A multi-stage state confirmation and control method and system for wearable device anomaly triggering

Abstract

This application provides a multi-stage state confirmation and control method and system for abnormal triggering of wearable devices, relating to the field of wearable health monitoring and abnormal event response control technology. This method is applied to key health monitoring scenarios on campuses. The method includes: collecting photoplethysmography (PPG) signals, acceleration signals, blood oxygen signals, and location information of target students; collecting and detecting user physiological data based on individual physiological baselines; entering an alert state when abnormal candidate conditions are met; calculating motion artifact scores and spatial-behavioral deviation scores; calling a risk pattern classifier to output abnormal probability values, risk pattern labels, and label confidence levels; and generating disease-specific treatment markers based on basic medical history. The disease-specific treatment markers are used to execute disease-specific score correction, determine the priority of fixed review units, filter contraindicated treatment actions, and determine event escalation rules. A final weighted total score is obtained based on the basic weighted score and the disease-specific score, and state transition and fixed review are performed based on the final weighted total score. The technical solution provided by this application can reduce the false alarm rate of abnormal monitoring in campus scenarios, improve the accuracy of confirming medical history-related abnormal events, the efficiency of fixed review, and the reliability of high-risk event response.

Description

Technical Field

[0001] This application relates to the field of wearable health monitoring and abnormal event response control technology, and in particular to a multi-stage state confirmation and control method and system for abnormal triggering of wearable devices. Background Technology

[0002] Currently, with the development of wearable devices, edge computing technology, and campus information infrastructure, solutions for daily health monitoring of students based on smart bracelets, wristbands, and other terminals are gradually being applied to campus safety management and student health protection scenarios. In existing technologies, wearable data collection units typically collect physiological or behavioral data such as heart rate, blood oxygen, and activity levels, and send this data to a backend system. The backend system then identifies abnormal states based on fixed thresholds, unified rules, or simplified models, and sends alerts to teachers, school doctors, or parents when indicators exceed limits.

[0003] In existing technologies, some solutions also combine location information, historical records, or activity recognition results to assist in the judgment of abnormal alarms, thereby improving the anomaly monitoring capability to a certain extent. For students with known underlying medical history or those requiring intensive health monitoring, existing technologies typically still adopt the same review path and handling logic as for ordinary students, lacking anomaly confirmation mechanisms based on medical history differences, fixed review unit selection mechanisms, and handling contraindication control mechanisms. This can lead to high-risk events related to medical history not being prioritized in a timely manner, or the handling suggestions pushed on-site not matching the medical history type, resulting in decreased review efficiency, unstable response links, and even increased risk of on-site misoperation, thus affecting the user experience and safety assurance effectiveness in key health monitoring scenarios on campus. Summary of the Invention

[0004] This application provides a multi-stage state confirmation and control method for abnormal triggering of wearable devices, which addresses the problem in existing campus wearable health monitoring solutions that struggle to confirm abnormal candidate events with low false alarms and trigger adaptive verification and response control based on individual physiological baselines, multi-source verification, and medical history strategies.

[0005] The technical solution includes collecting user data, such as physiological state and spatial location data, through the user's personal device.

[0006] Anomaly candidate detection is performed based on the features of the physiological state and spatial location data and the individual physiological state baseline.

[0007] For abnormal candidates, calculate motion artifact scores and spatial-behavior deviation scores. When the motion artifact score is higher than a preset threshold and the spatial-behavior deviation score is lower than a preset support threshold, the event is rolled back to the normal state.

[0008] The non-rollback events are classified using a pattern classifier, and anomaly probability values ​​and risk pattern labels are output.

[0009] For non-rollback events, when the user's basic medical history contains a medical history type corresponding to the risk pattern label, a disease-specific treatment label is generated, including disease correction parameters, fixed review unit priority parameters, candidate treatment action priority parameters, prohibited action set, and escalation threshold fine-tuning parameters.

[0010] For events that do not regress, a basic weighted score is calculated based on geographical location score, activity status score, 24-hour historical record score, group deviation score, and normalized spatial-behavioral deviation score. The final weighted total score is obtained by combining the disease-specific correction score and the environmental interference reduction score.

[0011] The weighted total score is compared with the first total score threshold and the second total score threshold to determine whether the event belongs to one of the following states: normal state, manual confirmation state, or confirmed abnormal state. Based on the disease-specific treatment mark, fixed review unit scheduling and / or prohibited action filtering and / or escalation rules are executed.

[0012] In the above or some embodiments, the individual physiological baseline includes at least one of the following: resting heart rate baseline, low-activity heart rate baseline, moderate-activity heart rate baseline, resting blood oxygen baseline, heart rate variability baseline, and recovery time baseline; wherein, the edge server unit generates one or more of the following thresholds based on the individual physiological baseline: upper limit threshold for heart rate, abnormal blood oxygen threshold, abnormal heart rate variability threshold, and recovery timeout threshold for the user's current scenario; the abnormal candidate conditions include at least one of the following: abnormal heart rate, abnormal heart rate variability, abnormal heart rate, abnormal blood oxygen, or hardware-level emergency sentinel triggering.

[0013] In the above or some embodiments, the calculation of motion artifact scores and spatial-behavior deviation scores for events in the initial alert state includes:

[0014] The motion artifact score is calculated based on at least two of the following indicators: acceleration amplitude integral, photoplethysmography pulse wave signal quality index, pulse cycle stability, and detectability.

[0015] Calculate the space-behavior deviation score based on the combination relationship between the target student's space category and current behavior category;

[0016] When the motion artifact score is higher than a preset threshold and the space-behavior deviation score does not support anomalies, and there are no abnormal blood oxygen levels or hardware-level emergency sentinel support conditions, the event will be rolled back to the normal state.

[0017] In the above or some embodiments, the risk pattern classifier is deployed on the edge server unit side, and the input features of the risk pattern classifier include at least a portion of the following: current window average heart rate, heart rate change rate, heart rate variability index, interpeak variability, pulse wave quality index, current blood oxygen value, blood oxygen change trend, activity intensity, activity category, spatial category, historical abnormality markers, and individual physiological state baseline deviation features.

[0018] The risk pattern labels include at least one of the following: rhythm abnormality risk pattern, cardiac load abnormality risk pattern, respiratory abnormality risk pattern, nonspecific abnormality risk pattern, and interference / low quality pending review pattern.

[0019] In the above or some embodiments, the disease-specific treatment markers include: event identifier, student identifier, medical history type, candidate risk level, fixed review unit priority list, candidate treatment action set, contraindicated action set, scoring correction rule identifier, escalation rule identifier, and effective duration;

[0020] The basic medical history includes at least one of rhythm-related medical history, cardiac load-related medical history, and respiratory-related medical history;

[0021] When a target student has multiple underlying medical histories, the edge server unit determines the primary medical history strategy based on the mapping relationship between the current risk pattern label and the medical history, and uses the non-primary medical history strategy as an additional constraint for prohibited action control or supplementary prompt information.

[0022] In the above or some embodiments, the fixed review unit reviews the event; the fixed review units are sorted according to medical history matching priority, online status, target student distance, device occupancy status, most recent self-inspection results, most recent data quality score and expected sampling delay; scheduling requests are sent to candidate fixed review units in sequence according to the sorting results; when the current candidate fixed review unit times out in any stage of the order acceptance stage, connection stage or sampling completion stage, the process is switched to the next candidate fixed review unit; when the scheduling of a preset number of candidate fixed review units fails, the process is switched to manual confirmation priority process or the continuous sampling of wearable information collection unit is extended as an alternative review method.

[0023] In the above or some embodiments, the calculation of the basic weighted score includes: summing the geographic location score, activity status score, 24-hour historical record score, group deviation score, and normalized spatial-behavioral deviation score according to preset weights to obtain the basic weighted score; wherein, the group deviation score includes: calculating the standardized heart rate deviation value of the current student, selecting other students located in the same spatial grid with the same or adjacent activity categories, effectively wearing masks, not currently entering a high-risk abnormal process, and whose data quality meets the standards as a reference group, using the median of the standardized heart rate deviation values ​​of the reference group as the center value, and determining the group deviation score based on the interval in which the difference between the current student and the center value falls; when the number of people in the reference group is less than a preset number, the group deviation score is set to a preset neutral score, and the median of the historical standardized heart rate deviation value of individuals in the same scene is used to form an auxiliary reference deviation.

