Personnel health abnormity monitoring method and system

Abstract

The invention provides a personnel health abnormity monitoring method, which comprises the following steps of: 1, dividing a day into six key life stage time periods, detecting the existence state of a check-in person at a time interval of 5 minutes, recording the existence state as 1, recording the absence state as 0, and counting the sum of the existence values in each time period; 2, carrying out standardization processing on existing values of all time periods, carrying out hierarchical clustering by adopting a Ward minimum variance method, and iteratively merging clusters until six typical life modes are generated; 3, constructing a deep feedforward neural network, and verifying the physiological rationality of a clustering result; and 4, estimating the change of the health condition according to the change trend of the life mode, and eliminating invalid data in combination with an empty room mark to realize health anomaly detection. And dynamic monitoring and anomaly detection of the health condition are realized by analyzing the life behavior mode of the facility check-in and combining a clustering algorithm and neural network verification. The invention further provides a system corresponding to the method.

Description

Technical Field

[0001] This invention belongs to the field of intelligent health monitoring technology, specifically relating to a method and system for monitoring abnormal health conditions in personnel based on time-based existence pattern clustering. Background Technology

[0002] In the field of personnel health monitoring, building a comprehensive and dynamic health assessment system has always been the goal pursued by the industry. However, existing technologies still have multi-dimensional limitations. These problems are interconnected and jointly restrict the improvement of monitoring accuracy and application value.

[0003] First, existing monitoring methods mostly focus on short-term capture of single physiological indicators such as heart rate and respiratory rate. While this type of data can reflect the physiological state at a particular moment, it is difficult to connect it into a long-term health trajectory. For example, the quality of sleep depends not only on the duration of a single sleep episode but also on the regularity of sleep patterns over several weeks; the health of exercise habits also needs to be judged in conjunction with the fluctuation trend of daily activity intensity and duration. Due to the lack of long-term tracking of lifestyle patterns, short-term monitoring of single indicators cannot construct a complete health profile and is insufficient to predict chronic health risks caused by gradual changes in behavioral habits.

[0004] Secondly, even with attempts to incorporate behavioral pattern analysis into some technologies, the accuracy of their clustering algorithms still needs improvement. Existing solutions often use improved algorithms for preliminary classification of behavioral data, but the misclassification rate increases significantly when faced with multidimensional behavioral information (such as activity intensity, duration, and spatial distribution). Especially when distinguishing between low-intensity activities (such as sedentary office work and light housework), the differences in characteristics between the two types of behavior are already subtle. Coupled with interference from environmental noise (such as sensor signal fluctuations), the algorithm is very likely to classify similar behaviors as outliers, or vice versa, leading to misinterpretation of behavioral patterns and thus affecting the assessment of health status.

[0005] More importantly, existing systems suffer from rigidity in their analysis of circadian rhythms. Current mainstream solutions often use fixed time windows to divide periods (e.g., 6:00-12:00 is uniformly marked as morning). This "one-size-fits-all" approach completely ignores individual differences in biological clocks: shift workers' peak activity may occur in the early morning, and the elderly's lunch break may be 2-3 hours longer than that of younger people; the core activity periods of these two groups can shift by 4-6 hours. Furthermore, the impact of key life stages such as meals, medication, and rehabilitation training on health status (e.g., postprandial blood sugar fluctuations, post-exercise heart rate changes) is not considered in the time window division. This time division method, detached from individual realities, leads to systematic biases in the analysis of time-series behavioral data, rendering subsequent health assessments unreliable.

[0006] In summary, the shortcomings of existing technologies in areas such as long-term health profiling, precise clustering of multidimensional behaviors, and personalized rhythm analysis significantly reduce the timeliness and accuracy of health anomaly monitoring. There is an urgent need for an innovative solution that can integrate key life stage segmentation, precise clustering algorithms, and physiological data verification to overcome the current technological bottlenecks. Summary of the Invention

[0007] To address the aforementioned issues, this invention provides a method and system for monitoring abnormal health conditions of individuals based on time-based existence pattern clustering. By analyzing the lifestyle patterns of facility residents and combining clustering algorithms with neural network verification, dynamic monitoring and abnormal detection of health conditions can be achieved.

[0008] In a first aspect, the present invention provides a method for monitoring abnormal health conditions in personnel, specifically including: Step 1: Divide the data into time periods and collect data to build the basic data.

[0009] Time Segmentation: The day is divided into 6 key time periods strongly associated with daily life scenarios, including from waking up to breakfast (6:00-8:30), from breakfast to the morning (8:30-11:00), around lunch (11:00-13:30), from lunch to dinner (13:30-16:00), around dinner (16:00-18:30), and from dinner to bedtime (18:30-21:00), avoiding the rigidity of fixed time windows.

[0010] Data collection: At 5-minute intervals, the presence status of guests in a specific area (such as the bed) is detected by sensors (presence is recorded as 1, absence as 0), and the sum of presence values ​​within each time period is calculated (e.g., there are 30 5-minute intervals in the 6:00-8:30 time period, and the presence value is the cumulative value of 1 within that time period).

