An artificial intelligence-based epidemic prevention monitoring joint defense system and method

Abstract

The application discloses a kind of epidemic prevention monitoring joint defense system and method based on artificial intelligence in medical health information technology field, the premise that the additional physical sensing equipment is not needed in the application, collection explicit basic data and implicit behavior association data, eliminate data heterogeneity and meet privacy compliance requirements by unified space-time mapping and anonymization processing;Parallel extraction first data based on state transition entropy and multi-modal abnormal score, and second data based on hidden Markov model, composite risk potential index is generated by using sliding time window and joint Bayesian estimation fusion, and risk determination benchmark is dynamically revised in combination with environmental fluctuation penalty factor;According to the fusion risk classification result, dynamic adjustment instruction is applied to multi-modal weight adaptive mechanism and hidden state inference early warning trigger logic, while having graceful degradation defense and structured configuration parameter external storage capability, realize low-cost, high-robust, generalizable closed-loop linkage epidemic prevention monitoring.

Description

Technical Field

[0001] This invention relates to the field of medical and health information technology, and in particular to an artificial intelligence-based method for joint prevention and control of the epidemic. Background Technology

[0002] Current joint prevention and control methods for epidemic prevention and control mainly rely on multi-dimensional perception dynamic adaptive networks, which integrate body temperature data, facial images and environmental information to achieve multi-task learning for identity recognition and abnormal body temperature detection.

[0003] The existing technology, with announcement number CN118299067B and titled "An Artificial Intelligence-Based Epidemic Prevention and Control Monitoring System and Method," captures the body temperature data and facial images of passing personnel, collects environmental data, and designs a multi-dimensional perception dynamic adaptive network to monitor individual body temperature and perform facial recognition, obtaining identity recognition results and abnormal body temperature detection results. Secondly, it introduces a risk assessment model, based on the identity recognition results and abnormal body temperature detection results, integrating the individual's historical health records and feature-engineered environmental data to assess the individual's health risk level, and notifying relevant personnel and managers of the individual's health risk level. This invention solves the technical problems of low efficiency and limited accuracy in identifying and recognizing individuals with abnormal body temperatures; the inability to accurately capture the comprehensive characteristics of an individual's health status, making it difficult to accurately assess the individual's health risk level; and insufficient information sharing and joint prevention and control mechanisms.

[0004] However, existing solutions suffer from significant bottlenecks in terms of performance and application scenarios. First, while the fusion of multimodal data improves recognition accuracy, it suffers from response latency issues, particularly in high-density crowd environments where real-time processing capabilities are insufficient, making it difficult to meet the demands of rapid and dynamic epidemic prevention and control monitoring. Second, environmental adaptability is limited—facial image recognition accuracy drops significantly in low light, occlusion, or extreme weather conditions, compromising the accuracy of risk assessment. Furthermore, the model relies on large-scale, high-quality, multi-source data, and insufficient handling of data heterogeneity and privacy protection limits its generalization ability across multiple scenarios and regions. Finally, the high cost of hardware and sensor integration in current systems hinders their widespread deployment in primary healthcare and public places.

[0005] In summary, these limitations collectively reveal significant technical gaps in AI-based epidemic prevention and control methods, particularly in terms of efficient real-time response, multi-environment stability, data fusion and generalization capabilities, and system cost control. There is an urgent need to achieve a balance between performance and application through more flexible data fusion strategies and lightweight model design. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides an artificial intelligence-based joint prevention and control method for epidemic prevention and monitoring. The objective of this invention is achieved as follows:

[0007] Without intrusion or the addition of additional physical sensing devices, multi-source basic monitoring data within the target epidemic prevention area is acquired in real time; the multi-source basic monitoring data is anonymized and spatiotemporally aligned to eliminate data heterogeneity and meet privacy protection compliance requirements.

[0008] Based on the processed multi-source basic monitoring data, multi-dimensional feature extraction is performed in parallel: the first data output by the dynamic risk discrimination model constructed based on the state transition entropy of complex systems and multimodal anomaly scoring is obtained; at the same time, the second data generated based on the latent state inference of social behavior time series is obtained.

[0009] By utilizing a pre-built multidimensional critical risk interaction mapping engine, the first data and the second data are subjected to joint Bayesian estimation fusion processing to eliminate the computational delay and single feature bottleneck in high-density crowd flow environment, and generate a composite risk potential index that characterizes the synergistic adaptive effect of multidimensional critical risk.

[0010] A risk assessment benchmark based on dynamic adjustment of environmental fluctuation factors is preset, and the composite risk potential index is compared and evaluated with the risk assessment benchmark to output the fusion risk classification result at the current moment.

