A method for early warning and intervention on the risk of falling of the elderly

Abstract

The application discloses a kind of early warning and intervention method about old person's fall risk, including the risk profile library and risk baseline library of fall, data acquisition and preprocessing are carried out to target old person, the dynamic entropy value quantization model of fall risk is constructed to carry out risk quantification and pre-judgment and early warning, when early warning triggers, the adaptive intervention of old person's autonomous response ability is verified in combination with risk level, whole-link closed-loop intervention nursing is carried out to fall event, and the dynamic entropy value quantization model is iteratively optimized according to federal learning framework.The application can advance the early warning of target old person's fall by dynamic entropy value quantization model, realize the advance prevention before falling, can prevent falling in advance, and further cooperate adaptive intervention, the real-time autonomous response ability of old person is used as the core of intervention strategy, the risk determination precision is insufficient, false alarm and high false negative rate are solved, the rescue lag of solitary old person is avoided, and the effectiveness of intervention is improved.

Description

Technical Field

[0001] This invention belongs to the field of health monitoring and medical assistance technology, specifically relating to a method for early warning and intervention regarding the risk of falls in the elderly. Background Technology

[0002] Falls are the leading cause of injury-related death among people aged 65 and above in my country. Over 70% of falls occur at home, and more than 50% of elderly people living alone who fall are unable to receive timely assistance, greatly increasing the risk of secondary injuries, lifelong disability, or even death. As the global aging population intensifies, fall prevention is receiving increasing attention. Currently, fall monitoring, fall risk warning, and fall intervention have become key research areas.

[0003] Most existing fall monitoring devices for the elderly focus on post-fall alarms, only recognizing events that have already occurred. They cannot capture precursor signals, predict risk evolution, or provide timely warnings, resulting in significant delays and a lack of preventative capabilities. Furthermore, these devices typically employ universal fixed thresholds or static grading rules, failing to adapt to individual health differences among the elderly. They cannot address variations in gait and physiological characteristics, nor can they distinguish between falls and everyday actions such as sitting, lying down, or climbing stairs. This leads to insufficient accuracy in risk assessment, high false alarm and false alarm rates, and poor compliance among the elderly due to false alarms, while high false alarm rates pose serious safety hazards. Moreover, the fixed warning and intervention logic of these devices cannot adapt to the real-time state of the elderly, neglecting their ability to respond independently. This often results in frequent false alarms disturbing family members or delayed rescue when the elderly are unresponsive, leading to extremely low intervention effectiveness. Summary of the Invention

[0004] To address the problems existing in the prior art, a method for early warning and intervention regarding fall risk in the elderly is provided, aiming to solve the problem that existing devices cannot provide real-time intervention for fall risk based on the health differences and real-time status of different elderly people.

[0005] The technical solution adopted by this invention to solve its technical problem is: A method for early warning and intervention regarding fall risk in the elderly includes the following steps: S1. Construct a risk profile and risk baseline database for the target elderly population regarding falls and assess the corresponding initial fall risk level; S2. Collect and preprocess multi-source data from the target elderly through wearable devices; S3. Construct a dynamic entropy value quantification model for fall risk to quantify and predict fall risk and issue early warnings. The quantification of fall risk involves dividing fall risk into different risk entropies and calculating the total risk entropy value of the target elderly person in real time. Based on the total risk entropy value, fall risk is divided into different risk levels. The prediction of fall risk involves comparing the rate of change of the total risk entropy value with a preset risk escalation threshold. When the rate of change exceeds the preset risk escalation threshold, an early warning for the corresponding risk level is triggered in advance. S4. When an early warning is triggered, verify the target elderly person's autonomous response ability and conduct adaptive intervention based on the risk level.

[0006] This solution provides a closed-loop intervention and care system for fall incidents, encompassing prevention, rescue, and post-incident care. By using a dynamic entropy quantification model to derive the total risk entropy value, and comparing the rate of change of the total risk entropy value with the risk escalation threshold, it is possible to predict fall risks in advance, thereby avoiding delayed warnings and ensuring timely intervention in fall risk.

[0007] Preferably, in step S3, the risk entropy includes basic risk entropy, instantaneous behavioral risk entropy, physiological stress risk entropy, and environmental risk entropy. The risk entropy is calculated in real time, and the entropy values ​​of the risk entropy are all normalized to the 0-1 range. The formula for calculating the total risk entropy value is: Total risk entropy value = Basic risk entropy × W1 + Instantaneous behavioral risk entropy × W2 + Physiological stress risk entropy × W3 + Environmental risk entropy × W4, where W1, W2, W3, and W4 are the dynamic weights corresponding to the four risk entropies. The dynamic weights are adjusted according to the initial fall risk level of the target elderly person and the real-time scenario, and W1 + W2 + W3 + W4 = 1. The weight W1 increases as the initial fall risk level of the target elderly increases; the weight W2 increases by 0.1-0.2 when the target elderly exhibit gait abnormalities or significant trunk swaying; the weight W3 increases by 0.1-0.2 when the target elderly exhibit postural blood pressure changes or abnormal heart rate fluctuations; and the weight W4 increases by 0.05-0.15 when the target elderly are in nighttime or high-risk areas. When the rate of change of the total risk entropy exceeds the preset risk escalation threshold, it can predict that the target elderly person's risk of falling will escalate within 15-60 seconds, and trigger early warning and intervention for the corresponding risk level after the escalation, thus achieving early prediction of fall events.

[0008] In this scheme, the weights are dynamically adjusted based on the initial fall risk level and the real-time status of the target elderly person. This makes the total risk entropy value more accurate in quantifying the current fall risk, thereby improving the early prediction of risk level escalation. The warning window is moved forward from 3-5 seconds before a fall to 15-60 seconds, truly achieving proactive prevention before a fall and reducing the fall incidence rate from the root.

[0009] Preferably, the risk levels are divided into low risk, medium risk, high risk, and very high risk. Low risk corresponds to a total risk entropy value less than 0.3, indicating no risk of fall and the elderly person is in normal condition. Medium risk corresponds to a total risk entropy value greater than or equal to 0.3 and less than 0.6, indicating a potential risk of fall requiring attention and intervention. High risk corresponds to a total risk entropy value greater than or equal to 0.6 and less than 0.9, indicating an extremely high probability of fall requiring immediate intervention. Very high risk corresponds to a total risk entropy value greater than or equal to 0. 9. A fall has been determined to have occurred, requiring emergency rescue. The triggering of the extremely high risk requires the simultaneous fulfillment of three judgment rules: the peak value of the acceleration impact exceeds a preset threshold, the sudden drop in center of gravity height exceeds a preset threshold, and the locked falling posture lasts for more than a preset duration. At the same time, a manual cancellation window of 15-30 seconds is set. The target elderly person can manually cancel the alarm by pressing the one-click emergency call button. The duration of the manual cancellation window is set according to the risk level: 30 seconds for low and medium risk, 20 seconds for high risk, and 15 seconds for extremely high risk.

[0010] This solution divides the total risk entropy value into four levels: low, medium, high, and extremely high, and establishes a three-tiered judgment rule for triggering extremely high risk. This allows for accurate identification of actual falls and excludes everyday actions such as sitting down or bending over quickly, effectively preventing false triggering of fall alarms. Combined with the setting of a manual cancellation window, it ensures emergency rescue in extremely high-risk situations while protecting the elderly's right to self-correction and avoiding the waste of social resources caused by false alarms.

[0011] Preferably, in step S4, the verification process is that when the system triggers a risk warning of any level, the auxiliary terminal of the wearable device immediately initiates an autonomous response capability verification; the verification response method includes at least one of three types: voice interaction response, one-click SOS physical button response, and preset posture command feedback; the verification response time is set according to the current risk level: 15 seconds for low risk, 10 seconds for medium risk, 5 seconds for high risk, and 3 seconds for extremely high risk; The autonomous response capability is assessed in real time based on the response time and completion rate of the target elderly person within the verification window, and is divided into three levels: complete response, partial response, and no response. The classification rules for the autonomous response capability levels are as follows: complete response means completing a valid response within the first half of the verification window time; partial response means completing a valid response at the end of the verification window or only completing part of the instruction; and no response means not completing any valid response at the end of the verification window. The adaptive intervention dynamically adjusts the risk level and information push scope of the warning based on a combination of the risk level at the time of warning triggering and the level of autonomous response capability. Among them, the medium-risk combined with full response means that only the wearable device's local gentle vibration and voice reminder are triggered, without triggering external linkage; The medium-risk combination of partial response or no response: if the risk level of the warning is upgraded to high risk, the warning information will be pushed to emergency contacts and community caregivers; The high-risk combination with a complete response: triggers a strong alert from wearable devices and smart home integration, but does not trigger emergency rescue; The high-risk combination of partial response or no response: The risk level of the warning is upgraded to extremely high risk, triggering emergency rescue linkage; The extremely high risk combined with a full response: push alarm information to the first-level emergency contact and wait for the elderly person to confirm whether to escalate the rescue; The extremely high risk combined with partial or no response: immediately trigger the highest level of emergency rescue, and synchronize all information to emergency contacts, elderly care service institutions and emergency stations.