[0011] Step 2: Based on the collected existence value data, cluster analysis is used to uncover stable lifestyle patterns of individuals, providing behavioral evidence for health assessment.

[0012] Data preprocessing: Standardize the existing values ​​for each time period (standardized existing value = original existing value / maximum possible existing value within the time period) to eliminate the influence of different time period lengths, so that the data is in the range [0,1] and ensure the effectiveness of clustering.

[0013] Hierarchical clustering: The Ward minimum variance method was used for clustering, and six typical lifestyle patterns were generated through the following steps: 1) Initialization: Treat the daily existing value vectors as independent clusters; 2) Calculate the distance: using the formula Calculate the inter-cluster merging variance increment; 3) Iterative merging: Select the cluster pairs with the smallest variance increment to merge, update the cluster center and distance matrix, until 6 clusters (i.e. 6 modes) are obtained.

[0014] Pattern definition: The six patterns cover behavioral characteristics ranging from healthy and active to severely disabled (e.g., pattern 6 is healthy and active, pattern 4 is low-activity / abnormal ambulation, etc.), and are associated with corresponding health risks (e.g., pattern 4 indicates the risk of dementia).

[0015] Step 3: To verify the scientific validity of the clustering pattern, physiological data (respiratory values) are introduced for cross-validation to ensure that the behavioral pattern is consistent with the physiological state.

[0016] Data preparation: Respiratory values ​​were collected every 5 minutes from 6:00 to 21:00 (180 dimensions in total), and after Z-score standardization, they were used as input to the neural network; the 6 pattern labels generated by clustering were converted into one-hot encoding as the output target.

[0017] Network design: A deep feedforward neural network is adopted, with the following structure: input layer (180 neurons) → hidden layer 1 (288 ReLU neurons) → hidden layer 2 (100 ReLU neurons) → output layer (6 Softmax neurons); training parameters are: 3000 epochs, 32 batches, Adam optimizer (learning rate 0.001), and classification cross-entropy loss function.

[0018] Validation criteria: If the validation set accuracy is ≥98%, the standard deviation of the same pattern of respiratory values ​​is <1.5 breaths / minute, and the proportion of the diagonal line of the confusion matrix is ​​>98%, then the clustering pattern is valid.

[0019] Step 4: Based on the validated pattern, determine the health trend by tracking its dynamic changes to complete the monitoring loop.

[0020] Continuously record the evolution of individual lifestyle patterns (e.g., from pattern 6 (healthy and active) to pattern 4 (low activity / abnormally away from home)), and combine this with empty room marking to exclude invalid data; When a change in the pattern matches the characteristics of health deterioration (such as the appearance of typical behavior in pattern 4), an abnormal warning is triggered, prompting medical staff to intervene.

[0021] Secondly, the present invention provides a system for monitoring abnormal health conditions in personnel, the system comprising: Data acquisition module: used to divide the time period into 6 key life stages, detect the presence status of the resident in a specific area at 5-minute intervals, and count the total presence value of each time period; Clustering analysis module: used to standardize existing value data, and uses hierarchical clustering algorithm to generate 6 typical lifestyle patterns; Neural network validation module: used to build and train a deep feedforward neural network to validate the consistency between clustering results and respiratory value data; Dynamic monitoring module: Used to predict health trends based on changes in lifestyle and to monitor health abnormalities.

[0022] Furthermore, the data acquisition module includes: The sensor unit is used to detect the presence state; The data preprocessing unit is used to count existing values ​​and perform standardization.

[0023] Furthermore, the clustering analysis module includes: Initialization unit: Treat the daily existing value vectors during the monitoring period as independent clusters; Matrix calculation unit: The Ward minimum variance method is used to calculate the inter-cluster merging variance increment; Iterative merging unit: The Ward minimum variance method in hierarchical clustering is used as the merging strategy. The cluster pairs with the smallest merging variance increment are selected for merging, and the cluster center and distance matrix are updated until 6 clusters are obtained.

[0024] Furthermore, the neural network verification module includes: Data standardization unit, used to perform Z-score standardization on respiratory value data; Model training unit, used to build and train neural networks; The results validation unit is used to validate the prediction accuracy and respiratory value convergence.

[0025] The effects described in the invention are merely those of the embodiments, and not all the effects of the invention. One of the above technical solutions has the following advantages or beneficial effects: 1. Time-based value design: reconstructing the relationship between time and behavior.

[0026] Existing technologies treat time as a physical quantity independent of behavior, using fixed time windows (e.g., 6:00-12:00 is uniformly defined as morning), which is completely detached from individual life realities and leads to systematic bias in time-series behavioral data. In contrast, this application proposes a "key life stage" segmentation logic, dividing the day into 6 time periods strongly tied to physiological needs (e.g., from waking up to breakfast, around lunchtime, etc.).

[0027] The creativity of this application design lies in: 1) For the first time, time division is associated with the core life scenarios of "mealing, activities and rest", so that the existence value (the frequency of occurrence in a specific area) directly reflects health-related behaviors (such as the correlation between the ability to get out of bed during mealtime and swallowing function).