[0011] Based on the fusion risk classification results and composite risk potential indicators, dynamic adjustment commands are applied to the multimodal weight adaptive mechanism of the dynamic risk discrimination model and the early warning triggering logic of the hidden state inference, thereby realizing forward-looking early warning and epidemic prevention scheduling for coordinated control.

[0012] Preferably, explicit basic data, including body temperature fluctuation curves, identity recognition confidence levels, and ambient temperature, humidity, and light intensity, are collected through infrared temperature measurement arrays, environmental sensing nodes, and identity verification terminals already deployed within the epidemic prevention area.

[0013] By extracting the intersection frequency of the desensitized regional personnel access control logs and movement trajectories, implicit behavioral correlation data representing group interactions are collected; global timestamps and spatial grid numbers are uniformly assigned to the explicit basic data and implicit behavioral correlation data, and heterogeneous data streams are mapped to a unified spatiotemporal coordinate system to form an anonymous multidimensional data set that satisfies the principle of minimum necessity.

[0014] Preferably, the process of acquiring the first data and the second data includes:

[0015] For the generation of the first data: extract the probability of state change of explicit basic data within a preset time window, and calculate the state transition entropy that represents the complexity of state switching; at the same time, extract the product of body temperature deviation and environmental mutation rate, and combine it with the decay of identity recognition confidence to calculate multimodal anomaly score.

[0016] The first data is composed of the state transition entropy and the multimodal anomaly score;

[0017] Regarding the generation of the second data: the implicit behavior association data is input into a pre-trained hidden Markov model, and the clustering frequency and crossover duration of personnel activity trajectories are used as the observation sequence. The probability of implicit transition of an individual's health risk status during the incubation period is calculated by a probability inference algorithm.

[0018] The second data is the output probability distribution of the hidden state of social behavior.

[0019] Preferably, the multidimensional critical risk interaction mapping engine uses a sliding time window technique to extract and synchronize the first data and the second data;

[0020] The calculation logic for generating the composite risk potential index using joint Bayesian estimation fusion processing is as follows: the hidden state probability distribution in the second data is used as the system prior probability, and the multimodal anomaly score in the first data is converted into the likelihood of the current observation.

[0021] The posterior probability is obtained by multiplying the prior probability by the likelihood and calculating the normalized ratio of the product over all possible states. The numerical result of the posterior probability is defined as the composite risk potential index, which objectively quantifies the overlapping risk of physical anomaly manifestation and behavioral latent accumulation.

[0022] Preferably, the environmental temperature, humidity, and light intensity are extracted from the explicit basic data, and an environmental fluctuation penalty factor is calculated; based on the environmental fluctuation penalty factor, the static risk baseline is dynamically offset and corrected to obtain the risk assessment benchmark; the specific adjustment logic is as follows:

[0023] When insufficient light or extreme weather causes an overall decrease in visual recognition confidence, the penalty factor increases, triggering the risk judgment benchmark to be lowered by a preset ratio, thereby reducing the reliance on environmentally sensitive data.

[0024] The absolute difference between the composite risk potential index and the risk judgment benchmark is compared. Based on the preset numerical range of the absolute difference, the fusion risk classification result is divided into normal state, latent concern state, and critical intervention state.

[0025] Preferably, the specific linkage control mechanism for applying the dynamic adjustment command includes:

[0026] When the system is identified as being in a state of latent concern, the first type of dynamic adjustment instruction is issued: increase the weight ratio of physiological feature signals such as body temperature fluctuations in the multimodal fusion of the dynamic risk discrimination model, and reduce the time step of feature extraction, so as to improve the system's sensitivity to early subtle physical anomalies.

[0027] Simultaneously, a second type of dynamic adjustment instruction is issued: the risk divergence coefficient in the state transition matrix of the hidden Markov model is increased, the critical threshold for triggering high-level early warnings in subsequent social behavior is lowered, and a strategy shift from ex-post detection to critical state early warning is achieved.

[0028] Preferably, it also includes a graceful degradation defense mechanism for high-density or harsh environment scenarios:

[0029] The engine monitors the temporal changes in the identity recognition confidence level in the first data in real time. If it detects that the confidence level is below the effective threshold for multiple consecutive time periods due to severe obstruction or high-density crowds, the engine determines that the physical perception channel is blocked.

[0030] At this point, the routine assessment in step four is intercepted, and the degradation mode is automatically triggered. The likelihood contribution of the first data in the Bayesian estimation is temporarily blocked, and the hidden state inference probability of the second data is directly mapped to the composite risk potential index. At the same time, a physical perception blind spot alarm command is sent to the remote command center to ensure that joint defense monitoring is not interrupted under extreme conditions.

[0031] Preferably, based on the linkage control mechanism, the epidemic prevention and control monitoring system further performs physical environment scheduling:

[0032] Based on the fusion risk classification results, for individuals or spatial grids determined to be in a critical intervention state, path redirection instructions are automatically sent to wearable devices, mobile quarantine terminals or electronic guidance screens within the area.

