and predict nuclear biochemical data to obtain more accurate results.

[0047] Some embodiments disclose a nuclear, chemical, and biological environment monitoring and early warning system based on multi-source data fusion and a Bayesian maximum entropy model. Through Bayesian maximum entropy spatiotemporal modeling and prediction theory, and based on detected multi-dimensional nuclear, chemical, and biological environment data, this system provides early warning of nuclear, chemical, and biological environmental crises, including:

[0048] (1) Based on the Gaussian plume model, a physical prior concentration field model driven by physical mechanism is constructed to describe the pollutant diffusion trend. Generally, real-time meteorological data (wind speed, wind direction, atmospheric stability) and potential source term parameters are integrated to drive the Gaussian plume model and generate a physical prior concentration field describing the pollutant diffusion trend. This field provides the system with an initial spatial pollutant distribution estimate that conforms to the laws of fluid dynamics, laying the physical foundation for the entire analysis and early warning.

[0049] In some embodiments, a Gaussian smoke rain model is used to calculate the attenuation factor for different physicochemical properties of nuclear, chemical, and biological contaminants.

[0050] Typically, the physical prior concentration field is expressed as:

[0051]

[0052]

[0053]

[0054]

[0055] in, The radiation dose rate at the predicted location (x, y, z) at time t; : Chemical concentration at position (x,y,z) at time t, volume / mass concentration; The predicted biotoxic agent equivalent concentration at location (x, y, z) at time t; Cumulative concentrations of various pollutants; Initial radiation source intensity; Initialize chemical and industrial toxic gas concentrations; Initial concentration of biological agent; : Radiation source attenuation factor; :Attenuation factors of poisonous gases and industrial toxic agents; Biological agent attenuation factor; : These are the diffusion parameters in the crosswind and vertical directions, respectively, and are functions of downwind distance x, atmospheric stability, and surface roughness. It is a unit of length; Effective velocity vector, in units of length / time;

[0056] :

[0057] The transverse Gaussian distribution term describes the diffusion pattern of matter in the transverse (y-direction) direction and conforms to a normal distribution. ': Relative to lateral distance, it describes the change in emission source strength;

[0058] :

[0059] The vertical Gaussian distribution term describes the diffusion of matter in the vertical z-direction; z: vertical coordinate; Effective emission height refers to the actual height of the emission source or the effective height after considering the elevation. : Terrain correction factor, which takes into account both natural terrain and man-made structures, and considers the influence of terrain such as mountains and valleys on diffusion. It is a function of spatial location (x, y) and is a dimensionless parameter. It may be 0 in flat terrain and the concentration value is adjusted in complex terrain.

[0060] ;

[0061] Among them, V d,rad : Dry settling velocity of radioactive materials, m / s, 0.005–0.015 m / s for gaseous radioactive materials, and 0.008–0.025 m / s for aerosol radioactive materials; H mix Atmospheric mixing layer height, in meters (m). During the day, it ranges from 400 to 1500 meters, depending on atmospheric stability; at night, it is approximately 300 meters. rad Radioactive erosion coefficient, ranging from 0.00008 to 0.0003, with the specific value depending on atmospheric stability; b rad Rainfall intensity index, ranging from 1.4 to 1.6, with the specific value depending on atmospheric stability; I: rainfall intensity, mm / h; T: rainfall intensity. 1 / 2 : half-life of radionuclide, s; ln2: natural logarithm constant, ≈0.693; t: diffusion time, s;

[0062] ;

[0063] Among them, V d,chem Dry deposition rate of chemical agents, m / s; H mix Atmospheric mixing layer height, in meters (m), ranging from 400 to 1500 m during the day and approximately 300 m at night; k chem : Chemical solubility coefficient; I: Rainfall intensity, mm / h; H rain Characteristic rain-washing height, in meters (m), typically taken as 500 meters; k hydro : Hydrolysis rate constant, s⁻¹; RH: Relative humidity, [0~1]; kphoto : Photolysis rate constant, s⁻¹; UV: Ultraviolet intensity index, [0~1]; t: Diffusion time, s;