[0024] In the above or some embodiments, the calculation of the basic weighted score includes: summing the geographic location score, activity status score, 24-hour historical record score, group deviation score, and normalized spatial-behavioral deviation score according to preset weights to obtain the basic weighted score; wherein, the group deviation score includes: calculating the standardized heart rate deviation value of the current student, selecting other students located in the same spatial grid with the same or adjacent activity categories, effectively wearing masks, not currently entering a high-risk abnormal process, and whose data quality meets the standards as a reference group, using the median of the standardized heart rate deviation values ​​of the reference group as the center value, and determining the group deviation score based on the interval in which the difference between the current student and the center value falls; when the number of people in the reference group is less than a preset number, the group deviation score is set to a preset neutral score, and the median of the historical standardized heart rate deviation value of individuals in the same scene is used to form an auxiliary reference deviation.

[0025] This invention also discloses a multi-stage confirmation and control system for abnormal events, including a wearable information acquisition unit for collecting students' physiological and physical data; an edge server unit for executing the method described; a fixed verification unit for secondary verification of abnormal events; an environmental sensor for providing air quality, temperature, humidity, or noise environmental parameters; and a management terminal for recording structured feedback results and viewing the event handling status; wherein the edge server unit is communicatively connected to the wearable information acquisition unit, the fixed verification unit, the environmental sensor, and the management terminal.

[0026] In the above or some embodiments, the edge server unit includes an event processing engine, a disease-specific treatment strategy library, and a feedback update engine; the disease-specific treatment strategy library is used to store disease correction parameters, fixed review unit priority parameters, candidate treatment action priority parameters, prohibited action sets, and escalation threshold fine-tuning parameters according to medical history type and risk mode; the feedback update engine is used to perform limited updates on the corresponding strategy parameter groups in the disease-specific treatment strategy library based on the feedback results after the event ends.

[0027] The technical solutions provided in this application, based on individual physiological baselines, motion artifact exclusion, spatial context verification, risk pattern classification, and disease-specific treatment marker-driven disease correction scoring, fixed review unit selection, forbidden action filtering, and escalation rule control, can achieve progressive confirmation of abnormal events, reducing false alarm rates caused by motion, environment, and group disturbances; and can improve the accuracy of confirming medical history-related abnormal events, the efficiency of fixed review, and the reliability of handling high-risk events. It is understood that the beneficial effects of multi-link redundant communication, long-distance low-power emergency communication, access control linkage, environmental device linkage, path planning, and feedback updates can be found in the relevant descriptions of the specific implementation methods in this application, and will not be repeated here. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the campus student anomaly monitoring architecture according to an embodiment of the present invention.

[0029] Figure 2 This is a flowchart illustrating the overall process of anomaly monitoring in an embodiment of the present invention.

[0030] Figure 3 This is a schematic diagram of a multi-stage state confirmation and control system for wearable devices triggered by abnormalities, according to an embodiment of the present invention. Detailed Implementation

[0031] It should be understood that in the description of the embodiments of this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. In this embodiment, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, features defined with "first" and "second" can explicitly or implicitly include one or more of that feature. In the description of this embodiment, unless otherwise stated, "multiple" means two or more.

[0032] The technical solutions provided in the embodiments of this application will be described below with reference to specific implementation methods.

[0033] An embodiment of a multi-stage state confirmation and control system for wearable devices triggered by abnormalities:

[0034] In the embodiments of this application, at least a wearable information collection unit, an edge server unit, a fixed verification unit, a monitoring terminal, and an environmental linkage unit are included.

[0035] The wearable information acquisition unit is preferably a smart bracelet, used to acquire photoplethysmography (PPG) signals, triaxial acceleration signals, optional blood oxygen signals, and location information, and supports at least one communication method among Bluetooth, cellular networks, and wireless networks. In some embodiments, the smart bracelet also has a built-in long-range low-power communication module, including but not limited to LoRa communication modules, NB-IoT communication modules, and low-power cellular modules in LTE-M / Cat.1 / Cat.1 bis.

[0036] The edge server unit is used to receive data sent by the smart bracelet and perform individual physiological state baseline management, step execution control, risk pattern classification, disease-specific treatment strategy generation, fixed review unit scheduling, hierarchical response control, linkage control, and feedback updates.

[0037] The fixed verification unit includes at least one of the following: ECG verification terminal, blood oxygen verification terminal, vital sign detection terminal, respiratory monitoring equipment, and environmental detection equipment, and is used to verify abnormal candidate events under the scheduling of the edge server unit.

[0038] The monitoring terminal includes at least one of the following: school doctor terminal, teacher terminal, homeroom teacher terminal, parent terminal, security guard terminal, and management terminal, used to receive alarms, perform manual confirmation, and view information within the scope of permissions.

[0039] The environmental linkage unit includes at least one of the following: access control equipment, fresh air equipment, air purification equipment, local exhaust ventilation equipment, campus broadcasting equipment, and path planning infrastructure. The access control equipment can be used for access control and area isolation; the fresh air equipment can be used to increase the fresh air volume in the target space; the air purification equipment can be used to improve the air quality in the target space; the local exhaust ventilation equipment can be used to directionally exhaust abnormal air in a localized area; and the campus broadcasting equipment can be used to issue review guidance, handling prompts, or regional safety alerts.

[0040] In the above or some embodiments, the establishment of individual physiological state baselines and threshold generation are as follows: The edge server unit establishes an individual physiological state baseline database for each student. This database includes at least: resting heart rate baseline, low-activity heart rate baseline, moderate-activity heart rate baseline, resting blood oxygen baseline, heart rate variability baseline, recovery time baseline, behavioral reference distributions under different spatial categories, medical history tags, and historical data quality indicators. The different spatial categories include classrooms, playgrounds, dormitories, canteens, medical rooms, indoor corridors, etc. For each spatial category, the edge server unit statistically analyzes one or more of the following distributions for the target student: activity level distribution, dwell time distribution, heart rate interval distribution, and day / night time distribution, forming a behavioral reference distribution. The historical data quality indicators include at least one of the following: signal loss rate, low-quality signal ratio, wearing interruption frequency, motion artifact triggering frequency, and transmission packet loss rate, used to evaluate the reliability of data within the corresponding time window and participate in subsequent threshold correction or alarm confidence calculation.

[0041] In some embodiments, the first seven days for new users serve as the initial learning period. The edge server unit selects data in a non-alarm state, with effective wear, non-strenuous exercise, and a daily effective duration of not less than a preset duration to establish an initial baseline. If the initial learning period data is insufficient, an age-group default baseline is used as an alternative, and this default baseline is gradually replaced as sufficient individual data is obtained subsequently. Effective wear refers to a data state where the smart bracelet maintains continuous contact with the wearer's skin, and the photoplethysmography (PPG) signal quality index, contact state signal, and motion artifact index meet preset conditions. Non-strenuous exercise refers to activities classified as stationary, walking, low-intensity activity, or low-to-medium intensity activity based on triaxial acceleration, cadence, and activity intensity classification results, excluding running, jumping, ball games, and short-duration strenuous sprinting. A non-alarm state refers to a data state where abnormal events within the corresponding time window have not entered a manual confirmation state or an abnormality confirmation state, or where the final weighted total score has not reached a first total score threshold. The preset duration can be set to any value among 6 hours, 8 hours, or 10 hours of cumulative effective data per day.

[0042] The edge server unit extracts multiple time windows that meet the resting scenario conditions during the initial learning period and calculates the average heart rate value for each time window. Outliers above the upper quartile plus a preset multiple of the interquartile range and below the lower quartile minus a preset multiple of the interquartile range are removed. The median or weighted average of the remaining samples is used as the resting heart rate baseline. Based on the activity recognition results, the data is divided into low-activity and medium-activity scenarios. The heart rate sample distribution under each scenario is statistically analyzed, and the median, quartile interval, or weighted average of the heart rate samples for each scenario is used as the heart rate baseline for the corresponding activity level. Blood oxygen data that meets the requirements of effective wearing, resting state, and data quality above a threshold are statistically analyzed. After removing data points that significantly deviate from the physiologically reasonable range for short periods, the median or mode of the high-frequency stable interval is used as the resting blood oxygen baseline. Based on continuous pulse interval sequences during nighttime resting or low-activity periods, at least one of the following heart rate variability indicators—RMSSD, SDNN, and pNN50—is calculated, and the results over multiple days are statistically analyzed to obtain the heart rate variability baseline. When a recovery segment is identified after a moderate to high activity, the time required for the heart rate to drop from the activity peak to the target recovery range is calculated, and the median or quantile value of multiple recovery times is taken as the baseline recovery time.