[0033] By dispersing the clustering of high-risk gathering events, the risk of gathering caused by high-density crowds can be mitigated from a physical space perspective, and medical response resources can be pre-allocated and configured before the target personnel arrive at the re-examination area.

[0034] Preferably, the Bayesian joint estimation parameters, risk assessment benchmark threshold, and weight adjustment step size are all predefined and extracted into structured configuration parameters;

[0035] The structured configuration parameters are stored independently in an external data carrier in the format of a spreadsheet or a comma-separated value file. In the initial stage of the method execution, the parameters in the external data carrier are read in batches into memory and directly assigned to the calculation logic through the data loading module.

[0036] This allows non-programmers to quickly calibrate and deploy systems based on the epidemic prevention requirements of different medical institutions or public places without having to reconstruct the underlying algorithm model.

[0037] An artificial intelligence-based epidemic prevention and control monitoring system includes:

[0038] The compliant data acquisition module is used to acquire multi-source basic monitoring data within the target epidemic prevention area in real time without intrusion or the addition of additional physical sensing devices; the multi-source basic monitoring data is anonymized and spatiotemporally aligned to eliminate data heterogeneity and meet privacy protection compliance requirements.

[0039] The multidimensional feature extraction module performs multidimensional feature extraction in parallel based on the processed multi-source basic monitoring data: it obtains the first data output by the dynamic risk discrimination model constructed based on the state transition entropy of complex systems and multimodal anomaly scores; at the same time, it obtains the second data generated based on the latent state inference of social behavior time series.

[0040] The risk interaction fusion module utilizes a pre-built multidimensional critical risk interaction mapping engine to perform joint Bayesian estimation fusion processing on the first and second data to eliminate computational delays and single-feature bottlenecks in high-density crowd environments and generate a composite risk potential index that characterizes the synergistic adaptive effect of multidimensional critical risks.

[0041] The dynamic risk assessment module presets a risk judgment benchmark based on dynamic adjustment of environmental fluctuation factors, compares and evaluates the composite risk potential index with the risk judgment benchmark, and outputs the fusion risk classification result at the current moment.

[0042] The linkage early warning and scheduling module, based on the fused risk classification results and composite risk potential indicators, applies dynamic adjustment commands to the multimodal weight adaptive mechanism of the dynamic risk discrimination model and the early warning triggering logic of the hidden state inference, thereby realizing forward-looking early warning and epidemic prevention scheduling of linkage control.

[0043] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0044] To address the real-time processing bottlenecks and insufficient environmental adaptability in high-density crowd environments, this solution utilizes existing facilities (such as infrared temperature measurement arrays, environmental sensing nodes, and identity verification terminals) to collect multi-source data without adding extra physical sensing equipment. This reduces hardware deployment costs and system complexity from the outset, overcoming the limitation of existing technologies being difficult to widely deploy in grassroots and public places due to high sensor integration costs. Simultaneously, by performing multi-dimensional feature extraction in parallel, a dynamic risk discrimination model is constructed based on state transition entropy and multimodal anomaly scores (including body temperature deviation, environmental mutation rate, and identity recognition confidence decay) to output the first data. Furthermore, a hidden state inference is performed based on a hidden Markov model to infer the frequency of clustering and intersection duration of personnel activity trajectories. The second data is then processed using the sliding time window technique and joint Bayesian estimation in the multidimensional critical risk interaction mapping engine (using the hidden state probability distribution of the second data as the prior probability, the multimodal anomaly score of the first data as the likelihood, and the posterior probability as a composite risk potential index). This effectively eliminates the computational delay and single feature bottleneck in high-density crowd environments. In addition, by introducing an environmental fluctuation penalty factor (calculated from environmental temperature, humidity and light intensity), the static risk baseline is dynamically offset and corrected. When insufficient light or extreme weather causes an overall decrease in visual recognition confidence, the risk judgment benchmark is automatically lowered, thereby avoiding excessive reliance on environmentally sensitive data and significantly improving the robustness and realism of the system's evaluation in complex scenarios such as occlusion and poor lighting.