[0064] ;

[0065] Among them, V d,bio : Dry deposition rate of biological agents, m / s; H mix Atmospheric mixing layer height, in meters (m), ranging from 400 to 1500 meters during the day (the exact value depends on atmospheric stability), and approximately 300 meters at night; k. bio : Bacterial agent washout coefficient; c bio Rainfall intensity index; I: Rainfall intensity, mm / h; k UV UV inactivation coefficient, s⁻¹; UV index UV index, [0–11+]; A: Arrhenius prefactor, s⁻¹; Ea: activation energy, J / mol; R: ideal gas constant, 8.314 J / mol·K; T: ambient temperature, K; T ref Reference temperature, K; k RH : Humidity deactivation coefficient, s⁻¹; RH: Ambient relative humidity, [0~1]; RH opt : Optimal relative humidity, [0~1]; t: Diffusion time, s;

[0066] ;

[0067] Among them, H max H min : Highest and lowest elevations within the analysis area, in meters; L: Characteristic length scale, 1000 meters; A rough Roughness area ratio = actual surface area / projected area; : Weighting coefficients, 0.1 and 0.3 respectively; D: Distance from the prediction point to the nearest building, m; W: Average street width, m.

[0068] (2) Data fusion and uncertainty quantification are performed using the Bayesian maximum entropy model to obtain the posterior probability distribution; specifically including:

[0069] (2-1) Establish the physical prior probability distribution, transforming the physical prior concentration field into a prior probability distribution within a Bayesian framework. Typically, the physical prior field output from the Gaussian plume model in the previous step is transformed into a prior probability distribution within a Bayesian framework. This distribution represents the best initial estimate of the risk status at various points in space based on physical laws before incorporating any real-time observation data. Specifically, this includes:

[0070] (2-1-1) Define a spatial random field. Define the spatial pollutant concentration field to be studied as a spatial random field X; where X is a set of random variables X(s), each random variable X(s) represents the unknown pollutant concentration at the spatial location s=(x,y,z), confirming that the pollutant concentration value is uncertain at any spatial point.

[0071] (2-1-2) Establish the form and mean of the prior distribution. Assume that the random field X in the space follows a multivariate Gaussian distribution, which is completely determined by its mean vector. Covariance Matrix Determined; mean vector Each component of the vector

[0072] The value is directly assigned as the calculation result of the Gaussian plume model at the corresponding position:

[0073]

[0074] Typically, in the absence of any field observation data, the best estimate of the concentration value based on physical laws is the output value of the Gaussian plume model, embedding the prediction depth of the physical model into the core of the probabilistic framework.

[0075] (2-1-3) Quantifying prior uncertainty and constructing the covariance matrix; the construction of the covariance matrix quantifies two types of uncertainty: the uncertainty of the estimate at each location itself (point variance), and the correlation between concentration estimates at different spatial locations (spatial covariance); the methods for calculating point variance and spatial covariance include:

[0076] a) Calculation of point variance Var(X(s))

[0077] The uncertainty of the physical model itself is quantified using an uncertain ship model. For each location, the point variance of the concentration estimate is calculated using the following formula:

[0078] ;

[0079] Among them, Var(Q total ), Var(σ y ), Var(σ) z ), It is the variance of the input parameters of the Gaussian plume model, representing the degree of uncertainty of the initial values ​​such as source strength, diffusion parameters, and wind speed. These values ​​can be initialized based on sensor accuracy, historical data statistics, or expert experience. Based on this formula, it can be seen that the more unstable the output of the physical model at a certain point, that is, the larger the variance, the more "flat" the probability distribution corresponding to it in the Bayesian prior distribution, that is, the lower the reliability of the initial estimate at that point.