[0043] The edge server unit generates individualized thresholds based on the individual's physiological baseline. For example, the resting heart rate abnormality thresholds include a high heart rate upper limit threshold and a low heart rate lower limit threshold; wherein the high heart rate upper limit threshold can be obtained by multiplying the resting heart rate baseline by a first proportional coefficient, and the low heart rate lower limit threshold can be obtained by multiplying the resting heart rate baseline by a second proportional coefficient; when the current heart rate continuously exceeds the high heart rate upper limit threshold or continuously falls below the low heart rate lower limit threshold for a preset duration, it is determined as a candidate for a resting heart rate abnormality; the blood oxygen abnormality threshold simultaneously considers the individual's relative decrease and absolute safety lower limit; when the current blood oxygen value decreases more than a preset decrease value relative to the resting blood oxygen baseline, and the current blood oxygen value... When the oxygen level is below the absolute safety threshold or continuously below the relative threshold for a preset duration, it is identified as a candidate for abnormal blood oxygenation. The abnormal heart rate variability threshold is determined based on the ratio between the current RMSSD value and the baseline RMSSD value. When the ratio is lower than the third proportional coefficient or higher than the fourth proportional coefficient, it is identified as a candidate for abnormal autonomic nervous system regulation. The recovery timeout threshold is used to characterize the heart rate recovery ability after the activity. When the time it takes for the current heart rate to fall back to within the baseline heart rate plus the preset recovery offset after the activity exceeds the recovery time baseline multiplied by the fifth proportional coefficient, it is identified as a candidate for abnormal recovery. Each of the proportional coefficients can be a fixed empirical coefficient or a coefficient adaptively adjusted based on the target student's historical stability index.

[0044] In some embodiments, after the initial baseline is established, the edge server unit dynamically updates the individual physiological state baseline database using a sliding time window update mechanism. During the update, only data that meets the criteria of being in a non-alarm state, being effectively worn, having data quality above a threshold, and not occurring during periods marked by acute illness or special events is included. For newly added data, exponentially weighted moving average, sliding median update, or quantile update methods can be used to iteratively correct each baseline. To avoid abnormal data contaminating the baseline, if newly added data continuously deviates from the current baseline by more than a preset deviation threshold, the integration of that data into the baseline update is paused, pending manual confirmation or reassessment with subsequent stable data.

[0045] In this embodiment, the system adopts a two-layer control architecture that separates the state machine layer and the response level layer.

[0046] The state machine layer describes the abnormal event confirmation process, including normal state, initial alert state, verification state, pattern analysis state, manual confirmation state, confirmed abnormal state, and event termination state. The response level layer describes the handling intensity, including no response, Level 1 response, Level 2 response, and Level 3 response.

[0047] In this embodiment, the state machine layer is responsible for determining whether an event is a genuine anomaly or a highly reliable anomaly, while the response level layer is responsible for determining the notification and linkage strength. Typically, events transition in the following order: normal state, initial alert state, verification state, pattern analysis state, manual confirmation state, confirmed anomaly state, and event termination state. When high-risk conditions are met, the event can directly transition from the pattern analysis state or manual confirmation state to the confirmed anomaly state, and simultaneously enter the third-level response phase.

[0048] A multi-stage state confirmation and control method for abnormal triggering of wearable devices, also known as a state machine layer control embodiment:

[0049] Step S101: User physiological parameters, status, and location data acquisition and detection. In this embodiment, the smart bracelet collects data at a preset sampling frequency and uploads or synchronizes the data to the edge server unit with a preset basic analysis window and sliding step size. The edge server unit first performs validity detection on the data in the current window. If the conditions such as valid wearing, effective pulse wave sampling ratio not less than a preset ratio, acceleration data continuity not less than a preset ratio, effective blood oxygen reading meeting requirements, and location confidence level meeting the standard are met, the data in the window is marked as valid; otherwise, the system enters a degraded mode for unknown location or insufficient data, only recording the status or prompting to wear it again, without triggering anomaly judgment.

[0050] The edge server unit uses the analysis window corresponding to the current event as the baseline time window to perform time alignment on heart rate, blood oxygen, activity, spatial, and historical marker-related data. For continuous features, they are preferably converted into comparable statistics, variability, or deviations from the individual's physiological baseline. For categorical features, they are preferably converted into numerical labels, one-hot encoded vectors, or ordinal encoded values. For Boolean markers, they are preferably converted into 0 / 1 values. For missing features, they are preferably represented by default padding values, the most recent valid value, mean padding, or a combination of missing indicator bits. In this embodiment, the edge server unit filters, detects peaks, removes abnormal intervals, and smooths the median of the pulse wave signal to obtain the current heart rate estimate; and calculates the current heart rate change rate, the current heart rate variability index, and the current blood oxygen index.

[0051] The event transitions from the normal state to the initial alert state when any of the following conditions are met: the current heart rate estimate is higher than the current scene's upper limit heart rate threshold; the current heart rate variability is higher than the preset variability threshold; the current heart rate variability index is lower than the heart rate variability baseline multiplied by a preset proportional coefficient; the current blood oxygen value meets the blood oxygen abnormality candidate conditions; or hardware-level emergency triggering can be implemented, for example, by setting an emergency button.

[0052] In some embodiments, when the pulse wave quality is insufficient but the blood oxygen pulse rate is available and the acceleration is stable, the blood oxygen pulse rate estimate or the most recent effective heart rate extrapolation value is allowed as a substitute input; if an effective heart rate still cannot be obtained, the current window will not enter the user physiological data acquisition and detection process.

[0053] Step S102: Motion artifact and spatial verification. In this embodiment, events in the initial alert state enter the verification state, and the edge server unit calculates the motion artifact score and the spatial-behavior deviation score.

[0054] The motion artifact score is obtained by weighting the integral of acceleration amplitude, pulse wave signal quality index, pulse cycle stability, and detectability rate. The edge server unit normalizes the above indicators and calculates the motion artifact score according to preset weights. When the motion artifact score is higher than a first threshold, it is judged as a significant motion artifact; when the motion artifact score is between the first and second thresholds, it is judged as a suspected motion artifact and a low-confidence label is added; when the motion artifact score is lower than the second threshold, it is judged as a non-major artifact.

[0055] In some embodiments, the motion artifact scoring Calculate according to the following steps:

[0056] First, calculate the acceleration amplitude sequence within the current analysis window. :

[0057] in, , , These represent the triaxial accelerations at the 1st... The sampled values ​​at each sampling point. To reduce the influence of the gravity component, the triaxial acceleration can be high-pass filtered first, or the calculation can be performed without the gravity acceleration component.

[0058] Furthermore, the integral index of acceleration amplitude is calculated. :

[0059] in, This indicates the number of acceleration sampling points within the current window. This represents the average acceleration amplitude within the current window. In some embodiments, it can also be expressed in the following alternative form:

[0060] in, This represents the nominal value of gravitational acceleration.

[0061] Secondly, calculate the pulse wave signal quality index. In some embodiments, the pulse wave signal quality index is determined based on pulse wave amplitude stability, baseline drift, and peak shape integrity, and is specifically expressed as follows:

[0062]

[0063] in, , , For the preset weights, and satisfying ; The amplitude stability component is preferably calculated based on the degree of variation in the peak values ​​of adjacent pulse waves. The baseline stability component is preferably calculated based on the proportion of low-frequency drift before and after filtering. The peak integrity component is preferably calculated based on the identifiability of the rising, falling, and main peaks of the pulse wave. In some embodiments, The range of values ​​is limited to after normalization. The higher the value, the higher the quality of the pulse wave.

[0064] Next, calculate the pulse cycle stability index. Let the effective pulse interval sequence detected within the current window be... The degree of dispersion of the pulse cycle is then expressed as:

[0065] in, This represents the standard deviation of the pulse interval sequence. This represents the mean of the pulse interval sequence. Correspondingly, the pulse cycle stability index is defined as:

[0066] in, This is a preset reference discrete threshold. The range of values ​​after normalization is A higher value indicates a more stable pulse cycle. Further, the detectability rate is calculated. Let the theoretical number of pulses to be detected within the current window be... The number of pulses that were effectively detected was ,but:

[0067] in, The value can be obtained based on the window duration and the current rough heart rate estimate, or based on the interpeak prediction value. The range of values ​​is preferably limited to: .

[0068] After obtaining the above indicators, they are normalized. For example:

[0069]

[0070] in, Indicates the upper limit of the reference value for the integral of the acceleration amplitude; , , , The range of values ​​is Furthermore, a higher value indicates a higher risk of motion artifacts.

[0071] In some embodiments, the motion artifact scoring Obtained using a weighted summation method:

[0072]

[0073] in, , , , Let non-negative weights be preset, and satisfy:

[0074]

[0075] For example, 0.35 is acceptable. 0.30 is acceptable. 0.20 is acceptable. A value of 0.15 is acceptable. In other embodiments, the weight may also be adaptively adjusted based on the device model, wearing position, or historical data quality, but the adjusted weight should still be limited to a preset range.