[0045] To address the issues of data heterogeneity, privacy protection, and insufficient system generalization and closed-loop linkage capabilities, this solution uniformly assigns global timestamps and spatial grid numbers to explicit basic data such as body temperature fluctuation curves, identity recognition confidence levels, ambient temperature and humidity, and light intensity acquired by infrared thermometer arrays, as well as implicit behavioral correlation data such as access control logs and movement trajectory intersection frequencies. This data is mapped to a unified spatiotemporal coordinate system and anonymized according to the principle of minimum necessity, thus completely eliminating data heterogeneity and meeting privacy protection compliance requirements. Furthermore, based on the fusion of risk classification results (normal state, implicit attention state, and critical intervention state) and composite risk potential indicators, dynamic adjustments are made to the multimodal weight adaptive mechanism of the dynamic risk discrimination model (increasing the weight ratio of physiological features such as body temperature fluctuations and reducing the feature extraction time step in the implicit attention state) and the risk divergence coefficient and early warning trigger threshold of the hidden Markov model's state transition matrix. The adjustment commands enable a strategy shift from post-detection to critical state early warning. Simultaneously, the system possesses a graceful degradation defense mechanism: when the identity recognition confidence level falls below the effective threshold due to severe obstruction or high-density crowds over multiple consecutive time periods, it automatically masks the likelihood contribution of the first data and directly maps the hidden state inference probability of the second data to a composite risk potential index. It then sends a physical perception blind spot alarm command to the remote command center, ensuring uninterrupted joint monitoring under extreme conditions. Furthermore, structured configuration parameters such as Bayesian joint estimation parameters, risk judgment benchmark thresholds, and weight adjustment step sizes are extracted and stored independently in spreadsheets or comma-separated value files as external data carriers. These are then batch-read into memory and directly assigned values ​​via the data loading module. This allows non-programmers to quickly complete calibration and deployment based on the epidemic prevention requirements of different medical institutions or public places, greatly improving the system's generalization ability and ease of application across multiple scenarios and regions. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0047] Figure 1 This is a schematic diagram illustrating the application of the present invention;

[0048] Figure 2 This is a schematic diagram of the method flow of the present invention;

[0049] Figure 3 This is a schematic diagram of the process framework for steps three to five in this invention;

[0050] Figure 4 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Please see Figures 1 to 4 This invention provides an artificial intelligence-based method for joint prevention and control of epidemic prevention and monitoring, the specific steps of which include:

[0053] Step 1: Acquire multi-source basic monitoring data in the target epidemic prevention area in real time without intrusion or adding additional physical sensing devices; perform anonymization and spatiotemporal alignment processing on the multi-source basic monitoring data to eliminate data heterogeneity and meet privacy protection compliance requirements.

[0054] Step 2: Based on the processed multi-source basic monitoring data, perform multi-dimensional feature extraction in parallel: obtain the first data output by the dynamic risk discrimination model constructed based on the state transition entropy of complex systems and multimodal anomaly scoring; at the same time, obtain the second data generated based on the latent state inference of social behavior time series.

[0055] Step 3: Using a pre-built multidimensional critical risk interaction mapping engine, perform joint Bayesian estimation fusion processing on the first and second data to eliminate computational delay and single feature bottleneck in high-density crowd flow environment, and generate a composite risk potential index that characterizes the synergistic adaptive effect of multidimensional critical risk.

[0056] Step 4: Preset a risk assessment benchmark based on dynamic adjustment of environmental fluctuation factors, compare and evaluate the composite risk potential index with the risk assessment benchmark, and output the fusion risk classification result at the current moment;

[0057] Step 5: Based on the fusion risk classification results and composite risk potential indicators, apply dynamic adjustment commands to the multimodal weight adaptive mechanism of the dynamic risk discrimination model in Step 2 and the early warning triggering logic of the hidden state inference, thereby realizing forward-looking early warning and epidemic prevention scheduling of linkage control.

[0058] Step one specifically includes: collecting explicit basic data, including body temperature fluctuation curves, identity recognition confidence levels, and environmental temperature, humidity, and light intensity, through infrared temperature measurement arrays, environmental sensing nodes, and identity verification terminals already deployed within the epidemic prevention area; collecting implicit behavioral correlation data representing group interactions by extracting the intersection frequency of desensitized regional personnel access control logs and movement trajectories; uniformly assigning global timestamps and spatial grid numbers to the explicit basic data and implicit behavioral correlation data, mapping heterogeneous data streams to a unified spatiotemporal coordinate system, and forming an anonymous multidimensional data set that satisfies the principle of minimum necessity.

[0059] The specific logic for obtaining the first and second data in step two includes: For the generation of the first data: extracting the probability of state changes in explicit basic data within a preset time window, calculating the state transition entropy representing the complexity of state switching; simultaneously extracting the product of body temperature deviation and environmental mutation rate, and combining it with the decay of identity recognition confidence, calculating a multimodal anomaly score; the first data is thus composed of the state transition entropy and the multimodal anomaly score. For the generation of the second data: inputting the implicit behavioral association data into a pre-trained Hidden Markov Model, using the clustering frequency and crossover duration of personnel activity trajectories as the observation sequence, and calculating the implicit transition probability of an individual's health risk state during the incubation period using a probabilistic inference algorithm; the second data is the output probability distribution of implicit social behavioral states.