[0080] Among them, Var(Q total The variance of total source intensity / concentration represents the uncertainty in estimating the total intensity (total release amount or total initial concentration) of the initial release from a pollution event. This is one of the largest sources of uncertainty in model predictions; specific sources and determination methods include:

[0081] Sensor extrapolation: When the location of the release source is not directly monitored by the sensor, the source strength extrapolated by the back diffusion model based on the early readings of a few sensors downwind has a high degree of uncertainty.

[0082] Expert experience range: When data is scarce, the range of source strength estimates given by experts based on the type of event (such as small leaks, large explosions);

[0083] Historical case statistics: Analyze similar historical events and count the deviation between their actual source strength and the initial estimate, using this as a variance estimate;

[0084] Default setting: When the system is initialized, a relatively large relative variance can be set (for example, assuming that the source strength estimate has 50% to 100% uncertainty).

[0085] and The variance of the lateral and vertical diffusion parameters represents the inaccuracy in the classification of atmospheric turbulence states (i.e., atmospheric stability) and the resulting errors in estimating horizontal and vertical diffusion capabilities. Specific sources and determination methods include:

[0086] Atmospheric stability classification error: There is subjectivity and error in determining Pasqual stability levels based on limited meteorological data (such as wind speed and sunshine). Different levels correspond to different σ values. y σ z The values ​​differed significantly;

[0087] Uncertainty in the diffusion formula: Even if the stability level is determined, the formula used to calculate σ... y(x) and σ z(x) Empirical formulas (such as Briggs' formula) also have a range of errors;

[0088] Local effects: Rough urban underlying surfaces and complex terrain can significantly alter diffusion patterns, making general parameters less applicable and increasing uncertainty in these areas;

[0089] Determination method: Typically based on research findings in atmospheric diffusion, a coefficient of variation (e.g., σ) is assigned to the diffusion parameters at each stability level. y If the coefficient of variation is 30%, then the variance can be calculated as Var(σ). y ) = (0.3×σ y )2 .

[0090] The variance of the effective wind speed represents the uncertainty in estimating the average wind speed driving pollutant transport. Specific sources and determination methods include:

[0091] Wind speed sensor measurement error: Anemometers deployed on-site have inherent instrument errors.

[0092] Spatial representativeness error: Single-point wind speed measurements cannot fully represent the average wind conditions along the entire diffusion path, especially in the vertical direction.

[0093] Spatiotemporal interpolation error: When using weather forecasts or sparse weather station data, interpolation to obtain wind speeds at specific locations will introduce errors.

[0094] Determination method: The calibration accuracy of the wind speed sensor can be directly used, for example, ±0.5m / s, then the variance is (0.5). 2 Alternatively, a relative error can be set based on the data source (e.g., forecast data may have larger errors), such as 10% to 20%.

[0095] b) Spatial covariance Cov(X(s) i ),X(s j The modeling of )) uses a parameterized covariance function;

[0096] To describe spatial correlation, a parameterized covariance function, i.e., a kernel function, can be used; in some embodiments, an exponential covariance function is used as follows:

[0097]

[0098] Where σ² represents a global point variance scaling parameter. In practice, its value is usually determined as follows: applying the aforementioned point variance calculation formula, the values ​​of discrete locations s in space are calculated. i Prior point variance Var(X(s) i Then, take the average or maximum variance of all these location points as the value of σ².

[0099] L: Spatial correlation length; L is a key parameter that defines the rate of decay of spatial correlation; beyond L, the correlation between the concentrations of two points decreases significantly. This parameter can be set according to the physical characteristics of pollutant diffusion and the geographical features of the monitoring area.

[0100] The covariance function ensures that concentration estimates of physically adjacent points are more correlated, making the prior distribution spatially continuous and smooth, consistent with the physical intuition of pollutant diffusion.