[0028] 2) Compared to a fixed time window, this design allows fluctuations in the presence value at different times (such as a sharp drop in the presence value before and after dinner) to directly map health abnormalities (such as loss of appetite), thus solving the problem of the disconnect between time series data and health indicators.

[0029] 2. Hierarchical clustering algorithm: to achieve accurate profiling of behavioral patterns.

[0030] Existing algorithms (such as the improved K-means) have poor clustering performance on multidimensional behavioral data, especially when distinguishing low-intensity activities (such as sedentary behavior and light housework). They are prone to misjudgment due to noise interference and cannot establish a reliable correlation between behavioral patterns and health status.

[0031] To address the insufficient adaptability of existing clustering algorithms to non-uniform behavioral data, this application employs hierarchical clustering based on Ward's minimum variance method. Its innovation lies in: 1) Algorithm targeting: By calculating the variance increment between clusters, the K-means algorithm's clustering bias for "low-intensity activities and abnormal behavior" is resolved, enabling the six typical patterns (such as "abnormal room leaving behavior" in pattern 4) to accurately correspond to health risks.

[0032] 2) Clinical value of patterns: The patterns generated by clustering are not only a classification of behavioral characteristics, but also directly related to health risks (such as pattern 2 corresponding to end-stage diseases), realizing a leap from "behavioral description" to "health interpretation", which is a technical depth that existing clustering algorithms have not reached.

[0033] 3. Neural network verification mechanism: Constructing a two-dimensional verification of behavior and physiology.

[0034] Existing technologies for monitoring single physiological indicators are disconnected from behavioral pattern analysis. The lack of a mechanism for verifying behavioral patterns with physiological data leads to clustering results that may deviate from the actual health status, making it impossible to achieve an effective closed loop from "data collection" to "health early warning".

[0035] This application introduces neural network verification of respiratory value patterns, and its inventiveness lies in: 1) Cross-modal validation logic: A deep feedforward network is trained using 180-dimensional respiratory data (collected every 5 minutes from 6:00 to 21:00), requiring a prediction accuracy of ≥98% and a standard deviation of respiratory rate of the same mode of breathing of <1.5 times / minute. For the first time, a quantitative correlation of "behavioral pattern physiological indicators" is established to ensure the physiological rationality of the clustering results.

[0036] 2) Reliability assurance: Compared with the existing technology of "separation of behavior and physiological monitoring", this mechanism enables abnormal early warning to meet the dual conditions of "abnormal behavior characteristics" and "abnormal physiological data", which significantly reduces the false alarm rate.

[0037] 4. Dynamic monitoring closed loop: Enables time-series tracking of early anomalies.

[0038] Existing technologies mostly rely on static state assessments. This application constructs a complete monitoring closed loop through pattern evolution analysis: when an individual's pattern shifts from "healthy and active (pattern 6)" to "low activity / abnormal room absence (pattern 4)," invalid data is excluded by combining empty room markings, triggering a tiered early warning. This design of "time-series change + multi-dimensional verification" solves the problem of existing technologies lagging in identifying early characteristics of dementia and chronic diseases.

[0039] The above description is merely an overview of the technical solution disclosed herein. In order to better understand the technical means of this disclosure and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this disclosure more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0040] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a flowchart illustrating the method of the present invention; Figure 2 This is a flowchart illustrating the neural network verification process in the method of this invention; Figure 3 This is a schematic diagram illustrating the construction of the neural network model in the method of this invention; Figure 4 This is a schematic diagram of the loss curve in the method of the present invention. Detailed Implementation

[0042] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0043] It should be understood that the following specific examples illustrate the implementation of this disclosure, and those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific implementation methods, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0044] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0045] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this disclosure. The drawings only show the components related to this disclosure and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0046] Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0047] This invention discloses a method and system for monitoring abnormal health conditions in personnel based on time-based existence pattern clustering. It is applicable to scenarios such as elderly care facilities and hospital rehabilitation departments. By analyzing the lifestyle patterns and physiological data of facility residents, it achieves dynamic monitoring of health abnormalities. The following detailed description of the implementation of this invention, along with specific operational procedures and system structure, provides a comprehensive overview.

[0048] The core principle of this method lies in achieving health anomaly monitoring through a three-tiered logic of "behavioral pattern clustering, physiological data verification, and dynamic trend analysis": First, by dividing key life stages into time periods, the presence values ​​of residents in specific areas are collected to capture their daily behavioral patterns; second, a hierarchical clustering algorithm is used to extract typical life patterns, and the reliability of the patterns is verified by combining deep feedforward neural networks with respiratory value data; finally, by tracking the trend of pattern changes, health anomalies (such as wandering behavior in the early stages of dementia, exacerbation of chronic diseases, etc.) are identified. This method is suitable for dynamic monitoring of the health status of residents (especially the elderly) in scenarios such as nursing homes and community care centers.