[0012] In this solution, adaptive intervention can be carried out by combining the autonomous response capability level with the current risk level. This allows the intensity of intervention and the scope of notification to be dynamically adjusted according to the actual condition of the elderly. While avoiding false alarms, it ensures that the highest level of rescue is implemented when there is a real emergency, and eliminates the misjudgment of an elderly person's coma as a low-risk situation, thus achieving accurate on-demand response.

[0013] Preferably, the intervention is further divided into pre-event intervention and in-event intervention. The pre-event intervention is for medium-risk and high-risk scenarios where a fall has not yet occurred, and is a preventative intervention. The intervention measures are implemented in three levels: the first level is real-time behavioral intervention using wearable devices, which plays targeted voice guidance to correct abnormal behaviors of the target elderly person and prevent the risk of fall from escalating; the second level is smart home linkage intervention, which uses a home edge network to link smart home devices and eliminate environmental fall triggers; the third level is long-term health intervention, which generates a fall prevention plan based on long-term total risk entropy value change data and pushes it to the target elderly person, emergency contacts, and community medical staff. The real-time behavioral intervention is voice guidance based on risk triggering reasons, and the smart home linkage intervention includes turning on all lights in high-risk areas, unlocking smart door locks, and activating smart handrails. The aforementioned in-process intervention is for scenarios with extremely high risk where a fall has already occurred. The intervention measures include: firstly, playing self-rescue guidance voice prompts on a wearable device to prevent the elderly person from blindly getting up and causing secondary injury; simultaneously collecting the elderly person's real-time vital signs data and pushing it to emergency contacts and emergency medical services; automatically dialing preset tiered emergency contact numbers in a loop, with hands-free two-way calling enabled throughout; and pushing the elderly person's precise location, home address, access control information, and core health record information to rescue personnel to shorten the rescue arrival time.

[0014] This solution prevents falls by eliminating the causes of falls beforehand through voice correction and home-based intervention; it also shortens rescue time and prevents secondary injuries by pushing accurate location and access control information during the fall, and reduces the possibility of recurrence by eliminating the potential causes of falls, rather than just providing intervention and care for a single fall event.

[0015] Preferably, in step S1, the risk database consists of all the static basic information of the target elderly person, including weight, history of falls, history of chronic diseases, history of medications that affect human balance and blood pressure regulation, history of drug allergies, and emergency contact person. The risk baseline library consists of normal threshold baselines specific to the target elderly. These normal threshold baselines are established based on daily life data collected by wearable devices on the target elderly over 7-14 consecutive days. The normal threshold baselines include kinematic baselines, physiological characteristic baselines, daily activity rhythm baselines, and home environment risk maps. The initial fall risk level is determined using the Morse Fall Assessment Scale by combining the target elderly person's static basic information with the normal threshold baseline. The initial fall risk level is the initial weight of the risk factor in the dynamic entropy value quantification model, providing a benchmark for real-time risk assessment.

[0016] In this solution, a dedicated health and risk database is established for the target elderly by using a risk profile database composed of static basic information and a normal threshold baseline established by daily data from wearable devices for 7-14 days. The initial risk level is then determined by combining the Morse Fall Assessment Scale, which makes the benchmark weight of subsequent risk calculations more suitable for the target elderly, improves the accuracy of risk assessment, and significantly reduces misjudgments and omissions caused by individual differences.

[0017] Preferably, the kinematic baseline includes normal gait parameters, resting or dynamic posture angle thresholds, trunk sway amplitude thresholds, and normal range of center of gravity height variation; the physiological characteristic baseline includes the normal resting or dynamic heart rate range, the normal heart rate variability range, the normal blood oxygen saturation threshold, the diurnal blood pressure rhythm range, and the maximum threshold for postural blood pressure variation; the daily activity rhythm baseline includes the time pattern of the target elderly person's daily activities and the activity range threshold; the home environment risk map marks high-risk areas and safe areas within the target elderly person's home environment; wherein, the normal gait parameters include cadence, stride length, gait cycle, and support phase ratio; the static or dynamic posture angles include pitch angle, roll angle, and yaw angle; the high-risk areas include the bathroom, stairwell, and kitchen, and the safe areas include the living room and bedroom.

[0018] In this solution, by refining the baseline data of kinematics, physiological characteristics, activity rhythms, and home environment risk maps, the human body state is overlaid with the environmental scene for calculation, providing a reference for subsequent risk entropy calculation and enhancing the sensitivity of fall risk identification.

[0019] Preferably, in step S2, the wearable device consists of a waist-to-abdomen main terminal and a wristband-style auxiliary terminal. The main terminal is worn at the waist and abdomen, integrating a 9-axis IMU inertial measurement unit, a barometer, a dual-mode positioning module, and a one-button emergency call physical button. The auxiliary terminal is worn on the wrist, integrating a PPG photoplethysmography sensor, a skin conductance sensor, a voice interaction module, and a sound pickup and playback module. The main terminal and the auxiliary terminal synchronize with millisecond-level timestamps to complete the synchronous acquisition of multi-sensor data. The timestamp synchronization accuracy between the main terminal and the auxiliary terminal is not less than 10ms. The 9-axis IMU inertial measurement unit includes a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer. The dual-mode positioning module includes GPS, a Beidou outdoor positioning module, and an indoor UWB positioning module. The kinematic data acquired by the 9-axis IMU inertial measurement unit is denoised using a Kalman filter algorithm, and the physiological signals acquired by the PPG sensor are smoothed and denoised using a moving average filter algorithm to remove invalid data caused by sensor jitter and poor contact. The wearable device can perform timestamp alignment and numerical normalization of multi-sensor data. The numerical normalization process uses the min-max normalization method to perform real-time difference calibration between the preprocessed data and the risk baseline library, identifying abnormal scenarios such as device detachment, abnormal wearing, and signal interruption, and triggering local wearing reminders. The corresponding abnormal data is not included in subsequent risk assessment calculations.

[0020] In this solution, wearable devices consisting of a main terminal on the waist and an auxiliary terminal on the wristband enable the simultaneous acquisition of inertial kinematic data and physiological data such as photoplethysmography (PPG) pulse waves and electrodermal activity. Furthermore, invalid data such as device detachment and poor contact are identified and eliminated through Kalman filtering and moving average filtering algorithms, thus preventing false alarms caused by wearing issues at the source. This ensures the accuracy of subsequent model input data and improves the reliability of subsequent risk assessments.

[0021] Preferably, the basic risk entropy is calculated based on the target elderly person's risk profile and is a static basic risk value. The update cycle is 1 month or when the target elderly person's health status undergoes a significant change. The basic risk entropy is obtained by linearly normalizing the score results of the Morse Fall Assessment Scale to the 0-1 interval. The instantaneous behavioral risk entropy is obtained by calculating the degree of deviation from the kinematic baseline and the rate of change of the time series based on real-time kinematic data collected by a 9-axis IMU. The update frequency is 100Hz. The calculation of instantaneous behavioral risk entropy includes four core features: gait parameter deviation, posture angle deviation, trunk sway amplitude, and center of gravity height change rate. The physiological stress risk entropy is calculated based on real-time physiological data collected by PPG and skin conductance sensors, and the degree of deviation and rate of change of the data from the physiological characteristic baseline is calculated. The update frequency is 20Hz. The calculation of physiological stress risk entropy includes five core features: heart rate deviation, heart rate variability change amplitude, postural blood pressure change amplitude, blood oxygen saturation deviation, and skin conductance response change amplitude. The environmental risk entropy is calculated based on the real-time location data of the dual-mode positioning module, the home environment risk map, and the degree of danger in the current scene, with an update frequency of 1Hz.