[0101] (2-1-4) Complete the prior distribution construction; through the above steps, the prior probability distribution is constructed. Expressed as:

[0102] ;

[0103] That is, X obeys a set of rules. For the mean, The covariance matrix is ​​a multivariate Gaussian distribution;

[0104] (2-2) Multi-source heterogeneous data fusion transforms the detected data and prior knowledge into constraints on the random field X, including:

[0105] Hard data fusion: High-precision detection data is introduced as a strong constraint to modify prior data; for location... High-precision sensor measurement values ​​at the location Model it as a deterministic constraint: Or a Gaussian distribution constraint with minimal variance: in, Approaching 0;

[0106] Soft data integration: quantifying uncertain information into probabilistic soft constraints; for low-precision sensor data, quantifying it into an interval constraint. or a distribution with large variance ;

[0107] Positioning expert experience The concentration is likely higher than the location. "Quantified into probability inequality constraints:"

[0108] ,in Confidence level;

[0109] In some embodiments, soft data includes interval estimates of low-precision sensor data and probability inequality constraints based on expert experience.

[0110] (2-3) Posterior probability update and optimization;

[0111] Applying the Bayesian maximum entropy principle, we seek the posterior probability distribution with the maximum entropy, while satisfying all data constraints in (2.2). :

[0112]

[0113] in The set of all probability density functions that satisfy the constraints; the output posterior probability distribution. It is an exponential family distribution with maximum entropy; simultaneously, the spatial covariance structure is optimized, constrained by the following exponential covariance function model:

[0114]

[0115] in, The upper limit of point variance, Spatial related length;

[0116] Finally, from the posterior probability distribution Extract the posterior concentration estimate for each spatial location. (Mean) and Estimation Uncertainty (Standard deviation).

[0117] (3) Based on the posterior probability distribution, a dynamic risk map is synthesized to trigger multi-level early warnings; specifically including:

[0118] (3.1) Construct a multidimensional data cube Synthesize dynamic risk maps; where each grid cell Include:

[0119] Posterior estimated concentration: ;

[0120] Probability of exceeding the safety threshold:

[0121] ,in, This is the posterior cumulative distribution function;

[0122] Estimate uncertainty: (Coefficient of variation).

[0123] (3.2) A pre-set two-factor decision matrix is ​​used to achieve multi-level early warning triggering; the two-factor decision matrix defines the early warning level L and the risk probability. Functional relationship with uncertainty U: The specific rules for the decision function M are defined in the following table:

[0124]

[0125] By traversing the risk map and applying the above rules to each grid cell, an early warning level is automatically assigned.

[0126] In some embodiments, the probability of risk exceeding a threshold and the estimated uncertainty are calculated based on the posterior distribution, and then mapped to an early warning level through a preset two-factor decision matrix.

[0127] (4) Introduce an expert feedback mechanism to continuously learn and optimize model parameters and early warning rules; specifically including:

[0128] (4-1) Automatic parameter calibration: Construct a historical case library, and use Bayesian inference and MCMC sampling methods to invert and optimize key physical parameters in the Gaussian plume model.

[0129] ,in For the first The model parameter vector for this event. This corresponds to the ground truth data;

[0130] Parameters are updated using Bayesian inference: ;

[0131] in, Let be the prior distribution of the parameters. Let be the likelihood function, usually assumed to be ; Output for the model;

[0132] Using the Markov chain Monte Carlo sampling method from the posterior distribution Samples are drawn from the sample and their mean is used to update the parameters of the Gaussian plume model: .

[0133] (4-2) Rule and Knowledge Base Optimization

[0134] Through expert feedback interfaces, logistic regression is used to optimize early warning thresholds, and natural language processing (NLP) techniques are used to transform expert text feedback into new probabilistic soft data rules to update the knowledge base; specifically including:

[0135] Threshold optimization: Collect expert correction data on warning levels Using a logistic regression model, establish the corrected rank and ( The relationship between )

[0136]

[0137] Adjust the threshold P in the decision matrix based on the model results. red P orange ,... and U red U orange ...;

[0138] Soft data knowledge base expansion: Extract entity-relation triples from expert text feedback T using natural language processing techniques. This is then transformed into new soft data rules. For example, extracting (urban canyon, diffusion parameter, should be doubled) generates the following rule:

[0139] When in an "urban canyon" terrain:

[0140]

[0141] in It has a relatively small variance.