[0060] Step three specifically includes: the multidimensional critical risk interaction mapping engine uses a sliding time window technique to extract and synchronize the first and second data; the calculation logic for generating the composite risk potential index using joint Bayesian estimation fusion processing is as follows: the hidden state probability distribution in the second data is used as the system prior probability, and the multimodal anomaly score in the first data is converted into the likelihood of the current observation; the posterior probability is obtained by multiplying the prior probability and the likelihood, and calculating the normalized ratio of the product in all possible states; the numerical result of the posterior probability is defined as the composite risk potential index, which objectively quantifies the dual overlapping risk of physical anomaly manifestation and behavioral hidden state accumulation.

[0061] Step four specifically includes: extracting environmental temperature and humidity and light intensity from the explicit basic data, and calculating the environmental fluctuation penalty factor; dynamically shifting and correcting the static risk baseline based on the environmental fluctuation penalty factor to obtain the risk judgment benchmark; the adjustment logic is as follows: when insufficient light or extreme weather causes an overall decrease in visual recognition confidence, the penalty factor increases, triggering the risk judgment benchmark to be lowered by a preset ratio, thereby reducing the dependence on environmentally sensitive data; comparing the absolute difference between the composite risk potential index and the risk judgment benchmark, and dividing the fused risk classification result into normal state, latent concern state, and critical intervention state according to the preset numerical range of the difference.

[0062] The specific linkage control mechanism for applying dynamic adjustment instructions in step five includes: when a latent concern state is determined, a first type of dynamic adjustment instruction is issued: increasing the weight ratio of physiological feature signals such as body temperature fluctuations in the multimodal fusion of the dynamic risk discrimination model, and reducing the time step of feature extraction, so as to improve the system's sensitivity to early weak physical anomalies; at the same time, a second type of dynamic adjustment instruction is issued: increasing the risk divergence coefficient in the state transition matrix of the hidden Markov model, reducing the critical threshold for triggering high-level early warnings in subsequent social behavior time series, and realizing a strategy shift from post-event detection to critical state early warning.

[0063] Step five also includes a graceful degradation defense mechanism for high-density or harsh environmental scenarios: real-time monitoring of the temporal changes in the identity recognition confidence in the first data. If the confidence is detected to be below the effective threshold for multiple consecutive time periods due to severe obstruction or high-density crowds, the engine determines that the physical perception channel is blocked. At this time, the routine assessment in step four is intercepted, the system automatically triggers the degradation mode, temporarily blocks the likelihood contribution of the first data in Bayesian estimation, directly maps the hidden state inference probability of the second data to a composite risk potential index, and simultaneously sends a physical perception blind spot alarm command to the remote command center to ensure that joint defense monitoring is not interrupted under extreme conditions.

[0064] Based on the linkage control mechanism in step five, the epidemic prevention and monitoring system further performs physical environment scheduling: based on the fusion risk classification results, for individuals or spatial grids determined to be in a critical intervention state, path redirection instructions are automatically sent to wearable devices, mobile quarantine terminals or electronic guidance screens in the area; by dispersing the clustering of high-risk gathering events, the risk of gathering caused by high-density crowds is alleviated from the physical space level, and the pre-allocation and configuration of medical response resources are completed in advance before the target personnel arrive at the re-inspection area.

[0065] To address the issues of weak generalization capability across multiple scenarios and high deployment costs across regions, the Bayesian joint estimation parameters in step three, the risk assessment benchmark threshold in step four, and the weight adjustment step size in step five are all predefined and extracted as structured configuration parameters.

[0066] The structured configuration parameters are stored independently in an external data carrier in the format of a spreadsheet or a comma-separated value file. In the initial stage of the method execution, the parameters in the external data carrier are read in batches into memory and directly assigned to the calculation logic through the data loading module. This allows non-programmers to complete rapid calibration and deployment according to the epidemic prevention requirements of different medical institutions or public places without having to reconstruct the underlying algorithm model.

[0067] Furthermore, in this embodiment, explicit basic data, including body temperature fluctuation curves, identity recognition confidence levels, ambient temperature and humidity, and light intensity, are collected through infrared temperature measurement arrays, environmental sensing nodes, and identity verification terminals already deployed within the epidemic prevention area. Simultaneously, implicit behavioral correlation data representing group interactions are collected by extracting anonymized access control logs and movement trajectory intersection frequencies of personnel in the area. To ensure data spatiotemporal consistency and privacy protection, a unified global timestamp and spatial grid number are assigned to all explicit basic data and implicit behavioral correlation data, mapping heterogeneous data streams to the same spatiotemporal coordinate system. This forms an anonymous multidimensional data set that satisfies the principle of minimum necessary data, ensuring spatiotemporal alignment and privacy compliance among the data.