[0022] This solution breaks down fall risk into four categories of risk entropy: basic risk, instantaneous behavioral risk, physiological stress risk, and environmental risk. The total risk entropy value is calculated based on archival information, degree of movement deviation, degree of physiological deviation, and current environment, so that it can encompass all causes of falls and more comprehensively and in real time reflect the risk of falls in the elderly.

[0023] Preferably, the iterative optimization of the dynamic entropy quantification model adopts a federated learning framework to construct a model iteration system, and uses real fall event data and intervention effect feedback data to optimize the weights and judgment thresholds of the dynamic entropy quantification model of fall risk. The federated learning framework adopts a horizontal federated learning architecture, where each terminal completes model training locally and only uploads the encrypted model gradient parameters to the cloud for aggregation, without transmitting the original privacy data throughout the process. The entire process of federated learning uses end-to-end encrypted transmission of data, and cloud storage uses de-identified encryption processing. The original data is preferentially stored locally on the wearable device and at the home edge gateway. The cloud only uploads encrypted feature data and does not store the original privacy data of the target elderly, providing the target elderly with complete control over the data and supporting the ability to turn off data collection, adjust warning and linkage permissions, and delete data stored in the cloud at any time.

[0024] In this solution, the federated learning framework enables continuous cross-device evolution of the model without disclosing the privacy of the elderly’s original movement, physiology, location, etc., breaking the dilemma that accurate models must sacrifice privacy. This allows the model to continuously iterate and optimize using fall and intervention feedback data, while ensuring the user’s control over the data.

[0025] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention uses a dynamic entropy quantification model and the rate of change of total risk entropy to predict the trend of risk evolution, providing early warning of falls in the target elderly and allowing time for subsequent intervention. This achieves proactive prevention before falls occur, reducing the incidence of falls at the source. It can prevent falls from happening in advance, going beyond just providing quick rescue after a fall. Combined with adaptive intervention based on the elderly's self-response ability and risk level, it takes the elderly's real-time self-response ability as the core of the intervention strategy, achieving dynamic and precise intervention. This solves the problems of insufficient risk assessment accuracy and high false alarm and false negative rates in traditional solutions when considering the different health differences and real-time status of the elderly. It also avoids the problem of delayed rescue for elderly people living alone due to lack of response, greatly improving the effectiveness of intervention for different elderly people.

[0026] 2. This invention adapts to the individual health differences of the elderly through a full-process early warning and intervention approach, achieving precise quantification of fall risk and significantly reducing false alarm and false negative rates. Simultaneously, the combination of a dynamic entropy quantification model and federated learning constructs a personalized fall prevention system that becomes increasingly accurate and tailored with use, as well as a personalized risk assessment system that is unique to each individual. Each fall event and its intervention effect are used to optimize the weights and prediction thresholds of the dynamic entropy model. Furthermore, the rehabilitation data generated afterward updates the individual risk baseline. This allows the system's perception of fall risk for a specific elderly person to become increasingly accurate over time, and intervention measures to become more aligned with their physical and mental changes, continuously strengthening the preventative effect and avoiding the rigidity of interventions or their inability to adapt to the elderly person's real-time condition.

[0027] 3. This invention constructs a closed-loop management system for the entire process through pre-event early warning intervention, in-event rescue intervention, and model iteration. It not only realizes the closed loop of early warning, intervention, and rescue for a single fall risk event, but also realizes the iterative closed loop of data collection, model evaluation, scheme optimization, and effect feedback for long-term risk management. This not only reduces the occurrence of fall events, but also reduces the risk of secondary injury and recurrence after a fall. It can continuously reduce the basic fall risk and fall recurrence rate of the elderly, and realize the full-cycle management of fall risk in the elderly. Attached Figure Description

[0028] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a schematic diagram of the overall process in this invention; Figure 2 This is a schematic diagram of the adaptive intervention process in this invention; Detailed Implementation

[0029] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar components or components having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0030] like Figure 1-2 As shown, this embodiment provides a method for early warning and intervention regarding the risk of falls in the elderly, including the following steps: S1. Construct a risk profile and baseline database of falls for the target elderly population and assess the corresponding initial fall risk level; this serves as a benchmark for real-time fall risk assessment and provides a basis for subsequent real-time fall risk assessments. S2. Collect and preprocess multi-source data from the target elderly through wearable devices; S3. Construct a dynamic entropy value quantification model for fall risk to quantify and predict fall risk and issue early warnings. The quantification of fall risk involves dividing fall risk into different risk entropies and calculating the total risk entropy value of the target elderly person in real time. Based on the total risk entropy value, fall risk is divided into different risk levels. The prediction of fall risk involves comparing the rate of change of the total risk entropy value with the preset risk jump threshold and issuing early warnings for the corresponding risk level based on the comparison results. S4. When an early warning is triggered, verify the target elderly person's autonomous response ability and conduct adaptive intervention based on the risk level.

[0031] This solution provides a closed-loop intervention and care system for fall incidents, encompassing prevention, rescue, and post-incident care. By using a dynamic entropy quantification model to derive the total risk entropy value, and comparing the rate of change of the total risk entropy value with the risk escalation threshold, it is possible to predict fall risks in advance, thereby avoiding delayed warnings and ensuring timely intervention in fall risk.

[0032] In step S1, the risk profile database consists of all the static basic information of the target elderly person. The static basic information includes at least age, height, weight, history of falls, history of chronic diseases, medication history that affects balance and blood pressure regulation, history of drug allergies, emergency contact level information, information of the local community elderly care service institution, and information of the nearest medical emergency institution. The risk profile database constructs standardized and interoperable full-dimensional health data. The history of falls includes the number of falls, the time of falls, the cause of falls, and the degree of injury. The history of chronic diseases includes diseases that affect balance, such as hypertension, diabetes, Parkinson's disease, stroke, osteoporosis, and bone and joint diseases. The medication history includes the names, dosages, and times of administration of drugs such as sedatives, antihypertensives, hypoglycemics, antidepressants, and diuretics. The emergency contact level information is divided into three levels: first level: immediate family members; second level: collateral family members / neighbors; and third level: community police officers / emergency personnel, including contact information, residential distance, and response priority. The information of the nearest medical emergency institution includes address, contact number, and emergency response range.

[0033] The risk baseline database consists of a personalized normal threshold baseline system specific to the target elderly. The personalized normal threshold baseline system is established based on daily life data collected by wearable devices on the target elderly over 7-14 consecutive days, excluding data from abnormal periods such as illness or physical discomfort. The personalized normal threshold baseline system includes at least kinematic baseline, physiological characteristic baseline, daily activity rhythm baseline, and home environment risk map.

[0034] The initial fall risk level is determined using the Morse Fall Assessment Scale, which combines the target elderly person's static basic information with a personalized normal threshold baseline system. The scale score ranges from 0 to 125 points, corresponding to four levels: low risk (<25 points), medium risk (25-44 points), high risk (45-64 points), and very high risk (≥65 points). The initial fall risk level serves as the initial weight for multi-dimensional risk factors in the subsequent risk quantification model, providing a benchmark for real-time risk assessment.

[0035] In this solution, a dedicated health and risk database is established for the target elderly by using a risk profile database composed of static basic information and a normal threshold baseline established by daily data from wearable devices for 7-14 days. The initial risk level is then determined by combining the Morse Fall Assessment Scale, which makes the benchmark weight of subsequent risk calculations more suitable for the target elderly, improves the accuracy of risk assessment, and significantly reduces misjudgments and omissions caused by individual differences.

[0036] The kinematic baseline includes normal gait parameters, resting or dynamic posture angle thresholds, trunk sway amplitude thresholds, normal range of center of gravity height variation, and kinematic characteristic thresholds for daily activities. Among these, normal gait parameters include at least cadence, stride length, gait cycle, stance phase ratio, and swing phase ratio; resting or dynamic posture angles include the normal range of pitch angle, roll angle, and yaw angle; and daily activities include sitting down, lying down, getting up, and climbing stairs.

[0037] The baseline physiological characteristics include the normal range of resting or dynamic heart rate, the normal range of heart rate variability, the normal threshold of blood oxygen saturation, the diurnal rhythm range of blood pressure, the maximum threshold of postural blood pressure change, and the normal range of resting skin conductance.

[0038] The daily activity rhythm baseline includes the time pattern of the target elderly person's daily activities, the activity range threshold, the activity movement pattern in the home environment, and the daily routine.