[0076] In some specific embodiments, the following determination rule is adopted: when When, it is judged as a clear motion artifact; when When, it is judged as a suspected motion artifact; when When this happens, it is determined to be a non-primary artifact. For example, within an 8-second analysis window, if the calculated... , , , ,but:

[0077] At this point, the window is identified as a suspected motion artifact and marked with a low confidence level, rather than being rolled back directly.

[0078] In some embodiments, if the pulse wave signal quality index If unavailable, peak loss rate is used. Alternative ,in:

[0079] The motion artifact score was then corrected to:

[0080] in, To replace weights and satisfy normalization constraints, if the number of valid pulses in the current window is lower than a preset lower limit, the window is directly marked as a low-quality window to be reviewed, and no abnormal rollback conclusion is made based solely on this window.

[0081] The space-behavior deviation score is jointly determined by the student's current space category and behavior category. For example, when a student is in a classroom or dormitory and is sitting, lying down, or in a low-activity state, the edge server unit assigns a negative deviation score to that window; when a student is in a classroom or dormitory and is engaged in continuous vigorous activity, the edge server unit assigns a positive deviation score; when a student is on the playground and is continuously running, the edge server unit assigns a negative deviation score; when a student is on the playground but is stationary and has a continuously abnormally high heart rate, the edge server unit assigns a positive deviation score; for unknown areas, the edge server unit may assign a neutral score.

[0082] The spatial-behavioral deviation score The calculation is performed based on the combination of spatial and behavioral categories. First, the edge server unit maps the student's current location to a spatial category based on the positioning results. The space categories include at least one of the following: classroom, dormitory, playground, corridor, canteen, medical room, and unknown area. Simultaneously, the edge server unit determines the behavior category based on acceleration characteristics, gait frequency characteristics, and body movement percentage. The behavior category includes at least one of the following: lying down, sitting, walking, moderate activity, continuous running, and falling / abnormal posture.

[0083] In some embodiments, the spatial-behavioral deviation score Determined by the following formula:

[0084]

[0085] in: This represents the pre-defined spatial-behavioral combination base score; This represents a correction score based on the degree of deviation of heart rate from the scene baseline; This represents a modified score based on at least one of the contextual factors, such as time period, course status, or area accessibility.

[0086] In some embodiments, the preset space-behavior combination base score The score is obtained using a lookup table. For example, a basic score table can be pre-constructed. The score is calculated based on the degree of deviation of the current heart rate from the baseline of the individual's physiological state in the corresponding scenario. .

[0087] in, This indicates the average heart rate for the current window. This represents the individual's baseline heart rate under corresponding spatial and behavioral conditions. In some embodiments, The segmentation is determined as follows: when hour, ;when hour, ;when hour, ;when hour, .

[0088] In some embodiments, if the current blood oxygen level decreases relative to the resting blood oxygen level by a preset proportion, or the recovery time exceeds a recovery timeout threshold, then it is allowed to... Additional support points are provided. For example: During class time, while continuously running in a classroom or dormitory... During the free play period on the playground and the behavior involved continuous running, The reliability is insufficient if the location is unknown or the position is not specified. The presence of supportive conditions for decreased blood oxygenation In the device switching or positioning drift window, .

[0089] Therefore, the spatial-behavioral deviation score is expressed as:

[0090] In some embodiments, to facilitate unified processing with other scoring methods, Cut off to a preset range, for example ,Right now:

[0091] In specific judgments, the following rules are adopted: When When, it is judged as a spatial-behavioral deviation support anomaly; when When, it is determined to be a spatial-behavioral deviation from neutrality; when When this occurs, it is determined that the spatial-behavioral deviation does not support the anomaly.

[0092] For example, if the target student is in a classroom and the behavior category is sitting still, then The current average heart rate is 110 beats / minute, and the baseline heart rate for the individual in the corresponding scenario is 78 beats / minute. Therefore:

[0093] Pick If it is during normal class hours and there is no location drift, then Thus we get:

[0094] At this point, the spatial-behavioral deviation score is neutral and does not independently support anomalies.

[0095] In some embodiments, if the confidence level of the behavior category identification result is lower than a preset threshold, then... De-weighting, for example:

[0096] in, The value ranges from 0.3 to 0.7. In other embodiments, if the space category is unavailable, then... Set to 0, and only based on and Generate neutral to weak support scores to avoid misjudgments due to missing positioning.

[0097] When the motion artifact score is higher than the preset threshold and the spatial-behavioral deviation score does not support the anomaly, and there are no strong risk supporting conditions such as abnormal blood oxygenation or hardware-level emergency sentinels, the edge server unit will roll back the event to the normal state; otherwise, the event will be migrated to the pattern analysis state.

[0098] In some embodiments, if the pulse wave signal quality index is unavailable, a combination of high acceleration and peak loss rate can be used as an alternative.

[0099] Step S103: Pattern Recognition and Disease-Specific Treatment Label Generation. In this embodiment, the edge server unit invokes a risk pattern classifier deployed on the edge side for events entering the pattern analysis state. The classifier uses at least a portion of the following input features as the feature vector X_event: current window average heart rate, heart rate change rate, heart rate variability index, interpeak variability, pulse wave quality index, current blood oxygen value, blood oxygen change trend, activity intensity, activity category, window body movement percentage, spatial category, time period, historical 24-hour true anomaly label, historical 24-hour false alarm label, and individual physiological state baseline deviation characteristics.

[0100] In some embodiments, the risk pattern classifier is preferably a lightweight gradient boosting tree model or a shallow neural network model; in scenarios with limited computing power, a logistic regression and expert rule fusion model, a decision tree model, or a pure rule engine can be used as alternatives.

[0101] The classifier output includes at least anomaly probability values, risk pattern labels, and label confidence scores. The risk pattern labels include at least rhythm abnormality risk patterns, cardiac load abnormality risk patterns, respiratory abnormality risk patterns, nonspecific abnormality risk patterns, and interference / low-quality pending verification patterns.

[0102] When the anomaly probability value reaches a preset threshold and a basic medical history label matching the risk pattern label exists in the student's medical record, the edge server unit generates a disease-specific treatment marker. This disease-specific treatment marker is preferably an event-level structured object, including at least an event identifier, student identifier, medical history type, candidate risk level, fixed review unit priority list, manual confirmation template identifier, candidate treatment action set, contraindicated action set, scoring correction rule identifier, escalation rule identifier, and effective duration. In some embodiments, it may also include environmental linkage suggestions, path planning preferences, assistive device prompts, and communication content pruning rules.

[0103] The disease-specific treatment markers are not only used for information recording, but also at least for: performing disease-specific correction scoring on the current event, determining the priority of fixed review units, filtering prohibited actions among candidate treatment actions, and determining event escalation rules, thereby forming a technical control chain driven by medical history matching results.

[0104] If a student has multiple underlying medical histories, the edge server unit prioritizes selecting the primary medical history strategy based on the mapping relationship between the current risk pattern label and the medical history. If multiple medical histories are matched, the primary medical history strategy is determined according to the preset risk priority, and the non-primary medical history strategy is used as an additional constraint, only for controlling prohibited actions or supplementing prompt information. In other embodiments, a parallel scoring scheme with multiple medical histories and a limited total correction score can also be adopted.

[0105] Step S104: Calculate the basic weighted score and the disease-specific correction score, and determine the final state of the event. In this embodiment, the edge server unit calculates the basic weighted score. The basic weighted score includes at least the geographic location score, activity status score, 24-hour historical record score, group deviation score, and spatial-behavioral deviation score.

[0106] In some embodiments, the basic weighted score Score based on geographical location Activity status score 24-hour historical record score Group deviation score and the normalized spatial-behavioral deviation score The scores are determined jointly. To facilitate a unified comparison of scores from different sources, it is preferable to first map each score to a unified dimension before performing a weighted summation.

[0107] For example, the basic weighted score is calculated according to the following formula:

[0108] in: To predetermine non-negative weights, the preferred option is one that satisfies... ; This represents the normalized spatial-behavioral deviation score. In some embodiments, 0.15 is acceptable. 0.15 is acceptable. 0.20 is acceptable. 0.20 is acceptable. A value of 0.30 is acceptable. In other embodiments, the weights can be fine-tuned according to different medical history types or school scenarios, but the normalization constraints described above are still preferred after adjustment.

[0109] Location score In some embodiments, the edge server unit determines a geographic location score based on at least one of the following: the risk attribute of the area to which the current geographic location belongs, the distance to the medical room or fixed review unit, and the historical abnormal density of the area. For example, the score can be assigned by spatial category mapping: the area near the medical room or covered by the fixed review unit. Classrooms and dormitories: Corridors and stairwells: Playground, remote outdoor areas: Unknown regions or insufficient location reliability: .

