Health monitoring, analyzing and early warning method and system for customs travelers

Abstract

The invention discloses a health monitoring analysis early warning method and system for customs travel personnel, and relates to the field of customs quarantine, and the method comprises the steps: obtaining biological characteristic monitoring data, health declaration data, travel track data and environment perception data; performing feature extraction on the biological feature monitoring data, the health declaration data, the travel track data and the environment perception data to obtain a plurality of monomer feature vectors; performing feature data fusion on all the monomer feature vectors to obtain fused feature data; and performing risk judgment based on the fused feature data and a dynamic threshold to generate early warning information. The method can improve the health risk identification precision, is adaptive to different epidemic situation stages and passenger population characteristics, and improves the sensitivity when the epidemic situation suddenly rises; excessive alarms are reduced when the risk is relatively low, so that missed alarm and false alarm are balanced, a management layer can also be supported to make a macroscopic decision on the overall condition of the crowd, and the prevention and control efficiency is further improved.

Description

Technical Field

[0001] The present invention relates to the field of customs quarantine, and more specifically, to a method and system for health monitoring, analysis and early warning of customs travelers. Background Art

[0002] With the frequent flow of people around the world and the wide variety of epidemics, customs, airports and other ports are facing increasing pressure on health monitoring. Traditional health monitoring methods mainly rely on single-point temperature detection. Infrared thermometers or forehead thermometers are usually used to quickly measure the temperature of passengers. Although this method is fast, it can only obtain single physiological data, and cannot detect other health abnormalities in time, and it is difficult to determine whether the passenger has a potential risk of infection. At the same time, the health declaration form is filled out by the passenger and then checked by the on-duty personnel. The health declaration data is usually text data with a low degree of structure, low information utilization, and difficult to conduct comprehensive analysis with other data sources. Ports or customs usually set a fixed threshold based on past experience or regulations of the competent authorities to determine whether the body temperature exceeds the standard or whether the symptoms are suspicious. This type of fixed threshold cannot be dynamically adjusted according to changes in the epidemic, characteristics of the passenger population and environmental risk factors, which is easy to lead to underreporting or false reporting, reducing the accuracy of epidemic prevention.

[0003] How to quickly integrate multi-source data and conduct dynamic intelligent assessment of health risks has become an urgent problem to be solved in the industry. Therefore, it is necessary to propose a health monitoring, analysis and early warning method and system for customs travelers. Summary of the invention

[0004] A series of simplified concepts are introduced in the Summary of the Invention section, which will be further described in detail in the Detailed Description of the Invention section. The Summary of the Invention section of this application does not mean to attempt to define the key features and essential technical features of the claimed technical solution, nor does it mean to attempt to determine the scope of protection of the claimed technical solution.

[0005] In the first aspect, the present application proposes a method for health monitoring, analysis and early warning of customs travelers, comprising: Obtain biometric monitoring data, health declaration data, travel trajectory data, and environmental perception data; Performing feature extraction on the biometric monitoring data, the health declaration data, the travel trajectory data, and the environmental perception data to obtain a plurality of individual feature vectors; Performing feature data fusion on all the above-mentioned individual feature vectors to obtain fused feature data; Risk judgment is made based on the above-mentioned fused feature data and dynamic thresholds to generate early warning information.

[0006] In a feasible implementation, the above-mentioned monomer feature vector includes a physiological feature vector, a health declaration vector, a travel trajectory vector and an environment perception vector. The above-mentioned biometric monitoring data, the above-mentioned health declaration data, the above-mentioned travel trajectory data and the above-mentioned environmental perception data are subjected to feature extraction to obtain multiple monomer feature vectors, including: Performing a first feature extraction on the above-mentioned biometric monitoring data based on a 1D convolutional neural network to obtain a physiological feature vector, wherein the above-mentioned biometric monitoring data includes body temperature data and gait data; Performing a second feature extraction on the health declaration data based on a hybrid coding strategy to obtain a health declaration vector; Performing spatial feature extraction, temporal feature extraction and first feature fusion operations on the above travel trajectory data to obtain a travel trajectory vector; The above environmental perception data is subjected to full connection layer feature extraction, ResNet feature extraction and second feature fusion operations to obtain an environmental perception vector.

[0007] In a feasible implementation manner, the above-mentioned feature data fusion of all the above-mentioned individual feature vectors to obtain fused feature data includes: Creating a passenger node according to the physiological feature vector, the health declaration vector, and the travel trajectory vector; Create an environment node based on the above environment perception vector; Create the passenger-environment side and the environment-environment side; Calculate the attention scores of the passenger and environment edges and the environment and environment edges respectively; Update the above passenger nodes and the above environment nodes based on the multi-head attention mechanism; The updated passenger node and the environment node are pooled to obtain the fused feature data.