[0039] The home environment risk map marks high-risk and safe areas in the target elderly’s home environment, based on the target elderly’s home layout and daily activity data; high-risk areas include at least the bathroom, stairwell and kitchen, and safe areas include at least the living room and bedroom. It also marks environmental risk information such as obstacles, anti-slip conditions and handrail installation locations in each area, and classifies them into corresponding danger levels.

[0040] In this solution, by refining the baseline data of kinematics, physiological characteristics, activity rhythms, and home environment risk maps, the human body state is overlaid with the environmental scene for calculation, providing a reference for subsequent risk entropy calculation and enhancing the sensitivity of fall risk identification.

[0041] In step S2, this step provides high-quality data input for the method, ensuring the comprehensiveness, synchronization, and accuracy of data collection. It also enables basic local functionality in scenarios without network connectivity. The wearable device consists of a waist-worn main terminal and a wrist-worn auxiliary terminal. The main terminal is worn at the waist and abdomen, integrating a 9-axis IMU inertial measurement unit, a barometer, a dual-mode positioning module, and a one-button emergency call function. The auxiliary terminal is worn on the wrist, integrating a PPG photoplethysmography sensor, a skin conductance sensor, a voice interaction module, and a sound pickup and playback module. The main terminal and the auxiliary terminal... The terminal achieves millisecond-level timestamp synchronization via Bluetooth 5.0 and above protocols, enabling simultaneous acquisition of data from multiple sensors. The waist and abdomen, specifically the navel, is the optimal monitoring point for the human body's center of gravity, resulting in the highest fall detection accuracy. The 9-axis IMU inertial measurement unit includes a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer. The dual-mode positioning module includes GPS, BeiDou outdoor positioning, and indoor UWB positioning modules. The timestamp synchronization accuracy between the main and auxiliary terminals is no less than 10ms. The wearable combination architecture of the waist-worn main terminal and the wristband-style auxiliary terminal enables simultaneous acquisition of multi-source data.

[0042] Specifically, the kinematic data acquired by the 9-axis IMU inertial measurement unit is denoised using a Kalman filter algorithm to eliminate signal deviations caused by daily human body shaking, sensor hardware noise, and environmental interference. An adaptive extended Kalman filter is used, and a moving average filter algorithm is applied to smooth and denoise the physiological signals acquired by the PPG sensor, removing outliers and invalid data caused by sensor shaking and poor contact. The window length of the moving average filter algorithm is set to 5-10 sampling points. At the wearable device, timestamp alignment and numerical normalization of multi-sensor data are performed. The numerical normalization uses a min-max normalization method to map all acquired data to the 0-1 range, eliminating calculation deviations caused by differences in the dimensions of different sensors. Real-time difference calibration is performed between the preprocessed data and the personalized normal threshold baseline system in the risk baseline library to calculate the deviation between the real-time data and the baseline. Abnormal scenarios such as device detachment, abnormal wearing, signal interruption, and low battery are identified in real time, triggering local vibration and voice wearing reminders. Corresponding abnormal data is not included in subsequent risk assessment calculations to avoid misjudgments caused by invalid data. The wearable device completes the preprocessing of multi-source data.

[0043] In this solution, wearable devices consisting of a main terminal on the waist and an auxiliary terminal on the wristband enable the simultaneous acquisition of inertial kinematic data and physiological data such as photoplethysmography (PPG) pulse waves and electrodermal activity. Furthermore, invalid data such as device detachment and poor contact are identified and eliminated through Kalman filtering and moving average filtering algorithms, thus preventing false alarms caused by wearing issues at the source. This ensures the accuracy of subsequent model input data and improves the reliability of subsequent risk assessments.

[0044] In step S3, the unquantifiable risk of falling is decomposed into four risk entropies that can be calculated in real time and are traceable. The risk entropies are calculated in real time. The four risk entropies are the basic risk entropy, the instantaneous behavioral risk entropy, the physiological stress risk entropy, and the environmental risk entropy. The entropy values ​​of the four risk entropies are all normalized to the 0-1 range. The higher the entropy value, the greater the risk of falling in the corresponding dimension.

[0045] The basic risk entropy is calculated based on the target elderly person's comprehensive risk profile and Morse Fall Assessment Scale score. It is a static basic risk value and is updated every month or immediately when there are significant changes in the target elderly person's health status, such as a new history of falls, worsening of chronic diseases, or adjustment of medication regimen. The basic risk entropy is obtained by linearly normalizing the Morse Fall Assessment Scale score to the 0-1 range. The higher the score, the greater the basic risk entropy value.

[0046] Instantaneous behavioral risk entropy is calculated based on real-time kinematic data acquired by a 9-axis IMU. It calculates the degree of deviation from the personalized kinematic baseline and the rate of change of the time series, normalizes it to the 0-1 range, and updates it at a frequency of 100Hz to ensure real-time capture of behavioral risks. The calculation of instantaneous behavioral risk entropy includes at least four core features: gait parameter deviation, posture angle deviation, trunk sway amplitude, and rate of change of center of gravity height.

[0047] Physiological stress risk entropy is calculated based on real-time physiological data collected by PPG and skin conductance sensors. It calculates the degree of deviation and rate of change of the data from the personalized physiological characteristic baseline, normalizes it to the 0-1 range, and updates it at a frequency of 20Hz. It matches the change characteristics of physiological signals. The calculation of physiological stress risk entropy includes at least five core features: heart rate deviation, heart rate variability, postural blood pressure, blood oxygen saturation deviation, and skin conductance response.

[0048] The environmental risk entropy is calculated based on the real-time location data of the dual-mode positioning module and the home environment risk map. It is combined with the degree of danger of the current scene, the activity time, and the deviation of the activity range to calculate the environmental risk value, which is then normalized to the 0-1 range. When in a high-risk area, during non-activity hours at night, or outside the daily activity range, the environmental risk entropy value increases accordingly. The update frequency is 1Hz, matching the update characteristics of the location data.

[0049] This solution breaks down fall risk into four categories of risk entropy: basic risk, instantaneous behavioral risk, physiological stress risk, and environmental risk. The total risk entropy value is calculated based on archival information, degree of movement deviation, degree of physiological deviation, and current environment, so that it can encompass all causes of falls and more comprehensively and in real time reflect the risk of falls in the elderly.

[0050] In step S3, the formula for calculating the total risk entropy value is: Total risk entropy value = Basic risk entropy × W1 + Instantaneous behavioral risk entropy × W2 + Physiological stress risk entropy × W3 + Environmental risk entropy × W4, where W1, W2, W3, and W4 are the dynamic weights corresponding to the four risk entropies. The dynamic weights are adaptively adjusted according to the initial fall risk level of the target elderly person and the real-time scenario, and W1 + W2 + W3 + W4 = 1.

[0051] The adaptive adjustment rule for dynamic weights W1-W4 is as follows: the higher the initial fall risk level, the greater the weight of W1 corresponding to the base risk entropy; in nighttime scenarios and high-risk area scenarios, the weight of W4 corresponding to the environmental risk entropy increases by 0.05-0.15; when a sudden change in postural blood pressure or abnormal fluctuation in heart rate is detected, the weight of W3 corresponding to the physiological stress risk entropy increases by 0.1-0.2; when abnormal gait or large trunk swaying is detected, the weight of W2 corresponding to the instantaneous behavioral risk entropy increases by 0.1-0.2, and the weights are always adjusted to maintain W1+W2+W3+W4=1.

[0052] The system can predict the evolution trend of fall risk in advance and calculate the time series change rate of the total risk entropy value in real time. When the change rate of the total risk entropy value exceeds the preset risk leap threshold, it can predict that the fall risk of the target elderly person will leap in risk level within 15-60 seconds and trigger the corresponding risk level warning and intervention in advance, realizing the pre-event prediction of fall events and completely breaking through the limitation of existing technologies that can only provide alarms after the fact.

[0053] This solution breaks down fall risk into four categories of risk entropy: basic risk, instantaneous behavioral risk, physiological stress risk, and environmental risk. The total risk entropy value is calculated based on archival information, degree of movement deviation, degree of physiological deviation, and current environment, so that it can encompass all causes of falls and more comprehensively and in real time reflect the risk of falls in the elderly.