[0068] Within a preset time window, for the generation of the first data, the probability of state changes in explicit basic data is extracted, and the state transition entropy is calculated to characterize the complexity of state switching. Furthermore, by multiplying body temperature deviation by the environmental mutation rate and combining it with the decay of identity recognition confidence, a multimodal anomaly score is calculated. The first data is thus synthesized from the state transition entropy and the multimodal anomaly score. For the generation of the second data, latent behavioral correlation data is input into a pre-trained Hidden Markov Model. Using the clustering frequency and crossover duration of personnel activity trajectories as the observation sequence, a probabilistic inference algorithm is used to calculate the latent transition probability of health risk states within the individual's incubation period. The second data is the probability distribution of the latent state of this social behavior.

[0069] The multidimensional critical risk interaction mapping engine, based on the sliding time window technique, intercepts and synchronizes the first and second data generated in step two. The latent state probability distribution in the second data is used as the prior probability, and the multimodal anomaly scores in the first data are converted into the likelihood of the current observation. The product of these probabilities is calculated using a joint Bayesian estimation method and normalized to obtain the posterior probability. This posterior probability value is defined as a composite risk potential index, which objectively measures the combined effect of physical anomalies and behavioral latent risks, achieving seamless fusion of multi-source risk information.

[0070] Environmental temperature, humidity, and light intensity are extracted from explicit baseline data to calculate an environmental fluctuation penalty factor. This factor is used to dynamically correct the deviation of a preset static risk baseline, forming a risk assessment benchmark. When insufficient light or extreme weather causes an overall decrease in visual recognition confidence, the environmental fluctuation penalty factor increases, prompting the risk assessment benchmark to be lowered by a preset proportion, thereby reducing reliance on environmentally sensitive data. By calculating the absolute difference between the composite risk potential index and the risk assessment benchmark, and based on a preset numerical range, the fused risk classification results are divided into normal state, latent concern state, and critical intervention state, achieving dynamic and high-precision classification of risk levels.

[0071] Upon identifying a latent state of concern, the system automatically issues the first type of dynamic adjustment instruction, which increases the weight of physiological feature signals such as body temperature fluctuations in the multimodal fusion of the dynamic risk discrimination model and reduces the time step of feature extraction to improve the sensitivity to early subtle anomalies. Simultaneously, the system issues the second type of dynamic adjustment instruction, which increases the risk divergence coefficient in the state transition matrix of the hidden Markov model, lowers the critical threshold for triggering high-level early warnings based on the time sequence of social behavior, and realizes the transformation of the early warning strategy from post-event detection to critical state early warning.

[0072] In addition, for high-density or harsh environments, when the confidence level of identity recognition is found to be lower than the set effective threshold for multiple consecutive time periods, it is determined that the physical perception channel is blocked. The degradation mode is actively triggered, the likelihood contribution of the first data in the routine assessment in step four is blocked, and the probability mapping composite risk potential index is directly inferred from the hidden state of the second data. At the same time, a physical perception blind spot alarm instruction is sent to the remote command center to ensure the continuity and security of joint defense monitoring in extreme environments.

[0073] Based on the above dynamic linkage control, further physical environment scheduling is carried out. For individuals or spatial grids that are determined to be in a critical intervention state, path redirection instructions are automatically sent to wearable devices, mobile quarantine terminals and electronic guidance screens in the area. This effectively disperses high-risk gatherings, alleviates the risks caused by dense crowds, and at the same time completes the pre-configuration response of medical resources in advance, comprehensively improving the efficiency and effectiveness of joint epidemic prevention and control.

[0074] An artificial intelligence-based epidemic prevention and control monitoring system includes:

[0075] The compliant data acquisition module is used to acquire multi-source basic monitoring data within the target epidemic prevention area in real time without intrusion or the addition of additional physical sensing devices; the multi-source basic monitoring data is anonymized and spatiotemporally aligned to eliminate data heterogeneity and meet privacy protection compliance requirements.

[0076] The multidimensional feature extraction module performs multidimensional feature extraction in parallel based on the processed multi-source basic monitoring data: it obtains the first data output by the dynamic risk discrimination model constructed based on the state transition entropy of complex systems and multimodal anomaly scores; at the same time, it obtains the second data generated based on the latent state inference of social behavior time series.

[0077] The risk interaction fusion module utilizes a pre-built multidimensional critical risk interaction mapping engine to perform joint Bayesian estimation fusion processing on the first and second data to eliminate computational delays and single-feature bottlenecks in high-density crowd environments and generate a composite risk potential index that characterizes the synergistic adaptive effect of multidimensional critical risks.

[0078] The dynamic risk assessment module presets a risk judgment benchmark based on dynamic adjustment of environmental fluctuation factors, compares and evaluates the composite risk potential index with the risk judgment benchmark, and outputs the fusion risk classification result at the current moment.