[0110] In some embodiments, if the distance from the current location to the nearest fixed verification unit is greater than a preset distance threshold, then additional steps are taken based on the above. Distance correction score.

[0111] Activity status score In some embodiments, the activity status score is determined based on both activity intensity and activity category, reflecting whether the current physiological abnormality is likely to be induced by normal movement. For example, values ​​are assigned according to the following rules: lying down or sitting: ;walk: Medium-level activities: Continuous running: Fall / abnormal posture: That is, a higher support score is assigned when abnormal candidates appear in a low-activity state, while a lower support score is assigned in a state where the heart rate can be explained by continuous running.

[0112] 24-hour historical record score In some embodiments, the 24-hour historical record score is determined by the number of true anomalies, the number of false alarms, and the time interval since the most recent true anomaly in the past 24 hours. For example, the original historical record value is first calculated:

[0113] in: This indicates the actual number of anomalies in the past 24 hours; This indicates the number of false alarms in the past 24 hours; This represents the proximity coefficient of the most recent real anomaly.

[0114] The nearest real anomaly proximity coefficient Determine as follows:

[0115] in: Indicates the time interval since the most recent real anomaly; Indicates the reference duration, such as 6 hours or 12 hours.

[0116] Then, Mapped to historical scores For example: when hour, ;when hour, ;when hour, ;when hour, .

[0117] Normalization of Spatial-Behavioral Deviation Scores In some embodiments, if the aforementioned spatial-behavioral deviation score The original value range is Then normalize as follows:

[0118] This allows it to participate in weighted calculations with other sub-scores under similar dimensions.

[0119] Group Deviation Score In some embodiments, the group deviation score Calculate as follows: First, calculate the current student's standardized heart rate deviation. Let the average heart rate in the current window be... The corresponding individual heart rate baseline for this scenario is The corresponding individual heart rate baseline fluctuation scale is ,but:

[0120] in, To prevent extremely small positive numbers with a denominator of zero.

[0121] Secondly, a reference group is constructed. The reference group preferably meets at least most of the following conditions: located in the same spatial grid; with consistent or adjacent activity categories; effectively worn; not currently entering a high-risk abnormal process; and meeting data quality standards.

[0122] Assume the reference group has a total of The student, regarding the number A number of reference students calculated their standardized heart rate deviation. In some embodiments, the median is used as the reference population center value:

[0123] Then, calculate the deviation of the current student relative to the reference group:

[0124] In some embodiments, absolute deviation values ​​may also be used:

[0125] To facilitate subsequent scoring, a segmented mapping method is preferred for obtaining the group deviation score. :when hour, ;when hour, ;when hour, ;when hour, .

[0126] In other embodiments, if the supporting role of anomalies "above the population center value" is emphasized more, then only when When a positive score is assigned, and when When Set to 0.

[0127] Alternative handling when the reference group is insufficient. If the reference group size... Insufficient preset quantity (For example, if there are fewer than 3 or 5 people), the edge server unit will set the group deviation score to a neutral value. For example, it can be set as follows: The historical median value of individuals in the same scenario was used as an auxiliary reference for substitution.

[0128] Specifically, let the deviation of the current student's historical standardized heart rate from the median value in the same scenario be denoted as _____. Then the auxiliary deviation can be calculated:

[0129] The It is preferred to use it only for auxiliary explanation or confidence adjustment, and not to directly replace it. The main score.

[0130] In some embodiments, the edge server unit also identifies environmental interference candidates when multiple terminals within the same spatial grid simultaneously experience severe body movement fluctuations or a decline in pulse wave quality within a short period of time, and generates an environmental interference index. If there is synchronous disturbance from multiple devices, and the environmental sensors, activity period, and fixed verification results support the conclusion of environmental interference, then the candidate event will be downweighted or rolled back; however, if multiple students simultaneously experience decreased blood oxygen, discomfort from manual feedback, abnormal fixed verification support, or air quality mutations that match asthma risk patterns, then the event will not be rolled back according to environmental interference.

[0131] Let the number of terminals within the same spatial grid that meet the conditions of effective wearability and data quality compliance be . Within a preset short time window Within the range, the number of terminals exhibiting both abnormal body movement and decreased pulse wave quality was [number missing]. Then the synchronization disturbance ratio is:

[0132] In some embodiments, environmental support factors It consists of three parts: environmental sensor anomalies, activity period matching, and fixed verification rejection anomalies, and is represented as follows:

[0133] in: This indicates whether the environmental sensor supports environmental interference; a value of 1 indicates support, and a value of 0 indicates otherwise. Indicates whether the current period is a time when there is a group activity, broadcast, or group movement to and from get out of class, which are prone to synchronization disturbances. If true, take 1; otherwise, take 0. This indicates whether the fixed review tends to reject genuine anomalies; it is set to 1 if true and 0 otherwise. For preset weights, the preferred option is one that satisfies... .

[0134] The environmental disturbance index, for example, is calculated using the following formula:

[0135] in, and The preset weights are preferred, and the following conditions are met: .

[0136] In some embodiments, when Greater than the environmental interference threshold If no of the following real anomalies support the current candidate event, the event will be downgraded or rolled back: multiple students simultaneously experience a decrease in blood oxygen; manual feedback indicates discomfort; fixed review supports anomalies; air quality mutations match respiratory-related risk patterns.

[0137] In other embodiments, only the anomaly probability value or label confidence level may be attenuated, without directly modifying it. .

[0138] In this embodiment, the edge server unit calculates the disease-corrected score based on the medical history type and risk pattern. The basic medical history preferably includes at least one of rhythm-related medical history, cardiac load-related medical history, and respiratory-related medical history. A history of premature ventricular contractions (PVCs) can be used as an example of rhythm-related medical history, a history of congenital heart disease can be used as an example of cardiac load-related medical history, and a history of asthma can be used as an example of respiratory-related medical history. For example, for students with a history of PVCs, a first correction score is assigned when they are at rest or in a low-activity state, exhibit a rhythm abnormality risk pattern across multiple consecutive windows, and the abnormality probability value reaches a first medical history threshold. For students with a history of congenital heart disease, a second correction score is assigned when they are in a low-activity or resting state, their current heart rate is consistently higher than a preset proportion of the resting baseline, and the recovery time exceeds a recovery timeout threshold. For students with a history of asthma, a third correction score is assigned based on the number of conditions met when they exhibit a respiratory abnormality risk pattern, environmental parameters exceed thresholds, a fixed review shows a downward trend in blood oxygen saturation, or the current blood oxygen value decreases relative to the baseline.

[0139] For students with a history of rhythm-related illnesses, if they are in a resting or low-activity state, and multiple consecutive windows show a rhythm abnormality risk pattern with an abnormality probability value reaching the first medical history threshold, then a first corrected score is assigned.

[0140] Indicatively, define: in: : 1 is taken when the condition of being at rest or with low activity is met; otherwise, 0 is taken. : 1 is set when the rhythm abnormality risk pattern is valid, and 0 is set otherwise; : The value is set to 1 if the abnormal probability value is greater than the first medical history threshold, and 0 otherwise. : Take 1 if multiple windows are active consecutively, otherwise take 0.

[0141] Correspondingly, let: That is, if it is true, it is assigned the value 2, and if it is false, it is assigned the value 0.

[0142] For students with a history of cardiac workload-related illnesses, a second correction score is assigned when they are in a low-activity or resting state, their current heart rate is consistently higher than the preset proportion of the resting baseline, and their recovery time exceeds the recovery timeout threshold.

[0143] For example, define: in: : 1 is set when low activity or resting state is established, otherwise 0 is set; : The value is 1 if the condition is met, otherwise it is 0. : Set to 1 if the recovery time exceeds the threshold, otherwise set to 0. Correspondingly, let: For students with a history of respiratory-related illnesses, when they exhibit a respiratory abnormality risk pattern and environmental parameters exceed thresholds, a fixed review of blood oxygen saturation showing a downward trend, or a current blood oxygen saturation value that is lower than the baseline meets the conditions, a third correction score is assigned based on the number of conditions met.

[0144] set up: : Set to 1 if the respiratory abnormality risk pattern is valid, otherwise set to 0; : Set to 1 when the environmental parameter exceeds the threshold, otherwise set to 0; : Set to 1 if the fixed review supports a downward trend in blood oxygenation, otherwise set to 0; : The value is 1 if the current blood oxygen level decreases relative to the baseline and meets the condition, otherwise it is 0.

[0145] Define conditional counting: Correspondingly, let: For example: only when the risk pattern holds and there is no additional support, When the risk model is valid and one additional condition is met, When the risk model is valid and two additional conditions are met, When the risk model is valid and three additional conditions are met, .