[0054] Risk levels are categorized based on total risk entropy values, into low, medium, high, and very high risk. The specific risk levels are as follows: Low risk (0-0.3 total risk entropy, less than 0.3, considered no risk of fall, elderly person in normal condition); Medium risk (0.3-0.6 total risk entropy, greater than or equal to 0.3 and less than 0.6, considered potential risk of fall requiring intervention); High risk (0.6-0.9 total risk entropy, greater than or equal to 0.6). If the total risk entropy is less than 0.9, it is considered to have an extremely high probability of falling and requires immediate intervention; extremely high risk corresponds to a total risk entropy value of 0.9-1.0, that is, a total risk entropy value greater than or equal to 0.9, and is considered to have occurred, requiring emergency rescue; among them, the triggering of extremely high risk requires the simultaneous fulfillment of three superimposed judgment rules: the peak value of acceleration impact exceeds the preset threshold, the sudden drop in center of gravity height exceeds the preset threshold, and the ground posture lock lasts for more than the preset duration. At the same time, a manual cancellation window of 15-30 seconds is set, and the target elderly person can manually cancel the alarm by pressing the one-click emergency call button.

[0055] The preset thresholds for this novel triple-combination judgment rule are: the preset threshold for the peak value of acceleration impact is 6g, the preset threshold for the sudden drop in center of gravity height is 30cm, and the preset threshold for the duration of the fall posture lock is 500ms; the duration of the manual cancellation window is set according to the risk level, set to 30 seconds for low and medium risk, 20 seconds for high risk, and 15 seconds for extremely high risk.

[0056] This solution divides the total risk entropy value into four levels: low, medium, high, and extremely high, and establishes a three-tiered judgment rule for triggering extremely high risk. This allows for accurate identification of actual falls and excludes everyday actions such as sitting down or bending over quickly, effectively preventing false triggering of fall alarms. Combined with the setting of a manual cancellation window, it ensures emergency rescue in extremely high-risk situations while protecting the elderly's right to self-correction and avoiding the waste of social resources caused by false alarms.

[0057] In step S4, another core inventive technical point of the present invention is to incorporate the real-time autonomous response ability of the target elderly into the core judgment dimension of intervention, so as to realize the dynamic adaptive adjustment of the early warning strategy. The verification process is that when the system triggers a risk warning of any level, the auxiliary terminal of the wearable device immediately initiates the autonomous response ability verification. The verification method includes at least one of three types: voice interaction response, one-click emergency call physical button response, and preset posture command feedback. The duration of the verification window is dynamically set according to the current risk level.

[0058] The verification window duration is set as follows: 15 seconds for low risk, 10 seconds for medium risk, 5 seconds for high risk, and 3 seconds for extremely high risk. The preset posture instruction feedback can be set to multiple consecutive nodding or raising hand actions, usually set to three times, to suit elderly people with limited hand mobility.

[0059] Autonomous response capability is assessed in real time based on the response time and completion rate of the target elderly person within the verification window, and is divided into three levels: complete response, partial response, and no response. The classification rules for autonomous response capability levels are as follows: Complete response means completing an effective response within the first half of the verification window, being able to clearly understand and execute response instructions, and is judged as being conscious and having full autonomous action and distress call capabilities; Partial response means completing an effective response or only completing some instructions at the end of the verification window, and is judged as being confused, having limited mobility, and only having partial autonomous response capability; No response means failing to complete any effective response at the end of the verification window, and is judged as being unconscious, having completely limited mobility, and having no autonomous distress call capability.

[0060] Adaptive intervention dynamically adjusts the risk level and information dissemination scope of an early warning based on a combination of the risk level triggered by the warning and the level of autonomous response capability. The execution rules for adaptive intervention are as follows: Medium-risk combined with full response: only triggers local gentle vibration and voice reminders on the wearable device, without triggering external linkages, to avoid unnecessary disturbances; Medium risk combined with partial response or no response: If the risk level of the warning is upgraded to high risk, the warning information is pushed to the first-level emergency contact and community caregiver, along with the real-time location of the target elderly person, the cause of the risk, and a remote two-way communication channel is opened. High-risk combined with full response: Triggering strong alerts from wearable devices and smart home linkages, such as continuous vibration, bright lights, and loud voice guidance on the wearable device side, and eliminating environmental risks through linkage with smart home devices via the home edge gateway without triggering emergency rescue; High risk combined with partial or no response: The risk level of the warning is upgraded to extremely high risk, triggering emergency rescue coordination; Extremely high risk combined with full response: push alarm information to the first-level emergency contact, play self-rescue guidance voice in a loop, wait for the elderly to confirm whether to upgrade rescue; if the target elderly confirms that no rescue is needed, the warning will be lifted; if there is no confirmation within 10 minutes, the full-link rescue linkage will be automatically upgraded. Extremely high risk combined with partial or no response: Immediately trigger the highest level of emergency response, and synchronize all information to emergency contacts, elderly care service institutions and emergency stations.

[0061] In this solution, adaptive intervention can be carried out by combining the autonomous response capability level with the current risk level. This allows the intensity of intervention and the scope of notification to be dynamically adjusted according to the actual condition of the elderly. While avoiding false alarms, it ensures that the highest level of rescue is implemented when there is a real emergency, and eliminates the misjudgment of an elderly person's coma as a low-risk situation, thus achieving accurate on-demand response.

[0062] In step S4, this step is the core value closed loop of the present invention, realizing the upgrade from passive alarm to active intervention and full-cycle care. The intervention is further divided into pre-intervention and in-process intervention. Pre-intervention belongs to the preventive intervention closed loop. Pre-intervention is for medium-risk and high-risk scenarios where a fall event has not occurred, and implements three-level progressive intervention measures to prevent the escalation of fall risk from the source.

[0063] The first level is real-time behavioral intervention using wearable devices. Wearable devices play targeted voice guidance based on the current risk triggering cause to correct abnormal behavior of the target elderly and prevent the risk of falls from escalating. For example, when there is a risk of orthostatic hypotension, the guidance is "Please get up slowly, sit quietly for 30 seconds before standing up", and when there is a risk of gait instability, the guidance is "You are walking unsteadily now, please slow down and sit down to rest while holding onto a fixed object nearby". The second level is smart home linkage intervention, which uses the home edge network to link home smart devices to eliminate environmental factors that may cause falls. This includes at least automatic full-on lighting in high-risk areas, automatic unlocking of smart door locks, automatic activation of smart handrails, and voice broadcast of anti-slip warnings for the ground. At the same time, it turns off smart devices related to ground obstacles to prevent the environmental risks from escalating. The third level is long-term health intervention, which generates personalized fall prevention plans based on long-term total risk entropy value changes. The plans include at least balance and muscle strength training programs, chronic disease management programs, medication adjustment reminders, and home aging-in-place modification programs. These plans are regularly sent to the target seniors, emergency contacts, and community healthcare workers to continuously reduce the basic fall risk of the target seniors.

[0064] Real-time behavioral intervention is based on personalized voice guidance triggered by risk; smart home linkage intervention includes at least turning on all lights in high-risk areas, unlocking smart door locks, and activating smart handrails.

[0065] In-process intervention is part of the emergency rescue closed loop. It is specifically designed for extremely high-risk scenarios where a fall has already occurred. The rescue measures involve implementing the entire emergency response process to minimize secondary injuries from the fall and shorten the rescue response time. First, play self-rescue guidance voice messages in a loop through wearable devices, such as "You have fallen. Please do not get up casually to avoid secondary injury. Rescue personnel are on their way. Please keep your breathing steady." This will prevent the elderly person from blindly getting up, which could lead to further fractures, brain hemorrhage, or other secondary injuries. At the same time, the real-time vital signs data of the target elderly person are continuously collected, such as real-time heart rate and blood oxygen, and pushed to emergency contacts and emergency medical institutions in real time to help rescuers predict the severity of the injury and prioritize the dispatch of emergency medical resources. It automatically and repeatedly dials preset emergency contact numbers at different levels, with hands-free two-way calling enabled throughout, facilitating remote communication between the target elderly person calling for help and rescue personnel. The system pushes the precise location, home address, access control information, and core health record information of the target elderly person to rescue personnel, shortening the rescue arrival time.