[0049] See Figure 1 The methods for monitoring abnormal health conditions in personnel include the following steps: Step 1: Time Period Division and Data Collection Time period definition: The day is divided into 6 key life stages, each corresponding to a specific life scenario, as follows:

[0050] Data Acquisition and Preprocessing: Presence value collection: The sensor records the presence status of the occupant in a specific area (such as the head of the bed, foot of the bed, toilet, bathroom, sofa, or other designated area) every 5 minutes, and calculates the sum of the presence values ​​within each time period. For example: If the occupant is in bed from 6:00 to 7:00 (12 5-minute intervals) during time period 1 (6:00-8:30) (presence value = 12), and is out of bed from 7:00 to 8:30 (18 5-minute intervals) (presence value = 0), then the original presence value for time period 1 is 12.

[0051] Respiratory value collection: Respiratory values ​​are collected synchronously every 5 minutes from 6:00 to 21:00 (a total of 180 data points). For example, the respiratory value sequence for a certain period is [14.2, 13.8, ..., 15.1] (unit: breaths / minute).

[0052] Data cleaning: Remove outliers caused by sensor malfunctions (such as values ​​that suddenly jump to 30 and remain there for more than one day), and mark vacant room data (rooms that have been unoccupied for a long time are marked as 0 values).

[0053] Step 2: Hierarchical Cluster Analysis To avoid numerical discrepancies caused by different time periods (e.g., time period 1 is 2.5 hours, time period 2 is 2.5 hours, etc.), the sum of the values ​​for each time period is standardized or normalized to ensure that the data are on the same scale. The standardized feature vectors range between [0,1], which facilitates cluster analysis.

[0054] The data standardization formula is as follows:

[0055] For example, if the original existing value of time period 1 is 12 and the maximum possible existing value is 30, then the standardized existing value is 12 / 30 = 0.4. The standardized data range is [0,1], forming a daily existing value vector (e.g., [0.4,0.3,0.5,0.2,0.6,0.7]).

[0056] Hierarchical clustering implementation: Initialize clusters: Treat the daily existing value vectors within the monitoring period as independent clusters (if monitoring is for N days, the initial number of clusters is N), and denote the cluster set as... , where N is the initial number of clusters (i.e., the total number of days).

[0057] Distance matrix calculation: The Ward minimum variance method is used to calculate the inter-cluster merging variance increment, as shown in the following formula:

[0058] in, For clusters The number of samples; For clusters The mean center vector; For clusters The number of samples; For clusters The mean center vector; The distance is Euclidean.

[0059] Iterative cluster merging: The Ward minimum variance method in hierarchical clustering is adopted as the merging strategy. By iteratively calculating the variance increment between clusters, adaptive cluster merging is achieved to generate new clusters.

[0060] (1) Traverse all cluster pairs and select from the distance matrix that make Minimum cluster pair ; (2) Merging and Generate new clusters ; (3) Update cluster center The mean vector of the merged clusters:

[0061] (4) Recalculate the new cluster The variance increment from merging with other clusters is used to update the distance matrix.

[0062] (5) Repeat the merging process until 6 clusters are obtained, corresponding to 6 typical lifestyles (mode 1-mode 6), which are defined as follows: Pattern 1: Meal-Triggered Bedridden Pattern Behavioral characteristics: The bedridden rate is >90% during fixed mealtimes (7:30 / 12:00 / 17:30) and at other times (6:00-21:00). The presence value decreases significantly during mealtimes (e.g., 30→18), while remaining high at other times. This reflects severe limitation of basic daily living activities.

[0063] Health-related: Signs of declining physiological function or postoperative recovery require attention to the risk of muscle atrophy and pressure sores. Typical manifestations of advanced chronic disease or severe disability require attention to the risk of pressure sores and swallowing dysfunction.

[0064] Pattern 2: Chronic Bedridden with Intermittent Mobility Pattern Behavioral characteristics: The 24-hour bed rest rate is >95%, with no regular out-of-bed behavior, and the standard deviation of the observed values ​​is <5. Respiratory test values ​​remain consistently low (neural network validation means are 30% lower than those of healthy individuals). While continuously bedridden, only minimal eating-related activities are maintained (daily out-of-bed time <30 minutes), and the fluctuation range of observed values ​​is <15%.

[0065] Health-related: Severe disability or signs of end-stage disease necessitate the activation of a hospice care plan. Indications of end-stage disease or severe motor neuron damage require assessment of metabolic syndrome risk in conjunction with nutritional indicators.

[0066] Pattern 3: Low-Intensity Indoor Ambulation Pattern Behavioral characteristics: During the day, the presence of respiratory values ​​fluctuated irregularly (standard deviation > 20), with frequent short-term out-of-bed behaviors (e.g., 1-2 times per hour). Respiratory value fluctuations were positively correlated with the duration of activity. Irregular indoor movement was observed, with presence values ​​exhibiting aperiodic fluctuations; bed rest time accounted for 60-80% of the time.

[0067] Health-related: Mild cognitive impairment or anxiety tendencies require enhanced behavioral interventions to prevent falls. Behavioral markers of mild cognitive impairment or depression.