[0142] Example 1

[0143] Example 1 discloses a nuclear, chemical, and biological environment monitoring and early warning system based on multi-source data fusion and a Bayesian maximum entropy model, such as... Figure 1 As shown, it includes: nuclear, biological and chemical environment monitoring and early warning equipment and an algorithm platform configured with multi-source data fusion and Bayesian maximum entropy model.

[0144] like Figure 2 As shown, the unmanned nuclear, biological, and chemical environment monitoring and early warning equipment integrates detectors, communication devices, three-dimensional point cloud scanning devices, power management, a main control platform, and installation and fixing accessories; among them, the environmental detectors include nuclear detection parts, chemical agent and industrial toxic gas detection parts, biological warfare agent detection parts, and environmental detection parts; the communication devices include Beidou positioning short message and wireless communication.

[0145] like Figure 3 As shown, the nuclear, chemical, and biological environment monitoring and early warning system based on multi-source data fusion and a Bayesian maximum entropy model includes 16 unattended nuclear, chemical, and biological environment monitoring and early warning devices, serving as 16 detection nodes. Multiple early warning devices can be deployed in a star network or a linear network.

[0146] Example 2

[0147] The nuclear, chemical, and biological environment monitoring and early warning method based on multi-source data fusion and Bayesian maximum entropy model disclosed in Example 2 is implemented using the early warning system disclosed in Example 1 of this invention, such as... Figure 4 As shown, it specifically includes:

[0148] Real-time collection of multi-dimensional nuclear, chemical, and biological environment data using nuclear, chemical, and biological environment monitoring devices;

[0149] A physical prior concentration field model describing the diffusion trend of pollutants was constructed using the Gaussian smoke and rain model.

[0150] Data fusion and uncertainty quantification are performed using the Bayesian maximum entropy model to obtain the posterior probability distribution.

[0151] Based on the posterior probability distribution, a dynamic risk map is synthesized and multi-level early warnings are triggered.

[0152] An expert feedback mechanism is introduced to continuously optimize model parameters and early warning rules.

[0153] The nuclear, chemical, and biological (NCB) environmental monitoring and early warning system and method disclosed in this invention, based on multi-source data fusion and a Bayesian maximum entropy model, integrates integrated environmental detection technology, nuclear detection technology, chemical detection technology, industrial toxic gas detection technology, and biological warfare agent detection technology, and can meet the needs of various application scenarios. The system features high integration and ease of use, as well as a high degree of intelligence. Based on a constructed physical prior concentration field model of pollutant diffusion trends, it utilizes a Bayesian maximum entropy model for data fusion and uncertainty quantification to obtain a posterior probability distribution, synthesizes a dynamic risk map and a multi-level early warning triggering mechanism, and combines the expert experience, relevant standards, and processing methods accumulated over many years of development in nuclear and chemical detection with real-time detection data to provide scientific and reasonable early warning and handling opinions for NCB crises. It has promising application prospects in the field of NCB environmental monitoring and early warning.

[0154] The technical solutions and technical details disclosed in the embodiments of this invention are merely illustrative of the inventive concept of this invention and do not constitute a limitation on the technical solutions of the embodiments of this invention. Any conventional changes, substitutions, or combinations made to the technical details disclosed in the embodiments of this invention have the same inventive concept as this invention and are within the protection scope of the claims of this invention.