[0079] The linkage early warning and scheduling module, based on the fused risk classification results and composite risk potential indicators, applies dynamic adjustment commands to the multimodal weight adaptive mechanism of the dynamic risk discrimination model and the early warning triggering logic of the hidden state inference, thereby realizing forward-looking early warning and epidemic prevention scheduling of linkage control.

[0080] The above description of the embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. It should be noted that those skilled in the art can make several improvements and modifications to the present invention without departing from the principles of the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims

1. A joint prevention and control method for epidemic prevention and monitoring based on artificial intelligence, characterized in that, The specific steps include: Without intrusion or the addition of additional physical sensing devices, multi-source basic monitoring data within the target epidemic prevention area is acquired in real time; the multi-source basic monitoring data is anonymized and spatiotemporally aligned to eliminate data heterogeneity and meet privacy protection compliance requirements. Based on the processed multi-source basic monitoring data, multi-dimensional feature extraction is performed in parallel: the first data output by the dynamic risk discrimination model constructed based on the state transition entropy of complex systems and multimodal anomaly scoring is obtained; at the same time, the second data generated based on the latent state inference of social behavior time series is obtained. By utilizing a pre-built multidimensional critical risk interaction mapping engine, the first data and the second data are subjected to joint Bayesian estimation fusion processing to eliminate the computational delay and single feature bottleneck in high-density crowd flow environment, and generate a composite risk potential index that characterizes the synergistic adaptive effect of multidimensional critical risk. A risk assessment benchmark based on dynamic adjustment of environmental fluctuation factors is preset, and the composite risk potential index is compared and evaluated with the risk assessment benchmark to output the fusion risk classification result at the current moment. Based on the fusion risk classification results and composite risk potential indicators, dynamic adjustment commands are applied to the multimodal weight adaptive mechanism of the dynamic risk discrimination model and the early warning triggering logic of the hidden state inference, thereby realizing forward-looking early warning and epidemic prevention scheduling for coordinated control.

2. The method for joint prevention and control of epidemic prevention and monitoring based on artificial intelligence according to claim 1, characterized in that: By using infrared temperature measurement arrays, environmental sensing nodes, and identity verification terminals already deployed within the epidemic prevention area, explicit basic data including body temperature fluctuation curves, identity recognition confidence levels, and environmental temperature, humidity, and light intensity are collected. By extracting the intersection frequency of the desensitized regional personnel access control logs and movement trajectories, implicit behavioral correlation data representing group interactions are collected; global timestamps and spatial grid numbers are uniformly assigned to the explicit basic data and implicit behavioral correlation data, and heterogeneous data streams are mapped to a unified spatiotemporal coordinate system to form an anonymous multidimensional data set that satisfies the principle of minimum necessity.

3. The joint prevention and control method for epidemic prevention and monitoring based on artificial intelligence according to claim 2, characterized in that: The process of acquiring the first data and the second data includes: For the generation of the first data: extract the probability of state change of explicit basic data within a preset time window, and calculate the state transition entropy that represents the complexity of state switching; at the same time, extract the product of body temperature deviation and environmental mutation rate, and combine it with the decay of identity recognition confidence to calculate multimodal anomaly score. The first data is composed of the state transition entropy and the multimodal anomaly score; Regarding the generation of the second data: the implicit behavior association data is input into a pre-trained hidden Markov model, and the clustering frequency and crossover duration of personnel activity trajectories are used as the observation sequence. The probability of implicit transition of an individual's health risk status during the incubation period is calculated by a probability inference algorithm. The second data is the output probability distribution of the hidden state of social behavior.

4. The joint prevention and control method for epidemic prevention and monitoring based on artificial intelligence according to claim 3, characterized in that: The multidimensional critical risk interaction mapping engine uses a sliding time window technique to extract and synchronize the first and second data. The calculation logic for generating the composite risk potential index using joint Bayesian estimation fusion processing is as follows: the hidden state probability distribution in the second data is used as the system prior probability, and the multimodal anomaly score in the first data is converted into the likelihood of the current observation. The posterior probability is obtained by multiplying the prior probability by the likelihood and calculating the normalized ratio of the product over all possible states. The numerical result of the posterior probability is defined as the composite risk potential index, which objectively quantifies the overlapping risk of physical anomaly manifestation and behavioral latent accumulation.

5. The joint prevention and control method for epidemic prevention and monitoring based on artificial intelligence according to claim 4, characterized in that: The environmental temperature, humidity, and light intensity are extracted from the explicit basic data, and the environmental fluctuation penalty factor is calculated. Based on the environmental fluctuation penalty factor, the static risk baseline is dynamically offset and corrected to obtain the risk judgment benchmark. The specific adjustment logic is as follows: When insufficient light or extreme weather causes an overall decrease in visual recognition confidence, the penalty factor increases, triggering the risk judgment benchmark to be lowered by a preset ratio, thereby reducing the reliance on environmentally sensitive data. The absolute difference between the composite risk potential index and the risk judgment benchmark is compared. Based on the preset numerical range of the absolute difference, the fusion risk classification result is divided into normal state, latent concern state, and critical intervention state.