[0146] In some embodiments, if the target student has multiple underlying medical histories, it is preferable to determine the primary medical history strategy based on the current risk pattern label. For example, the maximum value among the adjusted scores for each disease is taken as the current adjusted score for that disease. In other embodiments, a combination of a main medical history score and an additional medical history fine-tuning score may be used, but it is preferable to limit the total correction range to no more than a preset upper limit.

[0147] The edge server unit calculates the final weighted total score based on the basic weighted score and the disease-specific correction score. If the final weighted total score is lower than the first total score threshold, the event reverts to the normal state; if the final weighted total score is between the first and second total score thresholds, the event transitions to the manual confirmation state; if the final weighted total score is higher than the second total score threshold, the event directly transitions to the confirmed abnormal state and enters the third-level response.

[0148] Edge server units are based on a weighted average score. Disease-specific score correction Calculate the final weighted total score .

[0149] For example, the following method is adopted:

[0150] in: This indicates the weighting value for environmental interference; when there is no environmental interference weighting condition, ... When the environmental interference index reaches the preset threshold, The value can be 0.5, 1, or other preset values. In some other embodiments, a fixed review enhancement or a high-risk sentinel enhancement may be further introduced, but it is essentially still a modification of... Corrections.

[0151] In some embodiments, a first total score threshold is set. Second total score threshold ,in For example, take:

[0152] Correspondingly, the state transition rule is: when When, the event reverts to a normal state; when When the event is in progress, it is moved to a manual confirmation state; when At that point, the event is directly transferred to the confirmed abnormal state and enters the third-level response.

[0153] Step S105: Fixed Review, Manual Confirmation, and Tiered Response. In this embodiment, before an event enters the manual confirmation state or the confirmation abnormal state, the edge server unit prioritizes and schedules fixed review units according to medical history adaptation priority, online status, student distance, device occupancy status, most recent self-inspection results, most recent data quality score, and expected sampling delay. If the first candidate device times out at any stage of order acceptance, connection arrival, or sampling completion, the system switches to the next candidate device; if a preset number of devices fail consecutively, the system reverts to the manual confirmation priority process; if there is no fixed review unit, the system uses extended wristband continuous sampling and terminal manual confirmation as an alternative solution. The ordering of the fixed review units is controlled at least by the fixed review unit priority list in the disease-specific treatment marker.

[0154] The fixed verification unit is located in classrooms, dormitory management points, school clinics, stadium entrances, or other preset verification points to perform secondary verification of abnormal candidate events reported by the smart bracelet. The fixed verification unit includes at least a verification sensor module, an identity recognition module, a communication module, and a result output module. The verification sensor module collects at least one verification physiological parameter of the target student, including one or more of heart rate, blood oxygen saturation, pulse wave quality index, body temperature, respiratory rate, or blood pressure. When the edge server unit determines that the current abnormal candidate event meets the preset verification trigger conditions based on the data uploaded by the smart bracelet, it sends a verification request and verification instruction to the management terminal and / or fixed verification unit corresponding to the target student. After the target student arrives at the fixed verification unit, the unit first verifies their identity using one or more methods, such as campus card, QR code, facial recognition, student ID input, or close pairing with a smart bracelet. The unit then binds this verification operation to the corresponding student identifier and abnormal event identifier. Subsequently, the fixed verification unit outputs posture guidance information, prompting the target student to remain seated or standing still for a preset stable duration to reduce the impact of motion artifacts and poor contact on the verification results. The fixed verification unit acquires multiple frames of verification signals within a continuous preset sampling duration and performs quality assessment on the acquired verification signals. The quality assessment includes at least one or more of the following: contact status determination, pulse waveform continuity determination, signal-to-noise ratio determination, amplitude stability determination, and sampling integrity determination. If the verification signal does not meet the preset quality conditions, the fixed verification unit outputs a retest prompt and marks this verification as invalid. For verification signals that meet the quality conditions, the fixed verification unit calculates the core rate, blood oxygen saturation, and / or other verification parameters locally or via an edge server unit. The edge server unit uses at least one of the following as time anchors: the start time of sampling, the midpoint time of sampling, or the end time of sampling of the fixed verification unit. It extracts the smart bracelet parameters within a preset time window before and after the corresponding time anchor and performs a consistency comparison between the bracelet parameters and the verification parameters. This consistency comparison includes at least one of numerical difference comparison, anomaly direction consistency comparison, and trend consistency comparison. If the verification parameters measured by the fixed verification unit still exceed the corresponding individualized threshold and are consistent with the anomaly direction of the smart bracelet, or if the difference between the fixed verification parameters and the bracelet parameters is less than a preset difference threshold, the edge server unit determines that this verification supports anomalies. If the fixed verification parameters fall within the normal range and are inconsistent with the anomaly direction of the smart bracelet, the edge server unit determines that this verification does not support anomalies. If the verification signal quality is insufficient, the sampling is incomplete, or the identity binding fails, the edge server unit determines that this verification is invalid.The edge server unit performs abnormal event status migration control according to the review conclusion: for events supported as abnormal by the review, it migrates them from the manual confirmation stage to the confirmed abnormal stage; for events not supported as abnormal by the review, it demotes them to observation events or end events; for events with invalid reviews, it keeps them in the manual confirmation stage and controls the fixed review unit to perform a re-review or notifies the management personnel to perform a manual review. For events where the medical history label corresponds to a high-risk disease type and the comprehensive weighted total score is higher than the upgrade threshold, even if the fixed review result is in the critical range, the edge server unit can control the system to enter the upgrade disposal process.

[0155] In some embodiments, manual confirmation serves as an auxiliary input for event status migration and is not an essential limitation of the core technical concept of this application. The edge server unit automatically generates manual confirmation task objects according to the medical history type, risk mode, event level, and terminal permissions, and pushes question sets with different templates to the school doctor terminal, teacher terminal, head teacher terminal, or parent terminal. Effective manual confirmation must at least meet the following requirements: submitted within the time limit, the completion rate of answering key questions reaches the preset ratio, the terminal identity is legal and bound to the event identifier. If the confirmation results of different roles conflict, the valid result is determined according to the priority order of school doctor, fixed review result, teacher / class head teacher, and parent; if the high-priority result is missing, the majority agreement principle is adopted, and the observation is extended and the school doctor confirmation is upgraded if necessary.

[0156] In the embodiments of this application, the system determines the response level according to the final weighted total score, fixed review result, and manual confirmation result. The first-level response is applicable to low-risk confirmed events, and it performs school doctor notification, visible to the head teacher, and extended monitoring; the second-level response is applicable to medium-risk confirmed events, and it performs school doctor, head teacher, and parent notifications and initiates disease-specific disposal suggestions; the third-level response is applicable to high-risk confirmed events or extreme situations directly supported by hardware-level emergency sentinels, and it performs parallel redundant communication, emergency communication guarantee, linkage control, and strong reminder.

[0157] Extremely urgent events and results directly supported by hardware-level emergency sentinels have the highest priority, which is higher than the fixed review result and manual confirmation result; when manual confirmation supports abnormality in the case of obvious symptom complaints or deterioration of the on-site state, the priority is higher than that of the fixed review not supporting abnormality; when the fixed review supports abnormality with significant physiological parameter overlimits and manual confirmation is missing, the priority is higher than the initial low-risk judgment by the machine.

[0158] Step S106: Feedback on Treatment Results and Strategy Update. After the event, the school doctor can enter the final treatment result in the edge server unit management interface, including information such as whether it is a genuine anomaly, a false alarm, medical history matching, whether the on-site treatment was effective, and whether the patient was escalated to a hospital. The edge server unit updates the disease-specific treatment strategy parameters based on the feedback results. The school doctor or authorized treatment personnel do not directly trigger the strategy update in free text, but instead enter a structured feedback result object R_event through the edge server unit management interface. The feedback result object includes at least some of the following fields: event identifier, student identifier, medical history type, risk mode label, whether it is a genuine anomaly, whether the fixed review result supports anomalies, whether the manual confirmation result supports anomalies, whether the medical history matching is valid, whether the on-site treatment was effective, whether the patient was escalated to a hospital, whether the recovery after treatment met the standards, whether a false escalation occurred, feedback entry time, and feedback personnel identifier. After receiving the feedback result object, the edge server unit preferably performs field integrity verification, permission verification, event status closed-loop verification, and duplicate entry verification first. Only when the event has ended, the feedback source is valid, and it is not locked, is the policy update process allowed.