[0008] In a feasible implementation manner, the specific steps of determining the above dynamic threshold include: Obtain historical baseline threshold information, environmental correction information, epidemic model correction information, and population risk information; The above dynamic threshold is determined based on the above historical benchmark threshold information, environmental correction information, epidemic model correction information and group risk information.

[0009] In a feasible implementation manner, the above historical benchmark threshold information is determined based on a short-term threshold and a long-term threshold.

[0010] In a feasible implementation manner, the specific steps of determining the above epidemic model correction information include: Obtain infection density information based on the extended SEIR model; Determine the initial information of the epidemic model based on the above infection density information, historical threshold information and epidemic severity adjustment parameters; The initial information of the above epidemic model is adaptively optimized by threshold to obtain the revised information of the epidemic model.

[0011] In a feasible implementation manner, the adaptive threshold optimization of the initial information of the epidemic model to obtain the epidemic model correction information includes: Environmental modeling was performed using the extended SEIR model as the environment and historical health monitoring data as the experience pool; Predict the adjustment amount at the current moment by measuring the network; Construct a reward function based on the false positive and false negative index penalty term, threshold stability term, epidemic spread impact term, and policy constraint term; The above reward function is optimized according to the proximal strategy to obtain the correction information of the epidemic model.

[0012] In a feasible implementation manner, the specific steps of determining the above group risk information include: Acquiring contact intensity information between travelers, wherein the contact intensity information is an adjacency matrix of the traveler contact network; Determine group risk diffusion distribution information based on the above contact intensity information; Determine skewness information and kurtosis information based on the above group risk diffusion distribution information; The group risk information is determined based on the skewness information and the kurtosis information.

[0013] In a feasible implementation manner, the above-mentioned warning information includes warning information of customs clearance personnel on duty and warning information of customs clearance management personnel. The above risk judgment is based on the above fusion feature data and dynamic thresholds to generate early warning information, including: Performing individual risk judgment based on the above fusion feature data and the above dynamic threshold information of each person to obtain individual risk information; Generate the above-mentioned warning information for customs clearance personnel based on the above-mentioned personal risk information of each person; The above-mentioned customs clearance management personnel warning information is generated based on the above-mentioned personal risk information of all persons.

[0014] In the second aspect, the present application proposes a health monitoring, analysis and early warning system for customs travelers, comprising: A data acquisition unit, used to acquire biometric monitoring data, health declaration data, travel trajectory data, and environmental perception data; A feature extraction unit, used to extract features from the biometric monitoring data, the health declaration data, the travel trajectory data, and the environmental perception data to obtain a plurality of monomer feature vectors; A feature fusion unit, used for fusing feature data of all the above-mentioned individual feature vectors to obtain fused feature data; The risk judgment unit is used to make risk judgments based on the above-mentioned fused feature data and dynamic thresholds to generate warning information.

[0015] In summary, this application simultaneously collects and integrates multiple data sources such as physiological characteristics, health declarations, travel trajectories, environmental perception, etc., breaking the limitations of traditional single-dimensional detection. By extracting features from these data, the accuracy of health risk identification can be improved. This application introduces dynamic thresholds, which dynamically adjust the health risk determination criteria over time by integrating historical benchmarks, group risk distribution, epidemic transmission models, environmental corrections and other information. Compared with traditional fixed thresholds, dynamic thresholds can adapt to different epidemic stages and passenger population characteristics, improve sensitivity when the epidemic rises suddenly; reduce excessive alarms when the risk is low, thereby balancing missed reports and false alarms. The method proposed in this application can not only provide real-time risk warnings for individual passengers to customs clearance personnel, but also provide overall analysis for customs clearance managers. This multi-layer warning mode is conducive to dealing with individual risks at the first time, and can also support management to make macro decisions on the overall situation of the population, further improving the efficiency of prevention and control.

[0016] The present application proposes a method for health monitoring, analysis and early warning of customs travelers. Other advantages, objectives and features of the present application will be reflected in part through the following description, and will also be understood by technical personnel in the field through research and practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present specification. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings: Figure 1 A process diagram of a method for health monitoring, analysis and early warning of customs travelers provided in an embodiment of the present application; Figure 2 A structural schematic diagram of a health monitoring, analysis and early warning system for customs travelers provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices. The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0019] See also Figure 1 , which is a process diagram of a method for health monitoring, analysis and early warning of customs travelers provided in an embodiment of the present application, which may specifically include: S110, obtaining biometric monitoring data, health declaration data, travel trajectory data and environmental perception data; For example, the passenger's body temperature, gait (or heart rate, blood oxygen for wearable devices) and other biometric monitoring data are collected. These data can be obtained through sensors or wearable devices and can reflect the passenger's immediate health status.