Claims

1. A nuclear, chemical, and biological environment monitoring and early warning system based on multi-source data fusion and a Bayesian maximum entropy model, characterized in that, This includes nuclear, chemical, and biological environment monitoring and early warning equipment and algorithm platforms; among which, The nuclear, chemical, and biological environment monitoring and early warning equipment includes: a detection component for real-time acquisition of multi-dimensional nuclear, chemical, and biological environment data; a communication and positioning component for communication, positioning, and timing; a three-dimensional point cloud scanning component for acquiring three-dimensional spatial information of the deployment environment and constructing a digital base map; a main control platform for data acquisition, analysis, encapsulation, and transmission; and a power supply component for providing power assurance. The algorithm platform is configured with multi-source data fusion and a Bayesian maximum entropy model. It is configured to conduct early warning of nuclear, chemical, and biological environmental crises based on multi-dimensional nuclear, chemical, and biological environmental data detected by Bayesian maximum entropy spatiotemporal modeling and prediction theory.

2. The nuclear, chemical, and biological environment monitoring and early warning system based on multi-source data fusion and Bayesian maximum entropy model according to claim 1, characterized in that, The detection component includes: The environmental detection module is used to detect environmental data; Nuclear detection module, used to detect gamma rays; Chemical agents and industrial toxic gases detection module, used to detect the types of chemical agents and industrial toxic gases; Biological agent detection module, used to detect the type of biological agent.

3. The nuclear, chemical, and biological environment monitoring and early warning system based on multi-source data fusion and Bayesian maximum entropy model according to claim 1, characterized in that, The communication and positioning components include: Wireless communication module; integrates wireless communication technology and supports a decentralized self-organizing network architecture; Satellite navigation module; provides positioning information and precise time synchronization.

4. The nuclear, chemical, and biological environment monitoring and early warning system based on multi-source data fusion and Bayesian maximum entropy model according to claim 1, characterized in that, The method of using Bayesian maximum entropy spatiotemporal modeling and prediction theory, based on multi-dimensional nuclear, chemical, and biological environment data, to provide early warning of nuclear, chemical, and biological crises includes: (1) A physical prior concentration field model driven by physical mechanism is constructed based on the Gaussian smoke and rain model to describe the diffusion trend of pollutants; (2) Data fusion and uncertainty quantification are performed using the Bayesian maximum entropy model to obtain the posterior probability distribution; (3) Based on the posterior probability distribution, synthesize a dynamic risk map and trigger multi-level early warning; (4) Introduce an expert feedback mechanism to continuously optimize model parameters and early warning rules.

5. The nuclear, chemical, and biological environment monitoring and early warning system based on multi-source data fusion and Bayesian maximum entropy model according to claim 4, characterized in that, The physical prior concentration field model is expressed as follows: in, The radiation dose rate at the predicted location (x, y, z) at time t; : Chemical concentration at position (x,y,z) at time t, volume / mass concentration; The predicted biotoxic agent equivalent concentration at location (x, y, z) at time t; Cumulative concentrations of various pollutants; Initial radiation source intensity; Initialize chemical and industrial toxic gas concentrations; Initial concentration of biological agent; : Radiation source attenuation factor; :Attenuation factors of poisonous gases and industrial toxic agents; Biological agent attenuation factor; : These are the diffusion parameters in the crosswind and vertical directions, respectively, and are functions of downwind distance x, atmospheric stability, and surface roughness. It is a unit of length; Effective velocity vector, in units of length / time; The horizontal Gaussian distribution term describes the diffusion pattern of matter in the horizontal (y-direction) direction, which conforms to a normal distribution. ': Relative to the horizontal distance, it describes the change in source strength; : The Gaussian distribution term in the vertical direction, describing the diffusion of matter in the z-direction; z: vertical coordinate; Effective emission height; : Terrain correction factor, which is a function of spatial location (x, y); ； in, : Dry deposition rate of radioactive material, m / s, gaseous radioactive material: 0.005~0.015 m / s, aerosol radioactive material: 0.008~0.025 m / s; Hmix: Atmospheric mixing layer height, m, 400~1500m during the day, about 300m at night; arad: Radioactive scavenging coefficient, ranging from 0.00008~0.0003; brad: Rainfall intensity index, ranging from 1.4~1.6; I: Rainfall intensity, mm / h; T1 / 2: Radionuclide half-life, s; ln2: Natural logarithm constant, ≈0.693; t: Diffusion time, s; ； Among them, V d,chem Dry deposition rate of chemical agents, m / s; H mix Atmospheric mixing layer height, m, 400–1500 m during the day and approximately 300 m at night; k chem : Chemical solubility coefficient; I: Rainfall intensity, mm / h; H rain Characteristic rain-washing height, in meters (m), typically taken as 500 meters; k hydro : Hydrolysis rate constant, s⁻¹; RH: Relative humidity, [0~1]; k photo : Photolysis rate constant, s⁻¹; UV: Ultraviolet intensity index, [0~1]; t: Diffusion time, s; ； Among them, V d,bio : Dry deposition rate of biological agents, m / s; H mix Atmospheric mixing layer height, m, 400–1500 m during the day, approximately 300 m at night; k bio : Bacterial agent washout coefficient; c bio Rainfall intensity index; I: Rainfall intensity, mm / h; k UV UV inactivation coefficient, s⁻¹; UV index UV index, [0-11+]; A: Arrhenius prefactor, s⁻¹; Ea: activation energy, J / mol; R: ideal gas constant, 8.314 J / mol·K; T: ambient temperature, K; T ref Reference temperature, K; k RH : Humidity deactivation coefficient, s⁻¹; RH: Ambient relative humidity, [0-1]; R Hopt : Optimal relative humidity, [0-1]; t: Diffusion time, s; ； Among them, H max H min : Highest and lowest elevations within the analysis area, in meters; L: Characteristic length scale, 1000 meters; A rough Roughness area ratio = actual surface area / projected area; , : Weighting coefficients, 0.1 and 0.3 respectively; d: Distance from the prediction point to the nearest building, m; W: Average street width, m.