[0068] Pattern 4: Low Activity & Room Avoidance Pattern Behavioral characteristics: The presence value remained consistently low throughout the day (mean <15) and was not correlated with meal times. The neural network detected abnormally dispersed respiration values. Spatial use exhibited extreme characteristics—either consistently low activity (presence value <10 / day) or high-frequency, aimless cross-regional movement (number of movements per day >50).

[0069] Health-related: Dementia-related lingering behaviors or acute health events (such as infection and fever) trigger a Level 1 warning. Lingering symptoms of Alzheimer's disease or prodromal manifestations of delirium require initiating an assessment of dementia-related mental and behavioral symptoms.

[0070] Pattern 5: Diurnal Outing-Regularity Pattern Behavioral characteristics: During the morning period (8:30-11:00), the activity level returned to zero, and the bed rest rate was >80% after returning at noon. The respiratory rate decreased significantly after returning (Δ=-12%), consistent with the characteristics of cumulative fatigue. During the morning period (8:00-12:00), the activity level remained at zero, and stable indoor activity was observed in the afternoon.

[0071] Health-related: Early signs of compensatory decline in physical strength suggest a need for nutritional intake or cardiopulmonary function monitoring. Compensatory signs of declining physical strength while retaining instrumental activities of daily living suggest cardiopulmonary function monitoring.

[0072] Pattern 6: Healthy Mobility Pattern Behavioral characteristics: The presence value remained consistently low (mean <10) during the day (6:00-18:00), with regular bed rest at night. The fluctuation of respiratory values ​​conformed to the physiological rhythm. The mean presence value during the day (6:00-18:00) was <5, and the rhythm returned to normal at night (cosine similarity >0.8).

[0073] Health-related: Normal aging or healthy populations serve as a benchmark for the deterioration of other patterns. Healthy aging characteristics with good social participation, or behavioral projections of institutional adaptation difficulties, need to be differentiated in conjunction with psychological assessments.

[0074] Step 3: Neural Network Validation Verification objective: Since clustering is based on existing values, it is necessary to confirm its significance (i.e., whether the actual detection values ​​converge) using actual detection values ​​(average values ​​every 5 minutes). If convergence occurs, the method is considered correct.

[0075] By training a deep feedforward neural network (DNN), respiratory values ​​every 5 minutes were mapped to 6 lifestyle pattern labels generated by clustering, and the following was verified: 1) Consistency between neural network predicted labels and cluster labels (target error ≤ 2%).

[0076] 2) Convergence of respiratory values ​​under the same clustering pattern (the range of respiratory value fluctuations conforms to physiological laws).

[0077] See Figure 2 The specific verification method is as follows: 1. Data Preparation Input data format: Respiratory value data: 6:00-21:00, one data point every 5 minutes, a total of 180 dimensions (12×15 hours).

[0078] Example (breathing values ​​for a certain day): [12.5, 13.1, ..., 11.8], a total of 180 values, unit: breaths per minute.

[0079] 2. Data preprocessing: (1) Z-score standardization is used to eliminate dimensional differences in respiratory values. The standardization formula is:

[0080] in, The average respiratory rate over all days. The standard deviation of respiratory values ​​for all days.

[0081] It is important to note that Z-score standardization needs to be performed on each of the 180 values ​​to obtain the results. .

[0082] (2) Label encoding: Convert the category labels into one-hot vectors. The pattern labels (1-6) generated by hierarchical clustering need to be converted into one-hot encoding (binary encoding, where only one element in the vector is 1 to represent the current category, and the rest are 0).

[0083] For a total number of categories K=6 (patterns 1-6), if the original label is the first... kind( Its One-hot vector satisfy: ,in

[0084] get:

[0085] (3) Data set partitioning: training set (80%), validation set (20%).

[0086] 3. Neural Network Design: 1) Network Structure

[0087] Number of learning sessions: 3000 The results show that the neural network output graph, represented by the red line, is almost identical to the clustering results, therefore the clustering is considered correct. The consistency between the neural network output and the clustering results is higher than 98%, ensuring the reliability of the method.

[0088] 2) See model construction. Figure 3 The specific details are as follows: Data flow direction: ① Input: Standardized 180-dimensional respiratory data (data points every 5 minutes from 6:00 to 21:00); ②Hidden layer 1: 288 ReLU neurons process features; ③Hidden layer 2: 100 ReLU neurons further abstract features; ④ Output: 6-dimensional Softmax probability distribution (corresponding to one-hot encoding of 6 lifestyle patterns).

[0089] Training configuration: ① Training rounds: 3,000 epochs; ② Batch size: 32; ③ Optimizer: Adam (learning rate = 0.001); ④ Loss function: Categorical Cross-entropy.

[0090] Verification target: ① Prediction accuracy ≥ 98% (error ≤ 2%); ② The proportion of the diagonal in the confusion matrix is ​​>98%; ③ The standard deviation of the same mode of respiration is <1.5 breaths / minute.

[0091] Parameter calculation: Input layer → Hidden layer 1: 180×288+288=52128; Hidden layer 1 → Hidden layer 2: 288 × 100 + 100 = 28900; Hidden layer 2 → Output layer: 100×6+6=606; Total number of parameters: 52128 + 28900 + 606 = 81634.