[0020] Health declaration data refers to the health declaration form filled out by passengers before customs clearance, including past medical history, recent symptoms, possible contact history, etc. This part of the data can be structured (such as fixed multiple-choice questions) or contain unstructured text (free description by passengers).

[0021] Travel trajectory data records the travel information and activity locations of passengers within a certain time range. This data usually comes from air tickets, mobile device positioning (GPS or base station) or customs entry and exit records, and can be used to evaluate which areas the passengers have visited and how long they stayed.

[0022] Environmental perception data includes air quality, population density, public health indicators, etc. in the airport. It can also include real-time monitoring images (crowding, crowd mobility, etc. extracted after image processing).

[0023] S120, extracting features from the biometric monitoring data, the health declaration data, the travel trajectory data, and the environmental perception data to obtain a plurality of individual feature vectors; Exemplarily, the above four types of data are separately feature extracted for subsequent unified risk analysis. Biometric monitoring data can use 1D convolutional neural network (1D-CNN) to extract the features of temperature fluctuation or gait abnormality, and finally output a physiological feature vector, which can contain deep feature representations of multiple health indicators.

[0024] For unstructured text, health declaration data can be represented by text embedding (such as BERT, word vector) + RNN / LSTM / attention mechanism, etc.; for structured information (such as medical history, whether there is a fever, etc.), it can be directly mapped into a feature vector; and finally a health declaration vector is generated to summarize the individual's medical history and current health self-reported information.

[0025] The travel trajectory data analyzes spatial characteristics (passenger activity area) and temporal characteristics (length of stay, travel frequency) respectively; the trajectory information can be integrated into a travel trajectory vector using spatiotemporal methods such as graph models (GCN) or LSTM.

[0026] Environmental perception data is processed by multi-layer perceptron (MLP) or other algorithms to process sensor data. If images are included, high-level visual features can be extracted through convolutional neural networks such as ResNet, and finally integrated into an environmental perception vector. The S120 can convert the original four types of multimodal data into higher-level feature vectors for subsequent unified analysis.

[0027] S130, performing feature data fusion on all the above-mentioned individual feature vectors to obtain fused feature data; Exemplarily, four individual feature vectors (physiological characteristics, health declaration vector, travel trajectory vector, and environmental perception vector) are fused to finally output fused feature data, which is equivalent to a comprehensive risk representation that integrates multiple aspects of passenger information.

[0028] S140: Perform risk assessment based on the above fused feature data and dynamic thresholds to generate early warning information.

[0029] For example, a dynamic threshold that changes over time is determined by comprehensively using epidemic models, historical baseline thresholds, environmental corrections, group risk information, etc., so that the system can automatically adjust the risk level classification standards according to the development of the epidemic or changes in the scenario to reduce missed reports and false alarms. Warning information for customs clearance personnel is a risk reminder for individual passengers (such as: "Passenger A is at high risk, please undergo a health re-examination"). Warning letter for customs clearance management personnel: statistical or trend warnings for the overall health of passengers (such as: "The proportion of high-risk passengers has risen to 10%, and it is recommended to open a special isolation channel"). Through S140, the system converts risk assessment results into executable health warnings to help on-site staff or managers make appropriate intervention measures.

[0030] In summary, this application simultaneously collects and integrates multiple data sources such as physiological characteristics, health declarations, travel trajectories, environmental perception, etc., breaking the limitations of traditional single-dimensional detection. By extracting features from these data, the accuracy of health risk identification can be improved. This application introduces dynamic thresholds, which dynamically adjust the health risk determination criteria over time by integrating historical benchmarks, group risk distribution, epidemic transmission models, environmental corrections and other information. Compared with traditional fixed thresholds, dynamic thresholds can adapt to different epidemic stages and passenger population characteristics, improve sensitivity when the epidemic rises suddenly; reduce excessive alarms when the risk is low, thereby balancing missed reports and false alarms. The method proposed in this application can not only provide real-time risk warnings for individual passengers to customs clearance personnel, but also provide overall analysis for customs clearance managers. This multi-layer warning mode is conducive to dealing with individual risks at the first time, and can also support management to make macro decisions on the overall situation of the population, further improving the efficiency of prevention and control.

[0031] In a feasible implementation, the above-mentioned individual feature vectors include a physiological feature vector, a health declaration vector, a travel trajectory vector and an environment perception vector. The feature extraction of the biometric monitoring data, the health declaration data, the travel trajectory data and the environmental perception data to obtain multiple monomer feature vectors includes: Performing a first feature extraction on the above-mentioned biometric monitoring data based on a 1D convolutional neural network to obtain a physiological feature vector, wherein the above-mentioned biometric monitoring data includes body temperature data and gait data; Performing a second feature extraction on the health declaration data based on a hybrid coding strategy to obtain a health declaration vector; Performing spatial feature extraction, temporal feature extraction and first feature fusion operations on the above travel trajectory data to obtain a travel trajectory vector; The above environmental perception data is subjected to full connection layer feature extraction, ResNet feature extraction and second feature fusion operations to obtain an environmental perception vector.