[0066] This also includes post-incident care, which is part of the rehabilitation care closed loop. Post-incident care addresses the situation after the fall incident has been handled. The care measures involve implementing full-cycle rehabilitation management to reduce the risk of fall recurrence and form a long-term care closed loop. The system automatically generates a standardized fall event review report, which clearly identifies the time, location, triggering cause, changes in physiological data, and the entire rescue and treatment process. The report is simultaneously sent to the target elderly person's attending physician, community caregiver, and emergency contact person to pinpoint the root cause of the fall. Based on the review report, the community doctor, rehabilitation therapist and caregiver will develop a personalized rehabilitation care plan, including postoperative rehabilitation training, osteoporosis intervention plan, chronic disease management plan adjustment, medication plan optimization, psychological counseling and home aging-friendly modification suggestions. Community caregivers regularly visit elderly residents according to the plan to monitor their gait, balance, and physiological indicators, update the risk baseline database and risk level of the target elderly residents, and dynamically adjust the intervention plan to form a closed loop of long-term care. Regularly conduct health education on fall prevention to enhance the self-protection awareness of the target elderly and reduce the risk of fall recurrence from the root.

[0067] This solution aims to prevent falls by eliminating the causes of falls beforehand through voice correction and home-based intervention; to shorten rescue time and prevent secondary injuries by pushing accurate location and access control information during the fall; and to transform passive treatment into proactive rehabilitation care by generating a post-fall review report and adjusting the care plan in conjunction with community doctors. It also eliminates the hidden dangers that cause falls, reducing the possibility of recurrence, rather than just intervening in a single fall event.

[0068] The iterative optimization of the dynamic entropy quantification model employs a federated learning framework to construct the model iteration system. This system achieves joint incremental learning of multi-terminal models without disclosing the original privacy data of each terminal. Specifically, each wearable terminal trains its model locally, uploading only encrypted model gradient parameters to the cloud aggregation server. After global model aggregation is completed on the cloud, the updated model parameters are distributed to each terminal device. Real-world fall event data and intervention effect feedback data are used to optimize the weights and judgment thresholds of the multi-dimensional fall risk dynamic entropy quantification model, thereby continuously improving model accuracy and reducing false positive and false negative rates. The federated learning framework adopts a horizontal federated learning architecture, where each terminal trains its model locally, uploading only encrypted model gradient parameters to the cloud aggregation server, without transmitting any original privacy data throughout the process.

[0069] Federated learning employs end-to-end encrypted transmission of all data, and de-identified encryption for cloud storage. Raw data is preferentially stored locally on wearable devices and at the home edge gateway. Only encrypted feature data is uploaded to the cloud, without storing the target elderly person's original privacy data. The target elderly person is given complete control over their data, and can turn off data collection, adjust alerts and linkage permissions, and delete data stored in the cloud at any time. The entire process complies with the relevant laws and regulations on the management of sensitive medical and health data.

[0070] In this solution, the federated learning framework enables continuous cross-device evolution of the model without disclosing the privacy of the elderly’s original movement, physiology, location, etc., breaking the dilemma that accurate models must sacrifice privacy. This allows the model to continuously iterate and optimize using fall and intervention feedback data, while ensuring the user’s control over the data.

[0071] This plan uses Mr. Wang, a 78-year-old elderly man living alone with hypertension, as an example. It is applied to the scenario of home-based elderly care in an urban community. Mr. Wang has a history of one fall at home and has symptoms of orthostatic hypotension, as detailed below: Step 1: Construct a risk profile and risk baseline database for the target elderly population regarding falls and assess the corresponding initial fall risk level; 1. Collect all static basic information of Mr. Wang to construct a standardized, multi-dimensional health record related to fall risk: age 78, height 165cm, weight 62kg, history of one fall in the bathroom in 2024, history of hypertension for 12 years, long-term use of nifedipine antihypertensive drug in the morning, no history of drug allergies, emergency contact hierarchy is: first level daughter (living in the same city, 10 minutes away), second level nephew, third level community police officer; the linked institutions are the local XX Street Community Elderly Care Service Center, the nearby XX Community Health Service Center, and the XX Hospital Emergency Station.

[0072] 2. Have Mr. Wang wear a waist-mounted main terminal and a wristband-style auxiliary terminal to complete 10 consecutive days of daily full-scenario data collection and establish a personalized normal threshold baseline system.

[0073] Kinematic baseline: normal cadence 85±5 steps / minute, stride length 45±3cm, gait cycle 1.1±0.1s, stance phase ratio 62±3%; resting posture angles: pitch ±5°, roll ±3°, yaw ±10°; trunk sway amplitude ≤8°; normal range of center of gravity height variation ±15cm.

[0074] Baseline physiological characteristics: resting heart rate 65±8 beats / minute, normal range of heart rate variability 30-70ms, blood oxygen saturation ≥95%, morning systolic blood pressure 135±10mmHg, maximum threshold of postural blood pressure change ≤20mmHg.

[0075] Daily activity rhythm baseline: 6:00-21:00 is the main activity time, 22:00-5:00 the next day is the rest time, and the daily activity range is within the residential community.

[0076] Home environment risk map: The bathroom, kitchen, and balcony are marked as high-risk areas, while the living room and bedroom are marked as safe areas.

[0077] 3. The Morse Fall Assessment Scale was used to complete the initial risk level assessment. Wang scored 65 points, corresponding to a high initial risk level. The initial weights were set for the multi-dimensional fall risk dynamic entropy value quantification model: W1=0.2, W2=0.4, W3=0.25, W4=0.15.

[0078] Step 2: Collect and preprocess multi-source data from the target elderly using wearable devices; 1. The wearable device adopts a combined architecture of a main terminal on the waist and abdomen and an auxiliary terminal on the wristband. The main terminal is worn on Wang's waist and abdomen at the navel, the optimal monitoring point of the body's center of gravity. It integrates a 9-axis IMU inertial measurement unit, a high-precision barometer, a Beidou plus indoor UWB dual-mode positioning module, and a one-button emergency call physical button. The 9-axis IMU inertial measurement unit consists of a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer. The auxiliary terminal is worn on Wang's left wrist and integrates a PPG photoplethysmography sensor, a skin conductance sensor, a voice interaction module, and a sound pickup and playback module. The main and auxiliary terminals achieve 5ms-level timestamp synchronization via Bluetooth 5.2 protocol to complete the synchronous acquisition of multi-sensor data. 2. Lightweight data preprocessing at the device side: Adaptive extended Kalman filtering is used to reduce noise in the kinematic data acquired by the 9-axis IMU, eliminating interference from human body shaking and sensor noise; a moving average filtering algorithm with a window length of 8 is used to smooth the PPG physiological signal and remove outliers caused by poor contact; timestamp alignment of multi-sensor data is completed, and the min-max normalization method is used to map all data to the 0-1 interval, and real-time difference calibration is performed with the personalized baseline; real-time identification of normal device wearing and stable signal is performed, and all valid data are included in the subsequent risk assessment.

[0079] Step 3: Construct a dynamic entropy value quantification model for fall risk to quantify and predict fall risk and issue early warnings; At 6:10 AM, Mr. Wang got out of bed to go to the bathroom. The system calculated the risk entropy value in real time. 1. The basic risk entropy is 0.75, which is a static high-risk value obtained based on Morse score normalization; 2. When Wang stood up quickly, his postural systolic blood pressure dropped by 28 mmHg, his heart rate rose from 62 beats / minute to 88 beats / minute, his skin conductance increased significantly, and his physiological stress risk entropy was calculated to be 0.72 in real time. 3. When Wang stood up, his gait was unsteady, and the amplitude of his torso swaying increased to 18°, far exceeding the baseline threshold. His pitch angle deviation reached 12°, and his center of gravity fluctuated rapidly. The instantaneous behavioral risk entropy was calculated to be 0.78 in real time. 4. The positioning module identified that Wang entered the high-risk area of ​​the restroom, which was during the early morning when he was not fully active. The environmental risk entropy was calculated to be 0.6 in real time. 5. Real-time calculation of total risk entropy value = 0.75×0.2 + 0.78×0.4 + 0.72×0.25 + 0.6×0.15 = 0.732, corresponding to a high risk level; 6. The system detected that the total risk entropy value rose from 0.31 to 0.732 within 3 seconds, and the rate of change far exceeded the preset risk escalation threshold. It predicted that the risk would escalate to extremely high risk within 10 seconds and triggered an early warning and intervention. 7. Simultaneously set the manual cancellation window for high-risk situations to 20 seconds. If an extremely high-risk situation is subsequently triggered, the window will be automatically shortened to 15 seconds.