6. The nuclear, chemical, and biological environment monitoring and early warning system based on multi-source data fusion and Bayesian maximum entropy model according to claim 4, characterized in that, (2) Bayesian maximum entropy fusion and uncertainty quantification yield the posterior probability distribution, which specifically includes: (2-1) Establish the physical prior probability distribution; transform the physical prior concentration field into a prior probability distribution within a Bayesian framework, specifically including: (2-1-1) Define a spatial random field; define the spatial pollutant concentration field to be studied as a spatial random field X; where X is a set of random variables X(s), each random variable X(s) represents the unknown pollutant concentration at the spatial location s=(x,y,z); (2-1-2) Establish the form and mean of the prior distribution. Assume that the random field X in the space follows a multivariate Gaussian distribution, which is completely determined by its mean vector. Covariance Matrix Determined; mean vector Each component The value is directly assigned as the calculation result of the Gaussian plume model at the corresponding position: (2-1-3) Quantify prior uncertainty and construct a covariance matrix; including the point variance of the uncertainty estimate for each location and the spatial covariance between concentration estimates at different spatial locations; where, a) Formula for calculating point variance Var(X(s)): Among them, Var(Q total ), Var(σ y ), , It is the variance of the input parameters of the Gaussian plume model, representing the degree of uncertainty of the initial knowledge such as source strength, diffusion parameters, and wind speed; b) Spatial covariance Cov(X(s) i ),X(s j The modeling of )) uses a parametric covariance function; the exponential covariance function is: Where, σ 2 Point variance; L: Spatial correlation length; (2-1-4) Complete the construction of the prior distribution; the constructed prior probability distribution Expressed as: That is, X obeys a set of rules. For the mean, The covariance matrix is ​​a multivariate Gaussian distribution; (2-2) Multi-source heterogeneous data fusion transforms sensor detection data and prior knowledge into constraints on a random field x, including: Hard data fusion: for location High-precision sensor measurement values ​​at the location Model it as a deterministic constraint: Or a Gaussian distribution constraint with minimal variance: ;in, Approaching 0; Soft data integration: quantifying uncertain information into probabilistic soft constraints; for low-precision sensor data, quantifying it into an interval constraint. or a distribution with large variance Positioning expert experience The concentration is likely higher than the location. "Quantified into probability inequality constraints:" Where p is the confidence level; (2-3) Posterior probability update and optimization; Applying the Bayesian maximum entropy principle, we seek the posterior probability distribution with the maximum entropy, while satisfying all data constraints in (2.2). : in The set of all probability density functions that satisfy the constraints; the output posterior probability distribution. It is an exponential family distribution with maximum entropy; simultaneously, the spatial covariance structure is optimized, constrained by the following exponential covariance function model: in, The upper limit of point variance, Spatial related length; Finally, from the posterior probability distribution Extract the posterior concentration estimate for each spatial location. and estimation uncertainty .