[0159] In some embodiments, the system uses a weighted update method to update the parameter weights in the disease-specific treatment strategy. The edge server unit does not directly update the strategy parameters with a single field, but first calculates the event feedback score F_event based on the feedback result object. The event feedback score is used to characterize the consistency between the original disease-specific treatment strategy and the actual result in this event, as well as the effectiveness of the current treatment path. The event feedback score F_event is calculated from the true anomaly score, fixed review consistency score, manual confirmation consistency score, on-site treatment effectiveness score, reasonableness of escalation to hospital score, and erroneous escalation penalty score. The true anomaly score, fixed review consistency score, manual confirmation consistency score, on-site treatment effectiveness score, reasonableness of escalation to hospital score, and erroneous escalation penalty score are assigned according to a manually specified method. For example, if it is finally confirmed as a true anomaly, it means that the current risk identification and treatment chain has high reference value, and the true anomaly score can be set according to the following rules: true anomaly: S_true=1, false alarm: S_true=0.

[0160] The updatable parameter weights include correction parameter weights, fixed review unit priority weights, and treatment suggestion priority weights. The strategy parameters are not updated globally, but rather locally using "medical history type + risk pattern" as the index unit. That is, for each event, the corresponding parameter group in the strategy table is first located based on the medical history type and risk pattern, and then only some or all parameters in that parameter group are updated.

[0161] In some embodiments, the disease-specific correction parameter weights are used to control the extent to which disease-specific treatment markers enhance the final event score.

[0162] For example, for combinations of different medical history types and risk patterns, a set of disease correction parameter tables can be stored in the strategy library: during event scoring, the disease correction score is determined by the base correction value and the corresponding weight. After the update, if a certain medical history-pattern combination is frequently proven effective in real anomalies, the corresponding W_dis can be increased; conversely, if the combination causes false alarms multiple times, the corresponding W_dis can be decreased.

[0163] In some embodiments, the fixed review unit priority weight is used to control the ranking of different fixed review units under specific medical history types or risk patterns. Let the set of fixed review units be E={e1,e2,...,en}. For medical history types and risk patterns, the device priority weight reflects the degree to which the device is suitable for the current combination of medical history and risk pattern. For example, for respiratory-related medical histories, pulse oximetry and pulmonary function-related devices have higher weights; for rhythm-related medical histories, electrocardiogram (ECG) review devices have higher weights.

[0164] In some embodiments, the treatment recommendation priority weight is used to control the output order of candidate treatment actions in disease-specific treatment tags. Let the set of candidate treatment actions under a certain medical history type and risk model be A = {a1, a2, ..., am}. The treatment recommendation priority weight represents the priority weight of action a under that combination of medical history type and risk model. For example, for respiratory-related medical histories: "avoiding the triggering environment" has a higher weight; "suggesting the use of an auxiliary inhaler" has a higher weight; actions that conflict with contraindicated actions have their weights set to zero or are removed. If a certain treatment action is repeatedly reported as effective, its corresponding treatment recommendation priority weight can be gradually increased; if it is proven to be ineffective or prone to false escalation, its weight is decreased.

[0165] The priority weights of fixed review units and handling recommendations can be updated directly using the following method: The edge server unit updates the normalized parameters such as the priority weights of fixed review units and handling recommendations using an exponentially weighted moving average. Specifically, the update formula is as follows:

[0166] in: This represents the weight value before the update; This represents the normalized evaluation score of the current event feedback; The preset update coefficient has a preferred value range. ; This represents the updated weight value.

[0167] For non-normalized parameters such as disease-specific correction parameters and upgrade threshold fine-tuning parameters, since their actual business value range is inconsistent with the normalized feedback score dimension, it is preferable to maintain their internal normalized weights first. Then map it back to the actual business parameter value.

[0168] For the disease-specific correction parameter: let its allowed value range be... The updated actual disease correction parameter value is:

[0169] For the threshold fine-tuning parameter: let its allowable fine-tuning range be... The updated actual fine-tuning offset is:

[0170] The above methods effectively solved the problem of event feedback evaluation scores. The issue of inconsistent dimensions between parameters and actual business parameters has been addressed by implementing consistent adaptive updates for all types of parameters.

[0171] In this embodiment, the parameter weight update applies to at least one of the following: disease-specific correction parameters, fixed review unit priority weights, treatment recommendation priority, and escalation threshold fine-tuning parameters; the update does not change the basic medical history type label in the disease-specific intervention file. The classifier model update preferably employs offline training, versioned distribution, and rollbackable deployment methods, and is independent of individual physiological state baseline updates and strategy weight updates.

[0172] In other embodiments, when an event is in a Level 3 response state, the system may further execute a parallel communication strategy as an extended control method for handling high-risk events.

[0173] In this embodiment, when the event is in the third-level response state, the system executes a parallel communication strategy, including: dialing a preset emergency contact number or sending an emergency text message via a cellular network; pushing an alarm to the school clinic screen or management terminal via a wireless network; pushing an alarm to nearby teachers and security terminals via a Bluetooth gateway or campus LAN; and in embodiments with a campus broadcast interface, triggering a campus broadcast voice announcement.

[0174] For links that support acknowledgments, the system records the receipt confirmation status; for links that do not support acknowledgments, the system records the sending status and switches to the alternative notification link if no manual confirmation is received within a preset waiting time.

[0175] In some embodiments, the smart bracelet's built-in long-range low-power communication module is in a low-power state by default. When the system has entered Level 3 response, the cellular network signal strength is below a preset threshold, and the wireless network connection fails, the edge server unit or the smart bracelet activates the long-range low-power communication module and sends an emergency beacon containing at least the device identifier, anonymous student identifier, timestamp, last known location, event level, disease risk code, and battery status through a pre-deployed campus gateway. The smart bracelet then returns to the low-power state after sending the beacon or receiving confirmation from the gateway. If the campus does not deploy such a gateway, SMS short message codes or Bluetooth proximity broadcast alarm packets can be used as alternatives.

[0176] In other embodiments, when an event is in a Level 3 response state, the system may further perform at least one of the following linked operations: access control, environmental devices, route planning, and terminal strong reminders, as an extended implementation of disease-differentiated treatment.

[0177] In this embodiment, when the event is in the third-level response state, the system can also perform linked operations. For example, the system can send a location and status summary to a preset medical contact or emergency interface; in embodiments with a network access control interface, it can link the campus access control system to open the rescue channel; and it can trigger vibration and screen prompts on the smart bracelet.

[0178] For students with a history of premature ventricular contractions (PVCs), the system prioritizes displaying rhythm fluctuation trend graphs and fixed review results to the school doctor's terminal, and sends simplified abnormality summaries to the emergency contact terminal under authorized conditions.

[0179] For students with a history of congenital heart disease, the system, when interface conditions are available, prioritizes linking with the campus access control system, nearby access assistance facilities, and on-campus traffic guidance modules to generate suitable routes for low-load transportation. These routes can be generated based on a cost function that includes distance, number of staircases, access control resistance, and congestion level, with the number of staircases receiving a higher weight than in normal event scenarios. If map and access control interfaces are unavailable, the system outputs low-mobility rescue suggestions as alternative solutions to the teacher.

[0180] For students with a history of asthma, the system will prioritize connecting to the fresh air system, air purification system, or local exhaust system when the interface is available, and will display prompts to the teacher to move away from the irritant, assist in regulating breathing, and confirm the inhalation device; if the student is in the playground or a dusty area, the system will suggest moving to a predefined shelter or a well-ventilated area.

[0181] In this embodiment, the system further includes a taboo rule control module. This module maintains a mapping between medical history types and taboo actions. When the system generates candidate treatment suggestions, before sending them to the terminal, and during response escalation, it matches the candidate treatment actions. If a candidate action matches a taboo action, it is replaced with a medical history-specific safe action; otherwise, it is replaced with a general safe action. The system records at least the original candidate action, the rule matching identifier, the replacement action, and the execution time for subsequent feedback and updates.

[0182] For example, for high-risk events corresponding to a history of congenital heart disease, suggestions for rapid walking to transfer should be prohibited; for high-risk events corresponding to a history of asthma, suggestions for continuous observation in a stimulating environment for more than a preset duration should be prohibited; and for high-risk events corresponding to a history of premature ventricular contractions, suggestions for resuming exercise without review should be prohibited.