[0092] Model training: (1) Hyperparameter settings: Training epochs: 3000.

[0093] Batch Size: 32 (balancing memory and gradient stability).

[0094] Learning rate: 0.001.

[0095] (2) Training and monitoring the cross-entropy loss of the training and validation sets.

[0096] (3) The loss curve is as follows Figure 4 As shown in the figure, the horizontal axis of the loss curve represents the number of training iterations, and the vertical axis represents the model loss function value. The graph shows that the loss function value curve steadily decreases until it stabilizes, indicating that the model training effect is relatively good.

[0097] 4. Verification and Analysis: (1) Prediction and accuracy calculation (Target ≥ 98%)

[0098] (2) Convergence verification of the respiration value of the same cluster, the specific steps are as follows: For all samples in each pattern, calculate the mean and standard deviation of the respiratory values.

[0099] Check if the standard deviation is less than the preset threshold (e.g., standard deviation of mode 1 < 1.5 times / minute).

[0100] 5. Expected Results and Verification Standards: Accuracy: Validation set prediction accuracy ≥ 98% (error ≤ 2.0%).

[0101] Convergence: The standard deviation of respiratory values ​​in the same pattern is <1.5 breaths / minute.

[0102] Confusion matrix: Diagonal elements account for >98%, and the sum of off-diagonal elements is ≤2%.

[0103] 6. Summary of key logic: Collaborative verification principle: If the clustering results are valid, the respiration values ​​under the same pattern should have similar statistical characteristics (mean, range of fluctuation).

[0104] If a neural network can learn a nonlinear mapping between respiration values ​​and pattern labels and achieve high accuracy (>98%), it proves that the clustering labels are highly correlated with physiological data.

[0105] Error analysis: If the error is >2%, it needs to be checked: Is the clustering excessively merged (e.g., different health states are grouped into the same pattern)?

[0106] Is the respiratory data affected by noise (e.g., sensor malfunction)?

[0107] The above steps can be used to systematically verify the effectiveness of clustering results and provide a reliable basis for early warning of health anomalies.

[0108] Step 4: Dynamic monitoring of health status Pattern change tracking records the daily patterns of guests in real time, analyzes the trend of pattern evolution through sliding windows (such as 3-day windows), and excludes invalid data by combining vacant room marking (0 value).

[0109] For example: Pattern 6 → Pattern 4: Health status deteriorates; Pattern 4 → Pattern 6: Health status improves; Pattern 1 → Pattern 2: Functional declines further.

[0110] Early warning triggering and handling: Tiered early warnings are triggered based on pattern changes, with the specific rules as follows:

[0111] The warning information is pushed to medical staff via the APP, including the resident ID, details of changes in the mode of life, and abnormal respiratory values.

[0112] Example 1: Monitoring of healthy elderly people (Mode 6) Subjects: Mr. Zhang (78 years old, no chronic diseases, regular daily activities) is a resident of a nursing home.

[0113] Data collection: Values ​​for each time period: [5,8,6,7,4,9] (original value) → [0.17,0.27,0.20,0.23,0.13,0.30] (standardized); Respiratory rate: 12-16 breaths / minute, with regular fluctuations (slightly faster in the morning and stable in the afternoon).

[0114] Clustering result: Pattern 6 (healthy and active).

[0115] Neural network validation: Prediction accuracy 99.3%, respiratory rate standard deviation 1.0 breaths / minute, validation passed.

[0116] Dynamic monitoring: If mode 6 is maintained for 90 consecutive days, the system determines it to be "healthy and stable" and there will be no warning.

[0117] Example 2: Early warning of dementia (Mode 4) Target: The resident, Mr. Li (82 years old, has recently been wandering around at night).

[0118] Data collection: Values ​​for each time period: [10,12,8,11,9,7] (original value) → [0.33,0.40,0.27,0.37,0.30,0.23] (standardized), which is unrelated to meal time; Respiratory rate: 10-22 breaths / minute, standard deviation 2.1 breaths / minute (abnormally dispersed).

[0119] Clustering results: Pattern 4 (low activity / abnormal off-room type).

[0120] Neural network validation: Prediction accuracy is 98.7%, meeting the validation criteria.

[0121] Dynamic monitoring: When the system triggers a Level 1 warning, medical staff assess the patient and diagnose early-stage Alzheimer's disease, initiating drug intervention and behavioral guidance.

[0122] Example 3: Postoperative rehabilitation monitoring (Mode 1 → Mode 3) Subjects: The resident, Mr. Wang (65 years old, 2 weeks after hip replacement surgery).

[0123] Postoperative week 1 data: Existing values: [28,26,27,29,25,28] (original values) → [0.93,0.87,0.90,0.97,0.83,0.93] (after standardization); Clustering results: Pattern 1 (eating and leaving the bed).

[0124] Data from week 2 post-surgery: Existing values: [15,18,16,20,14,17] (original values) → [0.50,0.60,0.53,0.67,0.47,0.57] (standardized), with a standard deviation of 22 for existing values; clustering result: pattern 3 (indoor low-intensity activity type).