7. The nuclear, chemical, and biological environment monitoring and early warning system based on multi-source data fusion and Bayesian maximum entropy model according to claim 4, characterized in that, (3) Based on the posterior probability, a dynamic risk map is synthesized and multi-level early warnings are triggered, specifically including: (3.1) Construct a multidimensional data cube Synthesize dynamic risk maps; where each grid cell Include: Posterior estimated concentration ; Probability of exceeding the safety threshold: ,in This is the posterior cumulative distribution function; Estimate uncertainty: ; (3.2) A pre-set two-factor decision matrix is ​​used to achieve multi-level early warning triggering; the two-factor decision matrix defines the early warning level. Risk Probability and uncertainty Functional relationship: The specific rules for the decision function M are defined in the following table: By traversing the risk map and applying the above rules to each grid cell, an early warning level is automatically assigned.

8. The nuclear, chemical, and biological environment monitoring and early warning system based on multi-source data fusion and Bayesian maximum entropy model according to claim 4, characterized in that, (4) Introduce an expert feedback mechanism to continuously optimize model parameters and early warning rules, specifically including: (4-1) Automatic parameter calibration: Construct a historical case library, and use Bayesian inference and MCMC sampling methods to invert and optimize key physical parameters in the Gaussian plume model. in For the first The model parameter vector for this event. This corresponds to the ground truth data; Parameters are updated using Bayesian inference: in, Let be the prior distribution of the parameters. Let be the likelihood function, usually assumed to be ; Output for the model; Using the Markov chain Monte Carlo sampling method from the posterior distribution Samples are drawn from the sample and their mean is used to update the parameters of the Gaussian plume model: ; (4-2) Rule and Knowledge Base Optimization By using the expert feedback interface, logistic regression is used to optimize the early warning threshold, and natural language processing technology is used to transform expert text feedback into new probabilistic soft data rules to update the knowledge base.

9. The nuclear, chemical, and biological environment monitoring and early warning system based on multi-source data fusion and Bayesian maximum entropy model according to claim 8, characterized in that, (4-2) Rule and knowledge base optimization specifically includes: Threshold optimization: Collect expert correction data on warning levels Using a logistic regression model, establish the corrected rank and ( The relationship between ) Adjust the thresholds Pred, Porange, ... and Ured, Uorange, ... in the decision matrix based on the model results; Soft data knowledge base expansion: Extract entity-relation triples from expert text feedback T using natural language processing techniques. And transform it into new soft data rules.

10. A method for monitoring and early warning of nuclear, chemical, and biological environments based on multi-source data fusion and a Bayesian maximum entropy model, characterized in that... Implemented using the early warning system according to any one of claims 1 to 9, specifically including: Real-time collection of multi-dimensional nuclear, chemical, and biological environment data using nuclear, chemical, and biological environment monitoring devices; A physical prior concentration field model describing the diffusion trend of pollutants was constructed using the Gaussian smoke and rain model. The Bayesian maximum entropy model is used for data fusion and uncertainty quantification to obtain the posterior probability distribution; Based on the posterior probability distribution, a dynamic risk map is synthesized and multi-level early warnings are triggered. An expert feedback mechanism is introduced to continuously optimize model parameters and early warning rules.

Images