6. The method for joint prevention and control of epidemic prevention and monitoring based on artificial intelligence according to claim 5, characterized in that: The specific linkage control mechanism for applying the dynamic adjustment command includes: When the system is identified as being in a state of latent concern, the first type of dynamic adjustment instruction is issued: increase the weight ratio of physiological feature signals such as body temperature fluctuations in the multimodal fusion of the dynamic risk discrimination model, and reduce the time step of feature extraction, so as to improve the system's sensitivity to early subtle physical anomalies. Simultaneously, a second type of dynamic adjustment instruction is issued: the risk divergence coefficient in the state transition matrix of the hidden Markov model is increased, the critical threshold for triggering high-level early warnings in subsequent social behavior is lowered, and a strategy shift from ex-post detection to critical state early warning is achieved.

7. The method for joint prevention and control of epidemic prevention and monitoring based on artificial intelligence according to claim 6, characterized in that: It also includes graceful degradation defense mechanisms for systems in high-density or harsh environment scenarios: The engine monitors the temporal changes in the identity recognition confidence level in the first data in real time. If it detects that the confidence level is below the effective threshold for multiple consecutive time periods due to severe obstruction or high-density crowds, the engine determines that the physical perception channel is blocked. At this point, the routine assessment in step four is intercepted, and the degradation mode is automatically triggered. The likelihood contribution of the first data in the Bayesian estimation is temporarily blocked, and the hidden state inference probability of the second data is directly mapped to the composite risk potential index. At the same time, a physical perception blind spot alarm command is sent to the remote command center to ensure that joint defense monitoring is not interrupted under extreme conditions.

8. The joint prevention and control method for epidemic prevention and monitoring based on artificial intelligence according to claim 7, characterized in that: Based on the linkage control mechanism, the epidemic prevention and control monitoring system further performs physical environment scheduling: Based on the fusion risk classification results, for individuals or spatial grids determined to be in a critical intervention state, path redirection instructions are automatically sent to wearable devices, mobile quarantine terminals or electronic guidance screens within the area. By dispersing the clustering of high-risk gathering events, the risk of gathering caused by high-density crowds can be mitigated from a physical space perspective, and medical response resources can be pre-allocated and configured before the target personnel arrive at the re-examination area.

9. The joint prevention and control method for epidemic prevention and monitoring based on artificial intelligence according to claim 8, characterized in that: The Bayesian joint estimation parameters, risk assessment benchmark threshold, and weight adjustment step size are all predefined and extracted into structured configuration parameters. The structured configuration parameters are stored independently in an external data carrier in the format of a spreadsheet or a comma-separated value file. In the initial stage of the method execution, the parameters in the external data carrier are read in batches into memory and directly assigned to the calculation logic through the data loading module. This allows non-programmers to quickly calibrate and deploy systems based on the epidemic prevention requirements of different medical institutions or public places without having to reconstruct the underlying algorithm model.

10. An artificial intelligence-based epidemic prevention and control monitoring system, characterized in that: The system is used to execute the artificial intelligence-based joint prevention and control method for epidemic prevention and monitoring as described in any one of claims 1-9, comprising: The compliant data acquisition module is used to acquire multi-source basic monitoring data within the target epidemic prevention area in real time without intrusion or the addition of additional physical sensing devices; the multi-source basic monitoring data is anonymized and spatiotemporally aligned to eliminate data heterogeneity and meet privacy protection compliance requirements. The multidimensional feature extraction module performs multidimensional feature extraction in parallel based on the processed multi-source basic monitoring data: it obtains the first data output by the dynamic risk discrimination model constructed based on the state transition entropy of complex systems and multimodal anomaly scores; at the same time, it obtains the second data generated based on the latent state inference of social behavior time series. The risk interaction fusion module utilizes a pre-built multidimensional critical risk interaction mapping engine to perform joint Bayesian estimation fusion processing on the first and second data to eliminate computational delays and single-feature bottlenecks in high-density crowd environments and generate a composite risk potential index that characterizes the synergistic adaptive effect of multidimensional critical risks. The dynamic risk assessment module presets a risk judgment benchmark based on dynamic adjustment of environmental fluctuation factors, compares and evaluates the composite risk potential index with the risk judgment benchmark, and outputs the fusion risk classification result at the current moment. The linkage early warning and scheduling module, based on the fused risk classification results and composite risk potential indicators, applies dynamic adjustment commands to the multimodal weight adaptive mechanism of the dynamic risk discrimination model and the early warning triggering logic of the hidden state inference, thereby realizing forward-looking early warning and epidemic prevention scheduling of linkage control.

Images