[0125] Dynamic monitoring: Mode 1 → Mode 3, the system judged "good progress in rehabilitation", and the rehabilitation plan was adjusted to 30 minutes of indoor walking per day.

[0126] On the other hand, the present invention provides a system for monitoring abnormal human health, the system comprising: Data acquisition module: used to divide the time period into 6 key life stages, detect the presence status of the resident in a specific area at 5-minute intervals, and count the total presence value of each time period; Clustering Analysis Module: The clustering analysis module is deployed on the server side and implements hierarchical clustering algorithms using Python. It includes functions such as data standardization, cluster initialization, distance matrix calculation, and cluster merging, and finally generates 6 typical lifestyle patterns.

[0127] Neural network validation module: used to build and train a deep feedforward neural network to validate the consistency between clustering results and respiratory value data; Dynamic monitoring module: Used to predict health trends based on changes in lifestyle and to monitor health abnormalities.

[0128] Preferably, the data acquisition module includes: The sensor unit is used to detect presence status. Infrared sensors or millimeter-wave radars are deployed in specific areas (such as the bedside or room entrance) to detect the presence status of the occupant at 5-minute intervals (present is recorded as 1, and non-present is recorded as 0); respiratory data (unit: breaths / minute) are collected every 5 minutes from 6:00 to 21:00 and transmitted to the server via Bluetooth.

[0129] The data preprocessing unit is used to statistically analyze and standardize the existing values. It mainly cleans the original existing values ​​and respiration values ​​(removing abnormal values ​​from the sensors), calculates the sum of existing values ​​for each time period, and stores them as structured data (such as CSV format).

[0130] Preferably, the clustering analysis module includes: Initialization unit: Treat the daily existing value vectors during the monitoring period as independent clusters; Matrix calculation unit: The Ward minimum variance method is used to calculate the inter-cluster merging variance increment; Iterative merging unit: The Ward minimum variance method in hierarchical clustering is used as the merging strategy. The cluster pairs with the smallest merging variance increment are selected for merging, and the cluster center and distance matrix are updated until 6 clusters are obtained.

[0131] Preferably, the neural network verification module includes: Data standardization unit, used to perform Z-score standardization on respiratory value data; Model training unit, used to build and train neural networks; The results validation unit is used to validate the prediction accuracy and respiratory value convergence.

[0132] The dynamic monitoring and early warning module tracks changes in residents' lifestyles in real time, and combines this with vacant room markings (such as marking rooms that have been unoccupied for a long time as 0 values) to exclude invalid data. When the changes in the pattern are consistent with the characteristics of health deterioration, an early warning message is sent to medical staff via APP or SMS.

[0133] The above system is used to implement the method of monitoring abnormal health conditions in personnel. The specific steps and principles have been mentioned in the method of monitoring abnormal health conditions in personnel, so they will not be repeated here.

[0134] The basic principles of this disclosure have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this disclosure to the necessity of employing the aforementioned specific details for implementation.

[0135] It should also be noted that in the systems and methods of this disclosure, the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions to this disclosure.

[0136] Various changes, substitutions, and modifications can be made to the technology described herein without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of this disclosure is not limited to the specific aspects of the processes, machines, manufactures, events, means, methods, and actions described above. Currently existing or later-developed processes, machines, manufactures, events, means, methods, or actions that perform substantially the same function or achieve substantially the same result as the corresponding aspects described herein can be utilized. Therefore, the appended claims include such processes, machines, manufactures, events, means, methods, or actions within their scope.

[0137] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0138] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations therein.

[0139] Although the present invention has been described in detail in the specification, drawings and embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention; and all technical solutions and improvements that do not depart from the spirit and scope of the present invention are covered within the protection scope of the patent of the present invention.

Claims

1. A method for monitoring abnormal health conditions in personnel, characterized in that, Includes the following steps: Step 1: Time Segmentation and Data Collection: Divide the day into 6 key life stages: wake-up to breakfast (6:00-8:30), after breakfast to morning (8:30-11:00), around lunch (11:00-13:30), after lunch to before dinner (13:30-16:00), around dinner (16:00-18:30), and after dinner to bedtime (18:30-21:00). Check the presence status of the occupants at 5-minute intervals, recording 1 for presence and 0 for absence, and sum the presence values ​​within each time segment. Step 2: Hierarchical clustering analysis: Standardize the values ​​for each time period. The standardized feature vectors are in the range of [0,1]. Use Ward's minimum variance method for hierarchical clustering, and iteratively merge clusters until 6 typical lifestyle patterns are generated. Step 3: Neural Network Validation: Construct a deep feedforward neural network to learn the nonlinear mapping relationship between respiratory values ​​and lifestyle patterns, and verify the physiological rationality of the clustering results; Step 4: Dynamic monitoring of health status: Inferring changes in health status based on trends in lifestyle patterns to detect health abnormalities.