[0032] For example, physiological characteristic data D b include: The body temperature data T = {T1, T2, ..., Tn} and gait data G = {G1, G2, ..., Gn} use 1D convolutional neural network to extract time series features H b : Hb=σ(Wc*Db+bc) Among them, Wc is the convolution kernel, ∗ represents the convolution operation, σ is the nonlinear activation function (ReLU), and bc is the bias term. 1D-CNN can identify short-term features (such as abnormal body temperature fluctuations and abnormal gait) Health declaration data D hPresented in text form, it needs to be encoded through the embedding layer + RNN + attention mechanism: H h =Attention(RNN(Embed(D h ))) Among them, Embed(): converts text data into vector representation (such as using Word2Vec or BERT), RNN (recurrent neural network) is used to extract time series features of text, and Attention() is used to focus on important information (such as "recent fever" and "history of close contact").

[0033] Travel trajectory data It consists of the passenger's location (GPS), timestamp and area information, and spatial and temporal features need to be extracted.

[0034] Spatial features use GCN (graph neural network) to represent trajectory network :

[0035] in: is the adjacency matrix, is the trajectory node representation, is the model parameter.

[0036] Use LSTM (Long Short-Term Memory Network) to extract trajectory time features :

[0037] The trajectory features are fused to obtain the travel trajectory vector :

[0038] in, Indicates concatenating multiple vectors along a certain dimension.

[0039] Environmental perception data Consists of multiple environmental variables (air quality, human density, airport pollutants). Use MLP to extract linear relationships :

[0040] Using ResNet to extract high-level features :

[0041] Final environment perception vector for:

[0042] In summary, the method proposed in this embodiment breaks the limitations of traditional single or small data sources by extracting and fusing the features of physiological characteristics, health declaration data, travel trajectory and environmental perception, and can capture the possible health risks of passengers in a more comprehensive and three-dimensional way. By fusing data in different dimensions to form a comprehensive risk representation, the system can better grasp the abnormal points of the health status of passengers, reduce the omissions and false alarms caused by relying on a single feature, and significantly improve the depth and accuracy of health risk identification.

[0043] In a feasible implementation manner, the above-mentioned feature data fusion of all the above-mentioned individual feature vectors to obtain fused feature data includes: Creating a passenger node according to the physiological feature vector, the health declaration vector, and the travel trajectory vector; Create an environment node based on the above environment perception vector; Create the passenger-environment side and the environment-environment side; Calculate the attention scores of the passenger and environment edges and the environment and environment edges respectively; Update the above passenger nodes and the above environment nodes based on the multi-head attention mechanism; The updated passenger node and the environment node are pooled to obtain the fused feature data.

[0044] For example, the health status H of each passenger fusion Characterized by the following monomer feature vector: H fusion =[H b ,H h ,H t ] These features together constitute the passenger node V passenger : V passenger =H fusion The environmental node represents the health risk factors of an airport or port, which is represented by the environmental perception vector Extract environment node V env : V env =H e Passenger-Environment Edges (PEE) connect the passenger nodes when the passenger stays in a certain area. With this environment node The connection weights are based on the traveler's stay time and regional risk score:

[0045] in, is the connection weight between the passenger and the environment, Indicates the location information of passengers and the environment, Indicates passengers and travelers The square of the Euclidean distance between represents the influence of the control space distance, The duration of contact between travelers and the environment. is the total observation time.

[0046] Environment-Environment Edges (EEE) are the associations between different areas, such as different terminals at the same airport.

[0047] The strength of the connection is determined by the similarity of the two environmental areas (such as air quality, population density):

[0048] in, is the connection weight between the environment and the environment edge, Represents the Euclidean distance between two environmental region feature vectors.

[0049] In the Graph Attention Network (GAT), the attention mechanism is used to calculate the weight of the edge:

[0050] in, is the feature vector of node i, is the feature vector of node j, is the feature vector of node k, are the parameters of the attention network, represents feature concatenation, For Node Neighbor nodes, LeakyReLU is a leaky rectified linear unit, which is used to increase nonlinear transformation.

[0051] In order to consider different types of information more comprehensively, a multi-head attention mechanism is used for feature aggregation:

[0052] in: is the feature vector after aggregation, is the number of attention heads, For the The edge weights of the attention heads, For the The projection matrix of the attention head, is the feature of the node.