[0080] Step 4: When the warning is triggered, verify the target elderly person's autonomous response ability and conduct adaptive intervention based on the risk level; 1. When the system triggers a high-risk warning, the wristband auxiliary terminal immediately initiates an autonomous response verification and plays a voice message: "You are getting up too quickly, your blood pressure is fluctuating greatly, and you are walking unsteadily. Please hold onto the wall immediately, sit down slowly and rest. If you hear this, please press the button to respond." The high-risk verification window lasts for 5 seconds.

[0081] 2. Due to dizziness and blurred vision, Wang failed to complete any key presses, voice commands, or gestures within the 5-second verification window, and the system determined that there was no response.

[0082] 3. According to the adaptive intervention rules, if there is a high risk and no response, the risk level will automatically be upgraded to extremely high risk, and the highest level of emergency rescue linkage will be triggered immediately.

[0083] A closed-loop intervention and nursing care system covering pre-incident intervention, in-incident rescue, and post-incident care for fall incidents; 1. Closed-loop preventative intervention: Before the risk escalates, real-time behavioral intervention is implemented on the device: the system immediately plays voice guidance through auxiliary terminals in a loop: "Please do not move, immediately hold onto the sink next to you, slowly squat down or sit down, and do not get up quickly," to prevent the risk of falling from escalating.

[0084] Smart home linkage intervention: By linking home smart devices through the home edge gateway, it can automatically turn on all the lights in the bathroom and hallway, unlock the smart door lock of the entrance door, activate the smart handrail in the bathroom, and eliminate environmental factors that may cause falls.

[0085] Long-term health intervention: Following this incident, the system generated a personalized fall prevention plan based on long-term risk entropy data, adjusted Mr. Wang's blood pressure medication time to before bedtime, formulated a three-step morning standing method and a balance training plan three times a week, and sent these plans to Mr. Wang's daughter and community family doctor.

[0086] 2. Closed-loop emergency response during the incident: After the risk escalated, the self-rescue guidance system on the end played a loop of the voice message: "You are in a high-risk state for falling. Please hold onto a fixed object, sit down and rest slowly, and do not move around. Rescue personnel are on their way." This was to prevent Mr. Wang from making blind actions that could lead to a fall and secondary injury.

[0087] Emergency Response and Rescue: Immediately send alarm information to Wang's daughter, the community elderly care service center, and the XX community health service center, along with Wang's real-time accurate location, home address, access control password, core health records, and real-time blood pressure and heart rate data; automatically and repeatedly dial Wang's daughter's number, with hands-free two-way calling enabled throughout.

[0088] The community caregiver arrived at Wang's home within 5 minutes and helped Wang sit down to rest. After 10 minutes, her blood pressure was measured again and returned to 130 / 80 mmHg. It was confirmed that there had been no fall or physical injury, and the warning was lifted. Wang's daughter simultaneously checked on the elderly woman's condition remotely.

[0089] 3. Closed-loop post-treatment rehabilitation care: The system automatically generates a standardized risk event debriefing report, which clearly identifies the time, location, and triggering cause of the event, including the entire process of morning orthostatic hypotension, changes in physiological and behavioral data, and the rescue and disposal process, and is synchronized to the community health service center.

[0090] Community family doctors and rehabilitation therapists worked together to adjust Wang's antihypertensive medication regimen, develop a three-month balance and lower limb muscle strength training plan, and orthostatic hypotension prevention guidelines, while simultaneously conducting fall prevention health education.

[0091] Community caregivers visit Wang twice a week to monitor his gait, blood pressure, and balance. They update the personalized baseline database and risk level monthly and dynamically adjust the intervention plan. After this incident, Wang has not experienced any similar high-risk events, and his risk of falling has been significantly reduced.

[0092] The dynamic entropy quantization model is iteratively optimized based on the federated learning framework; 1. All data is transmitted with end-to-end AES256 encryption. Wang's original health data is stored on local wearable devices and home edge gateways. The cloud only stores encrypted feature data and does not store any original privacy data.

[0093] 2. Using a horizontal federated learning framework, the gradient parameters of the anonymized model for this event were uploaded to the cloud aggregation server and aggregated globally with the gradient parameters of other terminals. This optimized the weights and judgment thresholds of postural blood pressure changes and morning scenarios in the risk entropy model, improving the accuracy of risk prediction for similar scenarios. No user's original data was transmitted throughout the process.

[0094] 3. Wang can view his own risk data, turn off data collection, adjust emergency contacts and linkage permissions, and delete cloud storage data at any time through the accompanying WeChat mini-program.

[0095] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for early warning and intervention regarding the risk of falls in the elderly, characterized in that, Includes the following steps: S1. Construct a risk profile and risk baseline database for the target elderly population regarding falls and assess the corresponding initial fall risk level; S2. Collect and preprocess multi-source data from the target elderly through wearable devices; S3. Construct a dynamic entropy value quantification model for fall risk to quantify and predict fall risk and issue early warnings. The quantification of fall risk involves dividing fall risk into different risk entropies and calculating the total risk entropy value of the target elderly person in real time. Based on the total risk entropy value, fall risk is divided into different risk levels. The prediction of fall risk involves comparing the rate of change of the total risk entropy value with a preset risk escalation threshold. When the rate of change exceeds the preset risk escalation threshold, an early warning for the corresponding risk level is triggered in advance. S4. When an early warning is triggered, verify the target elderly person's autonomous response ability and conduct adaptive intervention based on the risk level.

2. The method for early warning and intervention regarding the risk of falls in the elderly according to claim 1, characterized in that, In step S3, the risk entropy includes basic risk entropy, instantaneous behavioral risk entropy, physiological stress risk entropy, and environmental risk entropy. The risk entropy is calculated in real time, and the entropy values ​​of the risk entropy are all normalized to the 0-1 range. The formula for calculating the total risk entropy value is: Total risk entropy value = Basic risk entropy × W1 + Instantaneous behavioral risk entropy × W2 + Physiological stress risk entropy × W3 + Environmental risk entropy × W4, where W1, W2, W3, and W4 are the dynamic weights corresponding to the four risk entropies. The dynamic weights are adjusted according to the initial fall risk level of the target elderly person and the real-time scenario, and W1 + W2 + W3 + W4 = 1. The weight W1 increases as the initial fall risk level of the target elderly increases; the weight W2 increases by 0.1-0.2 when the target elderly exhibit gait abnormalities or significant trunk swaying; the weight W3 increases by 0.1-0.2 when the target elderly exhibit postural blood pressure changes or abnormal heart rate fluctuations; and the weight W4 increases by 0.05-0.15 when the target elderly are in nighttime or high-risk areas. When the rate of change of the total risk entropy exceeds the preset risk escalation threshold, it can predict that the target elderly person's risk of falling will escalate within 15-60 seconds, and trigger early warning and intervention for the corresponding risk level after the escalation, thus achieving early prediction of fall events.

3. The method for early warning and intervention regarding the risk of falls in the elderly according to claim 2, characterized in that, The risk levels are divided into low risk, medium risk, high risk, and very high risk. Low risk corresponds to a total risk entropy value of less than 0.3, which is judged as having no risk of falling and the target elderly person is in normal condition. The medium risk level corresponds to a total risk entropy value greater than or equal to 0.3 and less than 0.6, indicating a potential risk of fall that requires attention and intervention. The high risk level corresponds to a total risk entropy value greater than or equal to 0.6 and less than 0.9, indicating an extremely high probability of fall that requires immediate intervention. The extremely high risk level corresponds to a total risk entropy value greater than or equal to 0.9, indicating that a fall has occurred and emergency rescue is required. The triggering of the extremely high risk requires the simultaneous fulfillment of three judgment rules: the peak value of the acceleration impact exceeds a preset threshold, the sudden drop in center of gravity height exceeds a preset threshold, and the ground posture lock lasts for more than a preset duration. At the same time, a manual cancellation window period of 15-30 seconds is set. The target elderly person can manually cancel the alarm by pressing the one-click emergency call button. The duration of the manual cancellation window period is set according to the risk level: 30 seconds for low and medium risk, 20 seconds for high risk, and 15 seconds for extremely high risk.