2. The method for monitoring abnormal health conditions in personnel according to claim 1, characterized in that, The hierarchical clustering analysis in step two includes: Initialize clusters: Treat the daily existing value vectors within the monitoring period as independent clusters, and denote the cluster set as... , where N is the initial number of clusters; Distance matrix calculation: The Ward minimum variance method is used to calculate the inter-cluster merging variance increment, as shown in the following formula: in, For clusters The number of samples; For clusters The mean center vector; For clusters The number of samples; For clusters The mean center vector; The distance is Euclidean. Iterative cluster merging: The Ward minimum variance method in hierarchical clustering is used as the merging strategy. The cluster pairs with the smallest merging variance increment are selected for merging, and the cluster center and distance matrix are updated until 6 clusters are obtained.

3. The method for monitoring abnormal health conditions in personnel according to claim 1, characterized in that, The six typical lifestyles are as follows: Mode 1 is the meal-out-of-bed type: only briefly leaving the bed during fixed meal times, with a bed rest rate >90% at other times, and the presence value drops significantly during meal times and remains high at other times; Mode 2 is characterized by prolonged bed rest with brief periods of activity: a 24-hour bed rest rate >95%, irregular bed-leaving behavior, a standard deviation of fluctuation in values ​​<5, and persistently low respiratory test values; Mode 3 is the indoor low-intensity activity type: daytime presence values ​​fluctuate irregularly, short-term bed-leaning behaviors occur frequently, and respiratory value fluctuations are positively correlated with the activity period; Mode 4 is low activity / abnormal atrial excursion: the respiratory value remains low throughout the entire time period, is not related to meal times, and the respiratory value is abnormally dispersed; Pattern 5 is characterized by regular daytime outings: values ​​return to zero during the morning, bed rest rate is >80% after returning at noon, and respiratory values ​​decrease significantly after returning; Mode 6 is the healthy and active type: the presence value remains low during the day, the patient stays in bed regularly at night, and the fluctuation of the respiratory value is in line with the physiological rhythm.

4. The method for monitoring abnormal health conditions in personnel according to claim 1, characterized in that, The specific implementation process of step three is as follows: Respiratory data were collected every 5 minutes from 6:00 to 21:00, forming a 180-dimensional vector, which was then standardized using Z-score to eliminate dimensional differences. The six typical lifestyle labels generated by hierarchical clustering are converted into one-hot encodings and used as the network output targets. The sets were divided into training and validation sets in an 8:2 ratio. A four-layer deep feedforward neural network was used, with 3000 training epochs, a batch size of 32, an optimizer of Adam, and a loss function of classification cross-entropy. The total number of parameters was 81634. Verification and result determination are carried out. If the core verification indicators are met, the lifestyle patterns generated by the clustering are consistent with the physiological data, and the results are valid. Otherwise, the clustering parameters need to be re-optimized or additional sample data needs to be added.

5. The method for monitoring abnormal health conditions in personnel according to claim 4, characterized in that, The architecture of a four-layer deep feedforward neural network includes: The input layer contains 180 neurons and is used to receive standardized 180-dimensional respiratory data. Hidden layer 1 contains 288 ReLU neurons and is used to extract basic respiratory value features; Hidden layer 2 contains 100 ReLU neurons and is used to abstract higher-order features; The output layer contains 6 Softmax neurons, which are used to output the probability distribution of the 6 modes.

6. The method for monitoring abnormal health conditions in personnel according to claim 4, characterized in that, The core validation metrics include: Prediction accuracy: ≥98% of the validation set predictions were correct; Respiratory convergence: Standard deviation of respiratory values ​​in the same mode <1.5 breaths / minute; Confusion matrix: Diagonal elements account for >98%.

7. A system for monitoring abnormal health conditions in personnel, characterized in that, The system includes: Data acquisition module: used to divide the time period into 6 key life stages, detect the presence status of the resident at 5-minute intervals, and count the total presence value of each time period; Clustering analysis module: used to standardize existing value data, and uses hierarchical clustering algorithm to generate 6 typical lifestyle patterns; Neural network validation module: used to build and train a deep feedforward neural network to validate the consistency between clustering results and respiratory value data; Dynamic monitoring module: Used to predict health trends based on changes in lifestyle and to monitor health abnormalities.

8. A personnel health abnormality monitoring system according to claim 7, characterized in that, The data acquisition module includes: The sensor unit is used to detect the presence state; The data preprocessing unit is used to count existing values ​​and perform standardization.

9. A personnel health abnormality monitoring system according to claim 7, characterized in that, The cluster analysis module includes: Initialization unit: Treat the daily existing value vectors during the monitoring period as independent clusters; Matrix calculation unit: The Ward minimum variance method is used to calculate the inter-cluster merging variance increment; Iterative merging unit: The Ward minimum variance method in hierarchical clustering is used as the merging strategy. The cluster pairs with the smallest merging variance increment are selected for merging, and the cluster center and distance matrix are updated until 6 clusters are obtained.

10. A personnel health abnormality monitoring system according to claim 7, characterized in that, The neural network verification module includes: Data standardization unit, used to perform Z-score standardization on respiratory value data; Model training unit, used to build and train neural networks; The results validation unit is used to validate the prediction accuracy and respiratory value convergence.

Images