[0053] Pool the updated traffic and environment nodes to obtain the final fusion feature data :

[0054] in, For pooling operation.

[0055] In a feasible implementation manner, the specific steps of determining the above dynamic threshold include: Obtain historical baseline threshold information, environmental correction information, epidemic model correction information, and population risk information; The above dynamic threshold is determined based on the above historical benchmark threshold information, environmental correction information, epidemic model correction information and group risk information.

[0056] Exemplary, dynamic threshold It can be calculated based on the following formula:

[0057] in, , , , are weight coefficients. The clip function is used to limit the threshold to the interval [Tmin, Tmax] allowed by the policy to ensure compliance, T max is the maximum threshold allowed by the policy, corresponding to the upper limit constraint. min is the minimum threshold allowed by the policy, corresponding to the lower limit constraint.

[0058] In a feasible implementation manner, the above historical benchmark threshold information is determined based on a short-term threshold and a long-term threshold. is the historical benchmark threshold information, is a short-term threshold based on the quantile of the fused features over the past 24 hours Calculation, reflecting recent risk volatility:

[0059] in, Indicates the time range Data distribution of fused features. , refers to the 90% quantile, that is, During this period, 90% of the distribution values ​​fall below this value and 10% fall above this value. EMA() is an exponential smoothing average, which is used to further smooth short-term data.

[0060] is a long-term threshold based on data from the past 7 days. Quantile median, capturing long-term trends:

[0061] in, Indicates the time range Data distribution of fused features. , refers to the 95% quantile, that is, During this period, 95% of the distribution values ​​fall below this value and 5% fall above this value. is the median.

[0062] The environmental correction information is calculated by the following formula:

[0063]

[0064] Where M represents the dimension or quantity of environmental factors. Table 1 shows the actual measured value of the jth environmental indicator at time t. and Respectively represent the historical mean and standard deviation of the jth environmental factor. j Represents the weight of the jth environmental indicator at the current moment. is the feature related to the jth environmental factor, W is a learnable parameter, is an activation function.

[0065] Correct information for epidemic models, Group risk information.

[0066] Specifically, the above epidemic model correction information The specific steps to determine include: Obtain infection density information based on the extended SEIR model; Determine the initial information of the epidemic model based on the above infection density information, historical threshold information and epidemic severity adjustment parameters; The initial information of the above epidemic model is adaptively optimized by threshold to obtain the revised information of the epidemic model.

[0067] Exemplarily, the extended SEIR model (SEIR-Q model) is obtained based on the following formula:

[0068] in, (Susceptible) indicates the number of individuals that are still not infected; (Exposed) refers to the number of individuals who have been exposed to the virus but have not yet developed the disease; (Infected) indicates the number of individuals currently in the infected state; (Recovered) indicates the number of individuals who have recovered or been isolated. , , and They represent the change rates of susceptible persons, latent persons, infected persons and recovered persons respectively; (Infection rate) refers to the probability that an infected person contacts a susceptible person and infects them within a unit of time, which is usually related to the contact frequency and environmental factors (ventilation, humidity, etc.). ,in, Represents the impact of epidemic prevention measures, for example: no epidemic prevention measures (such as not wearing masks, ),at this time Maintain a high level of effective epidemic prevention measures (such as strict isolation, ),at this time Rapid decline. is the latent conversion rate, defined as: ,in is the average incubation period (e.g., the incubation period of COVID-19 is approximately 5.1 days); (Recovery Rate / Removal Rate): refers to the rate at which infected people recover or are isolated: ,in is the average infection period (e.g., the COVID-19 infection period is approximately 14 days); The rate at which asymptomatic people are detected and isolated; The rate at which symptomatic people are detected and isolated.

[0069] Infection density Based on the following formula:

[0070] Historical Thresholds It represents the health warning threshold set in the past and can be calculated based on the following formula:

[0071] in, Representing the past A threshold of days, used to smooth short-term fluctuations.

[0072] Initial information of epidemic model Calculated by the following formula:

[0073] in, is the epidemic weight parameter used to control the weight of historical data and current epidemic data. If the value is large (close to 1), the threshold changes rapidly and is sensitive to epidemic fluctuations. If it is smaller (closer to 0), the threshold adjustment is more stable and mainly depends on the historical trend.

[0074] in, Can automatically adjust based on past data, It can be calculated based on the following formula:

[0075] in: represents the instantaneous rate of change of infection density, is a minimum value (to prevent the denominator from being zero). The rapid change indicates that the epidemic has occurred suddenly and should be improved. To enhance sensitivity. The slower the change, the more stable the epidemic is. To reduce over-adjustment. Perform adaptive threshold optimization to obtain epidemic model correction information.