4. A method for early warning and intervention regarding fall risk in the elderly according to claim 3, characterized in that, In step S4, the verification process is that when the system triggers a risk warning of any level, the wearable device immediately initiates an autonomous response capability verification; the verification response method includes at least one of three types: voice interaction response, one-click SOS physical button response, and preset posture command feedback; the verification response time is set according to the current risk level: 15 seconds for low risk, 10 seconds for medium risk, 5 seconds for high risk, and 3 seconds for extremely high risk. The autonomous response capability is assessed in real time based on the response time and completion rate of the target elderly person within the verification window, and is divided into three levels: complete response, partial response, and no response. The classification rules for the autonomous response capability levels are as follows: complete response means completing a valid response within the first half of the verification window time; partial response means completing a valid response or only completing part of the instructions at the end of the verification window; and no response means not completing any valid response at the end of the verification window. The adaptive intervention dynamically adjusts the risk level and information push scope of the warning based on a combination of the risk level at the time of warning triggering and the level of autonomous response capability. Among them, the medium-risk combined with full response means that only the wearable device's local gentle vibration and voice reminder are triggered, without triggering external linkage; The medium-risk combination of partial response or no response: if the risk level of the warning is upgraded to high risk, the warning information will be pushed to emergency contacts and community caregivers; The high-risk combination with a complete response: triggers a strong alert from wearable devices and smart home integration, but does not trigger emergency rescue; The high-risk combination of partial response or no response: The risk level of the warning is upgraded to extremely high risk, triggering emergency rescue linkage; The extremely high risk combined with a full response: push alarm information to emergency contacts and wait for the elderly person to confirm whether to escalate the rescue; The extremely high risk combined with partial or no response: immediately trigger the highest level of emergency rescue, and synchronize all information to emergency contacts, elderly care service institutions and emergency stations.

5. A method for early warning and intervention regarding fall risk in the elderly according to claim 4, characterized in that, The intervention is further divided into pre-event intervention and in-event intervention. Pre-event intervention targets medium-risk and high-risk scenarios where a fall has not yet occurred, and is a preventative intervention. The intervention measures are implemented in three levels: Level 1 is real-time behavioral intervention using wearable devices, which plays targeted voice guidance to correct abnormal behaviors of the target elderly person and prevent the risk of fall from escalating; Level 2 is smart home-linked intervention, which uses a home edge network to link smart home devices and eliminate environmental factors that could trigger falls; Level 3 is long-term health intervention, which generates a fall prevention plan based on long-term total risk entropy value changes and pushes it to the target elderly person, emergency contacts, and community medical staff. The real-time behavioral intervention is voice guidance based on risk triggers, and the smart home-linked intervention includes turning on all lights in high-risk areas, unlocking smart door locks, and activating smart handrails. The aforementioned in-process intervention is for scenarios with extremely high risk where a fall has already occurred. The intervention measures include: firstly, playing self-rescue guidance voice prompts on a wearable device to prevent the elderly person from blindly getting up and causing secondary injury; simultaneously collecting the elderly person's real-time vital signs data and pushing it to emergency contacts and emergency medical services; automatically dialing preset emergency contact numbers in a loop, with hands-free two-way calling enabled throughout; and pushing the elderly person's precise location, home address, access control information, and core health record information to rescue personnel to shorten the rescue arrival time.

6. A method for early warning and intervention regarding fall risk in the elderly according to claim 1, characterized in that, In step S1, the risk database consists of all the basic information of the target elderly person, including weight, history of falls, history of chronic diseases, history of medications that affect human balance and blood pressure regulation, history of drug allergies, and emergency contact person. The risk baseline library consists of normal threshold baselines specific to the target elderly. These normal threshold baselines are established based on daily life data collected by wearable devices on the target elderly over 7-14 consecutive days. The normal threshold baselines include kinematic baselines, physiological characteristic baselines, daily activity rhythm baselines, and home environment risk maps. The initial fall risk level is determined using the Morse Fall Assessment Scale by combining the target elderly person's static basic information with the normal threshold baseline. The initial fall risk level is the initial weight of the risk factor in the dynamic entropy value quantification model, providing a benchmark for real-time risk assessment.

7. A method for early warning and intervention regarding fall risk in the elderly according to claim 6, characterized in that, The kinematic baseline includes normal gait parameters, resting or dynamic posture angle thresholds, trunk sway amplitude thresholds, and normal range of center of gravity height variation; the physiological characteristic baseline includes the normal resting or dynamic heart rate range, the normal heart rate variability range, the normal blood oxygen saturation threshold, the diurnal blood pressure rhythm range, and the maximum threshold for postural blood pressure variation; the daily activity rhythm baseline includes the time pattern of the target elderly person's daily activities and the activity range threshold; the home environment risk map marks high-risk areas and safe areas within the target elderly person's home environment; wherein, the normal gait parameters include cadence, stride length, gait cycle, and stance phase ratio; the resting or dynamic posture angles include pitch angle, roll angle, and yaw angle; the high-risk areas include the bathroom, stairwell, and kitchen, and the safe areas include the living room and bedroom.

8. A method for early warning and intervention regarding fall risk in the elderly according to claim 1, characterized in that, In step S2, the wearable device consists of a waist-worn main terminal and a wristband-style auxiliary terminal. The main terminal is worn at the waist and abdomen, integrating a 9-axis IMU inertial measurement unit, a barometer, a dual-mode positioning module, and a one-button emergency call physical button. The auxiliary terminal is worn on the wrist, integrating a PPG photoplethysmography sensor, a skin conductance sensor, a voice interaction module, and a sound pickup and playback module. The main terminal and the auxiliary terminal synchronize with millisecond-level timestamps to complete the synchronous acquisition of multi-sensor data. The timestamp synchronization accuracy between the main terminal and the auxiliary terminal is no less than 10ms. The 9-axis IMU inertial measurement unit includes a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer. The dual-mode positioning module includes GPS, a Beidou outdoor positioning module, and an indoor UWB positioning module. The kinematic data acquired by the 9-axis IMU inertial measurement unit is denoised using a Kalman filter algorithm, and the physiological signals acquired by the PPG sensor are smoothed and denoised using a moving average filter algorithm to remove invalid data caused by sensor jitter and poor contact. The wearable device can perform timestamp alignment and numerical normalization of multi-sensor data. The numerical normalization process uses the min-max normalization method to perform real-time difference calibration between the preprocessed data and the risk baseline library, identifying abnormal scenarios such as device detachment, abnormal wearing, and signal interruption, and triggering local wearing reminders. The corresponding abnormal data is not included in subsequent risk assessment calculations.

9. A method for early warning and intervention regarding fall risk in the elderly according to claim 2, 6, or 8, characterized in that, The basic risk entropy is calculated based on the target elderly person's risk profile. It is a static basic risk value and is updated every month. When the target elderly person's health status undergoes a significant change, the basic risk entropy is obtained by linearly normalizing the score results of the Morse Fall Assessment Scale to the 0-1 range. The instantaneous behavioral risk entropy is obtained by calculating the degree of deviation from the kinematic baseline and the rate of change of the time series based on real-time kinematic data collected by a 9-axis IMU. The update frequency is 100Hz. The calculation of instantaneous behavioral risk entropy includes four core features: gait parameter deviation, posture angle deviation, trunk sway amplitude, and center of gravity height change rate. The physiological stress risk entropy is calculated based on real-time physiological data collected by PPG and skin conductance sensors, and the degree of deviation and rate of change of the data from the physiological characteristic baseline is calculated. The update frequency is 20Hz. The calculation of physiological stress risk entropy includes five core features: heart rate deviation, heart rate variability change amplitude, postural blood pressure change amplitude, blood oxygen saturation deviation, and skin conductance response change amplitude. The environmental risk entropy is calculated based on the real-time location data of the dual-mode positioning module, the home environment risk map, and the degree of danger in the current scene, with an update frequency of 1Hz.

10. A method for early warning and intervention regarding the risk of falls in the elderly, as described in claim 1, characterized in that, The iterative optimization of the dynamic entropy quantification model adopts a federated learning framework to construct the model iteration system, and uses real fall event data and intervention effect feedback data to optimize the weights and judgment thresholds of the dynamic entropy quantification model of fall risk. The federated learning framework adopts a horizontal federated learning architecture, where each terminal completes model training locally and only uploads the encrypted model gradient parameters to the cloud for aggregation, without transmitting the original privacy data throughout the process. The entire process of federated learning uses end-to-end encrypted transmission of data, and cloud storage uses de-identified encryption processing. The original data is preferentially stored locally on the wearable device and at the home edge gateway. The encrypted feature data is uploaded to the cloud, and the original privacy data of the target elderly is not stored. This provides the target elderly with complete control over the data and supports turning off data collection, adjusting warning and linkage permissions, and deleting cloud-stored data at any time.

Images