[0183] In practical application, when a student with a history of asthma is in the classroom, the edge server unit receives their current window pulse wave, acceleration, blood oxygen, and location data. Step S101, user physiological data collection and detection, confirms the presence of abnormal heart rate and abnormal blood oxygen, triggering an initial alert state. Subsequently, in step S102, the motion artifact score is below the significant artifact threshold, and the spatial-behavioral deviation score is positive, triggering a pattern analysis state. Step S103 shows the classifier outputting an anomaly probability of 0.78, the risk pattern label being a respiratory abnormality risk pattern with a confidence level of 0.69, and the presence of an asthma history in the medical record; therefore, the system generates a disease-specific treatment label. In step S104, the base weighted score is 2 points; simultaneously, PM2.5 and CO2 concentrations exceed their respective thresholds, and the fixed verification unit measures a blood oxygen level 4 percentage points lower than the individual's resting baseline; therefore, a disease-specific correction score of 2 points is applied, resulting in a final weighted total score of 4 points. The event then transitions to manual confirmation and enters the first or second level response candidate state. If the campus has deployed pulse oximetry equipment and a fresh air system, the system will prioritize scheduling the pulse oximetry equipment and coordinate with the fresh air system. If no such equipment has been deployed, the system will send a prompt to the teacher's terminal to move the student to a well-ventilated area and confirm the availability of a backup inhalation device. If the pulse oximetry result is consistent with the wristband analysis result, and the manual confirmation shows shortness of breath and interrupted speech, the system will further escalate the response level.

[0184] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

Claims

1. A multi-stage state confirmation and control method for abnormal triggering of wearable devices, comprising collecting user data on physiological state and spatial location using the user's personal device, characterized by: Anomaly candidate detection is performed based on the features of the physiological state and spatial location data and the individual physiological state baseline. For abnormal candidates, calculate motion artifact scores and spatial-behavior deviation scores. When the motion artifact score is higher than a preset threshold and the spatial-behavior deviation score is lower than a preset support threshold, the event is rolled back to the normal state. The non-rollback events are classified using a pattern classifier, and anomaly probability values ​​and risk pattern labels are output. For events that are not rolled back, when there is a medical history type in the user's basic medical history that corresponds to the risk pattern label, a disease-specific treatment label is generated, which includes disease correction parameters, fixed review unit priority parameters, candidate treatment action priority parameters, prohibited action set and escalation threshold fine-tuning parameters. For events that do not regress, a basic weighted score is calculated based on geographical location score, activity status score, 24-hour historical record score, group deviation score, and normalized spatial-behavioral deviation score. The final weighted total score is obtained by combining the disease-specific correction score and the environmental interference reduction score. The weighted total score is compared with the first total score threshold and the second total score threshold to determine whether the event belongs to a normal state, a manually confirmed state, or a confirmed abnormal state. Based on the disease-specific treatment mark, fixed review unit scheduling and / or prohibited action filtering and / or escalation rules are executed.

2. The multi-stage state confirmation and control method for abnormal triggering of wearable devices according to claim 1, characterized in that, The individual physiological baseline includes at least one of the following: resting heart rate baseline, low-activity heart rate baseline, moderate-activity heart rate baseline, resting blood oxygen baseline, heart rate variability baseline, and recovery time baseline; wherein, the edge server unit generates one or more of the following thresholds based on the individual physiological baseline: upper limit threshold for heart rate, abnormal blood oxygen threshold, abnormal heart rate variability threshold, and recovery timeout threshold for the user's current scenario; the abnormal candidate conditions include at least one of the following: abnormal heart rate, abnormal heart rate variability, abnormal heart rate, abnormal blood oxygen, or hardware-level emergency sentinel triggering.

3. The multi-stage state confirmation and control method for abnormal triggering of wearable devices according to claim 1, characterized in that, The calculation of motion artifact scores and spatial-behavioral deviation scores for events in the initial alert state includes: The motion artifact score is calculated based on at least two of the following indicators: acceleration amplitude integral, photoplethysmography pulse wave signal quality index, pulse cycle stability, and detectability. Calculate the space-behavior deviation score based on the combination relationship between the target student's space category and current behavior category; When the motion artifact score is higher than a preset threshold and the space-behavior deviation score does not support anomalies, and there are no abnormal blood oxygen levels or hardware-level emergency sentinel support conditions, the event will be rolled back to the normal state.

4. The multi-stage state confirmation and control method for abnormal triggering of wearable devices according to claim 1, characterized in that, The risk pattern classifier is deployed on the edge server unit side. The input features of the risk pattern classifier include at least a portion of the following: current window average heart rate, heart rate change rate, heart rate variability index, interpeak variability, pulse wave quality index, current blood oxygen value, blood oxygen change trend, activity intensity, activity category, spatial category, historical abnormality markers, and individual physiological state baseline deviation features. The risk pattern labels include at least one of the following: rhythm abnormality risk pattern, cardiac load abnormality risk pattern, respiratory abnormality risk pattern, nonspecific abnormality risk pattern, and interference / low quality pending review pattern.

5. The multi-stage state confirmation and control method for abnormal triggering of wearable devices according to claim 1, characterized in that, The disease-specific treatment markers include: event identifier, student identifier, medical history type, candidate risk level, fixed review unit priority list, candidate treatment action set, contraindicated action set, scoring correction rule identifier, escalation rule identifier, and effective duration; The basic medical history includes at least one of rhythm-related medical history, cardiac load-related medical history, and respiratory-related medical history; When a target student has multiple underlying medical histories, the edge server unit determines the primary medical history strategy based on the mapping relationship between the current risk pattern label and the medical history, and uses the non-primary medical history strategy as an additional constraint for prohibited action control or supplementary prompt information.

6. The multi-stage state confirmation and control method for abnormal triggering of wearable devices according to claim 1, characterized in that, The fixed review unit reviews the event; it sorts the fixed review units according to medical history matching priority, online status, target student distance, device occupancy status, most recent self-inspection results, most recent data quality score, and expected sampling delay; it sends scheduling requests to candidate fixed review units in sequence according to the sorting results; when the current candidate fixed review unit times out in any stage of the order acceptance stage, connection stage, or sampling completion stage, it switches to the next candidate fixed review unit; when a preset number of candidate fixed review units fail to be scheduled, it switches to the manual confirmation priority process or extends the continuous sampling of the wearable information collection unit as an alternative review method.

7. The multi-stage state confirmation and control method for abnormal triggering of wearable devices according to claim 1, characterized in that, The calculation of the basic weighted score includes: summing the geographic location score, activity status score, 24-hour historical record score, group deviation score, and normalized spatial-behavioral deviation score according to preset weights to obtain the basic weighted score; wherein, the group deviation score includes: calculating the standardized heart rate deviation value of the current student, selecting other students located in the same spatial grid with the same or adjacent activity categories, effectively wearing masks, not currently entering the high-risk abnormal process, and whose data quality meets the standards as a reference group, using the median of the standardized heart rate deviation values ​​of the reference group as the center value, and determining the group deviation score based on the interval in which the difference between the current student and the center value falls; when the number of people in the reference group is less than a preset number, the group deviation score is set to a preset neutral score, and the median of the historical standardized heart rate deviation value of individuals in the same scene is used to form an auxiliary reference deviation.

8. The multi-stage state confirmation and control method for abnormal triggering of wearable devices according to claim 1, characterized in that, The calculation of the basic weighted score includes: summing the geographic location score, activity status score, 24-hour historical record score, group deviation score, and normalized spatial-behavioral deviation score according to preset weights to obtain the basic weighted score; wherein, the group deviation score includes: calculating the standardized heart rate deviation value of the current student, selecting other students located in the same spatial grid with the same or adjacent activity categories, effectively wearing masks, not currently entering the high-risk abnormal process, and whose data quality meets the standards as a reference group, using the median of the standardized heart rate deviation values ​​of the reference group as the center value, and determining the group deviation score based on the interval in which the difference between the current student and the center value falls; when the number of people in the reference group is less than a preset number, the group deviation score is set to a preset neutral score, and the median of the historical standardized heart rate deviation value of individuals in the same scene is used to form an auxiliary reference deviation.

9. A multi-stage state confirmation and control system for wearable devices triggered by abnormalities, characterized in that, The device includes a wearable information acquisition unit for collecting students' physiological and physical data; an edge server unit for performing the method described in any one of claims 1 to 8; a fixed verification unit for performing secondary verification of abnormal events; and an environmental sensor for providing air quality, temperature, humidity, or noise environmental parameters. A management terminal for inputting structured feedback results and viewing the status of event handling; wherein, the edge server unit is communicatively connected to the wearable information acquisition unit, the fixed verification unit, the environmental sensor and the management terminal respectively.

10. The control system according to claim 9, characterized in that, The edge server unit includes an event processing engine, a disease-specific treatment strategy library, and a feedback update engine. The disease-specific treatment strategy library is used to store disease correction parameters, fixed review unit priority parameters, candidate treatment action priority parameters, contraindicated action sets, and escalation threshold fine-tuning parameters according to medical history type and risk pattern. The feedback update engine is used to perform a limited update on the corresponding strategy parameter group in the disease-specific treatment strategy library based on the feedback results after the event ends.

Images