[0076] Specifically, the adaptive threshold optimization of the initial information of the epidemic model to obtain the revised information of the epidemic model includes: Environmental modeling was performed using the extended SEIR model as the environment and historical health monitoring data as the experience pool; Predict the adjustment amount at the current moment by measuring the network; Construct a reward function based on the false positive and false negative index penalty term, threshold stability term, epidemic spread impact term, and policy constraint term; The above reward function is optimized according to the proximal strategy to obtain the correction information of the epidemic model.

[0077] For example, when modeling the environment, the extended SEIR model is used as the environment to simulate different epidemic spread scenarios, and historical health monitoring data is used as the experience pool to provide supervision signals.

[0078] Predict the current moment through the policy network The optimal threshold adjustment :

[0079] in, is the threshold at the current moment, is the threshold for the next moment.

[0080] The entropy regularization mechanism is used to ensure exploration and prevent the model from falling into local optimality.

[0081] Construct a reward function based on the false positive and false negative index penalty term, threshold stability term, epidemic spread impact term, and policy constraint term :

[0082] in, and Exponential penalties are used to prevent misjudgments from seriously affecting decision-making. Time series smoothing constraints are used to suppress sharp fluctuations in thresholds in the short term. Combined with Bayesian uncertainty estimation, decisions are made more sensitive to epidemic spread trends. Ensure that threshold adjustments meet policy boundary conditions.

[0083] Specific:

[0084]

[0085]

[0086]

[0087]

[0088] The final reward function for:

[0089] in, is the false alarm rate, is the underreporting rate, and are the first weight coefficient and the second weight coefficient, which are used to adjust and the intensity of punishment, and The first exponential coefficient and the second exponential coefficient are used to control the exponential growth rate. is the third weight coefficient, is the time window length, For time Health monitoring thresholds at all times, For time Health monitoring thresholds at all times, is the fourth weight coefficient, is the regularization term, Indicates the uncertainty of epidemic prediction, represents the rate of change of the number of infected people over time, is the expected number of infected people, is the square of the deviation between the current number of infections and the expected value, The variance represents the fluctuation of the number of infections. is the standard deviation, which is used to measure the actual fluctuation range of the number of infections; is the fifth weight coefficient, and The maximum and minimum health monitoring thresholds specified by the policy. To ensure No more than To ensure No less than .

[0090] In a feasible implementation manner, the specific steps of determining the above group risk information include: Acquiring contact intensity information between travelers, wherein the contact intensity information is an adjacency matrix of the traveler contact network; Determine group risk diffusion distribution information based on the above contact intensity information; Determine skewness information and kurtosis information based on the above group risk diffusion distribution information; The group risk information is determined based on the skewness information and the kurtosis information.

[0091] For example, the contact intensity of each passenger reflects the degree of close contact with other passengers. The passenger contact network is defined as the adjacency matrix of the network. The contact weights between different passengers are recorded:

[0092] in, Indicates passengers and travelers The contact strength between , They are the location information of two passengers. Used to control the effect of spatial distance on contact strength. It is the time overlap between two passengers in the same area. is the total time within the evaluation time window.

[0093] Leveraging Traveler Contact Network Calculate the risk propagation of the group. We use a graph convolutional network (GCN) to propagate risk:

[0094] in, For the The feature matrix of the layer contains the health risk information of passengers. To normalize the adjacency matrix:

[0095] For the The training parameters of layer GCN, is the ReLU activation function, and D is the degree matrix.

[0096] Skewness Used to measure the asymmetry of risk distribution:

[0097] in, is the risk distribution of passengers after the final GCN propagation, is the mean, is the standard deviation and E is the expectation.

[0098] Kurtosis A measure of how sharp the distribution is, i.e. whether there are extremely high-risk individuals:

[0099] Adjust the population health risk information based on skewness and kurtosis:

[0100] in, Represents the skewness based on T skew ) and kurtosis (T kurt ) is the amount of adjustment to the population health risk threshold.

[0101] In summary, the method provided in this embodiment achieves a three-dimensional characterization of the health status of passengers by integrating multimodal information such as physiological characteristics, health declarations, travel trajectories, and environmental perception. It avoids the traditional method of relying on only a single data source (such as body temperature or declaration), thereby significantly improving the comprehensiveness and accuracy of health abnormality identification. This embodiment introduces passenger and environment nodes and environment and environment nodes, and uses a multi-head attention mechanism to dynamically calculate the weight of the edge, which helps to capture high-risk environmental areas and the degree of interaction between passengers and the environment, and accurately locate possible high-risk contacts or aggregations. Through historical benchmarks, environmental corrections, and extended SEIR The dynamic threshold is determined by model correction and group risk information, and reinforcement learning is used for adaptive optimization. Compared with the traditional fixed threshold, it can effectively reduce missed reports and false alarms, and automatically adjust the risk judgment standard as the epidemic situation or passenger group characteristics change. In the reward function, multiple factors such as misjudgment penalty, threshold stability, epidemic spread impact and policy constraints are added to ensure that the stability of threshold adjustment can be maintained while pursuing detection accuracy (reducing missed reports and false alarms). Combined with the Clip function, the threshold is limited to the compliance range to ensure compliance with actual deployment requirements such as public health policies, laws and regulations. At the same time, by introducing skewness and kurtosis to measure the group risk distribution, it is possible to identify whether there are extremely high-risk individuals or concentrated outbreak trends; so that customs and epidemic prevention departments can more quickly discover super spreaders or outbreak risks and take targeted measures in advance. From multi-source data fusion to dynamic threshold optimization, and then to group risk distribution analysis, this embodiment forms a high-precision, adaptive, and comprehensive health early warning system that takes into account actual policy requirements. This system can not only accurately identify and warn high-risk passengers, but also flexibly adjust strategies according to epidemic changes and passenger flow characteristics, effectively improving the level of epidemic prevention and control and public health safety in port venues.

[0102] In a feasible implementation manner, the above-mentioned warning information includes warning information of customs clearance personnel on duty and warning information of customs clearance management personnel. The above risk judgment is based on the above fusion feature data and dynamic thresholds to generate early warning information, including: Performing individual risk judgment based on the above fusion feature data and the above dynamic threshold information of each person to obtain individual risk information; Generate the above-mentioned warning information for customs clearance personnel based on the above-mentioned personal risk information of each person; The above-mentioned customs clearance management personnel warning information is generated based on the above-mentioned personal risk information of all persons.

[0103] For example, the fusion feature data With dynamic threshold By comparison, we can get the risk level of passengers:

[0104] After the above calculations, each person’s personal risk information can be obtained.

[0105] The customs clearance officers are mainly responsible for on-site customs clearance and rapid screening of passengers, and need to make timely decisions based on the health status of each passenger. For low-risk (Green) passengers, the duty officer system can display "The passenger's health status is normal, no additional inspection is required" and the passenger can pass directly. For medium-risk (Yellow) passengers, the system prompts "The passenger is suspected to be abnormal, please conduct a secondary verification or a rapid health test." Duty officers can perform temperature re-examinations and rapid nucleic acid tests. For high-risk (Red) passengers: The system prompts "The health risk is high, and it is recommended to be isolated or deeply checked immediately." PCR testing, medical isolation observation, and notification of health departments to intervene can be performed.

[0106] Customs management personnel (such as customs management departments and airport command centers) pay more attention to the overall health situation of passengers in order to make macro-strategy adjustments.

[0107] Collect all passengers’ personal risk information and final risk scores for all passengers And the corresponding risk level. Count the number of high-risk passengers and medium-risk travelers And other indicators.

[0108] Calculate overall health risk index :

[0109] in, Indicates The health risk index of each traveler, Represents the total number of passengers.

[0110] Reflects the average health risk level of all travelers.

[0111] Calculate the proportion of high-risk passengers :

[0112] in, Represents the number of high-risk travelers, Represents the total number of passengers.

[0113] Use time series models (such as LSTM / Transformer) to predict the proportion of high-risk passengers or the average risk level in the next few hours or days, helping managers deploy epidemic prevention and control measures in advance.

[0114] In summary, this embodiment obtains the health risk score of a single passenger based on the fusion features and dynamic thresholds, and the on-duty personnel can quickly identify and take corresponding disposal measures. By summarizing the risk information of all passengers, managers can control the risk situation as a whole and formulate macro decisions such as diversion strategies, isolation plans, and resource allocation. When high risks or group risks rise, the on-duty layer and the management layer respond in conjunction under the same system to form a coordinated prevention and control mechanism.

[0115] like Figure 2 As shown, the present application proposes a health monitoring, analysis and early warning system 10 for customs travelers, comprising: The data acquisition unit 101 is used to acquire biometric monitoring data, health declaration data, travel trajectory data and environmental perception data; A feature extraction unit 102 is used to extract features from the biometric monitoring data, the health declaration data, the travel trajectory data, and the environmental perception data to obtain a plurality of monomer feature vectors; A feature fusion unit 103 is used to fuse feature data of all the above-mentioned individual feature vectors to obtain fused feature data; The risk judgment unit 104 is used to perform risk judgment based on the above-mentioned fused feature data and dynamic thresholds to generate warning information.

[0116] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for health monitoring, analysis and early warning of customs travelers, characterized in that: include: Obtain biometric monitoring data, health declaration data, travel trajectory data, and environmental perception data; Performing feature extraction on the biometric monitoring data, the health declaration data, the travel trajectory data, and the environmental perception data to obtain a plurality of monomer feature vectors; Performing feature data fusion on all the monomer feature vectors to obtain fused feature data; Risk judgment is performed based on the fused feature data and the dynamic threshold to generate early warning information.

2. The method for health monitoring, analysis and early warning of customs travelers according to claim 1 is characterized in that: The monomer feature vector includes a physiological feature vector, a health declaration vector, a travel trajectory vector and an environmental perception vector. The extracting features of the biometric monitoring data, the health declaration data, the travel trajectory data, and the environmental perception data to obtain a plurality of monomer feature vectors includes: Performing a first feature extraction on the biometric monitoring data based on a 1D convolutional neural network to obtain the physiological feature vector, wherein the biometric monitoring data includes body temperature data and gait data; Performing a second feature extraction on the health declaration data based on a hybrid coding strategy to obtain the health declaration vector; Performing spatial feature extraction, temporal feature extraction and first feature fusion operations on the travel trajectory data to obtain the travel trajectory vector; Performing full connection layer feature extraction, ResNet feature extraction and second feature fusion operations on the environmental perception data to obtain the environmental perception vector.

3. The method for health monitoring, analysis and early warning of customs travelers according to claim 2 is characterized in that: The step of fusing feature data of all the monomer feature vectors to obtain fused feature data includes: Creating a passenger node according to the physiological feature vector, the health declaration vector, and the travel trajectory vector; Creating an environment node according to the environment perception vector; Create the passenger-environment side and the environment-environment side; Calculating the attention scores of the passenger and the environment and the environment and the environment respectively; Update the passenger node and the environment node based on a multi-head attention mechanism; The updated passenger node and the environment node are pooled to obtain the fused feature data.

4. The method for health monitoring, analysis and early warning of customs travelers according to claim 1 is characterized in that: The specific steps of determining the dynamic threshold include: Obtain historical baseline threshold information, environmental correction information, epidemic model correction information, and population risk information; The dynamic threshold is determined based on the historical benchmark threshold information, environmental correction information, epidemic model correction information and group risk information.

5. The method for health monitoring, analysis and early warning of customs travelers according to claim 4 is characterized in that: The historical reference threshold information is determined based on a short-term threshold and a long-term threshold.

6. The method for health monitoring, analysis and early warning of customs travelers according to claim 4 is characterized in that: The specific steps of determining the epidemic model correction information include: Obtain infection density information based on the extended SEIR model; Determining initial information of the epidemic model according to the infection density information, historical threshold information and epidemic severity adjustment parameters; The initial information of the epidemic model is adaptively optimized by threshold value to obtain the revised information of the epidemic model.

7. The method for health monitoring, analysis and early warning of customs travelers according to claim 6 is characterized in that: The adaptive threshold optimization of the initial information of the epidemic model to obtain the epidemic model correction information includes: Environmental modeling is performed using the extended SEIR model as the environment and historical health monitoring data as the experience pool; Predict the adjustment amount at the current moment by measuring the network; Construct a reward function based on the false positive and false negative index penalty term, threshold stability term, epidemic spread impact term, and policy constraint term; The reward function is optimized according to the proximal strategy to obtain the epidemic model correction information.

8. The method for health monitoring, analysis and early warning of customs travelers according to claim 4 is characterized in that: The specific steps of determining the group risk information include: Acquiring contact intensity information between travelers, wherein the contact intensity information is an adjacency matrix of the traveler contact network; Determining group risk diffusion distribution information based on the contact intensity information; Determining skewness information and kurtosis information based on the group risk diffusion distribution information; The group risk information is determined according to the skewness information and the kurtosis information.

9. The method for health monitoring, analysis and early warning of customs travelers according to claim 1 is characterized in that: The warning information includes the warning information of customs clearance personnel on duty and the warning information of customs clearance management personnel. The risk judgment based on the fused feature data and the dynamic threshold to generate warning information includes: Performing personal risk judgment according to the fused feature data and the dynamic threshold information of each person to obtain personal risk information; Generate the customs clearance personnel warning information according to the personal risk information of each person; The customs clearance management personnel warning information is generated based on the personal risk information of all persons.

10. A health monitoring, analysis and early warning system for customs travelers, characterized in that: include: A data acquisition unit, used to acquire biometric monitoring data, health declaration data, travel trajectory data, and environmental perception data; A feature extraction unit, used for extracting features from the biometric monitoring data, the health declaration data, the travel trajectory data and the environmental perception data to obtain a plurality of monomer feature vectors; A feature fusion unit, used for fusing feature data of all the monomer feature vectors to obtain fused feature data; The risk judgment unit is used to make risk judgment based on the fused feature data and the dynamic threshold to generate warning information.

Images